Can a Repeated Opt-Out Reminder mitigate hypothetical bias in discrete choice experiments? An application to consumer valuation of novel food products

Can a Repeated Opt-Out Reminder mitigate hypothetical bias in discrete choice experiments? An... Abstract In this paper, we test whether a Repeated Opt-Out Reminder (ROOR) can mitigate hypothetical bias in stated discrete choice experiments (DCE). The data originate from a field experiment concerning consumer preferences for a novel food product made from cricket flour. Utilising a between-subject design with three treatments, we find significantly higher marginal willingness-to-pay values in hypothetical than in nonhypothetical settings, confirming the presence of hypothetical bias. Comparing this to a hypothetical setting where the ROOR is introduced, we find that the ROOR effectively mitigates hypothetical bias for one attribute and significantly reduces it for the rest of the attributes. 1. Introduction Since its advent in the early 1980s in marketing research, the stated discrete choice experiment (DCE) method has been used extensively in the environmental, agricultural, food, transport, health and marketing sectors to elicit the value of goods and services. However, critics have continually questioned the validity and the reliability of the results obtained from hypothetical experiments including stated DCE. If results from hypothetical experiments are to be useful for testing theories or to guide policy decisions, such approaches should yield reliable results. Nevertheless, this seems to not be generally the case (List and Gallet, 2001; Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Murphy et al., 2015a, 2015b), and one of the recurring issues tied to the stated DCE and stated preference methods in general is the issue of hypothetical bias. Due to the hypothetical nature of the choice questions asked, respondents may for various reasons overstate their Willingness-To-Pay (WTP) since their answers have no real consequences in terms of required payment for the good or service in question (e.g. Fox et al., 1996; Lusk and Schroeder, 2004; Harrison, 2006; Murphy et al., 2015a, 2015b; Vossler, Doyon and Rondeau, 2012; Vossler and Watson, 2013). There is a rich body of literature investigating hypothetical bias within the areas of environmental valuations employing the contingent valuation (CV) method to elicit the value of public goods (e.g. Carson et al., 1996; List and Shogren, 1998; Cummings and Taylor, 1999; Carlsson and Martinsson, 2001; List and Gallet, 2001; List, 2001; Little and Berrens, 2004; Murphy et al., 2015a, 2015b; Blumenschein et al., 2008). This issue is somewhat less explored in the DCE literature even though attention to hypothetical bias has increased in recent years. Given that people may behave differently depending on the type of the good and the experimental setup (Carson et al., 1996; Carson and Groves, 2007), the conclusions from external validity tests on CV studies based on public goods cannot be transferred directly to the DCE context nor to a private goods context. Hence, similar tests should be applied in DCE applications to thoroughly examine their validity. This is of particular importance as DCEs are extensively being used to assess demand for new products to guide policy development. In this regard, recent studies (e.g. Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Chang, Lusk and Norwood, 2009; Collins and Vossler, 2009; Loomis et al., 2009; Chowdhury et al., 2011; Vossler, Doyon and Rondeau, 2012; De-Magistris, Gracia and Nayga, 2013; Grebitus, Lusk and Nayga, 2013; Moser, Raffaelli and Notaro, 2014; Vossler and Watson, 2013) have examined the validity of DCEs empirically as well as theoretically. Most of these studies suggest that DCEs are indeed susceptible to hypothetical bias, even when they are used to value marketable private goods. Some investigations attempt to reduce hypothetical bias and increase explanatory power of empirical models by combining data from stated choice with reveal preference studies (e.g. Adamowicz, Louviere and Williams, 1994; Azevedo, Herriges and Kling, 2003; Brooks and Lusk, 2010). Others focus on increasing the realism of stated DCE studies by either employing hypothetical bias mitigation strategies (HBMS) (see Loomis, 2011, 2014 for a review) or by shifting to nonhypothetical choice experiments in order to reduce or completely eliminate hypothetical bias. Employing the latter, however, can be difficult since in addition to being costly and time consuming, the products of interest may not be available (Fox et al. 1996; De-Magistris, Gracia and Nayga, 2013). Therefore, HBMS may serve as important tools to remove hypothetical bias, and one such strategy, which has been often used in the literature, is the Cheap Talk (CT) script (Cummings and Taylor, 1999). However, the existing evidence with respect to the effectiveness of the CT script is mixed. While some have found that the CT script effectively mitigates hypothetical bias (Lusk, 2003; Carlsson, Frykblom and Lagerkvist, 2005; Murphy et al., 2015a, 2015b; List, Sinha and Taylor, 2006; Chowdhury et al., 2011; Tonsor and Shupp, 2011), others have found it to be effective only for certain sub-groups of respondents (List, 2001; Lusk, 2003; Aadland and Caplan, 2003, 2006; Barrage and Lee, 2010; Ami et al., 2011). Furthermore, the findings of Brown, Ajzen and Hrubes (2003) suggest that the CT is bid range sensitive, and Nayga, Woodward and Aiew (2006), Brummett, Nayga and Wu (2007) and Blumenschein et al. (2008) all found that the CT did not have any influence on the WTP values. A number of studies have tested the credibility of the CT script in DCE settings and their results generally show that the CT is ineffective in removing the hypothetical bias (List, Sinha and Taylor, 2006; Özdemir, Johnson and Hauber, 2009; Yue and Tong, 2009; Carlsson, García and Löfgren, 2010; Bosworth and Taylor, 2012; Moser, Raffaelli and Notaro, 2014). This might be explained by the fact that the CT script was originally developed for CV studies, and applying it to the DCEs may not be appropriate as there are contextual and structural differences between the two valuation approaches (Ladenburg and Olsen, 2014), thus, responses may differ markedly (List and Shogren, 1998; Carson and Groves, 2007). Given these inconsistencies surrounding the effectiveness of the CT script, researchers have proposed other HBMS. For instance, Jacquemet et al. (2013) implemented a Solemn Oath script, and De-Magistris, Gracia and Nayga (2013) tested an Honesty priming approach. Both approaches were found to be effective in removing hypothetical bias. Other researchers have instead relied on calibrating hypothetical values using respondents’ stated certainty in choice (Blumenschein et al., 2008). More recently, along the same lines, Ladenburg and Olsen (2014) suggested augmenting the CT script with an Opt-Out Reminder (OOR), which reminds respondents that if the prices of the experimentally designed alternatives are greater than what their household will pay, they should choose the opt-out alternative. In a split sample setup, they found that when the CT was applied with the OOR added, WTP estimates were significantly reduced compared to using the CT without the OOR. Varela et al. (2014) also tested the impact of presenting the CT with the OOR and contrary to Ladenburg and Olsen (2014), the OOR was not found to influence WTP. A possible explanation might be that Ladenburg and Olsen (2014) repeated the OOR before each single choice set whereas Varela et al. (2014) only presented it once in the scenario description. This seems to support Ladenburg and Olsen (2014) who speculate that, given the repeated choice nature of DCE, it may be of particular importance to repeat the reminder since respondents might otherwise forget about the reminder as they progress through the choice tasks. This article contributes to these developments in the literature in terms of extending the work by Ladenburg and Olsen (2014). The aim is to properly test the effectiveness of including a Repeated OOR (ROOR) in stated DCEs by directly testing its ability to mitigate hypothetical bias. Ladenburg and Olsen (2014) implemented the ROOR in conjunction with a CT script. In this study, we include only the ROOR to avoid confounding the effects of the ROOR and the CT. This allows for a more rigorous test of the pure effect of the ROOR. Another limitation of the experimental setup used in Ladenburg and Olsen (2014) as well as in Varela et al. (2014) is that they tested the OOR in a purely hypothetical setup. They did not include a nonhypothetical treatment which is necessary for establishing the extent to which hypothetical bias is present in the first place. Hence, they could not conclude whether the OOR actually reduced hypothetical bias. Therefore, in this article, we include an incentivised discrete choice experiment (REAL) treatment in addition to a fairly standard hypothetical DCE (HYPO) to investigate whether hypothetical bias exists at the outset. A third treatment presents the exact same hypothetical DCE except that the ROOR is added (ROOR), in order to examine the impact of the ROOR. Both Ladenburg and Olsen (2014) and Varela et al. (2014) presented the OOR script in combination with a CT script, thus confounding the effects of both scripts. To isolate the effect of the OOR in terms of mitigating hypothetical bias, we do not use a CT script but simply present respondents with the ROOR alone. With this article, we contribute to the literature in several aspects. First and foremost, to our knowledge, this is the first valid test of the effectiveness of the ROOR in terms of mitigating hypothetical bias. Second, the majority of studies investigating hypothetical bias have done so in a developed country context whereas only a couple of papers have considered it in a developing country context (Ehmke, Lusk and List, 2008; Chowdhury et al., 2011). One reason for this may be the presence of significant challenges in conducting valuation studies in developing countries (Durand-Morat, Wailes and Nayga, 2016). However, Ehmke, Lusk and List (2008) found hypothetical bias to be location dependent, implying that people from developing countries could have different susceptibility to hypothetical bias than those from developed countries. This can be related to the issue that the role of money may be viewed differently in these different regions (Gibson et al., 2016). In light of this, we conduct our study in a developing country, thus adding to the sparse DCE literature in such settings. This also represents a novel addition to the growing literature on nonhypothetical DCE (Carlsson and Martinsson, 2001; Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Alfnes et al., 2006; Johansson-Stenman and Svedsäter, 2008; Lusk, Fields and Prevatt, 2008; Chang, Lusk and Norwood, 2009; Loomis et al., 2009; Volinskiy et al., 2009; Yue and Tong, 2009; Grebitus, Lusk and Nayga, 2013; Michaud, Llerena and Joly, 2013; Gracia, 2014). Finally, while the aim of the article is mainly methodological, it also makes a novel contribution to the literature on consumers’ food preferences by estimating Kenyan consumers’ WTP for novel food products (NFPs) made with edible insects. 2. Background 2.1. Edible insects as a sustainable source of food The world population is projected to increase to 9 billion in 2050 (UN, 2015). Faced with climate change and environmental sustainability concerns, the current food production systems may not be able to meet the growing demand for food which is estimated to rise by 60 per cent in 2050 (FAO, 2009, 2013). In this respect, it is evident that the number of malnourished people in developing countries is (still) alarmingly high, although some nutritional improvements have been observed in recent years (Braun, 2007). Specifically, despite a globally positive change in the supply of food energy, total protein and animal protein in the past few decades, a decreasing trend has been observed in many developing regions (Pellet and Ghosh, 2004), especially in several sub-Saharan African (SSA) countries where food security has been and still is one of the biggest societal challenges. Increasing livestock production is one option to tackle the shortage of animal protein but it contributes significantly to the emission of greenhouse gasses (Steinfeld et al., 2006) while also requiring substantial inputs of water and crop-based feed. Both these required production inputs are expected to decrease in availability in SSA with future global warming. Increasing food security thus requires that either the productivity of the current agricultural sector increases e.g. through new technological developments or alternative sources of food that increase food security in a sustainable manner must be identified through research and development (Boland et al., 2013; van Huis, 2013). In line with this, the concept of producing edible insects for food has received substantially increasing attention recently (FAO, 2013). Insects can be a good source of important nutrients for human consumption (Bukkens, 1997; Christensen et al., 2006). As shown by Oonincx et al. (2010) and Halloran et al. (2017), edible insects can be considered environmentally friendly as they emit significantly lower amounts of greenhouse gasses than livestock production. Despite insects being traditionally consumed as food in large parts of Africa, Asia and Latin America, and historically all over the world, the many different aspects of eating insects, which has earlier been labelled with the collective term ‘entomophagy’ (Evans et al., 2015), have received rather limited attention in research (van Huis, 2013). Recently, however, the literature investigating e.g. food safety, nutritional value, consumer attitude and legislative aspects of producing insects for food has begun to increase very rapidly (e.g. Finke, 2013; van Huis, 2013; Looy, Dunkel and Wood, 2014; Ruby, Rozin and Chan, 2015; Alemu et al., 2017b). Most of the recent research focuses on such supply side issues. However, the demand side, i.e. consumers’ preferences for insects as food, will ultimately be equally important for a successful introduction of edible insects as a new and substantial component of the future food production sector. Therefore, gathering information about consumers’ preferences is necessary since decisions concerning establishment of new insect production sectors as well as regulatory, standardisation and quality control schemes are likely to be suboptimal if knowledge about potential demand in the market is not accounted for. 2.2. The use of DCEs in developing countries WTP for food products can be elicited using different approaches including DCEs and experimental auctions (Gracia, Loureiro and Nayga, 2011). DCEs are grounded in random utility theory and Lancaster’s characteristics demand theory (Lancaster, 1966; McFadden, 1986) most commonly used, and they rely on asking respondents to make choices in a manner that mimics their choices in actual purchasing situations. This makes them more familiar to consumers than experimental auctions (Alfnes et al., 2006). The application of DCEs to elicit values of goods and services is currently gaining popularity in developing countries (Bennett and Birol, 2010; Birol et al., 2015; Gibson et al., 2016). Notwithstanding this, most DCE studies, particularly consumer valuation of food attributes based on nonhypothetical DCEs have been conducted in developed countries (Alphonce and Alfnes, 2017). This is related to the challenges in conducting economic valuation studies in developing countries. Durand-Morat, Wailes and Nayga (2016) and Whittington (1998, 2002) highlight that the ability of respondents to follow and process information, the competences of local personnel, unavailability of proper population sampling frames, limited coverage of infrastructures, and poor design and implementation of survey scenarios are the main challenges. These issues need to be considered when conducting DCE valuation studies in developing countries. In this study, we exerted a great effort to obtain high-quality data by directly addressing these challenges in the survey design. The problem of hypothetical bias is a concern in any DCE, though it has not yet been thoroughly explored in developing country settings. To the authors’ knowledge, only two studies addressing this issue are available in the DCE literature so far. Aiming to assess the value of bottled mineral water, Ehmke, Lusk and List (2008) found that hypothetical bias differs across locations and it is larger in developed than in developing countries. The authors attributed this disparity to cultural differences. Chowdhury et al. (2011) found that Ugandan consumers’ WTP for biofortified foods are significantly higher in hypothetical treatments than in real treatments indicating the presence of hypothetical bias. In the current study, we aim to contribute to the sparse DCE literature in developing countries especially concerning hypothetical bias using data from consumer valuation of novel food products in Kenya. 2.3. Strategies to mitigate hypothetical bias in DCEs If hypothetical bias is present in DCEs – as is commonly found – the results will potentially be misleading for decision makers interested in utilising value estimates for e.g. policy evaluations or product development. The important question is then how to mitigate it. A primary solution would be to use nonhypothetical DCEs, e.g. using real products and real payments. These approaches are costly and they are often difficult to administer in the field, especially in experiments involving novel foods. In recognition of this, a range of different HBMS have been proposed in the DCE literature. These are broadly divided into two categories: ex-ante and ex-post HBMS (Whitehead and Cherry, 2007; Loomis, 2011, 2014). The ex-ante strategies are used to mitigate hypothetical bias at the design stage. According to Loomis (2011, 2014), the most common strategies are (i) developing a survey design that stimulates respondents to believe that their responses have real consequences in real situations (consequentiality) (e.g. Carson and Groves, 2007), (ii) using solemn oath and honesty priming approaches through which respondents are asked to swear to respond honestly to the study questions (e.g. De-Magistris, Gracia and Nayga, 2013; Jacquemet et al., 2013), (iii) using inferred valuation by asking respondents to state what they think other people would be willing to pay for the good in question (e.g. Lusk and Norwood, 2009), (iv) using the theory of cognitive dissonance to enable respondents to act in a trustworthy and rational manner (e.g. Alfnes, Yue and Jensen, 2010) and, finally (v) cheap talk (e.g. List, Sinha and Taylor, 2006). Cheap talk, which is probably the most commonly used HBMS, simply explains to respondents what hypothetical bias is. It was first applied by Cummings and Taylor (1999) and other researchers followed suit (e.g. Lusk, 2003; Carlsson, Frykblom and Lagerkvist, 2005; Chowdhury et al., 2011; Tonsor and Shupp, 2011). The ex-post strategies are used to mitigate hypothetical bias after the completion of the survey. One can find a detailed account of such approaches in Loomis (2014). The most common approach is to make use of follow-up questions on response uncertainty to adjust for hypothetical bias. Blumenschein et al. (2008) and Champ et al. (2009) showed that recoding uncertain responses mitigated the hypothetical bias in their respective studies. In a DCE set up, Ready, Champ and Lawton (2010) similarly found that calibrating choices using the uncertainty level of respondents resulted in hypothetical bias being mitigated. In this study, we focus on testing a recently proposed new ex-ante HMBS, namely the ROOR which until now has not been properly assessed for effectiveness. 3. Experimental design and procedure We test the effectiveness of the ROOR in a DCE field experiment in Kenya, where there is increasing interest in introducing edible insects into the food production sector. Specifically, we focus on consumers’ preferences for crickets as food since this type of insect is considered particularly promising in terms of developing viable production systems for household consumption as well as for large scale commercialisation in Kenya. 3.1. The product contexts and DCE design Rather than presenting respondents with whole crickets on a plate, traditional buns were chosen as the product context for the DCE since it was considered a more realistic market entry strategy to introduce crickets by adding cricket flour (CF) to buns which are generally considered affordable and available in most food outlets throughout Kenya. Buns are one of the bread types that form the major staple food in Kenya according to focus group interviews and key informant discussions. Furthermore, the practical manageability of product development and handling guided the choice of buns for this experiment. CF was mixed into wheat flour to bake the buns. The amounts of CF used to produce the buns were determined based on the recipe specification developed by Kinyuru, Kenji and Njoroge (2009). Serving as attribute levels in the DCE, the amounts of CF were varied at 0 per cent, 5 per cent and 10 per cent of the total flour. The product development process, the recipe formulation and the characteristics of the final bun products used for the field experiment are presented in Appendix A. Nonhypothetical DCEs involving new products require development of all the product combinations with all the relevant attributes. This is, however, typically constrained by budget and other logistic complications. For this study, we identified three attributes that would enable us to actually develop and produce all possible product combinations for the DCE. The first attribute represents the insect component in terms of the amount of CF in the buns (CF). The second attribute describes whether a portion of the wheat flour is fortified or not (Fortified), while the third attribute is the price of a bag of buns (Price). Insects like crickets may be consumed in their whole form or ground into flour. Besides representing the more realistic market entry strategy, we focus on the flour form because low protein foods such as maize and sorghum, which are popular in east Africa (Stevens and Winter-Nelson, 2008), can be easily enhanced with CF to increase their nutritional value. In addition, CF is already being produced on a rather large scale elsewhere in the world, hence, a similar production would seem achievable in Kenya. The Fortified attribute was included to assess Kenyan consumers’ preferences and WTP for fortified foods. The level for the fortified wheat flour was arbitrarily set to be 40 per cent of the total flour used to bake the different buns. Consumers in Kenya are familiar with food fortification because the Kenyan government through the Kenya National Food Fortification Alliance (KNFFA) has introduced a food fortification programme to reduce micronutrient deficiency. This programme covers foods including maize and wheat (KNFFA, 2011). A secondary purpose with including the fortified flour attribute was an attempt to (at least to some extent) separate out the pure effect of nutritional improvements. In other words, we wanted to give participants the option to choose buns that were nutritionally better than the standard alternative without necessarily forcing them to choose cricket flour. This way, if participants strongly rejected the CF, they might still exhibit preferences for better nutrition, thus avoiding confounding preferences for nutrition, which we expected to be positive, with preferences for other aspects of eating insects which might be negative, e.g. if respondents would find the idea of eating insects disgusting (Looy, Dunkel and Wood, 2014). The attributes and their levels are presented in Table 1. The amount of CF has three levels: 0 g (Standard with no CF), 6.25 g (Medium CF) and 12.5 g (High CF). This is equivalent to 0 per cent, 5 per cent and 10 per cent of the total wheat flour used per bag of buns. The Fortified attribute has two levels at 0 and 50 g. The total amount of flour is kept constant at 125 g per bag. Thus, when CF and/or fortified wheat flour