Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

Blumenschein,, Karen; Blomquist, Glenn, C.; Johannesson,, Magnus; Horn,, Nancy; Freeman,, Patricia

doi:10.1111/j.1468-0297.2007.02106.x

Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

Blumenschein,, Karen;Blomquist, Glenn, C.;Johannesson,, Magnus;Horn,, Nancy;Freeman,, Patricia 2008-01-01 00:00:00 Abstract Concern exists that hypothetical willingness to pay questions overestimate real willingness to pay. In a field experiment, we compare two methods of removing hypothetical bias, a cheap talk approach and a certainty approach, with real purchases. We find evidence of hypothetical bias for unadulterated contingent valuation. Contingent valuation with certainty statements removes the hypothetical bias, but the cheap talk approach has no significant impact. Our findings suggest that willingness to pay can be accurately estimated by adding a simple follow‐up question about the certainty of responses and that cheap talk is not a generally effective approach. Information about the willingness to pay for non‐market goods is crucial for understanding the welfare implications of different policies. Ideally we would like to rely on actual choices for inferring willingness to pay. But for many public programmes, especially those in the environmental and health area, revealed preference information is limited. This limitation can arise from the lack of markets or third‐party financing. Researchers have therefore tried to estimate willingness to pay based on stated preferences that are hypothetical choices. This methodology, known as contingent valuation, typically involves a survey or experiment in which there is a detailed description of the good being offered, a description of how the good would be provided, a method for eliciting preferences for the good, follow up debriefing questions and questions about socioeconomic characteristics (Portney, 1994; Carson, 2001). Contingent valuation is still somewhat controversial in economics (Diamond and Hausman, 1994; Carson et al., 2001; Ariely et al., 2003). At the heart of this controversy is the extent to which hypothetical choices correspond to real economic choices.1Carson et al. (1996) find that values from contingent valuation and revealed preference studies match fairly well, yet strong counterevidence exists. For example, laboratory experiments by Cummings et al. (1995, 1997) comparing real and hypothetical willingness to pay suggest that hypothetical responses sometimes substantially overestimate willingness to pay. Meta‐analysis by List and Gallet (2001) and reviews by Harrison (2006) and Harrison and Rutström (forthcoming) also suggest overestimation. This overstatement problem is known as hypothetical bias. The evidence on hypothetical bias has stimulated research into various techniques for removing this bias. Two basic approaches have evolved. Cummings and Taylor (1999) introduced a cheap talk approach. With this approach a cheap talk script precedes the elicitation of willingness to pay. Cheap talk scripts include an explicit discussion about hypothetical bias. Subjects are told what hypothetical bias is, that it is a common problem in hypothetical valuation questions, and why it might occur. Furthermore, subjects are asked to adjust for hypothetical bias in responding to the willingness to pay question. After the subject has read or heard the cheap talk script he/she responds to the willingness to pay question. Cummings and Taylor provided evidence of the effectiveness of the cheap talk approach in experimental referenda about donations to public goods. Subsequent studies on the cheap talk approach have found mixed results. Brown et al. (2003), using a similar design as Cummings and Taylor, found that the effectiveness of the cheap talk approach varied with the price level. In a second‐price auction for sports cards, List (2001) found that the cheap talk approach removed the hypothetical bias for nondealers, but not for dealers. Ajzen et al. (2004) found that a cheap talk corrective entreaty eliminated hypothetical bias on a referendum to contribute to a college scholarship fund. In a study by Murphy et al. (2005) on a voluntary contribution mechanism, the cheap talk approach did not fully remove the hypothetical bias. Aadland and Caplan (2006) develop a general cheap talk script that is short and neutral with regard to the direction of hypothetical bias and find that it exacerbates, rather than mitigates, the bias. Concurrently a second approach to mitigating hypothetical bias was developed. This approach is based on respondent certainty about willingness to pay. Two different versions of this approach have been used. In the first version a scale is used to assess the degree of certainty in hypothetical willingness to pay responses. Such a certainty scale was first used by Champ et al. (1997), who compared hypothetical dichotomous choice questions about donating a specified amount to a public good with actual donations to the public good. They assessed the certainty of the hypothetical donation responses on a 1–10 scale from very uncertain to very certain. They found that hypothetical donations significantly exceeded real donations but that there was no significant difference when only subjects who were very certain of their yes response were counted as real yes responses. Similar results using a certainty scale were also found by Ethier et al. (2000), Champ and Bishop (2001), Poe et al. (2002) and Vossler et al. (2003).2 A drawback of using a scale to assess certainty is that it is necessary to estimate the cut‐off level of certainty at which a hypothetical decision corresponds to a real decision. In the second version of the certainty approach, hypothetical willingness to pay responses are divided based on a follow‐up question with two degrees of certainty. An early form of this version was used by Johannesson et al. (1998) who divided hypothetical yes responses to buy a box of chocolates into ‘fairly sure’ and ‘absolutely sure’ yes responses. They found that the percentage of absolutely sure yes responses was lower than the proportion of real yes responses and the approach provided a conservative estimate of willingness to pay.3 This certainty question was subsequently adjusted by Blumenschein et al. (1998), who divided hypothetical yes responses into ‘probably sure’ and ‘definitely sure’ responses. Only yes responses about which subjects were ‘definitely sure’ were treated as yes responses. Yes responses about which subjects were only ‘probably sure’ were treated the same as no responses. This approach has been effective in removing hypothetical bias in both laboratory and field experiments (Blumenschein et al., 1998, 2001). Furthermore, by studying individuals who first respond to a hypothetical willingness to pay question and subsequently face a real purchase decision, it has been demonstrated that the degree of certainty in hypothetical yes responses is a very strong predictor of whether a hypothetical yes response corresponds to a real yes response (Blumenschein et al., 1998; Johannesson et al., 1999). In the current study we provide new evidence from a field experiment about the effectiveness of the cheap talk approach and the certainty approach in calibrating hypothetical willingness to pay responses. The purpose of the field experiment is to compare: real payments with stated hypothetical willingness to pay, real payments with stated hypothetical willingness to pay elicited with a cheap talk entreaty, and real payments with stated hypothetical willingness to pay adjusted based on follow up certainty statements. The main contribution of the article is the direct comparison of the cheap talk and certainty approaches in a field experiment with real payments at several prices. For a field experiment, we have a relatively large sample of more than 260 individuals, and in parametric analysis we control for more than 20 personal characteristics including income, health, health behaviour and time costs. The field experiment is carried out under the same conditions as a high quality contingent valuation study. Face‐to‐face interviews are used, and the subjects value a non‐trivial good that is not available on the market. In an environment of social concern for obesity and diabetes we chose a diabetes management programme as the good. It is primarily a private good, but spillovers exist from third party financing of health care. The subjects in the experiment are divided into three treatments groups: real, stated preference after cheap talk, and stated preference with certainty follow up. Subjects receive a real or hypothetical offer to purchase a real, but previously unavailable, service. By varying the price across subjects we are able to estimate the aggregate demand curve (Bishop and Heberlein, 1979). Our findings suggest that the cheap talk approach is not a generally effective method of removing hypothetical bias, but that it is possible to accurately estimate willingness to pay by adding a simple follow‐up question about the certainty of responses. 1. Experimental Design The experiment involved three different treatments referred to as the ‘real group’, ‘the hypothetical group’ and the ‘cheap talk group’. Subjects in the real group were actually given the opportunity to purchase a pharmacist‐provided diabetes management programme. Subjects in the hypothetical group received a dichotomous choice contingent valuation question about purchasing the same diabetes management programme followed by a question about the certainty of their hypothetical responses (with the categories: probably sure/definitely sure). When subjects answered the willingness to pay question they did not know that they would subsequently be asked about the certainty of their willingness to pay responses. Subjects in the cheap talk group received a dichotomous choice contingent valuation question preceded by a cheap talk script. In all three groups the price was varied between $15, $40 and $80. Each subject received only one price offer. The questionnaires were pilot‐tested in a focus group of diabetics in Lexington, Kentucky prior to the study. The focus group was also used to determine the prices used in the experiment. The data collection was carried out as ‘face‐to‐face’ interviews, and the same trained interviewer carried out all the interviews.4 The study had approval from the University of Kentucky Medical Institutional Review Board. Details of the experiment are further described below. 1.1. The Good The good used in our experiment was a pharmacist provided, type‐2 diabetes management programme. This programme was designed to assist type‐2 diabetics in attaining optimal management of their diabetes, thereby enhancing their life quality and decreasing their utilisation of expensive health care services such as Emergency Department visits and hospitalisations. The profession of pharmacy is slowly progressing towards a ‘patient care’ focus of practice rather than the historical ‘drug dispensing’ focus. Pharmacists in a variety of practice settings are developing new clinical services and disease management programmes to help patients achieve desired health outcomes. These services range from blood lipid management and diabetes education to smoking cessation programmes and cancer risk assessment (Bluml et al., 2000; Rodriguez de Bittner and Haines, 1997; Kennedy and Small, 2002; Giles et al., 2001). Although new pharmacist‐provided patient care and disease management programmes are being developed, these services are rarely provided in the community pharmacies where most people encounter pharmacists (Posey, 2003). Thus, it is very unlikely that the general public has a pre‐conceived market price for such services. Furthermore, in the current health care environment in the US, pharmacist‐provided disease management is rarely included as a benefit in health plans. The scope of a comprehensive disease management programme, such as the one in this experiment, exceeds the ethical obligations and expectations for required pharmacist service provision; therefore, it is not unethical for a pharmacist to deny the programme to individuals who do not wish to pay for the service. 1.2. Subject Recruitment Subjects were recruited from nine pharmacies in the state of Kentucky in the US. Three pharmacies were included in each of the experimental groups. The three pharmacies in the real group agreed to deliver the diabetes management programme to any subject that purchased the service at the price offered. All of the pharmacists in the real group had received extensive training from the American Pharmacy Services Corporation Foundation for Education and Research on providing the diabetes management programme prior to the implementation of this study. Only one price was used at each pharmacy, as the pharmacists were unwilling to charge patients within the same pharmacy different prices. The pharmacists identified their type‐2 diabetes patients who were age 18 or older and who had received a prescription for a type‐2 diabetes medication in the past 6 months. Potential subjects were contacted by phone. After confirming the diagnosis of diabetes they were asked if they would participate in a scientific study that involved an interview of approximately 15–20 minutes. Individuals who agreed to participate in the study were given a mutually convenient appointment time for the interview, which was carried out in the pharmacy. As compensation, each subject received $25 upon completion of the survey. The interviews took place between May 1 and July 23, 2003. In total 267 patients were interviewed (90 in the real group, 91 in the hypothetical group and 86 in the cheap talk group). 1.3. The Questionnaires Subjects were first given a questionnaire with background questions to fill in. In all experimental groups, a written description of the pharmacist‐provided diabetes management programme was then given to the subject. (The description of the diabetes management programme is given in Appendix 1.) The interviewer read the programme description to the subject while the subject read along on his/her own copy. The interviewer responded to any questions the subject had regarding the service. Next, the interviewer gave the subject a written copy of the survey. The interviewer read the valuation/purchase question (including the cheap talk script in that treatment) to the subject and the subject marked his/her response on the survey form. Subsequently in the hypothetical group the interviewer also read the certainty follow up question to the subject, and again, the subject marked his/her response on the survey form.5 1.3.1. Real Group Subjects in the real group were given the opportunity to purchase the diabetes management programme at their pharmacy. After receiving a description of the diabetes management programme the following question was posed:6 ‘You are now being offered the opportunity to purchase the diabetes disease management service that was just described to you. All of the services that were described to you would be provided for one flat rate. If you choose to purchase the service, you will have to use some of your household income to pay for it here and now with cash, cheque or credit card. Will you buy this service here and now at a price of $40? Please circle your answer below.’ 1.3.2. Hypothetical Group Subjects in the hypothetical group received a hypothetical dichotomous choice contingent valuation question. The question consisted of a description of the diabetes management programme after which the following question was posed: ‘Assume that you are being offered the opportunity to purchase the diabetes disease management service that was just described to you. All of the services that were described to you would be provided for one flat rate. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. Would you buy this service here and now at a price of $40? Please circle your answer below.’ The dichotomous choice contingent valuation question was followed by a question in which the subjects were asked if they were ‘probably sure’ or ‘definitely sure’ of their yes (no) answer. This question appeared on the page following the willingness to pay question and was phrased in the following way: ‘If you answered YES (NO), are you ‘probably sure’ or ‘definitely sure’ that you would (not) buy the diabetes management service here and now at a price of $40? Please circle your answer below.’ 1.3.3. Cheap Talk Group Subjects in the cheap talk group received a hypothetical dichotomous choice contingent valuation question preceded by a cheap talk script. The cheap talk script explained hypothetical bias and asked subjects to state what they would really do. We modelled our cheap talk script after the script used in the Cummings and Taylor (1999) study, with some adaptation due to the differences in the valuation question (a dichotomous purchase question versus a dichotomous referendum question) and the good (a predominately private, health good versus a public, environmental good). Cummings and Taylor (1999) used two versions of the cheap talk script; one with numbers to illustrate hypothetical bias and one without specific numbers. These two scripts yielded statistically indistinguishable results in their study. We used the script without numbers, as it is more generally applicable.7 Such cheap talk has reduced hypothetical bias in some previous studies. (The full cheap talk script is given in Appendix 2.) The following excerpts from the script illustrate the approach: ‘…in a recent study, a group of patients were asked if they were willing to purchase a disease management service that is similar to the diabetes disease management service that I have just described to you. Payment was hypothetical for these patients, as it will be for you. No one had to pay money if they said they would buy the disease management service. Another similar group of patients also participated in this study. These patients were offered the opportunity to actually purchase the disease management service at the same price. If patients in this second group agreed to purchase the programme they really did have to pay money. On average, more patients said ‘yes’ when the purchase question was hypothetical than when it was real. We call this ‘hypothetical bias’. ‘Hypothetical bias’ is the difference that we continually see in the way people respond to hypothetical situations as compared to real situations… How can we get people to think about their response to a hypothetical purchase question and respond as if it was a real purchase decision, where if they agree to the purchase they will really have to pay the price? How do we get them to think about what it means to really pay money, if in fact they really aren't going to have to do it? Let me tell you why I think we continually see this hypothetical bias, why people behave differently in a hypothetical purchase situation than they do when the purchase situation is real. I think that when we respond to hypothetical purchase questions, we give some thought to what we might do, but we know we can always change our minds especially if we don't want to buy. But, when the purchase offer is real, and we would actually have to spend our money if we say yes, we think a different way. If I were in your shoes… I would think about how I feel about spending my money this way. When I got ready to answer the question, I would ask myself: if this were a real offer to purchase the diabetes service, and I had to pay $40 if I said yes: do I really want to spend my money this way? If I really did, I would say yes; if I did not, I would say no… In any case, I ask you to respond just exactly as you would respond if you were really going to face the consequences of your response: which is to pay money if you say yes. Please keep this in mind when responding to the question.’ After reading the cheap talk script the subjects responded to the dichotomous choice contingent valuation question. As in the hypothetical group, the dichotomous choice contingent valuation question was followed (on a separate page) by a question in which the subjects were asked if they were ‘probably sure’ or ‘definitely sure’ of their yes (no) answer. 