is added, the amount of wheat flour is reduced accordingly. Besides the quality attributes, a Price attribute is included with six levels. An initial investigation of actual market prices of buns showed that a single bun would typically cost around 10 Kenyan Shillings1 (KShs), rarely less than KShs 5. Since focus group discussions indicated that respondents would not respond negatively to the other attributes, it was thus decided to set the minimum price for a bag with three buns at KShs 20. Focus group discussions further suggested that a suitable choke price for the price range would be KShs 90, and the six price levels were subsequently set at KShs 20, 25, 30, 40, 60 and 90. Table 1. Product attributes and levels Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Table 1. Product attributes and levels Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Once the attributes and their levels had been determined, the DCE design was produced. It was subjected to standard efficient design procedures in line with state-of-the-art experimental design theory (Louviere, Hensher and Swait, 2000; Ferrini and Scarpa, 2007; Scarpa and Rose, 2008; ChoiceMetrics, 2012). Three alternatives are included in the design: two of them are deigned ones and the last one is a specific alternative which is constant across choice sets. The three attributes, i.e. the amount of CF, Fortified and Price, have three, two and six levels, respectively. Based on this, a full factorial design results in 36 alternatives. To avoid exceeding the cognitive capacity of respondents, a fractional factorial design was employed. The design procedure involved two stages. First, we produced a design using zero priors and conducted two rounds of pilot studies with 42 respondents. Second, the data from the pilot studies were estimated to obtain priors in terms of attribute parameter estimates. The priors were subsequently used to inform and update the D-efficient fractional factorial design which was generated using the Ngene software (ChoiceMetrics, 2012). The design produced 12 choice sets which were presented to participants. Figure 1 shows an example of one of the choice sets used. Fig. 1. View largeDownload slide An example of a choice set used in the survey. Fig. 1. View largeDownload slide An example of a choice set used in the survey. As is evident from Figure 1, we use a forced choice type of DCE by including an alternative with a standard bag of buns which is constant across all choice sets. This is motivated by the fact that we are dealing with a staple convenience consumer product, namely bread (Cova and Pace, 2006). Even though the context of shopping for bread is a familiar everyday experience for most consumers, the main attribute of interest in our survey, namely the addition of cricket flour, is adding complexity and difficulty to the choice tasks. In this case, including a no-buy option may prompt respondents to use it as a simplifying strategy without actually making the required trade-offs to reveal their true responses (see von Haefen, Massey and Adamowicz, 2005; Beshears et al., 2008; Burton and Rigby, 2008; Boxall, Adamowicz and Moon, 2009; Rigby, Alcon and Burton, 2010). Hence, we chose the forced choice setting to reduce the risk of respondents resorting to such heuristics. While it is strictly speaking always possible for consumers in the real world to not buy a specific product, we argue that the majority of consumers in our case would in fact not consider no-buy an option when shopping for bread in the real world. Focus group interviews supported this, and during the personal interviews, no respondents complained about being forced to buy bread. Furthermore, while it may be more common in food DCE studies to include a no-buy option, it is not uncommon to use forced choice settings (e.g. Carlsson, Frykblom and Lagerkvist, 2007a, 2007b; Kallas et al., 2013). 3.2. The experimental treatments and hypothesis testing Three experimental treatments were designed to investigate whether hypothetical bias persists and whether it can be mitigated by a ROOR. In the first treatment (HYPO), participants make purely hypothetical choices without having to pay for the bun products. In the second treatment (REAL), participants’ choices are no longer hypothetical as they are subjected to an incentivisation mechanism similar to that of Alfnes et al. (2006) and Gracia, Loureiro and Nayga (2011). The inclusion of this treatment enables testing the extent to which hypothetical bias is present. The lack of such a treatment was the key limitation of Ladenburg and Olsen (2014) and Varela et al. (2014) who could only speculate about the impact of the OOR on hypothetical bias. In the third treatment (ROOR), participants make hypothetical choices but now framed with a ROOR in which subjects are provided with the following opt-out reminder: ‘If the prices of Bag 1 and Bag 2 are higher than what you think your household will pay, you should choose Bag 3.’ This treatment is specifically designed to investigate if hypothetical bias can be mitigated by the ROOR. As the name suggests the ROOR is placed repeatedly before each choice set so as to take the repeated nature of the choices into account and help subjects to stay on the ‘true’ preference path (Ladenburg and Olsen, 2014). In this manner, information flow and consistent trade-offs across choice sets would be ensured. In addition, the ROOR enables participants to attend to both the designed and the opt-out alternatives appropriately, in which case participants should make the required trade-offs among the attributes to reach a decision that ideally mirrors their ‘true’ preferences. Moreover, as discussed in Ladenburg and Olsen (2014), unlike the CT script, the ROOR is not affected by conformity issues as it does not convey any implicit cues as to what others do and how they behave in relation to value revelations. The experiment is undertaken based on a between-subject design to control for behavioural shift and confounding effects which might result if a within-subject design was adopted (Lusk and Schroeder, 2004; Charness, Gneezy and Kuhn, 2012). In total, 334 participants took part in the survey. Participants were randomised into the three treatments, resulting in 109 participants in each of the HYPO and ROOR treatments, and 116 participants in the REAL treatment. The participants are either household heads or spouses as they are the decision makers in dietary selection for their household members. With this experimental setup, we are able to test the following four research hypotheses. First, based on our expectation that the ROOR addresses the tendency of forgetting the opt-out alternatives in DCEs, we hypothesise that the hypothetical setting will result in fewer opt-out choices than the incentivised setting, but when adding the ROOR the amount of opt-out choices will be similar to the incentivised setting: H1: Opt-out frequencyHYPO < Opt-out frequencyREAL = Opt-out frequencyROOR Second, we hypothesise that the hypothetical setting will be subject to hypothetical bias in terms of mean marginal WTP estimates being higher than in the incentivised setting: H2: WTPHYPO > WTPREAL Conditional on confirmation of H2, we hypothesise that by adding the ROOR we can reduce hypothetical bias in terms of achieving mean marginal WTP estimates that are lower than those of the standard hypothetical setting: H3: WTPHYPO > WTPROOR Finally, also conditional on H2 being confirmed, but representing a stronger test than H3, we hypothesise that adding the ROOR will not only reduce hypothetical bias, but even completely eliminate it. In other words, the ROOR treatment will generate mean marginal WTP estimates equivalent to the incentivised setting: H4: WTPROOR = WTPREAL 3.3. Experimental procedure Survey interviews were conducted face-to-face using trained interviewers. The personal interviews took place in three different parts of Kenya that were selected to ensure inclusion of respondents from both rural and urban areas, of various different ethnicities, and where some still maintained the tradition of eating insects while others had left it. Respondents were recruited using a stratified random sampling approach to further ensure variation across key socio-demographic variables. For further descriptions of the sampling procedure, see Alemu et al. (2017a). The actual interview procedure followed four steps. First, participants were welcomed and seated and one trained interviewer was assigned to each for conducting the personal interview. Second, they were given KShs 30 in cash for their participation in the experiment. Participants were then randomly assigned to one of the three experimental treatments described above. The participants in the REAL treatment received an additional remuneration of KShs 90 which they could use fully or partly to buy the chosen products in the DCE. The provision of the monetary remuneration before the start of the DCE process was intended to induce a sense of ownership2 and avoid the house money effect that may be associated with windfall earnings in experiments (Thaler and Johnson, 1990; Cárdenas et al., 2014). Third, the samples of the three bun products were provided to participants so that they could view, smell and taste them.3 This allows consumers to experience the new products and their attributes, thereby facilitating preference formation. This is a unique feature of our study as in most stated DCE studies proxies for taste have been used instead of giving consumers direct experience of the product in question (Chowdhury et al., 2011). Fourth, participants were then presented with the DCE questionnaire. In addition to the choice sets, it contained information4 about crickets as food as well as a description of the attributes. Additionally, participants in the REAL treatment received instructions, which are adapted from Johansson-Stenman and Svedsäter (2008), Chang, Lusk and Norwood (2009) and Mørkbak, Olsen and Campbell (2014), regarding the incentivisation mechanism (see Appendix B). In Lusk and Schroeder (2004), Chang, Lusk and Norwood (2009) and Mørkbak, Olsen and Campbell (2014) a single random number was drawn for a whole group of participants to identify the binding choice set for all participants in that group. As opposed to this, but in line with Alfnes et al. (2006) and Gracia, Loureiro and Nayga (2011), the identification of the binding choice set in this study was implemented by asking each individual participant to draw a numbered ball from 1 to 12 from a hat after they had answered the 12 choice tasks. Participants were then given the bag of buns they had chosen in the binding choice set and they had to pay the price of that bag. 4. Econometric specification The random utility theory (RUT) (McFadden, 1986) serves as a theoretical framework to specify utility expressions in DCEs. In what follows, we follow Train and Weeks (2005). According to RUT, the utility of individual n choosing alternative k among J given alternatives in choice situation t can be expressed as Unkt=−αnmnkt+βn′xnkt+ϵnkt. (1) The utility is expressed as a component of price, m, and non-price attributes, x. The parameters αn and vector βn′ are the coefficients of the price and non-price attributes. For logit models, the variance of the error term ϵn can be written as sn2 (π26 ), where sn is the scale parameter which differs across decision makers. Without loss of generality, equation (1) can be divided by sn as Unkt=(αnsn)mnkt+(βnsn)′xnkt+εnkt. (2) Letting λn= (αnsn ) and cn= (βnsn ), equation (2) becomes Unkt=−λnmnkt+cn′xnkt+εnkt (3) where the error term, εn, is distributed IID extreme value. Equation (3) describes utility in preference space. As suggested by Train and Weeks (2005), an alternative option is to employ a WTP space approach. The utility in preference space in equation (3) may be reformulated so that the estimated coefficients refer directly to the WTP values. This approach has been applied by Train and Weeks (2005), Sonnier, Ainslie and Otter (2007), Scarpa, Thiene and Train (2008), Balcombe, Chalak and Fraser (2009), Thiene and Scarpa (2009), Hensher and Greene (2011) and Hole and Kolstad (2012). Most of the evidences so far suggest that models based on the WTP space approach yield more plausible behavioural explanations. Specifically, the distributions of WTP values are found to be within reasonable margins than values based on the preference space formulation. Therefore, there is an increasing support for implementation of the WTP space specification when the goal is to obtain appropriate WTP estimates which would be used for policy guidance. As mentioned above, the WTP space is expressed by reformulating the preference space utility expression. Reparametrising equation (3) gives Unkt=−λnmnkt+(−λnwn)′xnkt+εnkt. (4) Equation (4) is called utility in WTP space.Train and Weeks (2005) mentioned that equations (3) and (4) are behaviourally equivalent. Nevertheless, one can see, and as mentioned before, WTP calculation based on the preference space specification is directly impacted by the distribution of λn and cn, and, in some cases, like normal distributions, the moments of the WTP distribution may be undefined and furthermore the WTP distribution may suffer from a fat tail problem causing identification problems. Heterogeneity in preferences can be accounted for by specifying mixed logit models, which can be derived in a number of different ways (Train, 2003). The most common specification is a random parameter logit (RPL) model5 in which attribute parameters are treated as random parameters. The random parameters may in principle follow any relevant distribution, but often a normal distribution is used for quality attribute parameters and a lognormal is used for the negative of the price attribute parameter (Train, 2003; Train and Sonnier, 2005). We employ the RPL model as focus group interviews had indicated preference heterogeneity across participants. In line with Scarpa, Thiene and Train (2008), we assume that λn=−exp(υn), where υn is the latent random factor of the price coefficient. In addition, let β represent all the random parameters estimated in the WTP space model, then equation (4) can be rewritten as Unkt=Vnkt(βn,xnkt)+εnkt (5) where Vnkis the indirect utility function, which is a function of the attributes of the alternatives, i.e. Vnk=βn′Xnkt. The (negative of the) coefficient on price ( λn) is assumed to follow a lognormal distribution in order to ensure a strictly positive impact of income on utility. The focus group discussions and pilot tests indicated that some people would have negative preferences for the non-price attributes of the products while others would have positive. As a result, the distributions of the coefficients for these attributes are assumed to follow a normal distribution. We allow the scale, which is accommodated through λn, to vary across decision makers by specifying all random parameters to be correlated with each other, i.e. no restrictions are placed on the variance–covariance matrix. As the estimated parameters, β′, vary over individuals according to the specified distributions with density f(β) (Train, 2003), the probability that an individual n chooses alternative k among J given alternatives in choice situation t can be derived as Pnk=∫(eβn′Xnkt∑jJeβ′xntj)f(β)dβ. (6) In line with, e.g. Thiene and Scarpa (2009) and Hole and Kolstad (2012), equation (6), which is based on the WTP space specification, is estimated using the MSL procedure by maximising the full simulated log-likelihood over the sequence of choices and over the sample. We estimate the models using the BIOGEME 2.2 software (Bierlaire, 2003), maximising the simulated likelihood function with the CFSQP algorithm (Lawrence, Zhou and Tits, 1997) to avoid the local optima problem (Scarpa, Thiene and Train, 2008). For the simulation procedure, 3,000 Modified Latin Hypercube Sampling draws (Hess, Train and Polak, 2006) were used since this amount of draws was found to produce stable parameter estimates. In line with, e.g. Lusk and Schroeder (2004), Tonsor and Shupp (2011), De-Magistris, Gracia, Nayga (2013) and Moser, Raffaelli and Notaro (2014), we use the complete combinatorial method (Poe, Giraud and Loomis, 2005) to test the different hypotheses outlined above. For this purpose, 1,000 WTP estimates are bootstrapped for each attribute in each sample based on the Krinsky and Robb (1986) method using 1,000 draws from multivariate normal distributions of the coefficient estimates and the variance–covariance matrix of the random parameters. Differences between the WTP estimates can also be tested by pooling data for the two treatments in a specific hypothesis test as in De-Magistris, Gracia and Nayga (2013) and Bazzani et al. (2017). One can then specify an extended utility function by interacting dummy variables representing treatment effects with attribute levels. Accordingly, equation (4) can be extended as Unkt=−λn[mnkt+wn′xnkt+ωn(xnkt∗dtreatment)]+εnkt (7) Two dummy variables are created since there are three treatments. They are denoted by dtreatment taking a value of 1 for the second treatment in a specific hypothesis test, 0 otherwise. Differences in WTP estimates between the two treatments will be captured directly by the coefficient ωn, and its sign will indicate the direction of the treatment effect. 5. Results The summary of sample socio-demographics presented in Table 2 shows no differences across treatments. Except the level of education between the ROOR and REAL treatments, chi-square tests failed to reject that the socio-demographic distributions differ across treatments, suggesting that participants in the three treatments are sufficiently similar in their characteristics to rule this potential factor out as cause of any preference differences found. Table 2. Summary statistics of participants’ characteristics by treatment (percentages) Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 aNo education, and drop outs from primary and secondary schools. Table 2. Summary statistics of participants’ characteristics by treatment (percentages) Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 aNo education, and drop outs from primary and secondary schools. As a first step in investigating possible preference differences across sample splits, Table 3 provides a comparison of the choice distributions. Interestingly, while the Standard buns were chosen in 14 per cent of the choices in the HYPO treatment, they were chosen in 27 per cent and 31 per cent of the choices in the ROOR and REAL treatments, respectively. These results are in line with Lusk and Schroeder (2004) and Bazzani et al. (2017) who also found participants to choose the opt-out option more frequently in a real setting than in a hypothetical setting, though they used a ‘none of these’ opt-out. The chi-square tests confirm that the choice frequencies differ significantly for HYPO versus ROOR and HYPO versus REAL. However, there seems to be no significant difference when comparing ROOR and REAL. In other words, we can confirm hypothesis H1. These results provide a first indication that the ROOR may indeed serve its purpose as it seems to generate a better correspondence between hypothetical and nonhypothetical choices. Table 3. Choice frequencies Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Note: The chi-square tests have been carried out based on the actual number of choices rather than based on percentages. Table 3. Choice frequencies Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Note: The chi-square tests have been carried out based on the actual number of choices rather than based on percentages. The estimation results are presented in Table 4. As noted before, the estimated parameters directly represent the implied WTP distributions and the coefficients of the attributes are interpreted as marginal WTP estimates. Preference equality across the three treatments is tested using a likelihood ratio test (Swait and Louviere, 1993; Louviere, Hensher and Swait, 2000; Lusk and Schroeder, 2004; Chowdhury et al., 2011; Moser, Raffaelli and Notaro, 2014). The null hypothesis is rejected (x2 = 152.8, p-value < 0.01) suggesting that the overall preference structure is not identical across the three treatments. Furthermore, all the estimated standard deviations of the random parameters6 are significant, indicating substantial preference heterogeneity across participants in all three treatments. Table 4. Model results and comparison of WTP estimates Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors; p-values obtained using the complete combinatorial method (Poe, Giraud and Loomis, 2005) based on 1,000 bootstrapped WTP estimates derived using Krinsky and Robb (1986) method; p-values represent results of the one sided test that the differences between the equivalent WTP estimates of two treatments are positive. Table 4. Model results and comparison of WTP estimates Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors; p-values obtained using the complete combinatorial method (Poe, Giraud and Loomis, 2005) based on 1,000 bootstrapped WTP estimates derived using Krinsky and Robb (1986) method; p-values represent results of the one sided test that the differences between the equivalent WTP estimates of two treatments are positive. The model results presented in Table 4 show that consumers are willing to pay a premium for buns containing CF. More interestingly, the WTP estimates differ across the treatments. Generally, the WTP values obtained from the HYPO sample are three to six times as large as the WTP values obtained from the REAL sample, suggesting that the HYPO WTPs suffer noticeably from hypothetical bias. Specifically, the WTP estimate for High CF in the HYPO sample is almost six times that of the REAL sample. In addition, the WTP estimate for the Medium CF is more than three times as large in the HYPO as in the REAL sample. Turning to the WTP for fortified relative to non-fortified wheat flour, the estimate obtained in the HYPO sample is more than four times that obtained from the REAL sample. In accordance with the significantly fewer choices of Bag 3, i.e. the opt-out, in the HYPO sample (see Table 3), the ASC estimate for Bag 3 is much more negative in the HYPO compared to the REAL, confirming that participants in the HYPO react much more negatively to the opt-out alternative, irrespective of the attribute levels of the offered alternatives. According to the test results in Table 4, we can reject the null hypothesis of equal WTP estimates across the HYPO and REAL treatments for all attributes. Given that the attribute WTPs obtained in the HYPO sample are all higher than those of the REAL treatment, we thus confirm hypothesis H2, i.e. hypothetical bias is present in the HYPO sample. Furthermore, the negative opt-out effect observed for the ASC estimate in the HYPO treatment, i.e. participants disproportionally often choosing the non-opt-out alternatives regardless of their attributes, is not present in the REAL treatment. This serves as another indication of hypothetical bias, corresponding to the findings in Table 3. Turning to the question of particular interest in this article namely the effect of the ROOR, the complete combinatorial test results show that for all attributes the ROOR treatment results in significantly lower WTP estimates than the HYPO treatment. Hence, we can also confirm hypothesis H3. For the High CF attribute, the WTP obtained in the ROOR treatment is though still significantly higher than in the REAL treatment, while for the Medium CF, the difference is not significant, and for the Fortified attribute, the difference is only borderline significant. The ASC estimates are not significantly different, but the fact that the ASC estimate is still significantly lower than zero in the ROOR indicates that the negative status quo effect is not completely removed. Hence, the slightly stronger hypothesis H4 can only be partly confirmed since hypothetical bias, despite being significantly reduced, is not completely removed for all attributes. Notwithstanding this, presenting the ROOR before each choice set appears to be highly effective, removing 87–98 per cent of the hypothetical bias in absolute terms across the range of attributes as well as for the negative opt-out effect. All the above results regarding hypothetical bias and the impact of the ROOR in reducing it are graphically described in Figure 2. Probability density functions of the mean marginal WTPs are plotted for each treatment using 10,000 draws from multivariate normal distributions of the WTPs. The densities representing the WTP distributions in REAL and ROOR treatments are tighter than those in the HYPO treatment, confirming the results presented above. Fig. 2. View largeDownload slide Distribution of mean marginal WTP values by treatments. Fig. 2. View largeDownload slide Distribution of mean marginal WTP values by treatments. An excerpt of the results from running an extended utility function model on the pooled data is presented in Table 5 (see Appendix D for the full results). The estimates pertaining to the extended part capturing the treatment effects directly, i.e. the vector w in equation (7), generally confirm the findings of Table 4. The signs of the estimated coefficients show negative effects of the second treatments on WTP estimates in a specific hypothesis test. This is in line with our alternative hypotheses. Table 5. Robustness test using an extended utility function Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Asterisk (*) indicates significance at 5 per cent level or lower. The table only shows mean estimates of s for the ω parameters. Table 5. Robustness test using an extended utility function Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Asterisk (*) indicates significance at 5 per cent level or lower. The table only shows mean estimates of s for the ω parameters. 