1.4. Hypotheses and Tests To test the null hypothesis of no hypothetical bias, the percentage of yes responses is compared between the experimental groups. Three comparisons are carried out: hypothetical versus real, hypothetical definitely sure versus real, and cheap talk versus real. The first comparison tests for the presence of hypothetical bias in a standard, dichotomous choice willingness to pay question and the other two comparisons test if the certainty approach and/or the cheap talk approach removes the hypothetical bias. A contingency table Pearson chi‐square test is used to compare the percentage of yes responses between the groups (D'Agostino et al., 1988). While this test is informative, the drawback of the non‐parametric test is that it does not control for any differences in personal characteristics between the groups. Therefore, we also carry out logistic regression analysis to test if the probability of a yes response differs between the groups, controlling for the background variables collected in the study.8 The price of the diabetes management programme is included in the regression analysis in order to derive the demand curve and estimate the mean willingness to pay. The mean willingness to pay is also estimated non‐parametrically using the method developed by Kriström (1990). In the regression analysis we control for a host of personal characteristics collected in the study. To test and control for an income effect we include annual household income, household size and an indicator for wealth (if the subject owns his or her residence).9 To control for differences in health and health behaviour we include the following variables: previous participation in a diabetes management programme, membership in a diabetes support group, time since diabetes was diagnosed, perceived diabetes severity (mild, moderate, or severe), if the subject has suffered from any of the following complications from diabetes: cardiovascular disease, renal disease, vision problems or neuropathies; if any family member has suffered from complications of diabetes, smoking, body mass index, an indicator for whether the subject knows his/her haemoglobin A1c level, and perceived general health status (excellent, very good, good, fair, poor).10 We also control for the following socioeconomic characteristics: age, gender, years of education and ethnic background. Finally, we control for an indicator of the time cost of participating in the diabetes management programme (the travel time to the pharmacy).11 2. Results 2.1. Background Characteristics Table 1 provides the background characteristics for the three experimental groups.12 There are great similarities among the experimental groups in terms of background characteristics such as household income, household size, home ownership, age and ethnic background. In only three cases is the difference between groups significant at the 5% level. We control for differences in the regression analysis. Table 1 Background Characteristics (standard deviations in parentheses) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) Open in new tab Table 1 Background Characteristics (standard deviations in parentheses) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) Open in new tab 2.2. Experimental Results The experimental results are shown in Table 2. In the real group 45% of the subjects bought the diabetes management programme at a price of $15, 23% bought the programme at a price of $40 and 10% of the subjects bought the programme at a price of $80. If there is hypothetical bias, then these percentages will be higher in the hypothetical group. We find this is clearly the case; the percentage of yes responses is higher at all prices in the hypothetical group. Overall the percentage of ‘buyers’ is about twice as high in the hypothetical group (45% versus 26%), and this difference is highly significant. Table 2 Number (%) of Yes Responses in the Experimental Groups Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 *p‐value of the difference compared to the yes responses in the real group. Open in new tab Table 2 Number (%) of Yes Responses in the Experimental Groups Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 *p‐value of the difference compared to the yes responses in the real group. Open in new tab The overall percentage of yes responses in the cheap talk group (45%) is the same as in the hypothetical group; the hypothetical bias is substantial in the cheap talk group also. The certainty approach seems to work better in terms of reducing hypothetical bias. The overall percentage of definitely sure yes responses is close to the percentage of real yes responses (24% versus 26%), and we cannot reject the null hypothesis of no difference in this case. The results of the parametric tests using logistic regression analysis are shown in Table 3.13 In the regressions we include dummy variables for the experimental groups. We run two regression equations for the two interpretations of yes responses in the hypothetical group. The first one considers all yes responses to be yes responses in the hypothetical group. The second one considers only the definitely sure yes responses to be yes responses, i.e., the probably sure yes responses are coded as ‘no’. In the first regression, the dummy variable for the hypothetical group is highly significant, and the marginal effect is 23.4 percentage points.14 Also the dummy variable for the cheap talk group is highly significant, with a marginal effect of 24.9 percentage points. In the second regression, the one that considers only definitely sure yes responses to be yes responses, the hypothetical dummy variable is close to zero and not significant. These results imply that the follow up certainty approach is effective in removing the hypothetical bias. Table 3 Results of Logistic Regression Analysis of the Effect of Experimental Group on the Probability of a Yes Response . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab Table 3 Results of Logistic Regression Analysis of the Effect of Experimental Group on the Probability of a Yes Response . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab We can see also that there is a highly significant effect of price in the regression equations, consistent with a downward sloping demand curve. The price coefficient measures the slope of the demand curve. It has been argued that the slopes will differ for hypothetical and real willingness to pay data, due to a higher variance in the hypothetical willingness to pay (Haab et al., 1999). In our data there is a tendency for the hypothetical responses to be closer to the real responses for higher prices than for lower prices whereas with cheap talk it is the other way round. These tendencies would suggest that the slopes of the demand curves may differ between the three experimental groups. Therefore, we tested adding interaction terms between the price and the experimental group dummy variables, allowing the slope of the aggregate demand function to vary between the three experimental groups. The interaction terms are, however, not significant at the 10% level (neither individually nor jointly), and we cannot reject the null hypothesis of equal slopes between the three experimental groups (for either equation 1 or equation 2 in Table 3).15 There is also a significant effect of income and the indicator for wealth. Higher income and wealth shift the demand curve outwards, consistent with economic theory. The negative effect of household size on demand might also reflect an income effect (as the income per household member decreases with household size). Membership in a diabetes support group is also associated with a significantly higher demand for the diabetes management programme, as is knowledge about the haemoglobin A1c level. These variables might reflect a greater concern for the impact of diabetes. There is also a tendency for an effect of the perceived general health status, with a higher demand for subjects with a worse overall health status. Patients with a worse health status may derive greater health gains from participating in the diabetes management programme. As suggested by the work of Viscusi and Evans (1990), a worse health status may also decrease the marginal utility of income, which will increase the willingness to pay for a given health gain. The follow‐up question about the certainty of yes/no responses was also included in the cheap talk group. The overall percentage of definitely sure yes responses is 30% in the cheap talk group, which according to a chi‐square test is not significantly different from the percentage of real yes responses (26%; p = 0.489) or the percentage of definitely sure yes responses in the hypothetical group (24%; p = 0.365). We also re‐estimated the second regression equation in Table 3 counting only definitely sure yes as yes responses in the cheap talk group. The cheap talk group dummy variable is not significant in that regression equation (coefficient = 0.406; p = 0.322; marginal effect = 0.068). We also tested pooling the definitely sure yes responses in the hypothetical group and the cheap talk group, including a dummy variable for this joint sample (comparing definitely sure yes responses for the pooled hypothetical and cheap talk groups with the real yes responses). This dummy variable is not significant (coefficient = 0.108; p = 0.760; marginal effect = 0.018). The overall percentage of definitely sure yes responses in the pooled sample is 27%, which according to a chi‐square test is not significantly different from the percentage of real yes responses (26%; p = 0.785). 2.3. Aggregate Demand Curves and Willingness To Pay In Figures 1 and 2 we show the non‐parametric and parametric aggregate demand curves. The parametric demand curves are based on the logistic regression equations in Table 3 and are estimated at the mean of the covariates. The Figures confirm the previous results. The hypothetical and cheap talk demand curves are relatively close and both overestimate the actual demand. In contrast to cheap talk, the certainty calibration through recoding leads to a demand curve that is close to the real demand curve. Fig. 2. Open in new tabDownload slide Parametric Demand Curves Fig. 2. Open in new tabDownload slide Parametric Demand Curves Fig. 1. Open in new tabDownload slide Non‐parametric Demand Curves Fig. 1. Open in new tabDownload slide Non‐parametric Demand Curves The mean willingness to pay can be estimated as the area below the demand curves, and the estimated means are shown in Table 4.16 Using the parametric method the mean real willingness to pay is $22 and the mean hypothetical willingness to pay is $42. The cheap talk approach yields a mean willingness to pay of $44, whereas the certainty approach yields a mean willingness to pay of $20. The non‐parametric and the parametric methods yield similar estimates of the mean willingness to pay especially for the real and definitely sure hypothetical willingness to pay. The difference of $2 between the real and definitely sure willingness to pay, for both the non‐parametric and parametric methods, is not statistically different from zero. Table 4 Mean Willingness to Pay in the Experimental Groups ($) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) *p‐value of the difference compared to the mean willingness to pay in the real group and 95% confidence interval of the difference. Open in new tab Table 4 Mean Willingness to Pay in the Experimental Groups ($) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) *p‐value of the difference compared to the mean willingness to pay in the real group and 95% confidence interval of the difference. Open in new tab 2.4. Sensitivity Analysis As an additional test of whether or not the definitely sure yes responses differ from the real yes responses, we compare the effects of all covariates between the real group and the definitely‐sure hypothetical group. This is done by adding interaction terms between the hypothetical group dummy variable and all the other variables; equivalent to running separate regression functions for the real group and the definitely sure hypothetical group.17 With a likelihood ratio test we test if the hypothetical dummy variable and the interaction terms improve the significance of the model. We cannot reject the null hypothesis of no significant difference between the regression models at the 10% level.18 Our results suggest that probably sure and definitely sure yes responses are quite different (i.e. probably sure yes responses correspond to real ‘no’ responses). To investigate this further, we estimate two separate regressions on the hypothetical group data. The first regression (n = 69) counts probably sure yes responses as yes and excludes the definitely sure yes responses, and the second regression (n = 72) counts definitely sure yes responses as yes and excludes the probably sure yes responses. If the two categories of yes responses are identical the estimated coefficients and explanatory power should be similar for the two regressions. The results of these two regressions, shown in Table 5, differ substantially. In the regression on probably sure yes responses, the McFadden pseudo‐R2 is 0.233, no variable is significant at the 5% level, and the regression equation is far from significant (chi‐square 18.895 (23 df); p = 0.707). In the regression on definitely sure yes responses the McFadden pseudo‐R2 is 0.584 and the regression equation is highly significant (chi‐square = 51.804 (23 df); p = 0.001). Both the price variable and the wealth indicator are also highly significant with the expected signs. Household income has the expected sign but is not quite significant (p = 0.187).19 Table 5 Results of Logistic Regression Analysis Comparing the Probably Sure and Definitely Sure Yes Responses in the Hypothetical Group . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab Table 5 Results of Logistic Regression Analysis Comparing the Probably Sure and Definitely Sure Yes Responses in the Hypothetical Group . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab In the experimental results in Table 2 the percentage of yes responses at the $80 price in the cheap talk group seems high; i.e. substantially higher than in the hypothetical group and somewhat higher than at the $40 price in the cheap talk group. We therefore investigated if the background characteristics differed in the $80 cheap talk group compared to the rest of the sample. There is a significant difference at the 5% level for only two of the over twenty background characteristics. The number of subjects that had previously participated in a diabetes management programme is higher in the cheap talk group at the $80 price (22% versus 8%) and the number of patients with vision problems is higher in this group (59% versus 36%). Both of these variables are controlled for in the regression analysis. In the regression analysis, the vision problem variable is far from significant and has a point estimate that is close to zero. The variable for previous participation in a diabetes management programme has a sizeable positive point estimate but is not quite significant. Six of the 27 subjects at the $80 price in the cheap talk group had previously participated in a diabetes management programme and four of them said yes in the hypothetical willingness to pay question. This may be one explanation for the high rate of yes responses in the cheap talk group at the $80 price. As an extra sensitivity analysis we therefore re‐estimated our results excluding all subjects that had previously participated in a diabetes management programme (24 subjects). After this exclusion the fraction of yes responses in the cheap talk group is 61% at $15, 37% at $40 and 33% at $80, and the overall fraction of yes responses is still significantly higher than in the real group (44% in the cheap talk group and 23% in the real group; p = 0.005).20 The effect of the cheap talk group is also still highly significant in the regression analysis (p = 0.012 in the first regression equation in Table 3 and p = 0.006 in the second regression equation in Table 3). 3. Discussion and Concluding Remarks Our field experiment yields two important findings for eliciting willingness to pay using contingent valuation. First, hypothetical bias was removed with the follow‐up certainty question; this approach yielded dichotomous choice responses that were indistinguishable from the real decisions. Second, the cheap talk approach was not effective in removing hypothetical bias. We discuss each of these findings in turn below. Prior evidence on the cheap talk approach is mixed. Cummings and Taylor (1999) found that the cheap talk approach was effective in removing hypothetical bias in experimental referendums about donations to public goods. However, in the recent studies by List (2001), Brown et al. (2003) and Murphy et al. (2005), the cheap talk approach was effective only in specific sub‐groups. The diabetics in our study have had diabetes for an average of about nine years. To the extent that the diabetics in our study are informed and experienced, like the card dealers in List's experiment, our finding that hypothetical bias in stated preferences for the diabetes management programme is not mitigated by cheap talk is similar to List's finding. In terms of validity, we think our findings are important evidence. Our field experiment is carried out in a similar fashion as an actual contingent valuation study. Face‐to‐face interviews (which are often recommended) are used and subjects value a non‐trivial good for which there is no available market price on which to anchor. The consistent effects of price and income also support the validity of our results. The follow up certainty approach used in this study yielded estimates of willingness to pay that matched real payments. These results are encouraging and they are consistent with previous evidence from both a laboratory experiment on a private good and a field experiment on a health care good (Blumenschein et al., 1998, 2001). This follow up certainty approach is also a very practical approach and appealing against Occam's razor. It only entails adding a simple follow‐up question with two degrees of certainty to a dichotomous choice contingent valuation study. Why do certainty statements that follow contingent valuation reduce hypothetical bias? One straightforward explanation is that definitely yes more closely resembles the response necessary to make a purchase in real market situations. The manner in which the offer is made requires a decision that must be made immediately. Only respondents who are ready to produce the cash, cheque, or credit/debit card get to purchase. If probably yes signals some interest but not enough to make the payment when offered the opportunity, then probably yes is tantamount to no. A sales representative will not let the person have the product unless payment is made. Another explanation of how certainty statements produce a good match between statements and actual purchase behaviour draws on social psychology. Sample and Warland (1973) view attitude as a predisposition to behaviour. They hypothesise and find that certainty of attitude is a moderator variable that produces better measurement of attitude and intention and better prediction of behaviour. Certainty is used to partition people into more homogenous groups. Attitude is a major predictor of both intentions and behaviour for the high‐certainty group. Fazio and Zanna (1978a,b) and Raden (1985) also explore the link between attitude and behaviour and state that individuals who hold attitudes with greater certainty and more confidence behave more consistently. More recently Ajzen (1991) offers a theory of planned behaviour in which attitudes, subjective norms and perceived behavioural control predict intentions and, in turn, intentions predict behaviour. Intentions capture motivational forces. The stronger the intention, the stronger is the link to behaviour.21 The Fujii and Gärling (2003) application of attitude theory to improve the accuracy of stated preferences for travel modes provides a clearer link yet between stated willingness to pay, certainty statements and behaviour. Their assessment is that the single most important insight from attitude theory is that behavioural intention, the commitment to act, is the best predictor of behaviour. Stated preference methods are designed to elicit stable preferences that reflect a well‐behaved utility function. Stated preference is interpreted as a behavioural intention. Fujii and Gärling's view is that preference is composed of two parts, core preference determined by the utility function and contingent preference that may depend on factors such as framing, response mode and anchoring. Studies that do not eliminate contingent preference embedded in stated preference data will produce results that do not match well with behaviour. They expect that weak intention cause errors in predicting behaviour. Their before‐ and after‐panel survey of use of a new subway line in Kyoto, Japan in 1997–8 shows that strength of intention to ride the new line (yes instead of yes to some degree) was a factor that substantially increased the match between intention and behaviour. Our findings that follow up certainty statements permit reclassification of yes responses so that stated willingness to pay and purchase behaviour match well is consistent with the view from social psychology that strength of attitude and intention improve the link between them and behaviour. Our findings are consistent too with the view that only strong, certain stated purchase intentions match well with the typical market purchase. Cheap talk appears to be effective in mitigating hypothetical bias in some applications and ineffective in others. Our evidence suggests that certainty statements will be more effective in removing hypothetical bias but further research is warranted. Field experiments on goods with varying degrees of publicness and different subject populations would be valuable in establishing the certainty approach as a consistently reliable way to improve contingent valuation and determining if follow up questions with two degrees of certainty is always the best form of the certainty approach. Footnotes 1 " Wallis and Friedman (1942) criticised the use of hypothetical choices in experiments and it has been debated ever since (Thaler, 1987; Kagel and Roth, 1995; Camerer and Hogarth, 1999). In experimental economics it is standard practice to provide monetary incentives, whereas the bulk of experimental work in psychology is based on hypothetical choices. For an overview of the methodological debate between psychologists and economists, see Hertwig and Ortmann (2001). 