6. Conclusions Eliciting the value of goods based on approaches that reflect the real behaviour of decision makers is a central element of valuation studies be it for non-market or market applications. Stated DCE approaches are widely used to elicit the value of goods and their attributes, e.g. in order to predict demand for new products. How to reduce or eliminate hypothetical bias remains an important challenge for DCE researchers. Nonhypothetical DCEs are obviously the preferred option; however, in many cases, product unavailability and the associated high costs of implementation impair researchers from relying on such approaches. As a result, hypothetical DCEs are still used extensively. A range of different strategies, e.g. consequential survey questions, honesty priming, inferred valuation, cognitive dissonance and solemn oath scripts, have been suggested as amendments to hypothetical DCEs, aiming to reduce or eliminate hypothetical bias (Carson and Groves, 2007; Lusk and Norwood, 2009; Jacquemet et al., 2013; Alfnes, Yue and Jensen, 2010; De-Magistris, Gracia and Nayga, 2013). Inspired by previous studies by Ladenburg and Olsen (2014) and Varela et al. (2014), we extend this line of research by investigating the effectiveness of a ROOR in terms of reducing or eliminating hypothetical bias. This study is based on a field experiment concerned with consumer valuation of novel food products made with cricket flour in Kenya. Three treatments – one hypothetical, one hypothetical with ROOR and one nonhypothetical incentivised DCE – are tested in a between-subject design aimed at determining, firstly, the magnitude of hypothetical bias and, secondly, the ability of the ROOR to remove it. The results confirm that the mean marginal WTP values based on the simple hypothetical DCE suffer severely from hypothetical bias with estimates being up to several hundred percentages higher than the WTP estimates obtained in the nonhypothetical treatment. These results are generally in line with Ehmke, Lusk and List (2008) and Chowdhury et al. (2011) which are the only studies where hypothetical bias has been investigated in developing countries. The results further indicate that the ROOR considerably reduces or even eliminates the hypothetical bias while also being capable of reducing or removing negative opt-out effects. The results based on the ROOR are considerably closer to the results based on the nonhypothetical sample than the results based on the hypothetical sample without the ROOR. The ROOR eliminates hypothetical bias for one attribute while strongly reducing, though not completely eliminating it for others, highlighting the promising potential of introducing it in DCEs. The results concerning the presence of hypothetical bias presented in this paper are in contrast with Carlsson and Martinsson (2001), Cameron et al. (2002), Lusk and Schroeder (2004) and Alfnes and Steine (2005) where the hypothesis of equal marginal WTP values in real and hypothetical settings could not be rejected. However, other authors reported results which are in agreement with our results (see Johansson-Stenman and Svedsäter, 2008; Chowdhury et al., 2011; Moser, Raffaelli and Notaro, 2014). It should be noted that direct comparison of our results with these studies may not be appropriate due to reasons associated with the nature of the goods being valued, the sampling design and the nature of the experimental settings (Harrison and List, 2004; Harrison, 2006; Johansson-Stenman and Svedsäter, 2008). Both Carlsson and Martinsson (2001) and Alfnes and Steine (2005) used a within-subject design and they reported that there are no differences in marginal WTP values between hypothetical and real treatments. In this study, we used a between-subject design and find significant differences in the results in keeping with, e.g. Johansson-Stenman and Svedsäter (2008) and Moser, Raffaelli and Notaro (2014). Johansson-Stenman and Svedsäter (2008) found that hypothetical bias is larger in a between-subject design than in a within-subject design, and they concluded that it should be tested using the former type of design. However, List and Gallet (2001) did not find any significant differences in hypothetical bias when comparing between-subject and within-subject designs in a meta-analysis. Considering the nature of the experimental setting, as opposed to our field study, the similar food choice experiments by Lusk and Schroede (2004) and Alfnes and Steine (2005) were conducted in laboratory settings. Since they did not find substantial hypothetical bias, it is tempting to suggest that ex-ante hypothetical mitigation measures are more important in field settings than in lab settings. However, List and Gallet (2001) did not find any significant differences between lab and field settings in this regard. They did however find hypothetical bias to generally be higher for public goods than private goods, which was further confirmed by e.g. Alfnes and Steine (2005) and Johansson-Stenman and Svedsäter (2012). Importantly, this does not mean that hypothetical bias concerns can be disregarded when evaluating private goods. Our results strongly underline this, as we find substantial hypothetical bias in a private good context. This suggests that DCE analysts should consider employing some kind of ex-ante (or ex-post) hypothetical bias mitigation measures regardless of the type of good. Providing economic incentives as we do in our REAL treatment is of course the preferable way to avoid hypothetical bias in a private good context as it clearly enhances valuation scenario consequentiality – in line with the recommendations in Johnston et al. (2017). However, in most DCE cases, e.g. when assessing private goods incorporating new product attributes that are not yet available or when assessing public goods, such economic incentivisation is not a possibility, and the analyst must consider other options. We show that the ROOR is worth considering, as it seems to encourage truthful responses, or at least responses that are quite well aligned with those obtained in the ideal economically incentivised setting. We can of course only conjecture that this result is transferable to public good settings where economic incentivisation is very rarely possible. Our results differ from Ladenburg and Olsen (2014) and Varela et al. (2014) in four important ways. First, their studies considered preferences for public goods whereas the current study considers a private good context in terms of food choice. Second, these studies did not establish the extent to which hypothetical bias existed, which may be seen as a major limitation since it essentially precluded the authors from making any conclusions concerning the effectiveness of the ROOR in terms of mitigating hypothetical bias. Our experimental setup allows us to conclude that hypothetical bias is indeed present and of a non-negligible magnitude. Third, while the ROOR did not influence responses in the case of Varela et al. (2014), it significantly reduced the marginal WTP estimates of only some of the attributes in Ladenburg and Olsen (2014). Contrary to this, the results of the current study reveal that the ROOR significantly reduces the hypothetical bias for all attributes and even completely eliminates it for one attribute. Fourth, they implemented the ROOR together with a CT script, which confounded the effects of the ROOR and the CT. Thus, compared to these, our study provides a more strict experimental setup allowing a much more rigorous test of the ROOR in isolation. It also provides a clearer result in terms of the impact of the ROOR on hypothetical bias. The results reported in this study have particular relevance to stated DCE since framing them with the ROOR does not pose further challenges in terms of monetary cost, product availability and or logistic burden that otherwise might be the case in nonhypothetical DCEs. In sum, the main results of this study show that not only financial incentives but also ROOR can lead to plausible behavioural manifestations. Therefore, in situations where it is not possible to employ nonhypothetical DCEs, framing hypothetical DCEs with a ROOR would seem to be a promising strategy. Our results, however, provide only first-hand evidence and are not conclusive by themselves. Therefore, future studies are important in order to replicate our results and provide further empirical evidence in relation to the ability of the ROOR to improve the external validity of stated DCEs. In this regard, several areas of research can be identified for further research. First, future studies may consider conducting research in different experimental settings and with different types of goods so as to validate whether the effect of ROOR is context-specific or not. This includes investigating whether the ROOR leads to similar results to the ones reported in this study in situations where participants do not experience the attributes. Second, the actual wording of the ROOR is not tested here. Future studies may test other wordings of the ROOR, e.g. by focusing on the designed alternatives rather than on the opt-out alternative. Third, further research could be testing whether the ROOR leads to the same results as the results from the current study when it is applied to experimental designs that include a no-choice alternative rather than a forced choice. Fourth, unlike De-Magistris, Gracia and Nayga (2013), we only considered the ROOR without including other ex-ante hypothetical bias mitigation strategies to avoid a potential confounding effect. It can be argued that examining the effectiveness of the ROOR as compared to other strategies such as CT is necessary to draw a more general conclusion. This may be accomplished using an experimental design that includes a wide range of treatments with different combinations of the various ex-ante strategies for mitigating hypothetical bias. Acknowledgements We thank the management of the Department of Food Science and Technology at the Jomo Kenyatta University of Agriculture and Technology for giving us the opportunity to bake the bun products at the Food Processing Workshop Unit. Finally, we thank seminar participants at the university of Copenhagen and University of Stirling for comments and suggestions. The usual disclaimer applies for any remaining errors. Funding This research is part of the GREEiNSECT (13-06KU) project which was funded by the Danish International Development Agency (DANIDA), Ministry of Foreign Affairs of Denmark. Footnotes 1 1 US Dollar ~ 90.50 KShs at the time of the experiment. 2 As pointed out by a reviewer, consumers’ WTP for the novel insect-based food products could be different in real world situations where they would have to pay directly from their own pockets and not by some endowment offered by an interviewer. If the endowment is perceived as windfall money, the REAL treatment may not be as incentive compatible as desired, and the WTP estimates obtained in this sample may consequently be an overstatement of the ‘true’ WTP. We have not tested this in the present study as it was clear from the early stages of survey planning, that asking respondents to bring money to the interview would be problematic from an ethical point of view since the survey is conducted in relatively poor areas of Kenya. 3 We have not controlled for participants having or not having eaten bread at home before the experiment. However, we tried to control for the hunger level of the participants and the taste of previous consumed foods at breakfast or lunch time by placing the interviews between 10.00 and 12.00 in the morning and 14.00 and 16.00 in the afternoon. 4 This information was provided before consumers were subject to the DCE to mirror real market situations in that consumers often process a range of information before they make purchase decisions. However, we have not tested the effects of information provision on consumers’ responses as it falls beyond the scope of this paper. 5 This model can be extended to accommodate an error component representing potential correlation between designed alternatives as proposed by Scarpa, Ferrini and Willis (2005), Scarpa, Spalatro and Canavari (2007a) and Scarpa, Willis and Acutt (2007b). However, given that the status quo option in the current experiment is defined as a standard alternative which in many aspects is similar to the experimentally designed alternatives, it was decided to use the fairly standard RPL model. 6 The estimates of the Cholesky matrices, which were obtained from the finite difference hessian, are reported in Appendix C. References Aadland , D. and Caplan , A. D. ( 2003 ). Willingness to pay for curbside recycling with detection and mitigation of hypothetical bias . American Journal of Agricultural Economics 85 : 492 – 502 . Google Scholar CrossRef Search ADS Aadland , D. and Caplan , A. D. ( 2006 ). Cheap talk reconsidered: new evidence from CVM . Journal of Economic Behavior and Organization 60 : 562 – 578 . Google Scholar CrossRef Search ADS Adamowicz , W. , Louviere , J. and Williams , M. ( 1994 ). Combining revealed and stated preference methods for valuing environmental amenities . Journal of Environmental Economics and Management 26 : 271 – 292 . Google Scholar CrossRef Search ADS Alemu , M. H. , Olsen , S. B. , Vedel , S. E. , Kinyuru , J. N. and Pambo , K. O. ( 2017 a). Can insects increase food security in developing countries? An analysis of Kenyan consumer preferences and demand for cricket flour buns . Food Security 9 : 471 – 484 . Google Scholar CrossRef Search ADS Alemu , M. H. , Olsen , S. B. , Vedel , S. E. , Pambo , K. O. and Owino , V. O. ( 2017 b). Combining product attributes with recommendation and shopping location attributes to assess consumer preferences for insect-based food products . Food Quality and Preference 55 : 45 – 57 . Google Scholar CrossRef Search ADS Alfnes , F. , Guttormsen , A. G. , Steine , G. and Kolstad , K. ( 2006 ). Consumers’ willingness to pay for the color of salmon: a choice experiment with real economic incentives . American Journal of Agricultural Economics 88 : 1050 – 1061 . Google Scholar CrossRef Search ADS Alfnes , F. and Steine , G. ( 2005 ). None-of-these bias in stated choice experiments. Discussion Paper DP-06/05. Department of Economics and Resource Management, Norwegian University of Life Sciences, Norway. Alfnes , F. , Yue , C. and Jensen , H. H. ( 2010 ). Cognitive dissonance as a means of reducing hypothetical bias . European Review of Agricultural Economics 37 : 147 – 163 . Google Scholar CrossRef Search ADS Alphonce , R. and Alfnes , F. ( 2017 ). Eliciting consumer WTP for food characteristics in a developing context: application of four valuation methods in an African market . Journal of Agricultural Economic 68 : 123 – 142 . Google Scholar CrossRef Search ADS Ami , D. , Aprahamian , F. , Chanel , O. and Luchini , S. ( 2011 ). A test of cheap talk in different hypothetical contexts: the case of air pollution . Environmental and Resource Economics 50 : 111 – 130 . Google Scholar CrossRef Search ADS Azevedo , C. D. , Herriges , J. A. and Kling , C. L. ( 2003 ). Combining revealed and stated preferences: consistency tests and their interpretations . American Journal of Agricultural Economics 85 : 525 – 537 . Google Scholar CrossRef Search ADS Balcombe , K. , Chalak , A. and Fraser , I. ( 2009 ). Model selection for the mixed logit with Bayesian estimation . Journal of Environmental Economics and Management 57 : 226 – 237 . Google Scholar CrossRef Search ADS Barrage , L. and Lee , M. S. ( 2010 ). A penny for your thoughts: inducing truth-telling in stated preference elicitation . Economics Letters 106 : 140 – 142 . Google Scholar CrossRef Search ADS Bazzani , C. , Caputo , V. , Nayga , R. M. and Canavari , M. ( 2017 ). Testing commitment cost theory in choice experiments . Economic Inquiry 55 : 383 – 396 . Google Scholar CrossRef Search ADS Bennett , J. and Birol , E. ( 2010 ). Choice Experiments in Developing Countries: Implementation, Challenges and Policy Implications . Cheltenham, UK; Northampton, USA : Edward Elgar . Google Scholar CrossRef Search ADS Beshears , J. , Choi , J. J. , Laibson , D. and Madrian , B. C. ( 2008 ). How are preferences revealed? Journal of Public Economics 92 : 1787 – 1794 . Google Scholar CrossRef Search ADS Bierlaire , M. ( 2003 ). BIOGEME: A Free Package for the Estimation of Discrete Choice Models. Paper presented at the 3rd Swiss Transport Research Conference Monte Verita, Ascona, March 2003. Birol , E. , Meenakshi , J. V. , Oparinde , A. , Perez , S. and Tomlins , K. ( 2015 ). Developing country consumers’ acceptance of biofortified foods: a synthesis . Food Security 7 : 555 – 568 . Google Scholar CrossRef Search ADS Blumenschein , K. , Blomquist , G. C. , Johannesson , M. , Horn , N. and Freeman , P. ( 2008 ). Eliciting willingness to pay without bias: evidence from a field experiment . The Economic Journal 118 : 114 – 137 . Google Scholar CrossRef Search ADS Boland , M. J. , Rae , A. N. , Vereijken , J. M. , Meuwissen , M. P. M. , Fischer , A. R. H. , van Boekel , M. A. J. S. , Rutherfurd , S. M. , Gruppen , H. , Moughan , P. J. and Hendriks , W. H. ( 2013 ). The future supply of animal-derived protein for human consumption . Trends in Food Science and Technology 29 : 62 – 73 . Google Scholar CrossRef Search ADS Bosworth , R. and Taylor , L. O. ( 2012 ). Hypothetical bias in choice experiments: is cheap talk effective at eliminating bias on the intensive and extensive margins of choice? The B.E. Journal of Economic Analysis and Policy 12 : 1 – 28 . Google Scholar CrossRef Search ADS Boxall , P. , Adamowicz , W. L. and Moon , A. ( 2009 ). Complexity in choice experiments: choice of the status quo alternative and implications for welfare measurement* . Australian Journal of Agricultural and Resource Economics 53 : 503 – 519 . Google Scholar CrossRef Search ADS Braun , J. V. ( 2007 ). Focus on the world’s poorest and hungry people: IFPRI 2006–2007 Annual Report Essay: International Food Policy Research Institute (IFPRI). Washington, DC: IFPRI. Brooks , K. and Lusk , J. L. ( 2010 ). Stated and revealed preferences for organic and cloned milk: combining choice experiment and scanner data . American Journal of Agricultural Economics 92 ( 4 ): 1229 – 1241 . Google Scholar CrossRef Search ADS Brown , T. C. , Ajzen , I. and Hrubes , D. ( 2003 ). Further tests of entreaties to avoid hypothetical bias in referendum contingent valuation . Journal of Environmental Economics and Management 46 : 353 – 361 . Google Scholar CrossRef Search ADS Brummett , R. G. , Nayga , R. and Wu , X. ( 2007 ). On the use of cheap talk in new product valuation . Economics bulletin 2 : 1 – 9 . Bukkens , S. G. F. ( 1997 ). The nutritional value of edible insects . Ecology of Food and Nutrition 36 : 287 – 319 . Google Scholar CrossRef Search ADS Burton , M. and Rigby , D. ( 2008 ). Hurdle and latent class approaches to serial non-participation in choice models . Environmental and Resource Economics 42 : 211 – 226 . Google Scholar CrossRef Search ADS Cameron , T. A. , Poe , G. L. , Ethier , R. G. and Schulze , W. D. ( 2002 ). Alternative non-market value-elicitation methods: are the underlying preferences the same? Journal of Environmental Economics and Management 44 : 391 – 425 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2005 ). Using cheap talk as a test of validity in choice experiments . Economics Letters 89 : 147 – 152 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2007 a). Consumer benefits of labels and bans on GM foods—choice experiments with Swedish consumers . American Journal of Agricultural Economics 89 : 152 – 161 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2007 b). Consumer willingness to pay for farm animal welfare: mobile abattoirs versus transportation to slaughter . European Review of Agricultural Economics 34 : 321 – 344 . Google Scholar CrossRef Search ADS Carlsson , F. , García , J. and Löfgren , Å. ( 2010 ). Conformity and the demand for environmental goods . Environmental and Resource Economics 47 : 407 – 421 . Google Scholar CrossRef Search ADS Carlsson , F. and Martinsson , P. ( 2001 ). Do hypothetical and actual marginal willingness to pay differ in choice experiments? Application to the valuation of the environment . Journal of Environmental Economics and Management 41 : 179 – 192 . Google Scholar CrossRef Search ADS Carson , R. T. , Flores , N. E. , Martin , K. M. and Wright , J. L. ( 1996 ). Contingent valuation and revealed preference methodologies: comparing the estimates for quasi-public goods . Land Economics 72 : 80 – 99 . Google Scholar CrossRef Search ADS Carson , R. and Groves , T. ( 2007 ). Incentive and informational properties of preference questions . Environmental and Resource Economics 37 : 181 – 210 . Google Scholar CrossRef Search ADS Champ , P. A. , Moore , R. and Bishop , R. C. ( 2009 ). A comparison of approaches to mitigate hypothetical bias . Agricultural and Resource Economics Review 38 : 166 – 180 . Google Scholar CrossRef Search ADS Chang , J. B. , Lusk , J. L. and Norwood , F. B. ( 2009 ). How closely do hypothetical surveys and laboratory experiments predict field behavior? American Journal of Agricultural Economics 91 : 518 – 534 . Google Scholar CrossRef Search ADS Charness , G. , Gneezy , U. and Kuhn , M. A. ( 2012 ). Experimental methods: between-subject and within-subject design . Journal of Economic Behavior and Organization 81 : 1 – 8 . Google Scholar CrossRef Search ADS ChoiceMetrics ( 2012 ). Ngene 1.1.1 User Manual and Reference Guide, Australia. Chowdhury , S. , Meenakshi , J. V. , Tomlins , K. I. and Owori , C. ( 2011 ). Are consumers in developing countries willing to pay more for micronutrient-dense biofortified foods? Evidence from a field experiment in Uganda . American Journal of Agricultural Economics 93 : 83 – 97 . Google Scholar CrossRef Search ADS Christensen , D. L. , Orech , F. O. , Mungai , M. N. , Larsen , T. , Friis , H. and Aagaard-Hansen , J. ( 2006 ). Entomophagy among the Luo of Kenya: a potential mineral source? International Journal of Food Sciences and Nutrition 57 : 198 – 203 . Google Scholar CrossRef Search ADS Collins , J. P. and Vossler , C. A. ( 2009 ). Incentive compatibility tests of choice experiment value elicitation questions . Journal of Environmental Economics and Management 58 ( 2 ): 226 – 235 . Google Scholar CrossRef Search ADS Cova , B. and Pace , S. ( 2006 ). Brand community of convenience products: new forms of customer empowerment – the case ‘my Nutella The Community’ . European Journal of Marketing 40 : 1087 – 1105 . Google Scholar CrossRef Search ADS Cummings , R. G. and Taylor , L. O. ( 1999 ). Unbiased value estimates for environmental goods: a cheap talk design for the contingent valuation method . American Economic Review 89 : 649 – 665 . Google Scholar CrossRef Search ADS Cárdenas , J. C. , De Roux , N. , Jaramillo , C. R. and Martinez , L. R. ( 2014 ). Is it my money or not? An experiment on risk aversion and the house-money effect . Experimental Economics 17 : 47 – 60 . Google Scholar CrossRef Search ADS De-Magistris , T. , Gracia , A. and Nayga , R. M. ( 2013 ). On the use of honesty priming tasks to mitigate hypothetical bias in choice experiments . American Journal of Agricultural Economics 95 : 1136 – 1154 . Google Scholar CrossRef Search ADS Ding , M. , Grewal , R. and Liechty , J. ( 2005 ). Incentive-aligned conjoint analysis . Journal of Marketing Research 42 : 67 – 82 . Google Scholar CrossRef Search ADS Durand-Morat , A. , Wailes , E. J. and Nayga , J. R. M. ( 2016 ). Challenges of conducting contingent valuation studies in developing countries . American Journal of Agricultural Economics 98 : 597 – 609 . Google Scholar CrossRef Search ADS Ehmke , M. D. , Lusk , J. L. and List , J. A. ( 2008 ). Is hypothetical bias a universal phenomenon? A multinational investigation . Land Economics 84 : 489 – 500 . Google Scholar CrossRef Search ADS Evans , J. , Alemu , M. H. , Flore , R. , Frøst , M. B. , Halloran , A. , Jensen , A. B. , Maciel-Vergara , G. , Meyer-Rochow , V. B. , Münke-Svendsen , C. , Olsen , S. B. , Payne , C. , Roos , N. , Rozin , P. , Tan , H. S. G. , van Huis , A. , Vantomme , P. and Eilenberg , J. ( 2015 ). ‘Entomophagy’: an evolving terminology in need of review . Journal of Insects as Food and Feed 1 : 293 – 305 . Google Scholar CrossRef Search ADS FAO ( 2009 ). The State of Food and Agriculture: Livestock in the Balance . Rome : Sales and Marketing group, Communication Division . FAO ( 2013 ). Edible Insects: Future Prospects for Food and Feed Security . Rome . Ferrini , S. and Scarpa , R. ( 2007 ). Designs with a priori information for nonmarket valuation with choice experiments: a Monte Carlo study . Journal of Environmental Economics and Management 53 : 342 – 363 . Google Scholar CrossRef Search ADS Finke , M. D. ( 2013 ). Complete nutrient content of four species of feeder insects . Zoo Biology 32 : 27 – 36 . Google Scholar CrossRef Search ADS Fox , J. A. , Hayes , D. J. , Shogren , J. F. and Kliebenstein , J. B. ( 1996 ). Experimental methods in consumer preference studies . Journal of Food Distribution Research 27 : 1 – 7 . Gibson , J. M. , Rigby , D. , Polya , D. A. and Russell , N. ( 2016 ). Discrete choice experiments in developing countries: willingness to pay versus willingness to work . Environmental and Resource Economics 65 : 697 – 721 . Google Scholar CrossRef Search ADS Gracia , A. ( 2014 ). Consumers’ preferences for a local food product: a real choice experiment . Empirical Economics 47 : 111 – 128 . Google Scholar CrossRef Search ADS Gracia , A. , Loureiro , M. L. and Nayga Jr , R. M. ( 2011 ). Are valuations from nonhypothetical choice experiments different from those of experimental auctions? American Journal of Agricultural Economics 93 : 1358 – 1373 . Google Scholar CrossRef Search ADS Grebitus , C. , Lusk , J. L. and Nayga Jr , R. M. ( 2013 ). Explaining differences in real and hypothetical experimental auctions and choice experiments with personality . Journal of Economic Psychology 36 : 11 – 26 . Google Scholar CrossRef Search ADS Halloran , A. , Hanboonsong , Y. , Roos , N. and Bruun , S. ( 2017 ). Life cycle assessment of cricket farming in north-eastern Thailand . Journal of Cleaner Production 156 : 83 – 94 . Google Scholar CrossRef Search ADS Harrison , G. ( 2006 ). Experimental evidence on alternative environmental valuation methods . Environmental and Resource Economics 34 : 125 – 162 . Google Scholar CrossRef Search ADS Harrison , G. W. and List , J. A. ( 2004 ). Field experiments . Journal of Economic Literature 42 : 1009 – 1055 . Google Scholar CrossRef Search ADS Hensher , D. A. and Greene , W. H. ( 2011 ). Valuation of travel time savings in WTP and preference space in the presence of taste and scale heterogeneity . Journal of Transport Economics and Policy 45 : 505 – 525 . Hess , S. , Train , K. E. and Polak , J. W. ( 2006 ). On the use of a Modified Latin Hypercube Sampling (MLHS) method in the estimation of a Mixed Logit Model for vehicle choice . Transportation Research Part B: Methodological 40 : 147 – 163 . Google Scholar CrossRef Search ADS Hole , A. and Kolstad , J. ( 2012 ). Mixed logit estimation of willingness to pay distributions: a comparison of models in preference and WTP space using data from a health-related choice experiment . Empirical Economics 42 : 445 – 469 . Google Scholar CrossRef Search ADS Jacquemet , N. , Joule , R.-V. , Luchini , S. and Shogren , J. F. ( 2013 ). Preference elicitation under oath . Journal of Environmental Economics and Management 65 : 110 – 132 . Google Scholar CrossRef Search ADS Johansson-Stenman , O. and Svedsäter , H. ( 2008 ). Measuring hypothetical bias in choice experiments: the importance of cognitive consistency . The B.E. Journal of Economic Analysis and Policy 8 , (Article 41). Johansson-Stenman , O. and Svedsäter , H. ( 2012 ). Self-image and valuation of moral goods: stated versus actual willingness to pay . Journal of Economic Behavior & Organization 84 : 879 – 891 . Google Scholar CrossRef Search ADS Johnston , R. J. , Boyle , K. J. , Adamowicz , W. , Bennett , J. , Brouwer , R. , Cameron , T. A. , Hanemann , W. M. , Hanley , N. , Ryan , M. , Scarpa , R. , Tourangeau , R. and Vossler , C. A. ( 2017 ). Contemporary guidance for stated preference studies . Journal of the Association of Environmental and Resource Economists 4 : 319 – 405 . Google Scholar CrossRef Search ADS Kallas , Z. , Escobar , C. and Gil , J. M. ( 2013 ). Analysis of consumers’ preferences for a special-occasion red wine: a dual response choice experiment approach . Food Quality and Preference 30 : 156 – 168 . Google Scholar CrossRef Search ADS Kinyuru , J. , Kenji , G. and Njoroge , M. ( 2009 ). Process development, nutrition and sensory qualities of wheat buns enriched with edible termites (Macrotermes subhylanus) from Lake Victoria region, Kenya . African Journal of Food, Agriculture, Nutrition and Development 9 : 1739 – 1750 . Google Scholar CrossRef Search ADS Krinsky , I. and Robb , A. L. ( 1986 ). On approximating the statistical properties of elasticities . The Review of Economics and Statistics 68 : 715 – 719 . Google Scholar CrossRef Search ADS Ladenburg , J. and Olsen , S. B. ( 2014 ). Augmenting short Cheap Talk scripts with a repeated Opt-Out Reminder in Choice Experiment surveys . Resource and Energy Economics 37 : 39 – 63 . Google Scholar CrossRef Search ADS Lancaster , K. J. ( 1966 ). A new approach to consumer theory . Journal of Political Economy 74 : 132 – 157 . Google Scholar CrossRef Search ADS Lawrence , C. , Zhou , J. L. and Tits , A. L. ( 1997 ). User’s Guide for CFSQP Version 2.5: A C Code for Solving (Large Scale) Constrained Nonlinear (Minimax) Optimization Problems, Generating Iterates Satisfying All Inequality Constraints. Technical Report TR-94-16r1, Institute for Systems Research, University of Maryland, College Park, MD 20742, 1997. List , J. A. ( 2001 ). Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards . American Economic Review 91 : 1498 – 1507 . Google Scholar CrossRef Search ADS List , J. and Gallet , C. ( 2001 ). What experimental protocol influence disparities between actual and hypothetical stated values? Environmental and Resource Economics 20 : 241 – 254 . Google Scholar CrossRef Search ADS List , J. A. and Shogren , J. F. ( 1998 ). Calibration of the difference between actual and hypothetical valuations in a field experiment . Journal of Economic Behavior & Organization 37 : 193 – 205 . Google Scholar CrossRef Search ADS List , J. A. , Sinha , P. and Taylor , M. H. ( 2006 ). Using choice experiments to value non-market goods and services: evidence from field experiments . Advances in Economic Analysis and Policy 6 : 1 – 37 (Article 2). Little , J. and Berrens , R. ( 2004 ). Explaining disparities between actual and hypothetical stated values: further investigation using meta-analysis . Economics Bulletin 3 : 1 – 13 . Loomis , J. ( 2011 ). What’s to know about hypothetical bias in stated preference valuation studies? Journal of Economic Surveys 25 : 363 – 370 . Google Scholar CrossRef Search ADS Loomis , J. B. ( 2014 ). Strategies for overcoming hypothetical bias in stated preference surveys . Journal of Agricultural and Resource Economics 39 : 34 – 46 . Loomis , J. , Bell , P. , Cooney , H. and Asmus , C. ( 2009 ). A comparison of actual and hypothetical willingness to pay of parents and non-parents for protecting infant health: the case of nitrates in drinking water . Journal of Agricultural and Applied Economics 41 : 697 – 712 . Google Scholar CrossRef Search ADS Looy , H. , Dunkel , F. and Wood , J. ( 2014 ). How then shall we eat? Insect-eating attitudes and sustainable foodways . Agriculture and Human Values 31 : 131 – 141 . Google Scholar CrossRef Search ADS Louviere , J. J. , Hensher , D. A. and Swait , J. D. ( 2000 ). Stated Choice Methods: Analysis and Applications . Cambridge, UK : Cambridge University Press . Google Scholar CrossRef Search ADS Lusk , J. L. ( 2003 ). Effects of cheap talk on consumer willingness-to-pay for golden rice . American Journal of Agricultural Economics 85 : 840 – 856 . Google Scholar CrossRef Search ADS Lusk , J. L. , Fields , D. and Prevatt , W. ( 2008 ). An incentive compatible conjoint ranking mechanism . American Journal of Agricultural Economics 90 : 487 – 498 . Google Scholar CrossRef Search ADS Lusk , J. L. and Norwood , F. B. ( 2009 ). An inferred valuation method . Land Economics 85 : 500 – 514 . Google Scholar CrossRef Search ADS Lusk , J. L. and Schroeder , T. C. ( 2004 ). Are choice experiments incentive compatible? A test with quality differentiated beef steaks . American Journal of Agricultural Economics 86 : 467 – 482 . Google Scholar CrossRef Search ADS McFadden , D. ( 1986 ). The choice theory approach to market research . Marketing Science 5 : 275 – 297 . Google Scholar CrossRef Search ADS Michaud , C. , Llerena , D. and Joly , I. ( 2013 ). Willingness to pay for environmental attributes of non-food agricultural products: a real choice experiment . European Review of Agricultural Economics 40 : 313 – 329 . Google Scholar CrossRef Search ADS Moser , R. , Raffaelli , R. and Notaro , S. ( 2014 ). Testing hypothetical bias with a real choice experiment using respondents’ own money . European Review of Agricultural Economics 41 : 25 – 46 . Google Scholar CrossRef Search ADS Murphy , J. , Allen , P. G. , Stevens , T. and Weatherhead , D. ( 2005 a). A meta-analysis of hypothetical bias in stated preference valuation . Environmental and Resource Economics 30 : 313 – 325 . Google Scholar CrossRef Search ADS Murphy , J. , Stevens , T. and Weatherhead , D. ( 2005 b). Is cheap talk effective at eliminating hypothetical bias in a provision point mechanism? Environmental and Resource Economics 30 : 327 – 343 . Google Scholar CrossRef Search ADS Mørkbak , M. R. , Olsen , S. B. and Campbell , D. ( 2014 ). Behavioral implications of providing real incentives in stated choice experiments . Journal of Economic Psychology 45 : 102 – 116 . Google Scholar CrossRef Search ADS Nayga , R. M. , Woodward , R. and Aiew , W. ( 2006 ). Willingness to pay for reduced risk of foodborne illness: a nonhypothetical field experiment . Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie 54 : 461 – 475 . Google Scholar CrossRef Search ADS Oonincx , D. G. A. B. , Itterbeeck , J. v. , Heetkamp , M. J. W. , van den Brand , H. , van Loon , J. J. A. and van Huis , A. ( 2010 ). An exploration on greenhouse gas and ammonia production by insect species suitable for animal or human consumption . PLoS One 5 : 1 – 7 . Google Scholar CrossRef Search ADS Özdemir , S. , Johnson , F. R. and Hauber , A. B. ( 2009 ). Hypothetical bias, cheap talk, and stated willingness to pay for health care . Journal of Health Economics 28 : 894 – 901 . Google Scholar CrossRef Search ADS Pellett , P. L. and Ghosh , S. ( 2004 ). Lysine fortification: past, present, and future . Food Nutr Bull 25 : 107 – 113 . Google Scholar CrossRef Search ADS Poe , G. L. , Giraud , K. L. and Loomis , J. B. ( 2005 ). Computational methods for measuring the difference of empirical distributions . American Journal of Agricultural Economics 87 : 353 – 365 . Google Scholar CrossRef Search ADS Ready , R. C. , Champ , P. A. and Lawton , J. L. ( 2010 ). Using respondent uncertainty to mitigate hypothetical bias in a stated choice experiment . Land Economics 86 : 363 – 381 . Google Scholar CrossRef Search ADS Rigby , D. , Alcon , F. and Burton , M. ( 2010 ). Supply uncertainty and the economic value of irrigation water . European Review of Agricultural Economics 37 : 97 – 117 . Google Scholar CrossRef Search ADS Ruby , M. B. , Rozin , P. and Chan , C. ( 2015 ). Determinants of willingness to eat insects in the USA and India . Journal of Insects as Food and Feed 1 : 215 – 225 . Google Scholar CrossRef Search ADS Scarpa , R. , Ferrini , S. and Willis , K. ( 2005 ). Performance of error component models for status-quo effects in choice experiments. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Netherlands : Springer , 247 – 273 . Google Scholar CrossRef Search ADS Scarpa , R. and Rose , J. M. ( 2008 ). Design efficiency for non-market valuation with choice modelling: how to measure it, what to report and why . Australian Journal of Agricultural and Resource Economics 52 : 253 – 282 . Google Scholar CrossRef Search ADS Scarpa , R. , Spalatro , F. and Canavari , M. ( 2007 a). Investigating preferences for environment friendly production practices. In: Canavari M. and Olson K. (eds) , Organic Food: Consumers’ Choice and Farmers’ Opportunities . New York : Springer , 115 – 124 . Scarpa , R. , Thiene , M. and Train , K. ( 2008 ). Utility in willingness to pay space: a tool to address confounding random scale effects in destination choice to the Alps . American Journal of Agricultural Economics 90 : 994 – 1010 . Google Scholar CrossRef Search ADS Scarpa , R. , Willis , K. G. and Acutt , M. ( 2007 b). Valuing externalities from water supply: Status quo, choice complexity and individual random effects in panel kernel logit analysis of choice experiments . Journal of Environmental Planning and Management 50 : 449 – 466 . Google Scholar CrossRef Search ADS Sonnier , G. , Ainslie , A. and Otter , T. ( 2007 ). Heterogeneity distributions of willingness-to-pay in choice models . Quantitative Marketing and Economics 5 : 313 – 331 . Google Scholar CrossRef Search ADS Steinfeld , H. , Gerber , P. , Wassenaar , T. , Castel , V. , Rosales , M. and De Haan , C. ( 2006 ). Livestock’s Long Shadow . Rome : FAO . Stevens , R. and Winter-Nelson. , A. ( 2008 ). Consumer acceptance of provitamin A-biofortified maize in Maputo, Mozambique . Food Policy 33 : 341 – 351 . Google Scholar CrossRef Search ADS Swait , J. and Louviere , J. ( 1993 ). The role of the scale parameter in the estimation and comparison of multinomial logit models . Journal of Marketing Research 30 : 305 – 314 . Google Scholar CrossRef Search ADS Thaler , R. H. and Johnson , E. J. ( 1990 ). Gambling with the house money and trying to break even: the effects of prior outcomes on risky choice . Management Science 36 : 643 – 660 . Google Scholar CrossRef Search ADS Thiene , M. and Scarpa , R. ( 2009 ). Deriving and testing efficient estimates of WTP distributions in destination choice models . Environmental and Resource Economics 44 : 379 – 395 . Google Scholar CrossRef Search ADS Tonsor , G. T. and Shupp , R. S. ( 2011 ). Cheap talk scripts and online choice experiments: looking beyond the mean . American Journal of Agricultural Economics 93 : 1015 – 1031 . Google Scholar CrossRef Search ADS Train , K. E. ( 2003 ). Discrete choice methods with simulation . New York : Cambridge university press . Google Scholar CrossRef Search ADS Train , K. and Sonnier , G. ( 2005 ). Mixed logit with bounded distributions of correlated partworths. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Dordrecht, The Netherlands : Springer , 117 – 134 . Google Scholar CrossRef Search ADS Train , K. and Weeks , M. ( 2005 ). Discrete choice models in preference space and willingness-to-pay space. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Dordrecht, The Netherlands : Dordrecht, Springer , 1 – 16 . UN ( 2015 ). World population prospects: The 2015 revision, key findings and advance tables . Working Paper No. ESA/P/WP.241. New York : United Nations . van Huis , A. ( 2013 ). Potential of insects as food and feed in assuring food security . Annual Review of Entomology 58 : 563 – 583 . Google Scholar CrossRef Search ADS Varela , E. , Mahieu , P.-A. , Giergiczny , M. , Riera , P. and Soliño , M. ( 2014 ). Testing the single opt-out reminder in choice experiments: an application to fuel break management in Spain . Journal of Forest Economics 20 : 212 – 222 . Google Scholar CrossRef Search ADS Volinskiy , D. , Adamowicz , W. L. , Veeman , M. and Srivastava , L. ( 2009 ). Does choice context affect the results from incentive-compatible experiments? The case of Non-GM and country-of-origin premia in canola oil . Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie 57 : 205 – 221 . Google Scholar CrossRef Search ADS von Haefen , R. H. , Massey , D. M. and Adamowicz , W. L. ( 2005 ). Serial nonparticipation in repeated discrete choice models . American Journal of Agricultural Economics 87 : 1061 – 1076 . Google Scholar CrossRef Search ADS Vossler , C. A. , Doyon , M. and Rondeau , D. ( 2012 ). Truth in consequentiality: theory and field evidence on discrete choice experiments . American Economic Journal: Microeconomics 4 : 145 – 171 . Google Scholar CrossRef Search ADS Vossler , C. A. and Watson , S. B. ( 2013 ). Understanding the consequences of consequentiality: testing the validity of stated preferences in the field . Journal of Economic Behavior and Organization 86 : 137 – 147 . Google Scholar CrossRef Search ADS Whitehead , J. C. and Cherry , T. L. ( 2007 ). Willingness to pay for a Green Energy program: a comparison of ex-ante and ex-post hypothetical bias mitigation approaches . Resource and Energy Economics 29 : 247 – 261 . Google Scholar CrossRef Search ADS Whittington , D. ( 1998 ). Administering contingent valuation surveys in developing countries . World Development 26 : 21 – 30 . Google Scholar CrossRef Search ADS Whittington , D. ( 2002 ). Improving the performance of contingent valuation studies in developing countries . Environmental and Resource Economics 22 : 323 – 367 . Google Scholar CrossRef Search ADS Yue , C. and Tong , C. ( 2009 ). Organic or local? Investigating consumer preference for fresh produce using a choice experiment with real economic incentives . HortScience 44 : 366 – 371 . Appendix A The buns were baked at the Food Processing Workshop Unit at the Jomo Kenyatta University of Agriculture and Technology in Kenya by professional technicians. Since Kinyuru, Kenji and Njoroge (2009) developed their recipe for buns from termite flour, we chose to validate this by first producing trial buns from CF. At this point, the product characteristics were assessed in cooperation with a nutrition scientist and several technicians. Following this, 10 participants that included five males and five females were recruited to evaluate the products for taste, texture, colour and smell. The recipe formulation that was used for the main baking is presented in Table A1. And the final bun products which were used for the field experiment are characterised as shown in Table A2. One can see that as the amount of the CF increases, the bun products become heavy, soft and brown. The increase in weight of the buns can be linked to the fact that the increase in the amount of CF lead to an increase in the amount of fat in the buns that can reduce water transpiration. Table A1. Ingredient composition of the buns Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Source: Alemu et al. (2017a). Note: ‘g’ refers to grams. Baking fat (7.5 g), salt (1.25 g), sugar (5 g), yeast (2.5 g) and acetic acid (0.125 milliliter) were added to each bun. Table A1. Ingredient composition of the buns Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Source: Alemu et al. (2017a). Note: ‘g’ refers to grams. Baking fat (7.5 g), salt (1.25 g), sugar (5 g), yeast (2.5 g) and acetic acid (0.125 milliliter) were added to each bun. Table A2. Texture and visual appearance of the buns Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Source: Alemu et al. (2017a). Table A2. Texture and visual appearance of the buns Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Source: Alemu et al. (2017a). Appendix B Subject instruction for the REAL treatment ‘You will be provided with 12 different choice scenarios within which three bags (Bags 1, 2 and 3) of buns are included. Bags 1 and 2 may contain buns made from wheat flour mixed with cricket flour. Some portion of the wheat flour can be fortified. Bag 3 contains only buns made from the wheat flour which was not fortified. In each scenario, you should choose ONE of the bags you would like to purchase (Bag 1 or 2) or you can choose Bag 3 if you would not like to purchase Bag 1 or 2. After you complete all 12 shopping scenarios, we will ask you to draw a number (1–12) from an envelope to determine which shopping scenario will be binding. In the envelope, are numbers 1–12. If the number 1 is drawn, then the first shopping scenario will be binding, and so on. For the binding scenario, we will look at the product you have chosen, give you your chosen product, and you will pay the listed price in that scenario. You should use the 90 KShs for the purchase. The most expensive alternatives cost 90 KShs. Although only one of the 12 shopping scenarios will be binding there is an equal chance of any shopping scenario being selected as binding, so think about each answer carefully.’ Subject instruction for the HYPO and ROOR treatments ‘You will be provided with 12 different choice scenarios within which three bags (Bags 1, 2 and 3) of buns are included. Bags 1 and 2 may contain buns made from wheat flour mixed with cricket flour. Some portion of the wheat flour can be fortified. Bag 3 contains only buns made from the wheat flour which was not fortified. In each scenario, you should choose ONE of the bags you would like to purchase (Bag 1 or 2) or you can choose Bag 3 if you would not like to purchase Bag 1 or 2. For each choice scenario, assume that you have the opportunity to, here and now, to purchase ONE and ONLY ONE of the bags at the listed prices. While you will not actually buy any products today or pay the posted prices, please respond to each choice scenario as if it were a real one and you would have to give up real money were one of the 12 scenarios to be selected as binding.’ Appendix C Table C1. HYPO Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Note: Z-values in parentheses. Table C1. HYPO Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Note: Z-values in parentheses. Table C2. ROOR Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Note: Z-values in parentheses. Table C2. ROOR Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Note: Z-values in parentheses. Table C3. REAL Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Note: Z-values in parentheses. Table C3. REAL Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Note: Z-values in parentheses. Table C4. Estimates of the ECLC model by treatment Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Note: Z-values in parentheses. Table C4. Estimates of the ECLC model by treatment Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Note: Z-values in parentheses. Appendix D Table D2. Estimates of the ω parameters based on pooled data (two treatments at a time) Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors. Table D2. Estimates of the ω parameters based on pooled data (two treatments at a time) Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors. Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices) http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png European Review of Agricultural Economics Oxford University Press