2 " A potentially important variant of this approach incorporates different degrees of certainty directly into the contingent valuation questions and adjusts the estimated willingness to pay (Johannesson et al., 1993; Ready et al., 1995, 2001; Alberini et al., 2003; Evans et al., 2003; Vossler and Poe, 2005). In the current study we choose to focus on dichotomous choice contingent valuation and a follow up certainty question with two degrees of certainty in order to isolate the effect of follow up certainty calibration. 3 " This study was also carried out in Swedish, making the exact translation of the certainty categories difficult. 4 " List et al. (2004) found that subject anonymity can affect willingness to pay for a public good using a referendum format. They argue that subjects may vote yes to publicly advertise their own goodwill. However, any social approval should be less important with the diabetes management programme used in our study. 5 " The follow up certainty question was asked after the willingness to pay question in the cheap talk group also. All data and questionnaires are available from the first author upon request. 6 " The price was varied between $15, $40 and $80 but below we illustrate the questions with a $40 price. 7 " A problem with using actual numbers to illustrate hypothetical bias is that the size of hypothetical bias (and thus numbers) may vary across studies and it therefore becomes unclear which numbers to use. There is also a risk that numbers will lead to anchoring if prices are given. 8 " Using a probit model instead of a logit model yielded similar results and does not change the reported conclusions. 9 " For household income the subject could either fill in the exact amount or mark one of the following categories: <$5,000, $5,000–10,000, $10,001–20,000, $20,001–30,000, $30,001–50,000, $50,001–100,000, $100,001–150,000, >$150,000. A continuous income measure was constructed by setting the income for each subject to the midpoint of the interval ($175,000 was used for the highest income category). 10 " Haemoglobin A1c (HbA1c), also referred to as glycosylated haemoglobin, is a useful indicator of how well the blood glucose level has been controlled in the recent past and is routinely used to monitor diabetic patients; complications of diabetes can be delayed or prevented if the HbA1c level can be kept close to 7%. 11 " There were missing data for household income for four subjects and for body mass index for two subjects. In the regression analysis we imputed the mean value of household income and body mass index in the sample for these subjects to avoid losing any observations. Excluding the six observations with missing data from the regression analysis leads to similar results, and does not change the conclusions reported below. 12 " For continuous variables a two‐sided t‐test was used to test for statistical differences between the groups. For the categorical variables a contingency table Pearson chi‐square test was used. 13 " The continuous variables (price, time with diabetes, body mass index, age, education, household size, household income, and travel time) are included without any transformation in Table 3. We also tested a logarithmic transformation for these variables, and that yielded similar results. 14 " The marginal effects are evaluated at the mean of the other covariates. 15 " The chi‐square value of the likelihood‐ratio test of the joint significance of the two interaction coefficients is 3.36 (2 df) for the first regression equation in Table 3 and 3.97 (2 df) for the second regression equation in Table 3. The critical value at the 10% level is 4.61. 16 " In the estimation of mean willingness to pay with the non‐parametric method it was assumed that the maximum willingness to pay was equal to the highest price ($80) used in the study, and that the proportion of subjects with zero willingness to pay was equal to the proportion of ‘no’ responses at the lowest price used in the study ($15). The variance of mean willingness to pay was estimated based on 2,000 bootstrap replications using the method of Tambour and Zethraeus (1998). The estimation of mean willingness to pay in the parametric method was based on estimating the area below the demand curve using the formula: −(1/β) ln(1 + eα), where β is the price coefficient in the logistic regression equation and α is the constant in the logistic regression equation (with the effect of all other covariates added to the constant). This formula restricts willingness to pay to be nonnegative as is appropriate for a private good that does not have to be consumed. See Johansson (1995, p. 113) for details. 17 " The interaction terms are added to the second regression model in Table 3 (‘Definitely sure yes responses in hypothetical group’) but with the cheap talk group excluded. 18 " The chi‐square value of the test is 32.11 (26 df) and the critical value at the 10% level is 35.56. 19 " These regressions include the same variables as the regression results in Table 3 (except the experimental group dummy variables), with the exception that two variables (diabetes support group and renal disease) are excluded due to a lack of variation in the dependent variable for one of the dummy variable categories. 20 " The p‐values comparing the cheap talk group and the real group at the individual prices are now 0.102 ($15), 0.234 ($40) and 0.052 ($80). 21 " Mitchell and Carson (1989, Chapter 8) offer an insightful discussion of contingent valuation in the context of attitude‐intention‐behaviour relationships studied in social psychology and marketing. Stated willingness to pay is interpreted as an intention to actually pay. They offer three factors that will increase the predictive power of intention: correspondence, proximity and familiarity. They do not discuss increasing the predictive power of stated preferences through the use of stated certainty. References Aadland , D. and Caplan , A.J. (2006). ‘Cheap talk revisited: new evidence from CVM’ , Journal of Economic Behavior and Organisation , vol. 60 ( 4 ) (August), pp. 562 – 78 . Crossref Search ADS WorldCat Ajzen , Icek. ( 1991 ). ‘The theory of planned behavior’ , Organizational Behavior and Human Decision Processes , vol. 50 ( 2 ) (December), pp. 179 – 211 . Google Scholar Crossref Search ADS WorldCat Ajzen , I. , Brown , T.C. and Carvajal , F. ( 2004 ). ‘Explaining the discrepancy between intentions and actions: the case of hypothetical bias in contingent valuation’ , Personality and Social Psychology Bulletin , vol. 30 ( 9 ) (September), pp. 1108 – 21 . Google Scholar Crossref Search ADS PubMed WorldCat Alberini , A. , Boyle , K. and Welsh , M. ( 2003 ). ‘Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty’ , Journal of Environmental Economics and Management , vol. 45 ( 1 ) (January), pp. 40 – 62 . Google Scholar Crossref Search ADS WorldCat Ariely , D. , Loewenstein , G. and Prelec , D. ( 2003 ). ‘‘‘Coherent arbitrariness’’: stable demand curves without stable preferences’ , Quarterly Journal of Economics , vol. 118 ( 1 ) (February), pp. 73 – 105 . Google Scholar Crossref Search ADS WorldCat Bishop , R.C. and Heberlein , T.A. ( 1979 ). ‘Measuring values of extramarket goods: are indirect measures biased?’ , American Journal of Agricultural Economics , vol. 61 ( 5 ) (December), pp. 926 – 30 . Google Scholar Crossref Search ADS WorldCat Blumenschein , K. , Johannesson , M., Blomquist , G.C., Liljas , B. and O'Conor R.M. ( 1998 ). ‘Experimental results on expressed certainty and hypothetical bias in contingent valuation’ , Southern Economic Journal , vol. 65 ( 1 ) (July), pp. 169 – 77 . Google Scholar Crossref Search ADS WorldCat Blumenschein , K. , Johannesson , M., Yokoyama , K.K. and Freeman , P.R. ( 2001 ). ‘Hypothetical versus real willingness to pay in the health care sector: results from a field experiment’ , Journal of Health Economics , vol. 20 ( 3 ) (May), pp. 441 – 57 . Google Scholar Crossref Search ADS PubMed WorldCat Bluml , B.M. , McKenney , J.M., and Cziraky , M.J. ( 2000 ). ‘Pharmaceutical care services and results in Project ImPACT: Hyperlipidemia’ , Journal of the American Pharmaceutical Association , vol. 40 ( 2 ) (March – April), pp. 157 – 65 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Brown , T.C. , Ajzen , I. and Hrubes , D. ( 2003 ). ‘Further test of entreaties to avoid hypothetical bias in referendum contingent valuation’ , Journal of Environmental Economics and Management , vol. 46 ( 2 ) (September), pp. 353 – 61 . Google Scholar Crossref Search ADS WorldCat Camerer , C.F. and Hogarth , R.M. ( 1999 ). ‘The effects of financial incentives in experiments: a review and capital‐labor‐production framework’ , Journal of Risk and Uncertainty , vol. 19 ( 1 ) (December), pp. 7 – 42 . Google Scholar Crossref Search ADS WorldCat Carson , R. T. , ( 2001 ). ‘Contingent valuation, resources and environmental’, in ( N. J. Smelser and P.B. Baltes, eds.), International Encyclopedia of Social and Behavioral Sciences , vol. 19 , pp. 13272 – 5 , Amsterdam: Elsevier Science . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 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 , vol. 72 ( 1 ) (February), pp. 80 – 99 . Google Scholar Crossref Search ADS WorldCat Carson , R.T. , Flores , N.E. and Meade , N.F. ( 2001 ). ‘Contingent valuation: controversies and evidence’ , Environmental and Resource Economics , vol. 19 ( 2 ) (June), pp. 173 – 210 . Google Scholar Crossref Search ADS WorldCat Champ , P.A. and Bishop , R.C. ( 2001 ). ‘Donation payment mechanisms and contingent valuation: an empirical study of hypothetical bias’ , Environmental and Resource Economics , vol. 19 ( 4 ) (August), pp. 383 – 402 . Google Scholar Crossref Search ADS WorldCat Champ , P.A. , Bishop , R.C., Brown , T.C. and McCollum , D.W. ( 1997 ). ‘Using donation mechanisms to value nonuse benefits from public goods’ , Journal of Environmental Economics and Management , vol. 33 ( 2 ) (June), pp. 151 – 62 . Google Scholar Crossref Search ADS WorldCat Cummings , R.G. , Elliott , S., Harrison , G.W. and Murphy , J. ( 1997 ). ‘Are hypothetical referenda incentive compatible?’ , Journal of Political Economy , vol. 105 ( 3 ) (June), pp. 609 – 21 . Google Scholar Crossref Search ADS WorldCat Cummings , R.G. , Harrison , G.W. and Rutström , E.E. ( 1995 ). ‘Homegrown values and hypothetical surveys: is the dichotomous choice approach incentive‐compatible?’ , American Economic Review , vol. 85 ( 1 ) (March), pp. 260 – 6 . OpenURL Placeholder Text WorldCat 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 , vol. 89 ( 3 ) (June), pp. 649 – 65 . Google Scholar Crossref Search ADS WorldCat D'Agostino , R.B. , Chase , W. and Belanger , B. ( 1988 ). ‘The appropriateness of some common procedures for testing the equality of two independent binomial populations’ , American Statistician , vol. 42 ( 3 ) (August), pp. 198 – 202 . OpenURL Placeholder Text WorldCat Diamond , P.A. and Hausman , J.A. ( 1994 ). ‘Contingent valuation: is some number better than no number?’ , Journal of Economic Perspectives , vol. 8 ( 4 ) (Fall), pp. 45 – 64 . Google Scholar Crossref Search ADS PubMed WorldCat Ethier , R.G. , Poe , G.L., Schulze , W.D. and Clark , J. ( 2000 ). ‘A comparison of hypothetical phone and mail contingent valuation responses for green‐pricing electricity programs’ , Land Economics , vol. 76 ( 1 ) (February), pp. 54 – 67 . Google Scholar Crossref Search ADS WorldCat Evans , M.F. , Flores , N.E. and Boyle , K.J. ( 2003 ). ‘Multiple‐bounded uncertainty choice data as probabilistic intentions’ , Land Economics , vol. 79 ( 4 ) (November), pp. 549 – 60 . Google Scholar Crossref Search ADS WorldCat Fazio , R.H. and Zanna , M.P. ( 1978a ). ‘Attitudinal qualities relating to the strength of the attitude‐behavior relationship’ , Journal of Experimental Social Psychology , vol. 14 ( 4 ) (July), pp. 398 – 408 . Google Scholar Crossref Search ADS WorldCat Fazio , R.H. and Zanna , M.P. ( 1978b ). ‘On the predictive validity of attitudes: the roles of direct experience and confidence’ , Journal of Personality , vol. 46 ( 2 ) (June), pp. 228 – 43 . Google Scholar Crossref Search ADS WorldCat Fujii , S. and Gärling , T. ( 2003 ). ‘Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis’ , Transportation Research Part A , vol. 37 ( 4 ) (May), pp. 389 – 402 . OpenURL Placeholder Text WorldCat Giles , J.T. , Kennedy , D.T., Dunn , E.C., Wallace W.L., Meadows , S.L. and Cafiero , A.C. ( 2001 ). ‘Results of a community pharmacy‐based breast cancer risk‐assessment and education program’ , Pharmacotherapy , vol. 21 ( 2 ) (February), pp. 243 – 53 . Google Scholar Crossref Search ADS PubMed WorldCat Haab , T.C. , Huang , J.‐C. and Whitehead , J.C. ( 1999 ). ‘Are hypothetical referenda incentive compatible? A comment’ , Journal of Political Economy , vol. 107 ( 1 ) (February), pp. 186 – 96 . Google Scholar Crossref Search ADS WorldCat Harrison , G.W. ( 2006 ). ‘Experimental evidence on alternative environmental valuation methods’ , Environmental and Resource Economics , vol. 34 ( 1 ) (May), pp. 125 – 62 . Google Scholar Crossref Search ADS WorldCat Harrison , G.W. and Rutström , E.E. (forthcoming). ‘Experimental evidence on the existence of hypothetical bias in value elicitation methods’, in ( C. Plott and V. L. Smith, eds.), Handbook of Experimental Economics Results , New York: Elsevier Science . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hertwig , R. and Ortmann , A. ( 2001 ). ‘Experimental practices in economics: a methodological challenge for psychologists?’ , Behavioral and Brain Sciences , vol. 24 ( 3 ) (June), pp. 383 – 403 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Johannesson , M. , Blomquist , G.C., Blumenschein , K., Johansson , P‐O., Liljas , B. and O'Conor , R.M. ( 1999 ). ‘Calibrating hypothetical willingness to pay responses’ , Journal of Risk and Uncertainty , vol. 18 ( 1 ) (April), pp. 21 – 32 . Google Scholar Crossref Search ADS WorldCat Johannesson , M. , Johansson , P‐O., Kriström , B. and Gerdtham , U‐G. ( 1993 ). ‘Willingness to pay for antihypertensive therapy: further results’ , Journal of Health Economics , vol. 12 ( 1 ) (April), pp. 95 – 108 . Google Scholar Crossref Search ADS PubMed WorldCat Johannesson , M. , Liljas , B. and Johansson , P‐O. ( 1998 ). ‘An experimental comparison of dichotomous choice contingent valuation questions and real purchase decisions’ , Applied Economics , vol. 30 ( 5 ) (May), pp. 643 – 7 . Google Scholar Crossref Search ADS WorldCat Johansson , P‐O. ( 1995 ). Evaluating Health Risks: An Economic Approach , Cambridge: Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kagel , J.H. and Roth , A.E. (eds.). ( 1995 ). The Handbook of Experimental Economics , Princeton, New Jersey: Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kennedy , D.T. and Small , R.E. ( 2002 ). ‘Development and implementation of a smoking cessation clinic in community pharmacy practice’ , Journal of the American Pharmaceutical Association , vol. 42 ( 1 ) (January – February), pp. 83 – 92 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Kriström , B. ( 1990 ). ‘A non‐parametric approach to the estimation of welfare measures in discrete response valuation studies’ , Land Economics , vol. 66 ( 2 ) (May), pp. 135 – 9 . Google Scholar Crossref Search ADS WorldCat List , J.A. ( 2001 ). ‘Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards’ , American Economic Review , vol. 91 ( 5 ) (December), pp. 1498 – 507 . Google Scholar Crossref Search ADS WorldCat List , J.A. , Berrens , R.P., Bohara , A.K. and Kerkvliet , J. ( 2004 ). ‘Examining the role of social isolation on stated preferences’ , American Economic Review , vol. 94 ( 3 ) (June), pp. 741 – 52 . Google Scholar Crossref Search ADS WorldCat List , J.A. and Gallet , C.A. ( 2001 ). ‘What experimental protocol influence disparities between actual and hypothetical stated values?’ , Environmental and Resource Economics , vol. 20 ( 3 ) (November), pp. 241 – 54 . Google Scholar Crossref Search ADS WorldCat Mitchell , R.C. and Carson , R.T. ( 1989 ). Using Surveys to Value Public Goods: The Contingent Valuation Method , Washington, DC: Resources for the Future . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Murphy , J.J. , Stevens , T. and Weatherhead , D. ( 2005 ). ‘Is cheap talk effective at eliminating hypothetical bias in a provision point mechanism?’ , Environmental and Resource Economics , vol. 30 ( 3 ) (March), pp. 327 – 43 . Google Scholar Crossref Search ADS WorldCat Poe , G.L. , Clark , J.E., Rondeau , D. and Schulze , W.D. ( 2002 ). ‘Provision point mechanisms and field validity tests of contingent valuation’ , Environmental and Resource Economics , vol. 23 ( 1 ) (September), pp. 105 – 31 . Google Scholar Crossref Search ADS WorldCat Portney , P.R. ( 1994 ). ‘The contingent valuation debate: why economists should care’ , Journal of Economic Perspectives , vol. 8 ( 4 ) (Fall), pp. 3 – 18 . Google Scholar Crossref Search ADS PubMed WorldCat Posey , L.M. ( 2003 ). ‘Proving that pharmaceutical care makes a difference in community pharmacy’ , Journal of the American Pharmaceutical Association , vol. 43 ( 2 ) (March‐April), pp. 136 – 9 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Raden , D. ( 1985 ). ‘Strength‐related attitude dimensions’ , Social Psychology Quarterly , vol. 48 ( 4 ) (December), pp. 312 – 30 . Google Scholar Crossref Search ADS WorldCat Ready , R.C. , Navrud , S. and Dubourg , W.R. ( 2001 ). ‘How do respondents with uncertain willingness to pay answer contingent valuation questions?’ , Land Economics , vol. 77 ( 3 ) (August), pp. 315 – 26 . Google Scholar Crossref Search ADS WorldCat Ready , R.C. , Whitehead , J.C. and Blomquist , G.C. ( 1995 ). ‘Contingent valuation when respondents are ambivalent’ , Journal of Environmental Economics and Management , vol. 29 ( 2 ) (September), pp. 181 – 96 . Google Scholar Crossref Search ADS WorldCat Rodriguez de Bittner , M. and Haines , S.T. ( 1997 ). ‘Pharmacy‐based diabetes management: a practical approach’ , Journal of the American Pharmaceutical Association , vol. 37 ( 4 ) (July‐August), pp. 443 – 55 . OpenURL Placeholder Text WorldCat Sample , J. and Warland , R. ( 1973 ). ‘Attitude and prediction of behavior’ , Social Forces , vol. 51 ( 3 ) (March), pp. 292 – 304 . Google Scholar Crossref Search ADS WorldCat Tambour , T. and Zethraeus , N. ( 1998 ). ‘Nonparametric willingness‐to‐pay measures and confidence statements’ , Medical Decision Making , vol. 18 ( 3 ) (July – September), pp. 330 – 6 . Google Scholar Crossref Search ADS PubMed WorldCat Thaler , R. ( 1987 ). ‘The psychology of choice and the assumptions of economics’, in ( A. E. Roth, ed.), Laboratory Experimentation in Economics: Six Points of View , pp. 99 – 130 , Cambridge: Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Viscusi , W.K. and Evans , W.N. ( 1990 ). ‘Utility functions that depend on health status: estimates and economic implications’ , American Economic Review , vol. 80 ( 3 ) (June), pp. 353 – 74 . OpenURL Placeholder Text WorldCat Vossler , C.A. , Ethier , R.G., Poe , G.L., and Welsh , M.P. ( 2003 ). ‘Payment certainty in discrete choice contingent valuation responses: results from a field validity test’ , Southern Economic Journal , vol. 69 ( 4 ) (April), pp. 886 – 902 . Google Scholar Crossref Search ADS WorldCat Vossler , C.A. and Poe , G.L. ( 2005 ). ‘Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty: a comment’ , Journal of Environmental Economics and Management , vol. 49 ( 1 ) (January), pp. 197 – 200 . Google Scholar Crossref Search ADS WorldCat Wallis , W.A. and Friedman , M. ( 1942 ). ‘The empirical derivation of indifference functions’, in ( O. Lange, F. McIntyre and T.O. Yntema, eds.), Studies in Mathematical Economics and Econometrics in Memory of Henry Schultz , pp. 175 – 89 , Chicago: University of Chicago Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Appendices Appendix 1: Description of the Diabetes Management Programme Pharmacist‐Provided Diabetes Management Programme Description Controlling diabetes is extremely important for your health. When diabetes is not properly controlled it can lead to complications such as nerve problems, heart problems, sexual problems, kidney failure, blindness, and amputations. These complications usually occur in people who have had uncontrolled diabetes for many years. Many studies