Can a Repeated Opt-Out Reminder mitigate hypothetical bias in discrete choice experiments? An application to consumer valuation of novel food products

Loading next page...
 
/lp/ou_press/can-a-repeated-opt-out-reminder-mitigate-hypothetical-bias-in-discrete-Z5gvviTKhn
Publisher
Oxford University Press
Copyright
© Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com
ISSN
0165-1587
eISSN
1464-3618
D.O.I.
10.1093/erae/jby009
Publisher site
See Article on Publisher Site

Abstract

Abstract In this paper, we test whether a Repeated Opt-Out Reminder (ROOR) can mitigate hypothetical bias in stated discrete choice experiments (DCE). The data originate from a field experiment concerning consumer preferences for a novel food product made from cricket flour. Utilising a between-subject design with three treatments, we find significantly higher marginal willingness-to-pay values in hypothetical than in nonhypothetical settings, confirming the presence of hypothetical bias. Comparing this to a hypothetical setting where the ROOR is introduced, we find that the ROOR effectively mitigates hypothetical bias for one attribute and significantly reduces it for the rest of the attributes. 1. Introduction Since its advent in the early 1980s in marketing research, the stated discrete choice experiment (DCE) method has been used extensively in the environmental, agricultural, food, transport, health and marketing sectors to elicit the value of goods and services. However, critics have continually questioned the validity and the reliability of the results obtained from hypothetical experiments including stated DCE. If results from hypothetical experiments are to be useful for testing theories or to guide policy decisions, such approaches should yield reliable results. Nevertheless, this seems to not be generally the case (List and Gallet, 2001; Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Murphy et al., 2015a, 2015b), and one of the recurring issues tied to the stated DCE and stated preference methods in general is the issue of hypothetical bias. Due to the hypothetical nature of the choice questions asked, respondents may for various reasons overstate their Willingness-To-Pay (WTP) since their answers have no real consequences in terms of required payment for the good or service in question (e.g. Fox et al., 1996; Lusk and Schroeder, 2004; Harrison, 2006; Murphy et al., 2015a, 2015b; Vossler, Doyon and Rondeau, 2012; Vossler and Watson, 2013). There is a rich body of literature investigating hypothetical bias within the areas of environmental valuations employing the contingent valuation (CV) method to elicit the value of public goods (e.g. Carson et al., 1996; List and Shogren, 1998; Cummings and Taylor, 1999; Carlsson and Martinsson, 2001; List and Gallet, 2001; List, 2001; Little and Berrens, 2004; Murphy et al., 2015a, 2015b; Blumenschein et al., 2008). This issue is somewhat less explored in the DCE literature even though attention to hypothetical bias has increased in recent years. Given that people may behave differently depending on the type of the good and the experimental setup (Carson et al., 1996; Carson and Groves, 2007), the conclusions from external validity tests on CV studies based on public goods cannot be transferred directly to the DCE context nor to a private goods context. Hence, similar tests should be applied in DCE applications to thoroughly examine their validity. This is of particular importance as DCEs are extensively being used to assess demand for new products to guide policy development. In this regard, recent studies (e.g. Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Chang, Lusk and Norwood, 2009; Collins and Vossler, 2009; Loomis et al., 2009; Chowdhury et al., 2011; Vossler, Doyon and Rondeau, 2012; De-Magistris, Gracia and Nayga, 2013; Grebitus, Lusk and Nayga, 2013; Moser, Raffaelli and Notaro, 2014; Vossler and Watson, 2013) have examined the validity of DCEs empirically as well as theoretically. Most of these studies suggest that DCEs are indeed susceptible to hypothetical bias, even when they are used to value marketable private goods. Some investigations attempt to reduce hypothetical bias and increase explanatory power of empirical models by combining data from stated choice with reveal preference studies (e.g. Adamowicz, Louviere and Williams, 1994; Azevedo, Herriges and Kling, 2003; Brooks and Lusk, 2010). Others focus on increasing the realism of stated DCE studies by either employing hypothetical bias mitigation strategies (HBMS) (see Loomis, 2011, 2014 for a review) or by shifting to nonhypothetical choice experiments in order to reduce or completely eliminate hypothetical bias. Employing the latter, however, can be difficult since in addition to being costly and time consuming, the products of interest may not be available (Fox et al. 1996; De-Magistris, Gracia and Nayga, 2013). Therefore, HBMS may serve as important tools to remove hypothetical bias, and one such strategy, which has been often used in the literature, is the Cheap Talk (CT) script (Cummings and Taylor, 1999). However, the existing evidence with respect to the effectiveness of the CT script is mixed. While some have found that the CT script effectively mitigates hypothetical bias (Lusk, 2003; Carlsson, Frykblom and Lagerkvist, 2005; Murphy et al., 2015a, 2015b; List, Sinha and Taylor, 2006; Chowdhury et al., 2011; Tonsor and Shupp, 2011), others have found it to be effective only for certain sub-groups of respondents (List, 2001; Lusk, 2003; Aadland and Caplan, 2003, 2006; Barrage and Lee, 2010; Ami et al., 2011). Furthermore, the findings of Brown, Ajzen and Hrubes (2003) suggest that the CT is bid range sensitive, and Nayga, Woodward and Aiew (2006), Brummett, Nayga and Wu (2007) and Blumenschein et al. (2008) all found that the CT did not have any influence on the WTP values. A number of studies have tested the credibility of the CT script in DCE settings and their results generally show that the CT is ineffective in removing the hypothetical bias (List, Sinha and Taylor, 2006; Özdemir, Johnson and Hauber, 2009; Yue and Tong, 2009; Carlsson, García and Löfgren, 2010; Bosworth and Taylor, 2012; Moser, Raffaelli and Notaro, 2014). This might be explained by the fact that the CT script was originally developed for CV studies, and applying it to the DCEs may not be appropriate as there are contextual and structural differences between the two valuation approaches (Ladenburg and Olsen, 2014), thus, responses may differ markedly (List and Shogren, 1998; Carson and Groves, 2007). Given these inconsistencies surrounding the effectiveness of the CT script, researchers have proposed other HBMS. For instance, Jacquemet et al. (2013) implemented a Solemn Oath script, and De-Magistris, Gracia and Nayga (2013) tested an Honesty priming approach. Both approaches were found to be effective in removing hypothetical bias. Other researchers have instead relied on calibrating hypothetical values using respondents’ stated certainty in choice (Blumenschein et al., 2008). More recently, along the same lines, Ladenburg and Olsen (2014) suggested augmenting the CT script with an Opt-Out Reminder (OOR), which reminds respondents that if the prices of the experimentally designed alternatives are greater than what their household will pay, they should choose the opt-out alternative. In a split sample setup, they found that when the CT was applied with the OOR added, WTP estimates were significantly reduced compared to using the CT without the OOR. Varela et al. (2014) also tested the impact of presenting the CT with the OOR and contrary to Ladenburg and Olsen (2014), the OOR was not found to influence WTP. A possible explanation might be that Ladenburg and Olsen (2014) repeated the OOR before each single choice set whereas Varela et al. (2014) only presented it once in the scenario description. This seems to support Ladenburg and Olsen (2014) who speculate that, given the repeated choice nature of DCE, it may be of particular importance to repeat the reminder since respondents might otherwise forget about the reminder as they progress through the choice tasks. This article contributes to these developments in the literature in terms of extending the work by Ladenburg and Olsen (2014). The aim is to properly test the effectiveness of including a Repeated OOR (ROOR) in stated DCEs by directly testing its ability to mitigate hypothetical bias. Ladenburg and Olsen (2014) implemented the ROOR in conjunction with a CT script. In this study, we include only the ROOR to avoid confounding the effects of the ROOR and the CT. This allows for a more rigorous test of the pure effect of the ROOR. Another limitation of the experimental setup used in Ladenburg and Olsen (2014) as well as in Varela et al. (2014) is that they tested the OOR in a purely hypothetical setup. They did not include a nonhypothetical treatment which is necessary for establishing the extent to which hypothetical bias is present in the first place. Hence, they could not conclude whether the OOR actually reduced hypothetical bias. Therefore, in this article, we include an incentivised discrete choice experiment (REAL) treatment in addition to a fairly standard hypothetical DCE (HYPO) to investigate whether hypothetical bias exists at the outset. A third treatment presents the exact same hypothetical DCE except that the ROOR is added (ROOR), in order to examine the impact of the ROOR. Both Ladenburg and Olsen (2014) and Varela et al. (2014) presented the OOR script in combination with a CT script, thus confounding the effects of both scripts. To isolate the effect of the OOR in terms of mitigating hypothetical bias, we do not use a CT script but simply present respondents with the ROOR alone. With this article, we contribute to the literature in several aspects. First and foremost, to our knowledge, this is the first valid test of the effectiveness of the ROOR in terms of mitigating hypothetical bias. Second, the majority of studies investigating hypothetical bias have done so in a developed country context whereas only a couple of papers have considered it in a developing country context (Ehmke, Lusk and List, 2008; Chowdhury et al., 2011). One reason for this may be the presence of significant challenges in conducting valuation studies in developing countries (Durand-Morat, Wailes and Nayga, 2016). However, Ehmke, Lusk and List (2008) found hypothetical bias to be location dependent, implying that people from developing countries could have different susceptibility to hypothetical bias than those from developed countries. This can be related to the issue that the role of money may be viewed differently in these different regions (Gibson et al., 2016). In light of this, we conduct our study in a developing country, thus adding to the sparse DCE literature in such settings. This also represents a novel addition to the growing literature on nonhypothetical DCE (Carlsson and Martinsson, 2001; Lusk and Schroeder, 2004; Ding, Grewal and Liechty, 2005; Alfnes et al., 2006; Johansson-Stenman and Svedsäter, 2008; Lusk, Fields and Prevatt, 2008; Chang, Lusk and Norwood, 2009; Loomis et al., 2009; Volinskiy et al., 2009; Yue and Tong, 2009; Grebitus, Lusk and Nayga, 2013; Michaud, Llerena and Joly, 2013; Gracia, 2014). Finally, while the aim of the article is mainly methodological, it also makes a novel contribution to the literature on consumers’ food preferences by estimating Kenyan consumers’ WTP for novel food products (NFPs) made with edible insects. 2. Background 2.1. Edible insects as a sustainable source of food The world population is projected to increase to 9 billion in 2050 (UN, 2015). Faced with climate change and environmental sustainability concerns, the current food production systems may not be able to meet the growing demand for food which is estimated to rise by 60 per cent in 2050 (FAO, 2009, 2013). In this respect, it is evident that the number of malnourished people in developing countries is (still) alarmingly high, although some nutritional improvements have been observed in recent years (Braun, 2007). Specifically, despite a globally positive change in the supply of food energy, total protein and animal protein in the past few decades, a decreasing trend has been observed in many developing regions (Pellet and Ghosh, 2004), especially in several sub-Saharan African (SSA) countries where food security has been and still is one of the biggest societal challenges. Increasing livestock production is one option to tackle the shortage of animal protein but it contributes significantly to the emission of greenhouse gasses (Steinfeld et al., 2006) while also requiring substantial inputs of water and crop-based feed. Both these required production inputs are expected to decrease in availability in SSA with future global warming. Increasing food security thus requires that either the productivity of the current agricultural sector increases e.g. through new technological developments or alternative sources of food that increase food security in a sustainable manner must be identified through research and development (Boland et al., 2013; van Huis, 2013). In line with this, the concept of producing edible insects for food has received substantially increasing attention recently (FAO, 2013). Insects can be a good source of important nutrients for human consumption (Bukkens, 1997; Christensen et al., 2006). As shown by Oonincx et al. (2010) and Halloran et al. (2017), edible insects can be considered environmentally friendly as they emit significantly lower amounts of greenhouse gasses than livestock production. Despite insects being traditionally consumed as food in large parts of Africa, Asia and Latin America, and historically all over the world, the many different aspects of eating insects, which has earlier been labelled with the collective term ‘entomophagy’ (Evans et al., 2015), have received rather limited attention in research (van Huis, 2013). Recently, however, the literature investigating e.g. food safety, nutritional value, consumer attitude and legislative aspects of producing insects for food has begun to increase very rapidly (e.g. Finke, 2013; van Huis, 2013; Looy, Dunkel and Wood, 2014; Ruby, Rozin and Chan, 2015; Alemu et al., 2017b). Most of the recent research focuses on such supply side issues. However, the demand side, i.e. consumers’ preferences for insects as food, will ultimately be equally important for a successful introduction of edible insects as a new and substantial component of the future food production sector. Therefore, gathering information about consumers’ preferences is necessary since decisions concerning establishment of new insect production sectors as well as regulatory, standardisation and quality control schemes are likely to be suboptimal if knowledge about potential demand in the market is not accounted for. 2.2. The use of DCEs in developing countries WTP for food products can be elicited using different approaches including DCEs and experimental auctions (Gracia, Loureiro and Nayga, 2011). DCEs are grounded in random utility theory and Lancaster’s characteristics demand theory (Lancaster, 1966; McFadden, 1986) most commonly used, and they rely on asking respondents to make choices in a manner that mimics their choices in actual purchasing situations. This makes them more familiar to consumers than experimental auctions (Alfnes et al., 2006). The application of DCEs to elicit values of goods and services is currently gaining popularity in developing countries (Bennett and Birol, 2010; Birol et al., 2015; Gibson et al., 2016). Notwithstanding this, most DCE studies, particularly consumer valuation of food attributes based on nonhypothetical DCEs have been conducted in developed countries (Alphonce and Alfnes, 2017). This is related to the challenges in conducting economic valuation studies in developing countries. Durand-Morat, Wailes and Nayga (2016) and Whittington (1998, 2002) highlight that the ability of respondents to follow and process information, the competences of local personnel, unavailability of proper population sampling frames, limited coverage of infrastructures, and poor design and implementation of survey scenarios are the main challenges. These issues need to be considered when conducting DCE valuation studies in developing countries. In this study, we exerted a great effort to obtain high-quality data by directly addressing these challenges in the survey design. The problem of hypothetical bias is a concern in any DCE, though it has not yet been thoroughly explored in developing country settings. To the authors’ knowledge, only two studies addressing this issue are available in the DCE literature so far. Aiming to assess the value of bottled mineral water, Ehmke, Lusk and List (2008) found that hypothetical bias differs across locations and it is larger in developed than in developing countries. The authors attributed this disparity to cultural differences. Chowdhury et al. (2011) found that Ugandan consumers’ WTP for biofortified foods are significantly higher in hypothetical treatments than in real treatments indicating the presence of hypothetical bias. In the current study, we aim to contribute to the sparse DCE literature in developing countries especially concerning hypothetical bias using data from consumer valuation of novel food products in Kenya. 2.3. Strategies to mitigate hypothetical bias in DCEs If hypothetical bias is present in DCEs – as is commonly found – the results will potentially be misleading for decision makers interested in utilising value estimates for e.g. policy evaluations or product development. The important question is then how to mitigate it. A primary solution would be to use nonhypothetical DCEs, e.g. using real products and real payments. These approaches are costly and they are often difficult to administer in the field, especially in experiments involving novel foods. In recognition of this, a range of different HBMS have been proposed in the DCE literature. These are broadly divided into two categories: ex-ante and ex-post HBMS (Whitehead and Cherry, 2007; Loomis, 2011, 2014). The ex-ante strategies are used to mitigate hypothetical bias at the design stage. According to Loomis (2011, 2014), the most common strategies are (i) developing a survey design that stimulates respondents to believe that their responses have real consequences in real situations (consequentiality) (e.g. Carson and Groves, 2007), (ii) using solemn oath and honesty priming approaches through which respondents are asked to swear to respond honestly to the study questions (e.g. De-Magistris, Gracia and Nayga, 2013; Jacquemet et al., 2013), (iii) using inferred valuation by asking respondents to state what they think other people would be willing to pay for the good in question (e.g. Lusk and Norwood, 2009), (iv) using the theory of cognitive dissonance to enable respondents to act in a trustworthy and rational manner (e.g. Alfnes, Yue and Jensen, 2010) and, finally (v) cheap talk (e.g. List, Sinha and Taylor, 2006). Cheap talk, which is probably the most commonly used HBMS, simply explains to respondents what hypothetical bias is. It was first applied by Cummings and Taylor (1999) and other researchers followed suit (e.g. Lusk, 2003; Carlsson, Frykblom and Lagerkvist, 2005; Chowdhury et al., 2011; Tonsor and Shupp, 2011). The ex-post strategies are used to mitigate hypothetical bias after the completion of the survey. One can find a detailed account of such approaches in Loomis (2014). The most common approach is to make use of follow-up questions on response uncertainty to adjust for hypothetical bias. Blumenschein et al. (2008) and Champ et al. (2009) showed that recoding uncertain responses mitigated the hypothetical bias in their respective studies. In a DCE set up, Ready, Champ and Lawton (2010) similarly found that calibrating choices using the uncertainty level of respondents resulted in hypothetical bias being mitigated. In this study, we focus on testing a recently proposed new ex-ante HMBS, namely the ROOR which until now has not been properly assessed for effectiveness. 3. Experimental design and procedure We test the effectiveness of the ROOR in a DCE field experiment in Kenya, where there is increasing interest in introducing edible insects into the food production sector. Specifically, we focus on consumers’ preferences for crickets as food since this type of insect is considered particularly promising in terms of developing viable production systems for household consumption as well as for large scale commercialisation in Kenya. 3.1. The product contexts and DCE design Rather than presenting respondents with whole crickets on a plate, traditional buns were chosen as the product context for the DCE since it was considered a more realistic market entry strategy to introduce crickets by adding cricket flour (CF) to buns which are generally considered affordable and available in most food outlets throughout Kenya. Buns are one of the bread types that form the major staple food in Kenya according to focus group interviews and key informant discussions. Furthermore, the practical manageability of product development and handling guided the choice of buns for this experiment. CF was mixed into wheat flour to bake the buns. The amounts of CF used to produce the buns were determined based on the recipe specification developed by Kinyuru, Kenji and Njoroge (2009). Serving as attribute levels in the DCE, the amounts of CF were varied at 0 per cent, 5 per cent and 10 per cent of the total flour. The product development process, the recipe formulation and the characteristics of the final bun products used for the field experiment are presented in Appendix A. Nonhypothetical DCEs involving new products require development of all the product combinations with all the relevant attributes. This is, however, typically constrained by budget and other logistic complications. For this study, we identified three attributes that would enable us to actually develop and produce all possible product combinations for the DCE. The first attribute represents the insect component in terms of the amount of CF in the buns (CF). The second attribute describes whether a portion of the wheat flour is fortified or not (Fortified), while the third attribute is the price of a bag of buns (Price). Insects like crickets may be consumed in their whole form or ground into flour. Besides representing the more realistic market entry strategy, we focus on the flour form because low protein foods such as maize and sorghum, which are popular in east Africa (Stevens and Winter-Nelson, 2008), can be easily enhanced with CF to increase their nutritional value. In addition, CF is already being produced on a rather large scale elsewhere in the world, hence, a similar production would seem achievable in Kenya. The Fortified attribute was included to assess Kenyan consumers’ preferences and WTP for fortified foods. The level for the fortified wheat flour was arbitrarily set to be 40 per cent of the total flour used to bake the different buns. Consumers in Kenya are familiar with food fortification because the Kenyan government through the Kenya National Food Fortification Alliance (KNFFA) has introduced a food fortification programme to reduce micronutrient deficiency. This programme covers foods including maize and wheat (KNFFA, 2011). A secondary purpose with including the fortified flour attribute was an attempt to (at least to some extent) separate out the pure effect of nutritional improvements. In other words, we wanted to give participants the option to choose buns that were nutritionally better than the standard alternative without necessarily forcing them to choose cricket flour. This way, if participants strongly rejected the CF, they might still exhibit preferences for better nutrition, thus avoiding confounding preferences for nutrition, which we expected to be positive, with preferences for other aspects of eating insects which might be negative, e.g. if respondents would find the idea of eating insects disgusting (Looy, Dunkel and Wood, 2014). The attributes and their levels are presented in Table 1. The amount of CF has three levels: 0 g (Standard with no CF), 6.25 g (Medium CF) and 12.5 g (High CF). This is equivalent to 0 per cent, 5 per cent and 10 per cent of the total wheat flour used per bag of buns. The Fortified attribute has two levels at 0 and 50 g. The total amount of flour is kept constant at 125 g per bag. Thus, when CF and/or fortified wheat flour is added, the amount of wheat flour is reduced accordingly. Besides the quality attributes, a Price attribute is included with six levels. An initial investigation of actual market prices of buns showed that a single bun would typically cost around 10 Kenyan Shillings1 (KShs), rarely less than KShs 5. Since focus group discussions indicated that respondents would not respond negatively to the other attributes, it was thus decided to set the minimum price for a bag with three buns at KShs 20. Focus group discussions further suggested that a suitable choke price for the price range would be KShs 90, and the six price levels were subsequently set at KShs 20, 25, 30, 40, 60 and 90. Table 1. Product attributes and levels Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Table 1. Product attributes and levels Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Attribute description Level Variable name Coding Amount of cricket flour 12.5 g or 10% High CF Dummy 6.25 g or 5% Medium CF Dummy None None Reference Whether some portion of the wheat flour is fortified or not 50 g Fortified Dummy 0 g None Reference The price of a bag of buns 20, 25, 30, 40, 60, 90 Price Continuous Once the attributes and their levels had been determined, the DCE design was produced. It was subjected to standard efficient design procedures in line with state-of-the-art experimental design theory (Louviere, Hensher and Swait, 2000; Ferrini and Scarpa, 2007; Scarpa and Rose, 2008; ChoiceMetrics, 2012). Three alternatives are included in the design: two of them are deigned ones and the last one is a specific alternative which is constant across choice sets. The three attributes, i.e. the amount of CF, Fortified and Price, have three, two and six levels, respectively. Based on this, a full factorial design results in 36 alternatives. To avoid exceeding the cognitive capacity of respondents, a fractional factorial design was employed. The design procedure involved two stages. First, we produced a design using zero priors and conducted two rounds of pilot studies with 42 respondents. Second, the data from the pilot studies were estimated to obtain priors in terms of attribute parameter estimates. The priors were subsequently used to inform and update the D-efficient fractional factorial design which was generated using the Ngene software (ChoiceMetrics, 2012). The design produced 12 choice sets which were presented to participants. Figure 1 shows an example of one of the choice sets used. Fig. 1. View largeDownload slide An example of a choice set used in the survey. Fig. 1. View largeDownload slide An example of a choice set used in the survey. As is evident from Figure 1, we use a forced choice type of DCE by including an alternative with a standard bag of buns which is constant across all choice sets. This is motivated by the fact that we are dealing with a staple convenience consumer product, namely bread (Cova and Pace, 2006). Even though the context of shopping for bread is a familiar everyday experience for most consumers, the main attribute of interest in our survey, namely the addition of cricket flour, is adding complexity and difficulty to the choice tasks. In this case, including a no-buy option may prompt respondents to use it as a simplifying strategy without actually making the required trade-offs to reveal their true responses (see von Haefen, Massey and Adamowicz, 2005; Beshears et al., 2008; Burton and Rigby, 2008; Boxall, Adamowicz and Moon, 2009; Rigby, Alcon and Burton, 2010). Hence, we chose the forced choice setting to reduce the risk of respondents resorting to such heuristics. While it is strictly speaking always possible for consumers in the real world to not buy a specific product, we argue that the majority of consumers in our case would in fact not consider no-buy an option when shopping for bread in the real world. Focus group interviews supported this, and during the personal interviews, no respondents complained about being forced to buy bread. Furthermore, while it may be more common in food DCE studies to include a no-buy option, it is not uncommon to use forced choice settings (e.g. Carlsson, Frykblom and Lagerkvist, 2007a, 2007b; Kallas et al., 2013). 