have shown keeping your blood sugar and haemoglobin A1c at recommended levels can delay or even prevent these complications. You can significantly improve your health, feel better and prevent these complications by maintaining an appropriate diet and exercise plan and taking your diabetes medicines properly. Your pharmacist can provide education and motivation to help you understand and control your diabetes. This diabetes programme was developed by pharmacists at the American Pharmacy Services Corporation. The programme incorporates all of the latest recommendations for optimal diabetes management from the American Diabetes Association and the Centers for Disease Control and Prevention. The programme will allow you and your pharmacist to work together as a team to reach your desired blood sugar and haemoglobin A1c levels through appropriate medication use, monitoring, and lifestyle changes. This programme will last three months and will consist of three formal appointments with your pharmacist. With your permission, your pharmacist will measure your blood sugar, haemoglobin A1c, and conduct other tests to assess your overall health at the beginning and end of the programme. This will allow your pharmacist to make recommendations to you and your physician to maximise the effectiveness of your diabetes therapy. You will have the opportunity to discuss any concerns or problems you have during any of the scheduled visits. The first visit with your pharmacist will last between 45 minutes and one hour. During this visit, your pharmacist will take a medication history and discuss with you the medicines you take for diabetes. You will take a short survey to assess symptoms associated with diabetes and you will complete a short questionnaire about the relationship of haemoglobin A1c to diabetes control. Your pharmacist will then discuss the results of these surveys with you. Your pharmacist will review the relationship of diabetes control to symptoms and blood sugar testing, the proper way to take your diabetes medicine, and target levels for blood sugar and haemoglobin A1c to prevent complications in the future. In addition your pharmacist will measure your blood sugar, haemoglobin A1c, blood pressure and weight. These measures will help your pharmacist assess the progress that you make during the three month programme. The second visit will last between 25 and 35 minutes. Your pharmacist will again ask you to complete a short survey to assess symptoms associated with diabetes. Then, your pharmacist will provide you with information about eating a healthy diet, maintaining regular exercise, and weight control if you are overweight. In addition, your pharmacist will measure your blood sugar and weight, discuss any problems you have encountered since the last visit, and answer any questions you may have. The third and final visit will last between 30 and 45 minutes. You will complete a short survey assessing your diabetes symptoms and control. Your pharmacist will measure your blood sugar, haemoglobin A1c, weight, and blood pressure. Your pharmacist will discuss the progress you have made since the first visit and will review potential complications associated with uncontrolled diabetes and provide tips to help you maintain the progress you have made during the programme. You will have the opportunity to discuss any problems you have encountered since the last visit and ask questions. At the end of the third month, you may choose to continue the diabetes management programme, and you can ask your pharmacist to schedule new appointments with you to best suit your individual needs. Appendix 2: The Cheap Talk Script (with a $40 price). Assume that you are being offered the opportunity to purchase the diabetes disease management service that was just described to you. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. I am going to ask you: ‘Would you buy this service at a price of $40.’ But, before you answer the question, I want to describe a problem that we have in studies like this one. This is a hypothetical purchase question – not a real one. If you say that you would purchase the diabetes service, you will not actually purchase the service or pay any money at the end of our interview. But, I would like for you to respond to the question as though your response would involve a real cash payment. And that is the problem. In most studies of this kind, folks seem to have a hard time doing this. They respond differently to a hypothetical purchase question, where they do not really have to pay money, than they do in a real purchase situation where they actually will have to pay money. For example, in a recent study, a group of patients were asked if they were willing to purchase a disease management service that is similar to the diabetes disease management service that I have just described to you. Payment was hypothetical for these patients, as it will be for you. No one had to pay money if they said they would buy the disease management service. Another similar group of patients also participated in this study. These patients were offered the opportunity to actually purchase the disease management service at the same price. If patients in this second group agreed to purchase the programme they really did have to pay money. On average, more patients said ‘yes’ when the purchase question was hypothetical than when it was real. We call this ‘hypothetical bias’. ‘Hypothetical bias’ is the difference that we continually see in the way people respond to hypothetical situations as compared to real situations – people seem to respond differently to purchase questions when they really don't have to pay money as a result of their response. In the real offer to purchase the disease management service, where people knew they would have to pay money if they said ‘yes’, fewer said ‘yes’ than when payment was hypothetical and people knew they would not pay anything if they said ‘yes’. How can we get people to think about their response to a hypothetical purchase question and respond as if it was a real purchase decision, where if they agree to the purchase they will really have to pay the price? How do we get them to think about what it means to really pay money, if in fact they really aren't going to have to do it? Let me tell you why I think we continually see this hypothetical bias, why people behave differently in a hypothetical purchase situation than they do when the purchase situation is real. I think that when we respond to hypothetical purchase questions, we give some thought to what we might do, but we know we can always change our minds especially if we don't want to buy. But, when the purchase offer is real, and we would actually have to spend our money if we say yes, we think a different way. When we are faced with the possibility of having to spend money, we think about our options: if I spend money on this, that is money I don't have to spend on other things. If I spend money on a diabetes management service, that is money I don't have to spend on groceries, go to a movie, or perhaps spend on some other way of improving my diabetes. So when the payment is real we answer in a way that takes into account the limited amount of money we have. We answer realising that we just don't have enough money to do everything we might like to do. This is just my opinion, of course, but it is what I think may be going on in hypothetical purchase questions. In any case, the only way that we know to get people like you to respond to our hypothetical purchase question just like you would respond if the purchase offer was real is to simply ask you: when you reply to the hypothetical purchase question below, please do the following: Think about what you are replying to. If this were real and you said yes, you would actually have to pay $40 right now – do you really want the diabetes management service enough that you would be willing to spend the money? If I were in your shoes, and I were asked whether or not I would purchase the diabetes management service that was just described at a price of $40, I would think about how I feel about spending my money this way. When I got ready to answer the question, I would ask myself: if this were a real offer to purchase the diabetes service, and I had to pay $40 if I said yes: do I really want to spend my money this way? If I really did, I would say yes; if I did not, I would say no – I wouldn't throw my money around. That is just my opinion, of course. You must do whatever you want to do. In any case, I ask you to respond just exactly as you would respond if you were really going to face the consequences of your response: which is to pay money if you say yes. Please keep this in mind when responding to the question. So, assume that you are being offered the opportunity to purchase the diabetes disease management service that was described to you. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. Would you buy this service here and now at a price of $ 40? Please circle your answer below. Yes No Author notes " An earlier draft of this article was presented at the iHEA 5th World Congress held July 10–13, 2005 in Barcelona. We are grateful to Jason Shogren and Laura Taylor for detailed suggestions on the design of our experiment and the cheap talk script. For comments, we thank Richard Bishop, Patricia Champ, Tommy Gärling, Glenn Harrison, John List, John Whitehead, Kip Viscusi, seminar participants at the University of Connecticut and Appalachian State University, Editor Leonardo Felli, and three anonymous referees. We also thank Niklas Zethraeus for research assistance and Ronald Langley for suggestions about certainty statements. Finally, we are indebted to the American Pharmacy Services Corporation and the following pharmacists who participated in this experiment: Matt Cull, Drane Stephens, Aaron McIntosh, Kathy Salyer, Steve and Alicia Dawson, Gary Hamm, Alyson and Leon Claywell, Jason Wallace and Melodie Hawkins. This research was supported by Cooperative Agreement Number E11/CCE421825 from the Centers for Disease Control and Prevention (CDC). Views and contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC. Magnus Johannesson received financial support from the Swedish Research Council. © The Author(s). Journal compilation © Royal Economic Society http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png The Economic Journal Oxford University Press http://www.deepdyve.com/lp/oxford-university-press/eliciting-willingness-to-pay-without-bias-evidence-from-a-field-ON3pk0WEI7

Loading next page...