3.2. The experimental treatments and hypothesis testing Three experimental treatments were designed to investigate whether hypothetical bias persists and whether it can be mitigated by a ROOR. In the first treatment (HYPO), participants make purely hypothetical choices without having to pay for the bun products. In the second treatment (REAL), participants’ choices are no longer hypothetical as they are subjected to an incentivisation mechanism similar to that of Alfnes et al. (2006) and Gracia, Loureiro and Nayga (2011). The inclusion of this treatment enables testing the extent to which hypothetical bias is present. The lack of such a treatment was the key limitation of Ladenburg and Olsen (2014) and Varela et al. (2014) who could only speculate about the impact of the OOR on hypothetical bias. In the third treatment (ROOR), participants make hypothetical choices but now framed with a ROOR in which subjects are provided with the following opt-out reminder: ‘If the prices of Bag 1 and Bag 2 are higher than what you think your household will pay, you should choose Bag 3.’ This treatment is specifically designed to investigate if hypothetical bias can be mitigated by the ROOR. As the name suggests the ROOR is placed repeatedly before each choice set so as to take the repeated nature of the choices into account and help subjects to stay on the ‘true’ preference path (Ladenburg and Olsen, 2014). In this manner, information flow and consistent trade-offs across choice sets would be ensured. In addition, the ROOR enables participants to attend to both the designed and the opt-out alternatives appropriately, in which case participants should make the required trade-offs among the attributes to reach a decision that ideally mirrors their ‘true’ preferences. Moreover, as discussed in Ladenburg and Olsen (2014), unlike the CT script, the ROOR is not affected by conformity issues as it does not convey any implicit cues as to what others do and how they behave in relation to value revelations. The experiment is undertaken based on a between-subject design to control for behavioural shift and confounding effects which might result if a within-subject design was adopted (Lusk and Schroeder, 2004; Charness, Gneezy and Kuhn, 2012). In total, 334 participants took part in the survey. Participants were randomised into the three treatments, resulting in 109 participants in each of the HYPO and ROOR treatments, and 116 participants in the REAL treatment. The participants are either household heads or spouses as they are the decision makers in dietary selection for their household members. With this experimental setup, we are able to test the following four research hypotheses. First, based on our expectation that the ROOR addresses the tendency of forgetting the opt-out alternatives in DCEs, we hypothesise that the hypothetical setting will result in fewer opt-out choices than the incentivised setting, but when adding the ROOR the amount of opt-out choices will be similar to the incentivised setting: H1: Opt-out frequencyHYPO < Opt-out frequencyREAL = Opt-out frequencyROOR Second, we hypothesise that the hypothetical setting will be subject to hypothetical bias in terms of mean marginal WTP estimates being higher than in the incentivised setting: H2: WTPHYPO > WTPREAL Conditional on confirmation of H2, we hypothesise that by adding the ROOR we can reduce hypothetical bias in terms of achieving mean marginal WTP estimates that are lower than those of the standard hypothetical setting: H3: WTPHYPO > WTPROOR Finally, also conditional on H2 being confirmed, but representing a stronger test than H3, we hypothesise that adding the ROOR will not only reduce hypothetical bias, but even completely eliminate it. In other words, the ROOR treatment will generate mean marginal WTP estimates equivalent to the incentivised setting: H4: WTPROOR = WTPREAL 3.3. Experimental procedure Survey interviews were conducted face-to-face using trained interviewers. The personal interviews took place in three different parts of Kenya that were selected to ensure inclusion of respondents from both rural and urban areas, of various different ethnicities, and where some still maintained the tradition of eating insects while others had left it. Respondents were recruited using a stratified random sampling approach to further ensure variation across key socio-demographic variables. For further descriptions of the sampling procedure, see Alemu et al. (2017a). The actual interview procedure followed four steps. First, participants were welcomed and seated and one trained interviewer was assigned to each for conducting the personal interview. Second, they were given KShs 30 in cash for their participation in the experiment. Participants were then randomly assigned to one of the three experimental treatments described above. The participants in the REAL treatment received an additional remuneration of KShs 90 which they could use fully or partly to buy the chosen products in the DCE. The provision of the monetary remuneration before the start of the DCE process was intended to induce a sense of ownership2 and avoid the house money effect that may be associated with windfall earnings in experiments (Thaler and Johnson, 1990; Cárdenas et al., 2014). Third, the samples of the three bun products were provided to participants so that they could view, smell and taste them.3 This allows consumers to experience the new products and their attributes, thereby facilitating preference formation. This is a unique feature of our study as in most stated DCE studies proxies for taste have been used instead of giving consumers direct experience of the product in question (Chowdhury et al., 2011). Fourth, participants were then presented with the DCE questionnaire. In addition to the choice sets, it contained information4 about crickets as food as well as a description of the attributes. Additionally, participants in the REAL treatment received instructions, which are adapted from Johansson-Stenman and Svedsäter (2008), Chang, Lusk and Norwood (2009) and Mørkbak, Olsen and Campbell (2014), regarding the incentivisation mechanism (see Appendix B). In Lusk and Schroeder (2004), Chang, Lusk and Norwood (2009) and Mørkbak, Olsen and Campbell (2014) a single random number was drawn for a whole group of participants to identify the binding choice set for all participants in that group. As opposed to this, but in line with Alfnes et al. (2006) and Gracia, Loureiro and Nayga (2011), the identification of the binding choice set in this study was implemented by asking each individual participant to draw a numbered ball from 1 to 12 from a hat after they had answered the 12 choice tasks. Participants were then given the bag of buns they had chosen in the binding choice set and they had to pay the price of that bag. 4. Econometric specification The random utility theory (RUT) (McFadden, 1986) serves as a theoretical framework to specify utility expressions in DCEs. In what follows, we follow Train and Weeks (2005). According to RUT, the utility of individual n choosing alternative k among J given alternatives in choice situation t can be expressed as Unkt=−αnmnkt+βn′xnkt+ϵnkt. (1) The utility is expressed as a component of price, m, and non-price attributes, x. The parameters αn and vector βn′ are the coefficients of the price and non-price attributes. For logit models, the variance of the error term ϵn can be written as sn2 (π26 ), where sn is the scale parameter which differs across decision makers. Without loss of generality, equation (1) can be divided by sn as Unkt=(αnsn)mnkt+(βnsn)′xnkt+εnkt. (2) Letting λn= (αnsn ) and cn= (βnsn ), equation (2) becomes Unkt=−λnmnkt+cn′xnkt+εnkt (3) where the error term, εn, is distributed IID extreme value. Equation (3) describes utility in preference space. As suggested by Train and Weeks (2005), an alternative option is to employ a WTP space approach. The utility in preference space in equation (3) may be reformulated so that the estimated coefficients refer directly to the WTP values. This approach has been applied by Train and Weeks (2005), Sonnier, Ainslie and Otter (2007), Scarpa, Thiene and Train (2008), Balcombe, Chalak and Fraser (2009), Thiene and Scarpa (2009), Hensher and Greene (2011) and Hole and Kolstad (2012). Most of the evidences so far suggest that models based on the WTP space approach yield more plausible behavioural explanations. Specifically, the distributions of WTP values are found to be within reasonable margins than values based on the preference space formulation. Therefore, there is an increasing support for implementation of the WTP space specification when the goal is to obtain appropriate WTP estimates which would be used for policy guidance. As mentioned above, the WTP space is expressed by reformulating the preference space utility expression. Reparametrising equation (3) gives Unkt=−λnmnkt+(−λnwn)′xnkt+εnkt. (4) Equation (4) is called utility in WTP space.Train and Weeks (2005) mentioned that equations (3) and (4) are behaviourally equivalent. Nevertheless, one can see, and as mentioned before, WTP calculation based on the preference space specification is directly impacted by the distribution of λn and cn, and, in some cases, like normal distributions, the moments of the WTP distribution may be undefined and furthermore the WTP distribution may suffer from a fat tail problem causing identification problems. Heterogeneity in preferences can be accounted for by specifying mixed logit models, which can be derived in a number of different ways (Train, 2003). The most common specification is a random parameter logit (RPL) model5 in which attribute parameters are treated as random parameters. The random parameters may in principle follow any relevant distribution, but often a normal distribution is used for quality attribute parameters and a lognormal is used for the negative of the price attribute parameter (Train, 2003; Train and Sonnier, 2005). We employ the RPL model as focus group interviews had indicated preference heterogeneity across participants. In line with Scarpa, Thiene and Train (2008), we assume that λn=−exp(υn), where υn is the latent random factor of the price coefficient. In addition, let β represent all the random parameters estimated in the WTP space model, then equation (4) can be rewritten as Unkt=Vnkt(βn,xnkt)+εnkt (5) where Vnkis the indirect utility function, which is a function of the attributes of the alternatives, i.e. Vnk=βn′Xnkt. The (negative of the) coefficient on price ( λn) is assumed to follow a lognormal distribution in order to ensure a strictly positive impact of income on utility. The focus group discussions and pilot tests indicated that some people would have negative preferences for the non-price attributes of the products while others would have positive. As a result, the distributions of the coefficients for these attributes are assumed to follow a normal distribution. We allow the scale, which is accommodated through λn, to vary across decision makers by specifying all random parameters to be correlated with each other, i.e. no restrictions are placed on the variance–covariance matrix. As the estimated parameters, β′, vary over individuals according to the specified distributions with density f(β) (Train, 2003), the probability that an individual n chooses alternative k among J given alternatives in choice situation t can be derived as Pnk=∫(eβn′Xnkt∑jJeβ′xntj)f(β)dβ. (6) In line with, e.g. Thiene and Scarpa (2009) and Hole and Kolstad (2012), equation (6), which is based on the WTP space specification, is estimated using the MSL procedure by maximising the full simulated log-likelihood over the sequence of choices and over the sample. We estimate the models using the BIOGEME 2.2 software (Bierlaire, 2003), maximising the simulated likelihood function with the CFSQP algorithm (Lawrence, Zhou and Tits, 1997) to avoid the local optima problem (Scarpa, Thiene and Train, 2008). For the simulation procedure, 3,000 Modified Latin Hypercube Sampling draws (Hess, Train and Polak, 2006) were used since this amount of draws was found to produce stable parameter estimates. In line with, e.g. Lusk and Schroeder (2004), Tonsor and Shupp (2011), De-Magistris, Gracia, Nayga (2013) and Moser, Raffaelli and Notaro (2014), we use the complete combinatorial method (Poe, Giraud and Loomis, 2005) to test the different hypotheses outlined above. For this purpose, 1,000 WTP estimates are bootstrapped for each attribute in each sample based on the Krinsky and Robb (1986) method using 1,000 draws from multivariate normal distributions of the coefficient estimates and the variance–covariance matrix of the random parameters. Differences between the WTP estimates can also be tested by pooling data for the two treatments in a specific hypothesis test as in De-Magistris, Gracia and Nayga (2013) and Bazzani et al. (2017). One can then specify an extended utility function by interacting dummy variables representing treatment effects with attribute levels. Accordingly, equation (4) can be extended as Unkt=−λn[mnkt+wn′xnkt+ωn(xnkt∗dtreatment)]+εnkt (7) Two dummy variables are created since there are three treatments. They are denoted by dtreatment taking a value of 1 for the second treatment in a specific hypothesis test, 0 otherwise. Differences in WTP estimates between the two treatments will be captured directly by the coefficient ωn, and its sign will indicate the direction of the treatment effect. 5. Results The summary of sample socio-demographics presented in Table 2 shows no differences across treatments. Except the level of education between the ROOR and REAL treatments, chi-square tests failed to reject that the socio-demographic distributions differ across treatments, suggesting that participants in the three treatments are sufficiently similar in their characteristics to rule this potential factor out as cause of any preference differences found. Table 2. Summary statistics of participants’ characteristics by treatment (percentages) Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 aNo education, and drop outs from primary and secondary schools. Table 2. Summary statistics of participants’ characteristics by treatment (percentages) Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 Variable HYPO (a) ROOR (b) REAL (c) x2-tests, p-values a vs. b a vs. c b vs. c Age  18–34 years 38.5 45.9 40.5  35–54 years 55.0 41.3 47.4 0.107 0.431 0.665  55–64 years 5.5 8.3 9.5  Above 64 years 1 4.5 2.6 Household size  1–4 persons 56.9 48.6 53.4  More than four persons 43.1 51.4 46.6 0.531 0.823 0.782 Gender  Female 53.2 52.3 50.1 0.999 0.827 0.935  Male 46.8 47.7 49.9 Region  Rural 60.1 60.1 64 0.999 0.716 0.716  Urban 39.9 39.9 36 Monthly income in KShs  <15,000 66 63 58 0.658 0.315 0.697  15,000–50,000 25 30 37  >50,000 9 7 5 Education  Primary school 31.2 43.1 27.6  Secondary school 36.7 31.2 28.5 0.152 0.178 0.009  Tertiary level 11 14.7 15.5  University education 6.4 2.8 3.5  Othera 14.7 8.2 24.6 Number of participants 109 109 116 aNo education, and drop outs from primary and secondary schools. As a first step in investigating possible preference differences across sample splits, Table 3 provides a comparison of the choice distributions. Interestingly, while the Standard buns were chosen in 14 per cent of the choices in the HYPO treatment, they were chosen in 27 per cent and 31 per cent of the choices in the ROOR and REAL treatments, respectively. These results are in line with Lusk and Schroeder (2004) and Bazzani et al. (2017) who also found participants to choose the opt-out option more frequently in a real setting than in a hypothetical setting, though they used a ‘none of these’ opt-out. The chi-square tests confirm that the choice frequencies differ significantly for HYPO versus ROOR and HYPO versus REAL. However, there seems to be no significant difference when comparing ROOR and REAL. In other words, we can confirm hypothesis H1. These results provide a first indication that the ROOR may indeed serve its purpose as it seems to generate a better correspondence between hypothetical and nonhypothetical choices. Table 3. Choice frequencies Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Note: The chi-square tests have been carried out based on the actual number of choices rather than based on percentages. Table 3. Choice frequencies Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Alternative HYPO (1) % ROOR (2) % REAL (3) % x2-test, p-value (1) vs. (2) (1) vs. (3) (2) vs. (3) Bag 1 43.1 34.5 33.9 <0.001 <0.001 0.136 Bag 2 42.7 38.2 35.5 Bag 3 (standard buns) 14.2 27.3 30.6 Note: The chi-square tests have been carried out based on the actual number of choices rather than based on percentages. The estimation results are presented in Table 4. As noted before, the estimated parameters directly represent the implied WTP distributions and the coefficients of the attributes are interpreted as marginal WTP estimates. Preference equality across the three treatments is tested using a likelihood ratio test (Swait and Louviere, 1993; Louviere, Hensher and Swait, 2000; Lusk and Schroeder, 2004; Chowdhury et al., 2011; Moser, Raffaelli and Notaro, 2014). The null hypothesis is rejected (x2 = 152.8, p-value < 0.01) suggesting that the overall preference structure is not identical across the three treatments. Furthermore, all the estimated standard deviations of the random parameters6 are significant, indicating substantial preference heterogeneity across participants in all three treatments. Table 4. Model results and comparison of WTP estimates Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors; p-values obtained using the complete combinatorial method (Poe, Giraud and Loomis, 2005) based on 1,000 bootstrapped WTP estimates derived using Krinsky and Robb (1986) method; p-values represent results of the one sided test that the differences between the equivalent WTP estimates of two treatments are positive. Table 4. Model results and comparison of WTP estimates Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Parameters HYPO (1) ROOR (2) REAL (3) (1) vs. (3) p-value (1) vs. (2) p-value (2) vs. (3) p-value Mean parameter estimates  ASC (standard bun) −32.1 (5.42)* −5.24 (2.24)* −3.40 (2.29) 0.000* 0.000* 0.287  High CF 143 (13.8)* 40.8 (3.76)* 25.7 (3.67)* 0.000* 0.000* 0.000*  Medium CF 138 (11.5)* 43.4 (3.33)* 41.7 (4.28)* 0.000* 0.000* 0.407  Fortified 64.1 (6.79)* 20.4 (2.95)* 14.4 (2.03)* 0.000* 0.000* 0.049*  Price −3.2 (0.185)* −2.17 (0.177)* −2.29 (0.131)* Standard deviation estimates  ASC (standard bun) 37.8 (5.04)* 12.5 (2.50)* 7.72 (2.58)* 0.000* 0.000* 0.104  High CF 118 (9.3)* 86.1 (8.51)* 53.5 (5.67)* 0.000* 0.002* 0.004*  Medium CF 45.8 (3.34)* 28.9 (2.56)* 11.1 (1.75)* 0.000* 0.000* 0.000*  Fortified 14.0 (2.21)* 4.34 (1.31)* 3.62 (1.36)* 0.000* 0.000* 0.320  Price 0.770 (0.251)* 0.742 (0.239)* 0.752 (0.247)* No. of observations 1,308 1,308 1,392 Null log-likelihood −1,437.0 −1,437.0 −1,529.3 Final log-likelihood −835.7 −825.5 −993.3 Adjusted ρ2 0.405 0.412 0.337 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors; p-values obtained using the complete combinatorial method (Poe, Giraud and Loomis, 2005) based on 1,000 bootstrapped WTP estimates derived using Krinsky and Robb (1986) method; p-values represent results of the one sided test that the differences between the equivalent WTP estimates of two treatments are positive. The model results presented in Table 4 show that consumers are willing to pay a premium for buns containing CF. More interestingly, the WTP estimates differ across the treatments. Generally, the WTP values obtained from the HYPO sample are three to six times as large as the WTP values obtained from the REAL sample, suggesting that the HYPO WTPs suffer noticeably from hypothetical bias. Specifically, the WTP estimate for High CF in the HYPO sample is almost six times that of the REAL sample. In addition, the WTP estimate for the Medium CF is more than three times as large in the HYPO as in the REAL sample. Turning to the WTP for fortified relative to non-fortified wheat flour, the estimate obtained in the HYPO sample is more than four times that obtained from the REAL sample. In accordance with the significantly fewer choices of Bag 3, i.e. the opt-out, in the HYPO sample (see Table 3), the ASC estimate for Bag 3 is much more negative in the HYPO compared to the REAL, confirming that participants in the HYPO react much more negatively to the opt-out alternative, irrespective of the attribute levels of the offered alternatives. According to the test results in Table 4, we can reject the null hypothesis of equal WTP estimates across the HYPO and REAL treatments for all attributes. Given that the attribute WTPs obtained in the HYPO sample are all higher than those of the REAL treatment, we thus confirm hypothesis H2, i.e. hypothetical bias is present in the HYPO sample. Furthermore, the negative opt-out effect observed for the ASC estimate in the HYPO treatment, i.e. participants disproportionally often choosing the non-opt-out alternatives regardless of their attributes, is not present in the REAL treatment. This serves as another indication of hypothetical bias, corresponding to the findings in Table 3. Turning to the question of particular interest in this article namely the effect of the ROOR, the complete combinatorial test results show that for all attributes the ROOR treatment results in significantly lower WTP estimates than the HYPO treatment. Hence, we can also confirm hypothesis H3. For the High CF attribute, the WTP obtained in the ROOR treatment is though still significantly higher than in the REAL treatment, while for the Medium CF, the difference is not significant, and for the Fortified attribute, the difference is only borderline significant. The ASC estimates are not significantly different, but the fact that the ASC estimate is still significantly lower than zero in the ROOR indicates that the negative status quo effect is not completely removed. Hence, the slightly stronger hypothesis H4 can only be partly confirmed since hypothetical bias, despite being significantly reduced, is not completely removed for all attributes. Notwithstanding this, presenting the ROOR before each choice set appears to be highly effective, removing 87–98 per cent of the hypothetical bias in absolute terms across the range of attributes as well as for the negative opt-out effect. All the above results regarding hypothetical bias and the impact of the ROOR in reducing it are graphically described in Figure 2. Probability density functions of the mean marginal WTPs are plotted for each treatment using 10,000 draws from multivariate normal distributions of the WTPs. The densities representing the WTP distributions in REAL and ROOR treatments are tighter than those in the HYPO treatment, confirming the results presented above. Fig. 2. View largeDownload slide Distribution of mean marginal WTP values by treatments. Fig. 2. View largeDownload slide Distribution of mean marginal WTP values by treatments. An excerpt of the results from running an extended utility function model on the pooled data is presented in Table 5 (see Appendix D for the full results). The estimates pertaining to the extended part capturing the treatment effects directly, i.e. the vector w in equation (7), generally confirm the findings of Table 4. The signs of the estimated coefficients show negative effects of the second treatments on WTP estimates in a specific hypothesis test. This is in line with our alternative hypotheses. Table 5. Robustness test using an extended utility function Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Asterisk (*) indicates significance at 5 per cent level or lower. The table only shows mean estimates of s for the ω parameters. Table 5. Robustness test using an extended utility function Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Hypothesis tests Estimate Standard error p-value H2: WTPHYPO > WTPREAL  ASC* dtreatment 11.7* 2.64 0.00  High CF * dtreatment −124* 5.77 0.00  Medium CF* dtreatment −102* 4.92 0.00  Fortified * dtreatment −50.9* 3.23 0.00 H3: WTPHYPO > WTPROOR  ASC* dtreatment 12.0 4.41 0.01  High CF * dtreatment −106* 8.88 0.00  Medium CF* dtreatment −92.2* 7.20 0.00  Fortified * dtreatment −46.7* 5.37 0.00 H4: WTPROOR = WTPREAL  ASC* dtreatment −2.98 4.55 0.51  High CF * dtreatment −26.8* 5.09 0.00  Medium CF* dtreatment −12.5 9.91 0.21  Fortified * dtreatment −6.97* 3.53 0.05 Asterisk (*) indicates significance at 5 per cent level or lower. The table only shows mean estimates of s for the ω parameters. 6. Conclusions Eliciting the value of goods based on approaches that reflect the real behaviour of decision makers is a central element of valuation studies be it for non-market or market applications. Stated DCE approaches are widely used to elicit the value of goods and their attributes, e.g. in order to predict demand for new products. How to reduce or eliminate hypothetical bias remains an important challenge for DCE researchers. Nonhypothetical DCEs are obviously the preferred option; however, in many cases, product unavailability and the associated high costs of implementation impair researchers from relying on such approaches. As a result, hypothetical DCEs are still used extensively. A range of different strategies, e.g. consequential survey questions, honesty priming, inferred valuation, cognitive dissonance and solemn oath scripts, have been suggested as amendments to hypothetical DCEs, aiming to reduce or eliminate hypothetical bias (Carson and Groves, 2007; Lusk and Norwood, 2009; Jacquemet et al., 2013; Alfnes, Yue and Jensen, 2010; De-Magistris, Gracia and Nayga, 2013). Inspired by previous studies by Ladenburg and Olsen (2014) and Varela et al. (2014), we extend this line of research by investigating the effectiveness of a ROOR in terms of reducing or eliminating hypothetical bias. This study is based on a field experiment concerned with consumer valuation of novel food products made with cricket flour in Kenya. Three treatments – one hypothetical, one hypothetical with ROOR and one nonhypothetical incentivised DCE – are tested in a between-subject design aimed at determining, firstly, the magnitude of hypothetical bias and, secondly, the ability of the ROOR to remove it. The results confirm that the mean marginal WTP values based on the simple hypothetical DCE suffer severely from hypothetical bias with estimates being up to several hundred percentages higher than the WTP estimates obtained in the nonhypothetical treatment. These results are generally in line with Ehmke, Lusk and List (2008) and Chowdhury et al. (2011) which are the only studies where hypothetical bias has been investigated in developing countries. The results further indicate that the ROOR considerably reduces or even eliminates the hypothetical bias while also being capable of reducing or removing negative opt-out effects. The results based on the ROOR are considerably closer to the results based on the nonhypothetical sample than the results based on the hypothetical sample without the ROOR. The ROOR eliminates hypothetical bias for one attribute while strongly reducing, though not completely eliminating it for others, highlighting the promising potential of introducing it in DCEs. The results concerning the presence of hypothetical bias presented in this paper are in contrast with Carlsson and Martinsson (2001), Cameron et al. (2002), Lusk and Schroeder (2004) and Alfnes and Steine (2005) where the hypothesis of equal marginal WTP values in real and hypothetical settings could not be rejected. However, other authors reported results which are in agreement with our results (see Johansson-Stenman and Svedsäter, 2008; Chowdhury et al., 2011; Moser, Raffaelli and Notaro, 2014). It should be noted that direct comparison of our results with these studies may not be appropriate due to reasons associated with the nature of the goods being valued, the sampling design and the nature of the experimental settings (Harrison and List, 2004; Harrison, 2006; Johansson-Stenman and Svedsäter, 2008). Both Carlsson and Martinsson (2001) and Alfnes and Steine (2005) used a within-subject design and they reported that there are no differences in marginal WTP values between hypothetical and real treatments. In this study, we used a between-subject design and find significant differences in the results in keeping with, e.g. Johansson-Stenman and Svedsäter (2008) and Moser, Raffaelli and Notaro (2014). Johansson-Stenman and Svedsäter (2008) found that hypothetical bias is larger in a between-subject design than in a within-subject design, and they concluded that it should be tested using the former type of design. However, List and Gallet (2001) did not find any significant differences in hypothetical bias when comparing between-subject and within-subject designs in a meta-analysis. Considering the nature of the experimental setting, as opposed to our field study, the similar food choice experiments by Lusk and Schroede (2004) and Alfnes and Steine (2005) were conducted in laboratory settings. Since they did not find substantial hypothetical bias, it is tempting to suggest that ex-ante hypothetical mitigation measures are more important in field settings than in lab settings. However, List and Gallet (2001) did not find any significant differences between lab and field settings in this regard. They did however find hypothetical bias to generally be higher for public goods than private goods, which was further confirmed by e.g. Alfnes and Steine (2005) and Johansson-Stenman and Svedsäter (2012). Importantly, this does not mean that hypothetical bias concerns can be disregarded when evaluating private goods. Our results strongly underline this, as we find substantial hypothetical bias in a private good context. This suggests that DCE analysts should consider employing some kind of ex-ante (or ex-post) hypothetical bias mitigation measures regardless of the type of good. Providing economic incentives as we do in our REAL treatment is of course the preferable way to avoid hypothetical bias in a private good context as it clearly enhances valuation scenario consequentiality – in line with the recommendations in Johnston et al. (2017). However, in most DCE cases, e.g. when assessing private goods incorporating new product attributes that are not yet available or when assessing public goods, such economic incentivisation is not a possibility, and the analyst must consider other options. We show that the ROOR is worth considering, as it seems to encourage truthful responses, or at least responses that are quite well aligned with those obtained in the ideal economically incentivised setting. We can of course only conjecture that this result is transferable to public good settings where economic incentivisation is very rarely possible. Our results differ from Ladenburg and Olsen (2014) and Varela et al. (2014) in four important ways. First, their studies considered preferences for public goods whereas the current study considers a private good context in terms of food choice. Second, these studies did not establish the extent to which hypothetical bias existed, which may be seen as a major limitation since it essentially precluded the authors from making any conclusions concerning the effectiveness of the ROOR in terms of mitigating hypothetical bias. Our experimental setup allows us to conclude that hypothetical bias is indeed present and of a non-negligible magnitude. Third, while the ROOR did not influence responses in the case of Varela et al. (2014), it significantly reduced the marginal WTP estimates of only some of the attributes in Ladenburg and Olsen (2014). Contrary to this, the results of the current study reveal that the ROOR significantly reduces the hypothetical bias for all attributes and even completely eliminates it for one attribute. Fourth, they implemented the ROOR together with a CT script, which confounded the effects of the ROOR and the CT. Thus, compared to these, our study provides a more strict experimental setup allowing a much more rigorous test of the ROOR in isolation. It also provides a clearer result in terms of the impact of the ROOR on hypothetical bias. The results reported in this study have particular relevance to stated DCE since framing them with the ROOR does not pose further challenges in terms of monetary cost, product availability and or logistic burden that otherwise might be the case in nonhypothetical DCEs. In sum, the main results of this study show that not only financial incentives but also ROOR can lead to plausible behavioural manifestations. Therefore, in situations where it is not possible to employ nonhypothetical DCEs, framing hypothetical DCEs with a ROOR would seem to be a promising strategy. Our results, however, provide only first-hand evidence and are not conclusive by themselves. Therefore, future studies are important in order to replicate our results and provide further empirical evidence in relation to the ability of the ROOR to improve the external validity of stated DCEs. In this regard, several areas of research can be identified for further research. First, future studies may consider conducting research in different experimental settings and with different types of goods so as to validate whether the effect of ROOR is context-specific or not. This includes investigating whether the ROOR leads to similar results to the ones reported in this study in situations where participants do not experience the attributes. Second, the actual wording of the ROOR is not tested here. Future studies may test other wordings of the ROOR, e.g. by focusing on the designed alternatives rather than on the opt-out alternative. Third, further research could be testing whether the ROOR leads to the same results as the results from the current study when it is applied to experimental designs that include a no-choice alternative rather than a forced choice. Fourth, unlike De-Magistris, Gracia and Nayga (2013), we only considered the ROOR without including other ex-ante hypothetical bias mitigation strategies to avoid a potential confounding effect. It can be argued that examining the effectiveness of the ROOR as compared to other strategies such as CT is necessary to draw a more general conclusion. This may be accomplished using an experimental design that includes a wide range of treatments with different combinations of the various ex-ante strategies for mitigating hypothetical bias. Acknowledgements We thank the management of the Department of Food Science and Technology at the Jomo Kenyatta University of Agriculture and Technology for giving us the opportunity to bake the bun products at the Food Processing Workshop Unit. Finally, we thank seminar participants at the university of Copenhagen and University of Stirling for comments and suggestions. The usual disclaimer applies for any remaining errors. Funding This research is part of the GREEiNSECT (13-06KU) project which was funded by the Danish International Development Agency (DANIDA), Ministry of Foreign Affairs of Denmark. Footnotes 1 1 US Dollar ~ 90.50 KShs at the time of the experiment. 2 As pointed out by a reviewer, consumers’ WTP for the novel insect-based food products could be different in real world situations where they would have to pay directly from their own pockets and not by some endowment offered by an interviewer. If the endowment is perceived as windfall money, the REAL treatment may not be as incentive compatible as desired, and the WTP estimates obtained in this sample may consequently be an overstatement of the ‘true’ WTP. We have not tested this in the present study as it was clear from the early stages of survey planning, that asking respondents to bring money to the interview would be problematic from an ethical point of view since the survey is conducted in relatively poor areas of Kenya. 3 We have not controlled for participants having or not having eaten bread at home before the experiment. However, we tried to control for the hunger level of the participants and the taste of previous consumed foods at breakfast or lunch time by placing the interviews between 10.00 and 12.00 in the morning and 14.00 and 16.00 in the afternoon. 4 This information was provided before consumers were subject to the DCE to mirror real market situations in that consumers often process a range of information before they make purchase decisions. However, we have not tested the effects of information provision on consumers’ responses as it falls beyond the scope of this paper. 5 This model can be extended to accommodate an error component representing potential correlation between designed alternatives as proposed by Scarpa, Ferrini and Willis (2005), Scarpa, Spalatro and Canavari (2007a) and Scarpa, Willis and Acutt (2007b). However, given that the status quo option in the current experiment is defined as a standard alternative which in many aspects is similar to the experimentally designed alternatives, it was decided to use the fairly standard RPL model. 6 The estimates of the Cholesky matrices, which were obtained from the finite difference hessian, are reported in Appendix C. References Aadland , D. and Caplan , A. D. ( 2003 ). Willingness to pay for curbside recycling with detection and mitigation of hypothetical bias . American Journal of Agricultural Economics 85 : 492 – 502 . Google Scholar CrossRef Search ADS Aadland , D. and Caplan , A. D. ( 2006 ). Cheap talk reconsidered: new evidence from CVM . Journal of Economic Behavior and Organization 60 : 562 – 578 . Google Scholar CrossRef Search ADS Adamowicz , W. , Louviere , J. and Williams , M. ( 1994 ). Combining revealed and stated preference methods for valuing environmental amenities . Journal of Environmental Economics and Management 26 : 271 – 292 . Google Scholar CrossRef Search ADS Alemu , M. H. , Olsen , S. B. , Vedel , S. E. , Kinyuru , J. N. and Pambo , K. O. ( 2017 a). Can insects increase food security in developing countries? An analysis of Kenyan consumer preferences and demand for cricket flour buns . Food Security 9 : 471 – 484 . Google Scholar CrossRef Search ADS Alemu , M. H. , Olsen , S. B. , Vedel , S. E. , Pambo , K. O. and Owino , V. O. ( 2017 b). Combining product attributes with recommendation and shopping location attributes to assess consumer preferences for insect-based food products . Food Quality and Preference 55 : 45 – 57 . Google Scholar CrossRef Search ADS Alfnes , F. , Guttormsen , A. G. , Steine , G. and Kolstad , K. ( 2006 ). Consumers’ willingness to pay for the color of salmon: a choice experiment with real economic incentives . American Journal of Agricultural Economics 88 : 1050 – 1061 . Google Scholar CrossRef Search ADS Alfnes , F. and Steine , G. ( 2005 ). None-of-these bias in stated choice experiments. Discussion Paper DP-06/05. Department of Economics and Resource Management, Norwegian University of Life Sciences, Norway. Alfnes , F. , Yue , C. and Jensen , H. H. ( 2010 ). Cognitive dissonance as a means of reducing hypothetical bias . European Review of Agricultural Economics 37 : 147 – 163 . Google Scholar CrossRef Search ADS Alphonce , R. and Alfnes , F. ( 2017 ). Eliciting consumer WTP for food characteristics in a developing context: application of four valuation methods in an African market . Journal of Agricultural Economic 68 : 123 – 142 . Google Scholar CrossRef Search ADS Ami , D. , Aprahamian , F. , Chanel , O. and Luchini , S. ( 2011 ). A test of cheap talk in different hypothetical contexts: the case of air pollution . Environmental and Resource Economics 50 : 111 – 130 . Google Scholar CrossRef Search ADS Azevedo , C. D. , Herriges , J. A. and Kling , C. L. ( 2003 ). Combining revealed and stated preferences: consistency tests and their interpretations . American Journal of Agricultural Economics 85 : 525 – 537 . Google Scholar CrossRef Search ADS Balcombe , K. , Chalak , A. and Fraser , I. ( 2009 ). Model selection for the mixed logit with Bayesian estimation . Journal of Environmental Economics and Management 57 : 226 – 237 . Google Scholar CrossRef Search ADS Barrage , L. and Lee , M. S. ( 2010 ). A penny for your thoughts: inducing truth-telling in stated preference elicitation . Economics Letters 106 : 140 – 142 . Google Scholar CrossRef Search ADS Bazzani , C. , Caputo , V. , Nayga , R. M. and Canavari , M. ( 2017 ). Testing commitment cost theory in choice experiments . Economic Inquiry 55 : 383 – 396 . Google Scholar CrossRef Search ADS Bennett , J. and Birol , E. ( 2010 ). Choice Experiments in Developing Countries: Implementation, Challenges and Policy Implications . Cheltenham, UK; Northampton, USA : Edward Elgar . Google Scholar CrossRef Search ADS Beshears , J. , Choi , J. J. , Laibson , D. and Madrian , B. C. ( 2008 ). How are preferences revealed? Journal of Public Economics 92 : 1787 – 1794 . Google Scholar CrossRef Search ADS Bierlaire , M. ( 2003 ). BIOGEME: A Free Package for the Estimation of Discrete Choice Models. Paper presented at the 3rd Swiss Transport Research Conference Monte Verita, Ascona, March 2003. Birol , E. , Meenakshi , J. V. , Oparinde , A. , Perez , S. and Tomlins , K. ( 2015 ). Developing country consumers’ acceptance of biofortified foods: a synthesis . Food Security 7 : 555 – 568 . Google Scholar CrossRef Search ADS Blumenschein , K. , Blomquist , G. C. , Johannesson , M. , Horn , N. and Freeman , P. ( 2008 ). Eliciting willingness to pay without bias: evidence from a field experiment . The Economic Journal 118 : 114 – 137 . Google Scholar CrossRef Search ADS Boland , M. J. , Rae , A. N. , Vereijken , J. M. , Meuwissen , M. P. M. , Fischer , A. R. H. , van Boekel , M. A. J. S. , Rutherfurd , S. M. , Gruppen , H. , Moughan , P. J. and Hendriks , W. H. ( 2013 ). The future supply of animal-derived protein for human consumption . Trends in Food Science and Technology 29 : 62 – 73 . Google Scholar CrossRef Search ADS Bosworth , R. and Taylor , L. O. ( 2012 ). Hypothetical bias in choice experiments: is cheap talk effective at eliminating bias on the intensive and extensive margins of choice? The B.E. Journal of Economic Analysis and Policy 12 : 1 – 28 . Google Scholar CrossRef Search ADS Boxall , P. , Adamowicz , W. L. and Moon , A. ( 2009 ). Complexity in choice experiments: choice of the status quo alternative and implications for welfare measurement* . Australian Journal of Agricultural and Resource Economics 53 : 503 – 519 . Google Scholar CrossRef Search ADS Braun , J. V. ( 2007 ). Focus on the world’s poorest and hungry people: IFPRI 2006–2007 Annual Report Essay: International Food Policy Research Institute (IFPRI). Washington, DC: IFPRI. Brooks , K. and Lusk , J. L. ( 2010 ). Stated and revealed preferences for organic and cloned milk: combining choice experiment and scanner data . American Journal of Agricultural Economics 92 ( 4 ): 1229 – 1241 . Google Scholar CrossRef Search ADS Brown , T. C. , Ajzen , I. and Hrubes , D. ( 2003 ). Further tests of entreaties to avoid hypothetical bias in referendum contingent valuation . Journal of Environmental Economics and Management 46 : 353 – 361 . Google Scholar CrossRef Search ADS Brummett , R. G. , Nayga , R. and Wu , X. ( 2007 ). On the use of cheap talk in new product valuation . Economics bulletin 2 : 1 – 9 . Bukkens , S. G. F. ( 1997 ). The nutritional value of edible insects . Ecology of Food and Nutrition 36 : 287 – 319 . Google Scholar CrossRef Search ADS Burton , M. and Rigby , D. ( 2008 ). Hurdle and latent class approaches to serial non-participation in choice models . Environmental and Resource Economics 42 : 211 – 226 . Google Scholar CrossRef Search ADS Cameron , T. A. , Poe , G. L. , Ethier , R. G. and Schulze , W. D. ( 2002 ). Alternative non-market value-elicitation methods: are the underlying preferences the same? Journal of Environmental Economics and Management 44 : 391 – 425 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2005 ). Using cheap talk as a test of validity in choice experiments . Economics Letters 89 : 147 – 152 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2007 a). Consumer benefits of labels and bans on GM foods—choice experiments with Swedish consumers . American Journal of Agricultural Economics 89 : 152 – 161 . Google Scholar CrossRef Search ADS Carlsson , F. , Frykblom , P. and Lagerkvist , C. J. ( 2007 b). Consumer willingness to pay for farm animal welfare: mobile abattoirs versus transportation to slaughter . European Review of Agricultural Economics 34 : 321 – 344 . Google Scholar CrossRef Search ADS Carlsson , F. , García , J. and Löfgren , Å. ( 2010 ). Conformity and the demand for environmental goods . Environmental and Resource Economics 47 : 407 – 421 . Google Scholar CrossRef Search ADS Carlsson , F. and Martinsson , P. ( 2001 ). Do hypothetical and actual marginal willingness to pay differ in choice experiments? Application to the valuation of the environment . Journal of Environmental Economics and Management 41 : 179 – 192 . Google Scholar CrossRef Search ADS Carson , R. T. , Flores , N. E. , Martin , K. M. and Wright , J. L. ( 1996 ). Contingent valuation and revealed preference methodologies: comparing the estimates for quasi-public goods . Land Economics 72 : 80 – 99 . Google Scholar CrossRef Search ADS Carson , R. and Groves , T. ( 2007 ). Incentive and informational properties of preference questions . Environmental and Resource Economics 37 : 181 – 210 . Google Scholar CrossRef Search ADS Champ , P. A. , Moore , R. and Bishop , R. C. ( 2009 ). A comparison of approaches to mitigate hypothetical bias . Agricultural and Resource Economics Review 38 : 166 – 180 . Google Scholar CrossRef Search ADS Chang , J. B. , Lusk , J. L. and Norwood , F. B. ( 2009 ). How closely do hypothetical surveys and laboratory experiments predict field behavior? American Journal of Agricultural Economics 91 : 518 – 534 . Google Scholar CrossRef Search ADS Charness , G. , Gneezy , U. and Kuhn , M. A. ( 2012 ). Experimental methods: between-subject and within-subject design . Journal of Economic Behavior and Organization 81 : 1 – 8 . Google Scholar CrossRef Search ADS ChoiceMetrics ( 2012 ). Ngene 1.1.1 User Manual and Reference Guide, Australia. Chowdhury , S. , Meenakshi , J. V. , Tomlins , K. I. and Owori , C. ( 2011 ). Are consumers in developing countries willing to pay more for micronutrient-dense biofortified foods? Evidence from a field experiment in Uganda . American Journal of Agricultural Economics 93 : 83 – 97 . Google Scholar CrossRef Search ADS Christensen , D. L. , Orech , F. O. , Mungai , M. N. , Larsen , T. , Friis , H. and Aagaard-Hansen , J. ( 2006 ). Entomophagy among the Luo of Kenya: a potential mineral source? International Journal of Food Sciences and Nutrition 57 : 198 – 203 . Google Scholar CrossRef Search ADS Collins , J. P. and Vossler , C. A. ( 2009 ). Incentive compatibility tests of choice experiment value elicitation questions . Journal of Environmental Economics and Management 58 ( 2 ): 226 – 235 . Google Scholar CrossRef Search ADS Cova , B. and Pace , S. ( 2006 ). Brand community of convenience products: new forms of customer empowerment – the case ‘my Nutella The Community’ . European Journal of Marketing 40 : 1087 – 1105 . Google Scholar CrossRef Search ADS Cummings , R. G. and Taylor , L. O. ( 1999 ). Unbiased value estimates for environmental goods: a cheap talk design for the contingent valuation method . American Economic Review 89 : 649 – 665 . Google Scholar CrossRef Search ADS Cárdenas , J. C. , De Roux , N. , Jaramillo , C. R. and Martinez , L. R. ( 2014 ). Is it my money or not? An experiment on risk aversion and the house-money effect . Experimental Economics 17 : 47 – 60 . Google Scholar CrossRef Search ADS De-Magistris , T. , Gracia , A. and Nayga , R. M. ( 2013 ). On the use of honesty priming tasks to mitigate hypothetical bias in choice experiments . American Journal of Agricultural Economics 95 : 1136 – 1154 . Google Scholar CrossRef Search ADS Ding , M. , Grewal , R. and Liechty , J. ( 2005 ). Incentive-aligned conjoint analysis . Journal of Marketing Research 42 : 67 – 82 . Google Scholar CrossRef Search ADS Durand-Morat , A. , Wailes , E. J. and Nayga , J. R. M. ( 2016 ). Challenges of conducting contingent valuation studies in developing countries . American Journal of Agricultural Economics 98 : 597 – 609 . Google Scholar CrossRef Search ADS Ehmke , M. D. , Lusk , J. L. and List , J. A. ( 2008 ). Is hypothetical bias a universal phenomenon? A multinational investigation . Land Economics 84 : 489 – 500 . Google Scholar CrossRef Search ADS Evans , J. , Alemu , M. H. , Flore , R. , Frøst , M. B. , Halloran , A. , Jensen , A. B. , Maciel-Vergara , G. , Meyer-Rochow , V. B. , Münke-Svendsen , C. , Olsen , S. B. , Payne , C. , Roos , N. , Rozin , P. , Tan , H. S. G. , van Huis , A. , Vantomme , P. and Eilenberg , J. ( 2015 ). ‘Entomophagy’: an evolving terminology in need of review . Journal of Insects as Food and Feed 1 : 293 – 305 . Google Scholar CrossRef Search ADS FAO ( 2009 ). The State of Food and Agriculture: Livestock in the Balance . Rome : Sales and Marketing group, Communication Division . FAO ( 2013 ). Edible Insects: Future Prospects for Food and Feed Security . Rome . Ferrini , S. and Scarpa , R. ( 2007 ). Designs with a priori information for nonmarket valuation with choice experiments: a Monte Carlo study . Journal of Environmental Economics and Management 53 : 342 – 363 . Google Scholar CrossRef Search ADS Finke , M. D. ( 2013 ). Complete nutrient content of four species of feeder insects . Zoo Biology 32 : 27 – 36 . Google Scholar CrossRef Search ADS Fox , J. A. , Hayes , D. J. , Shogren , J. F. and Kliebenstein , J. B. ( 1996 ). Experimental methods in consumer preference studies . Journal of Food Distribution Research 27 : 1 – 7 . Gibson , J. M. , Rigby , D. , Polya , D. A. and Russell , N. ( 2016 ). Discrete choice experiments in developing countries: willingness to pay versus willingness to work . Environmental and Resource Economics 65 : 697 – 721 . Google Scholar CrossRef Search ADS Gracia , A. ( 2014 ). Consumers’ preferences for a local food product: a real choice experiment . Empirical Economics 47 : 111 – 128 . Google Scholar CrossRef Search ADS Gracia , A. , Loureiro , M. L. and Nayga Jr , R. M. ( 2011 ). Are valuations from nonhypothetical choice experiments different from those of experimental auctions? American Journal of Agricultural Economics 93 : 1358 – 1373 . Google Scholar CrossRef Search ADS Grebitus , C. , Lusk , J. L. and Nayga Jr , R. M. ( 2013 ). Explaining differences in real and hypothetical experimental auctions and choice experiments with personality . Journal of Economic Psychology 36 : 11 – 26 . Google Scholar CrossRef Search ADS Halloran , A. , Hanboonsong , Y. , Roos , N. and Bruun , S. ( 2017 ). Life cycle assessment of cricket farming in north-eastern Thailand . Journal of Cleaner Production 156 : 83 – 94 . Google Scholar CrossRef Search ADS Harrison , G. ( 2006 ). Experimental evidence on alternative environmental valuation methods . Environmental and Resource Economics 34 : 125 – 162 . Google Scholar CrossRef Search ADS Harrison , G. W. and List , J. A. ( 2004 ). Field experiments . Journal of Economic Literature 42 : 1009 – 1055 . Google Scholar CrossRef Search ADS Hensher , D. A. and Greene , W. H. ( 2011 ). Valuation of travel time savings in WTP and preference space in the presence of taste and scale heterogeneity . Journal of Transport Economics and Policy 45 : 505 – 525 . Hess , S. , Train , K. E. and Polak , J. W. ( 2006 ). On the use of a Modified Latin Hypercube Sampling (MLHS) method in the estimation of a Mixed Logit Model for vehicle choice . Transportation Research Part B: Methodological 40 : 147 – 163 . Google Scholar CrossRef Search ADS Hole , A. and Kolstad , J. ( 2012 ). Mixed logit estimation of willingness to pay distributions: a comparison of models in preference and WTP space using data from a health-related choice experiment . Empirical Economics 42 : 445 – 469 . Google Scholar CrossRef Search ADS Jacquemet , N. , Joule , R.-V. , Luchini , S. and Shogren , J. F. ( 2013 ). Preference elicitation under oath . Journal of Environmental Economics and Management 65 : 110 – 132 . Google Scholar CrossRef Search ADS Johansson-Stenman , O. and Svedsäter , H. ( 2008 ). Measuring hypothetical bias in choice experiments: the importance of cognitive consistency . The B.E. Journal of Economic Analysis and Policy 8 , (Article 41). Johansson-Stenman , O. and Svedsäter , H. ( 2012 ). Self-image and valuation of moral goods: stated versus actual willingness to pay . Journal of Economic Behavior & Organization 84 : 879 – 891 . Google Scholar CrossRef Search ADS Johnston , R. J. , Boyle , K. J. , Adamowicz , W. , Bennett , J. , Brouwer , R. , Cameron , T. A. , Hanemann , W. M. , Hanley , N. , Ryan , M. , Scarpa , R. , Tourangeau , R. and Vossler , C. A. ( 2017 ). Contemporary guidance for stated preference studies . Journal of the Association of Environmental and Resource Economists 4 : 319 – 405 . Google Scholar CrossRef Search ADS Kallas , Z. , Escobar , C. and Gil , J. M. ( 2013 ). Analysis of consumers’ preferences for a special-occasion red wine: a dual response choice experiment approach . Food Quality and Preference 30 : 156 – 168 . Google Scholar CrossRef Search ADS Kinyuru , J. , Kenji , G. and Njoroge , M. ( 2009 ). Process development, nutrition and sensory qualities of wheat buns enriched with edible termites (Macrotermes subhylanus) from Lake Victoria region, Kenya . African Journal of Food, Agriculture, Nutrition and Development 9 : 1739 – 1750 . Google Scholar CrossRef Search ADS Krinsky , I. and Robb , A. L. ( 1986 ). On approximating the statistical properties of elasticities . The Review of Economics and Statistics 68 : 715 – 719 . Google Scholar CrossRef Search ADS Ladenburg , J. and Olsen , S. B. ( 2014 ). Augmenting short Cheap Talk scripts with a repeated Opt-Out Reminder in Choice Experiment surveys . Resource and Energy Economics 37 : 39 – 63 . Google Scholar CrossRef Search ADS Lancaster , K. J. ( 1966 ). A new approach to consumer theory . Journal of Political Economy 74 : 132 – 157 . Google Scholar CrossRef Search ADS Lawrence , C. , Zhou , J. L. and Tits , A. L. ( 1997 ). User’s Guide for CFSQP Version 2.5: A C Code for Solving (Large Scale) Constrained Nonlinear (Minimax) Optimization Problems, Generating Iterates Satisfying All Inequality Constraints. Technical Report TR-94-16r1, Institute for Systems Research, University of Maryland, College Park, MD 20742, 1997. List , J. A. ( 2001 ). Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards . American Economic Review 91 : 1498 – 1507 . Google Scholar CrossRef Search ADS List , J. and Gallet , C. ( 2001 ). What experimental protocol influence disparities between actual and hypothetical stated values? Environmental and Resource Economics 20 : 241 – 254 . Google Scholar CrossRef Search ADS List , J. A. and Shogren , J. F. ( 1998 ). Calibration of the difference between actual and hypothetical valuations in a field experiment . Journal of Economic Behavior & Organization 37 : 193 – 205 . Google Scholar CrossRef Search ADS List , J. A. , Sinha , P. and Taylor , M. H. ( 2006 ). Using choice experiments to value non-market goods and services: evidence from field experiments . Advances in Economic Analysis and Policy 6 : 1 – 37 (Article 2). Little , J. and Berrens , R. ( 2004 ). Explaining disparities between actual and hypothetical stated values: further investigation using meta-analysis . Economics Bulletin 3 : 1 – 13 . Loomis , J. ( 2011 ). What’s to know about hypothetical bias in stated preference valuation studies? Journal of Economic Surveys 25 : 363 – 370 . Google Scholar CrossRef Search ADS Loomis , J. B. ( 2014 ). Strategies for overcoming hypothetical bias in stated preference surveys . Journal of Agricultural and Resource Economics 39 : 34 – 46 . Loomis , J. , Bell , P. , Cooney , H. and Asmus , C. ( 2009 ). A comparison of actual and hypothetical willingness to pay of parents and non-parents for protecting infant health: the case of nitrates in drinking water . Journal of Agricultural and Applied Economics 41 : 697 – 712 . Google Scholar CrossRef Search ADS Looy , H. , Dunkel , F. and Wood , J. ( 2014 ). How then shall we eat? Insect-eating attitudes and sustainable foodways . Agriculture and Human Values 31 : 131 – 141 . Google Scholar CrossRef Search ADS Louviere , J. J. , Hensher , D. A. and Swait , J. D. ( 2000 ). Stated Choice Methods: Analysis and Applications . Cambridge, UK : Cambridge University Press . Google Scholar CrossRef Search ADS Lusk , J. L. ( 2003 ). Effects of cheap talk on consumer willingness-to-pay for golden rice . American Journal of Agricultural Economics 85 : 840 – 856 . Google Scholar CrossRef Search ADS Lusk , J. L. , Fields , D. and Prevatt , W. ( 2008 ). An incentive compatible conjoint ranking mechanism . American Journal of Agricultural Economics 90 : 487 – 498 . Google Scholar CrossRef Search ADS Lusk , J. L. and Norwood , F. B. ( 2009 ). An inferred valuation method . Land Economics 85 : 500 – 514 . Google Scholar CrossRef Search ADS Lusk , J. L. and Schroeder , T. C. ( 2004 ). Are choice experiments incentive compatible? A test with quality differentiated beef steaks . American Journal of Agricultural Economics 86 : 467 – 482 . Google Scholar CrossRef Search ADS McFadden , D. ( 1986 ). The choice theory approach to market research . Marketing Science 5 : 275 – 297 . Google Scholar CrossRef Search ADS Michaud , C. , Llerena , D. and Joly , I. ( 2013 ). Willingness to pay for environmental attributes of non-food agricultural products: a real choice experiment . European Review of Agricultural Economics 40 : 313 – 329 . Google Scholar CrossRef Search ADS Moser , R. , Raffaelli , R. and Notaro , S. ( 2014 ). Testing hypothetical bias with a real choice experiment using respondents’ own money . European Review of Agricultural Economics 41 : 25 – 46 . Google Scholar CrossRef Search ADS Murphy , J. , Allen , P. G. , Stevens , T. and Weatherhead , D. ( 2005 a). A meta-analysis of hypothetical bias in stated preference valuation . Environmental and Resource Economics 30 : 313 – 325 . Google Scholar CrossRef Search ADS Murphy , J. , Stevens , T. and Weatherhead , D. ( 2005 b). Is cheap talk effective at eliminating hypothetical bias in a provision point mechanism? Environmental and Resource Economics 30 : 327 – 343 . Google Scholar CrossRef Search ADS Mørkbak , M. R. , Olsen , S. B. and Campbell , D. ( 2014 ). Behavioral implications of providing real incentives in stated choice experiments . Journal of Economic Psychology 45 : 102 – 116 . Google Scholar CrossRef Search ADS Nayga , R. M. , Woodward , R. and Aiew , W. ( 2006 ). Willingness to pay for reduced risk of foodborne illness: a nonhypothetical field experiment . Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie 54 : 461 – 475 . Google Scholar CrossRef Search ADS Oonincx , D. G. A. B. , Itterbeeck , J. v. , Heetkamp , M. J. W. , van den Brand , H. , van Loon , J. J. A. and van Huis , A. ( 2010 ). An exploration on greenhouse gas and ammonia production by insect species suitable for animal or human consumption . PLoS One 5 : 1 – 7 . Google Scholar CrossRef Search ADS Özdemir , S. , Johnson , F. R. and Hauber , A. B. ( 2009 ). Hypothetical bias, cheap talk, and stated willingness to pay for health care . Journal of Health Economics 28 : 894 – 901 . Google Scholar CrossRef Search ADS Pellett , P. L. and Ghosh , S. ( 2004 ). Lysine fortification: past, present, and future . Food Nutr Bull 25 : 107 – 113 . Google Scholar CrossRef Search ADS Poe , G. L. , Giraud , K. L. and Loomis , J. B. ( 2005 ). Computational methods for measuring the difference of empirical distributions . American Journal of Agricultural Economics 87 : 353 – 365 . Google Scholar CrossRef Search ADS Ready , R. C. , Champ , P. A. and Lawton , J. L. ( 2010 ). Using respondent uncertainty to mitigate hypothetical bias in a stated choice experiment . Land Economics 86 : 363 – 381 . Google Scholar CrossRef Search ADS Rigby , D. , Alcon , F. and Burton , M. ( 2010 ). Supply uncertainty and the economic value of irrigation water . European Review of Agricultural Economics 37 : 97 – 117 . Google Scholar CrossRef Search ADS Ruby , M. B. , Rozin , P. and Chan , C. ( 2015 ). Determinants of willingness to eat insects in the USA and India . Journal of Insects as Food and Feed 1 : 215 – 225 . Google Scholar CrossRef Search ADS Scarpa , R. , Ferrini , S. and Willis , K. ( 2005 ). Performance of error component models for status-quo effects in choice experiments. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Netherlands : Springer , 247 – 273 . Google Scholar CrossRef Search ADS Scarpa , R. and Rose , J. M. ( 2008 ). Design efficiency for non-market valuation with choice modelling: how to measure it, what to report and why . Australian Journal of Agricultural and Resource Economics 52 : 253 – 282 . Google Scholar CrossRef Search ADS Scarpa , R. , Spalatro , F. and Canavari , M. ( 2007 a). Investigating preferences for environment friendly production practices. In: Canavari M. and Olson K. (eds) , Organic Food: Consumers’ Choice and Farmers’ Opportunities . New York : Springer , 115 – 124 . Scarpa , R. , Thiene , M. and Train , K. ( 2008 ). Utility in willingness to pay space: a tool to address confounding random scale effects in destination choice to the Alps . American Journal of Agricultural Economics 90 : 994 – 1010 . Google Scholar CrossRef Search ADS Scarpa , R. , Willis , K. G. and Acutt , M. ( 2007 b). Valuing externalities from water supply: Status quo, choice complexity and individual random effects in panel kernel logit analysis of choice experiments . Journal of Environmental Planning and Management 50 : 449 – 466 . Google Scholar CrossRef Search ADS Sonnier , G. , Ainslie , A. and Otter , T. ( 2007 ). Heterogeneity distributions of willingness-to-pay in choice models . Quantitative Marketing and Economics 5 : 313 – 331 . Google Scholar CrossRef Search ADS Steinfeld , H. , Gerber , P. , Wassenaar , T. , Castel , V. , Rosales , M. and De Haan , C. ( 2006 ). Livestock’s Long Shadow . Rome : FAO . Stevens , R. and Winter-Nelson. , A. ( 2008 ). Consumer acceptance of provitamin A-biofortified maize in Maputo, Mozambique . Food Policy 33 : 341 – 351 . Google Scholar CrossRef Search ADS Swait , J. and Louviere , J. ( 1993 ). The role of the scale parameter in the estimation and comparison of multinomial logit models . Journal of Marketing Research 30 : 305 – 314 . Google Scholar CrossRef Search ADS Thaler , R. H. and Johnson , E. J. ( 1990 ). Gambling with the house money and trying to break even: the effects of prior outcomes on risky choice . Management Science 36 : 643 – 660 . Google Scholar CrossRef Search ADS Thiene , M. and Scarpa , R. ( 2009 ). Deriving and testing efficient estimates of WTP distributions in destination choice models . Environmental and Resource Economics 44 : 379 – 395 . Google Scholar CrossRef Search ADS Tonsor , G. T. and Shupp , R. S. ( 2011 ). Cheap talk scripts and online choice experiments: looking beyond the mean . American Journal of Agricultural Economics 93 : 1015 – 1031 . Google Scholar CrossRef Search ADS Train , K. E. ( 2003 ). Discrete choice methods with simulation . New York : Cambridge university press . Google Scholar CrossRef Search ADS Train , K. and Sonnier , G. ( 2005 ). Mixed logit with bounded distributions of correlated partworths. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Dordrecht, The Netherlands : Springer , 117 – 134 . Google Scholar CrossRef Search ADS Train , K. and Weeks , M. ( 2005 ). Discrete choice models in preference space and willingness-to-pay space. In: Scarpa R. and Alberini A. (eds) , Applications of Simulation Methods in Environmental and Resource Economics . Dordrecht, The Netherlands : Dordrecht, Springer , 1 – 16 . UN ( 2015 ). World population prospects: The 2015 revision, key findings and advance tables . Working Paper No. ESA/P/WP.241. New York : United Nations . van Huis , A. ( 2013 ). Potential of insects as food and feed in assuring food security . Annual Review of Entomology 58 : 563 – 583 . Google Scholar CrossRef Search ADS Varela , E. , Mahieu , P.-A. , Giergiczny , M. , Riera , P. and Soliño , M. ( 2014 ). Testing the single opt-out reminder in choice experiments: an application to fuel break management in Spain . Journal of Forest Economics 20 : 212 – 222 . Google Scholar CrossRef Search ADS Volinskiy , D. , Adamowicz , W. L. , Veeman , M. and Srivastava , L. ( 2009 ). Does choice context affect the results from incentive-compatible experiments? The case of Non-GM and country-of-origin premia in canola oil . Canadian Journal of Agricultural Economics/Revue canadienne d’agroeconomie 57 : 205 – 221 . Google Scholar CrossRef Search ADS von Haefen , R. H. , Massey , D. M. and Adamowicz , W. L. ( 2005 ). Serial nonparticipation in repeated discrete choice models . American Journal of Agricultural Economics 87 : 1061 – 1076 . Google Scholar CrossRef Search ADS Vossler , C. A. , Doyon , M. and Rondeau , D. ( 2012 ). Truth in consequentiality: theory and field evidence on discrete choice experiments . American Economic Journal: Microeconomics 4 : 145 – 171 . Google Scholar CrossRef Search ADS Vossler , C. A. and Watson , S. B. ( 2013 ). Understanding the consequences of consequentiality: testing the validity of stated preferences in the field . Journal of Economic Behavior and Organization 86 : 137 – 147 . Google Scholar CrossRef Search ADS Whitehead , J. C. and Cherry , T. L. ( 2007 ). Willingness to pay for a Green Energy program: a comparison of ex-ante and ex-post hypothetical bias mitigation approaches . Resource and Energy Economics 29 : 247 – 261 . Google Scholar CrossRef Search ADS Whittington , D. ( 1998 ). Administering contingent valuation surveys in developing countries . World Development 26 : 21 – 30 . Google Scholar CrossRef Search ADS Whittington , D. ( 2002 ). Improving the performance of contingent valuation studies in developing countries . Environmental and Resource Economics 22 : 323 – 367 . Google Scholar CrossRef Search ADS Yue , C. and Tong , C. ( 2009 ). Organic or local? Investigating consumer preference for fresh produce using a choice experiment with real economic incentives . HortScience 44 : 366 – 371 . Appendix A The buns were baked at the Food Processing Workshop Unit at the Jomo Kenyatta University of Agriculture and Technology in Kenya by professional technicians. Since Kinyuru, Kenji and Njoroge (2009) developed their recipe for buns from termite flour, we chose to validate this by first producing trial buns from CF. At this point, the product characteristics were assessed in cooperation with a nutrition scientist and several technicians. Following this, 10 participants that included five males and five females were recruited to evaluate the products for taste, texture, colour and smell. The recipe formulation that was used for the main baking is presented in Table A1. And the final bun products which were used for the field experiment are characterised as shown in Table A2. One can see that as the amount of the CF increases, the bun products become heavy, soft and brown. The increase in weight of the buns can be linked to the fact that the increase in the amount of CF lead to an increase in the amount of fat in the buns that can reduce water transpiration. Table A1. Ingredient composition of the buns Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Source: Alemu et al. (2017a). Note: ‘g’ refers to grams. Baking fat (7.5 g), salt (1.25 g), sugar (5 g), yeast (2.5 g) and acetic acid (0.125 milliliter) were added to each bun. Table A1. Ingredient composition of the buns Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Buns Amount of wheat flour (g) Amount of cricket flour (g) Amount of fortified wheat flour (g) Standard bun 125 0 0 Fortified standard bun 75 0 50 Medium CF bun 118.75 6.25 0 Fortified medium CF bun 68.75 6.25 50 High CF bun 112.5 12.5 0 Fortified high CF bun 62.5 12.5 50 Source: Alemu et al. (2017a). Note: ‘g’ refers to grams. Baking fat (7.5 g), salt (1.25 g), sugar (5 g), yeast (2.5 g) and acetic acid (0.125 milliliter) were added to each bun. Table A2. Texture and visual appearance of the buns Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Source: Alemu et al. (2017a). Table A2. Texture and visual appearance of the buns Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Buns Final weight (g) Texture Appearance Standard bun 158 Firm White Fortified standard bun 158 Firm White Medium CF bun 159 Medium firm Medium brown Fortified medium CF bun 159 Medium firm Medium brown High CF bun 160 Soft Brown Fortified high CF bun 160 Soft Brown Source: Alemu et al. (2017a). Appendix B Subject instruction for the REAL treatment ‘You will be provided with 12 different choice scenarios within which three bags (Bags 1, 2 and 3) of buns are included. Bags 1 and 2 may contain buns made from wheat flour mixed with cricket flour. Some portion of the wheat flour can be fortified. Bag 3 contains only buns made from the wheat flour which was not fortified. In each scenario, you should choose ONE of the bags you would like to purchase (Bag 1 or 2) or you can choose Bag 3 if you would not like to purchase Bag 1 or 2. After you complete all 12 shopping scenarios, we will ask you to draw a number (1–12) from an envelope to determine which shopping scenario will be binding. In the envelope, are numbers 1–12. If the number 1 is drawn, then the first shopping scenario will be binding, and so on. For the binding scenario, we will look at the product you have chosen, give you your chosen product, and you will pay the listed price in that scenario. You should use the 90 KShs for the purchase. The most expensive alternatives cost 90 KShs. Although only one of the 12 shopping scenarios will be binding there is an equal chance of any shopping scenario being selected as binding, so think about each answer carefully.’ Subject instruction for the HYPO and ROOR treatments ‘You will be provided with 12 different choice scenarios within which three bags (Bags 1, 2 and 3) of buns are included. Bags 1 and 2 may contain buns made from wheat flour mixed with cricket flour. Some portion of the wheat flour can be fortified. Bag 3 contains only buns made from the wheat flour which was not fortified. In each scenario, you should choose ONE of the bags you would like to purchase (Bag 1 or 2) or you can choose Bag 3 if you would not like to purchase Bag 1 or 2. For each choice scenario, assume that you have the opportunity to, here and now, to purchase ONE and ONLY ONE of the bags at the listed prices. While you will not actually buy any products today or pay the posted prices, please respond to each choice scenario as if it were a real one and you would have to give up real money were one of the 12 scenarios to be selected as binding.’ Appendix C Table C1. HYPO Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Note: Z-values in parentheses. Table C1. HYPO Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 37,900 (99.5) High CF 28,791 (125) 35,700 (14.5) Medium CF 10,125 (56.1) 18,114 (10.6) 13,700 (8.8) Fortified 6,364 (36.2) 8,256 (10.4) 5,347 (7.64) 2,700 (6.9) ln(λ) −41.5 (−3.51) −139 (−4.9) −31 (−1.16) −12 (−0.899) 2.6 (3.8) Note: Z-values in parentheses. Table C2. ROOR Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Note: Z-values in parentheses. Table C2. ROOR Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 156 (2.5) High CF −83 (−2.8) 7460 (5.1) Medium CF −7 (−0.19) 6068 (5.9) 5800 (6.0) Fortified 187 (2.9) 2788 (5.8) 2923 (6.3) 1760 (5.9) ln(λ) 0.8 (0.3) −26 (−1.4) −12 (−0.73) −6 (−0.69) 1.46 (2.94) Note: Z-values in parentheses. Table C3. REAL Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Note: Z-values in parentheses. Table C3. REAL Cholesky matrix from WTP space estimates Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Parameters ASC High CF Medium CF Fortified ln(λ) ASC 3,770 (94.5) High CF 1,962 (71.0) 4,850 (7.3) Medium CF 688 (20.2) 3,121 (5.9) 26,80 (5.6) Fortified 575 (25.3) 1,059 (6.9) 677 (4.9) 261 (4.5) ln(λ) 9 (5.6) −31 (−3.0) −19 (−2.0) −5 (−1.64) 1.14 (2.81) Note: Z-values in parentheses. Table C4. Estimates of the ECLC model by treatment Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Note: Z-values in parentheses. Table C4. Estimates of the ECLC model by treatment Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Parameters HYPO ROOR REAL Estimate Estimate Estimate High CF 4.47 (14.42) 3.88 (14.79) 3.88 (14.69) Medium CF 3.84 (14.39) 2.99 (12.15) 3.08 (13.05) Fortified 2.69 (12.04) 2.19 (11.54) 2.34 (12.78) Price −0.074 (−12.14) −0.082 (−13.32) −0.079 (−16.06) Number of parameters 19 19 19 Final log-likelihood −876.2 −897.0 −1,078.2 Rho-square 0.545 0.569 0.474 AIC 1,790.4 1,832.0 2,194.4 BIC 1,841.6 1,883.1 2,246.8 AIC3 1,809.4 1,851.0 2,213.4 CAIC 1,860.6 1,902.1 2,265.8 Note: Z-values in parentheses. Appendix D Table D2. Estimates of the ω parameters based on pooled data (two treatments at a time) Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors. Table D2. Estimates of the ω parameters based on pooled data (two treatments at a time) Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Parameter HYPO and REAL HYPO and ROOR ROOR and REAL Mean parameter estimates  ASC* dtreatment 11.7 (2.64)* 12.0 (4.41)* −2.98 (4.55)  High CF * dtreatment −124 (5.01)* −106 (8.88)* −26.8 (5.09)*  Medium CF* dtreatment −102 (4.92)* −92.2 (7.20)* −12.5 (9.91)  Fortified * dtreatment −50.9 (3.23)* −46.7 (5.37)* −6.97 (3.53)* Standard deviation estimates  ASC* dtreatment 12.5 (1.52)* 12.1 (1.77)* 3.34 (1.50)*  High CF * dtreatment 21.8 (1.62)* 32.7 (3.06)* 19.8 (3.95)*  Medium CF* dtreatment 11.6 (1.73)* 31.1 (2.82)* 5.06 (2.17)*  Fortified * dtreatment 2.59 (1.15)* 10.7 (1.75)* 8.3 (2.46)*  No. of observations 2,700 2,616 2,700  Null log-likelihood −2,966.3 −2,873.9 −2,966.3  Final Log-likelihood −1841.0 −1,682.0 −2,022.1  Adjusted ρ2 0.364 0.399 0.312 Note: Asterisk (*) indicates significance at 5 per cent level or lower. Figures in parentheses are standard errors. Author notes Review coordinated by Iain Fraser © Oxford University Press and Foundation for the European Review of Agricultural Economics 2018; all rights reserved. For permissions, please e-mail: journals.permissions@oup.com This article is published and distributed under the terms of the Oxford University Press, Standard Journals Publication Model (https://academic.oup.com/journals/pages/about_us/legal/notices)

Journal

European Review of Agricultural EconomicsOxford University Press

Published: Apr 17, 2018

There are no references for this article.

You’re reading a free preview. Subscribe to read the entire article.


DeepDyve is your
personal research library

It’s your single place to instantly
discover and read the research
that matters to you.

Enjoy affordable access to
over 18 million articles from more than
15,000 peer-reviewed journals.

All for just $49/month

Explore the DeepDyve Library

Search

Query the DeepDyve database, plus search all of PubMed and Google Scholar seamlessly

Organize

Save any article or search result from DeepDyve, PubMed, and Google Scholar... all in one place.

Access

Get unlimited, online access to over 18 million full-text articles from more than 15,000 scientific journals.

Your journals are on DeepDyve

Read from thousands of the leading scholarly journals from SpringerNature, Elsevier, Wiley-Blackwell, Oxford University Press and more.

All the latest content is available, no embargo periods.

See the journals in your area

DeepDyve

Freelancer

DeepDyve

Pro

Price

FREE

$49/month
$360/year

Save searches from
Google Scholar,
PubMed

Create lists to
organize your research

Export lists, citations

Read DeepDyve articles

Abstract access only

Unlimited access to over
18 million full-text articles

Print

20 pages / month

PDF Discount

20% off