References (80)

Robert Mitchell, R. Carson (1989)
Using Surveys to Value Public Goods: The Contingent Valuation Method
R. Cummings, S. Elliott, G. Harrison, James Murphy (1997)
Are Hypothetical Referenda Incentive Compatible?
Journal of Political Economy, 105
(2004)
?Examining the role of social isolation on stated preferences?
American Economic Review, 94
D. Ariely, G. Loewenstein, D. Prelec (2006)
The Construction of Preference: “Coherent Arbitrariness”: Stable Demand Curves Without Stable Preferences
M. Johannesson, G. Blomquist, K. Blumenschein, P. Johansson, B. Liljas, R. O'Conor (1999)
Calibrating Hypothetical Willingness to Pay Responses
Journal of Risk and Uncertainty, 18
Christian Vossler, Robert Ethier, Gregory Poe, Michael Welsh (2003)
Payment Certainty in Discrete Choice Contingent Valuation Responses: Results from a Field Validity Test
Southern Economic Journal, 69
J. List, C. Gallet (2001)
What Experimental Protocol Influence Disparities Between Actual and Hypothetical Stated Values?
Environmental and Resource Economics, 20
David Aadland, A. Caplan (2006)
Cheap Talk Reconsidered: New Evidence From CVM
, 20
M. Bittner, Stuart Haines (1997)
Pharmacy-based diabetes management: a practical approach.
Journal of the American Pharmaceutical Association, NS37 4
Christian Vossler, Gregory Poe (2005)
Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty: a comment
Journal of Environmental Economics and Management, 49
R. Berrens, A. Bohara, Joe Kerkvliet, J. List (2004)
Examining the Role of Social Isolation on Stated Preferences
Artefactual Field Experiments
A. Roth (1987)
Laboratory experimentation in economics: List of contributors
R. Ready, S. Navrud, Rw. Dubourg (2001)
How Do Respondents with Uncertain Willingness to Pay Answer Contingent Valuation Questions?
Land Economics, 77
J. Kagel, A. Roth (1997)
Handbook of Experimental Economics
Henry Schultz, O. Lange, F. McIntyre, T. Yntema (1942)
Studies in mathematical economics and econometrics : in memory of Henry Schultz
Economica, 10
(1995)
citation_publisher=Cambridge University Press, ; Evaluating Health Risks: An Economic Approach
Michael Welsh, Gregory Poe (1998)
Elicitation Effects in Contingent Valuation: Comparisons to a Multiple Bounded Discrete Choice Approach
Journal of Environmental Economics and Management, 36
G. Harrison (2005)
Experimental Evidence on Alternative Environmental Valuation Methods
Environmental and Resource Economics, 34
R. Hertwig, A. Ortmann (2001)
Experimental practices in economics: A methodological challenge for psychologists?
Behavioral and Brain Sciences, 24
David Aadland, A. Caplan (2003)
Cheap Talk Revisited: New Evidence From Cvm
Gregory Poe, Jeremy Clark, D. Rondeau, W. Schulze (2002)
Provision Point Mechanisms and Field Validity Tests of Contingent Valuation
Environmental and Resource Economics, 23
B. Kriström (1990)
A non-parametric approach to the estimation of welfare measures in discrete response valuation studies.
Land Economics, 66
D. Raden (1985)
Strength-related attitude dimensions
Social Psychology Quarterly, 48
D. Kennedy, R. Small (2002)
Development and implementation of a smoking cessation clinic in community pharmacy practice.
Journal of the American Pharmaceutical Association, 42 1
citation_publisher=Elsevier Science, ; Handbook of Experimental Economics Results
R. Fazio, M. Zanna (1978)
On the predictive validity of attitudes: The roles of direct experience and confidence1
Journal of Personality, 46
J. List (2001)
Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards
Framed Field Experiments
P. Turner (1999)
Journal of the American Pharmaceutical Association Guidelines for Authors
Journal of the American Pharmaceutical Association, 39
T. Haab, Ju-Chin Huang, J. Whitehead (1999)
Are Hypothetical Referenda Incentive Compatible? A comment
Journal of Political Economy, 107
M. Tambour, N. Zethraeus (1998)
Nonparametric Willingness-to-pay Measures and Confidence Statements
Medical Decision Making, 18
K. Blumenschein, Magnus Johannesson, KK Yokoyama, P. Freeman (2001)
Hypothetical versus real willingness to pay in the health care sector: results from a field experiment.
Journal of health economics, 20 3
G. Harrison, E. Rutström (2008)
Chapter 81 Experimental Evidence on the Existence of Hypothetical Bias in Value Elicitation Methods
, 1
Robert Ethier, Gregory Poe, W. Schulze, Jeremy Clark (1997)
A Comparison of Hypothetical Phone and Mail Contingent Valuation Responses for Green-Pricing Electricity Programs
Land Economics, 76
P. Johansson (1995)
Evaluating Health Risks: An Economic Approach
R. Carson, Nicholas Flores, N. Meade (2000)
Contingent Valuation: Controversies and Evidence
Environmental and Resource Economics, 19
(1995)
citation_publisher=Princeton University Press, ; The Handbook of Experimental Economics
(2001)
citation_publisher=Elsevier Science, ; International Encyclopedia of Social and Behavioral Sciences
, 19
T. Brown, I. Ajzen, D. Hrubes (2003)
Further tests of entreaties to avoid hypothetical bias in referendum contingent valuation
Journal of Environmental Economics and Management, 46
R. D'Agostino, Warren Chase, A. Belanger (1988)
The Appropriateness of Some Common Procedures for Testing the Equality of Two Independent Binomial Populations
The American Statistician, 42
M. Johannesson, P. Johansson, B. Kriström, U. Gerdtham (1993)
Willingness to pay for antihypertensive therapy--further results.
Journal of health economics, 12 1
R. Carson (2001)
Resources and Environment: Contingent Valuation
Mary Evans, Nicholas Flores, K. Boyle (2003)
Multiple-Bounded Uncertainty Choice Data as Probabilistic Intentions
Land Economics, 79
Joel Giles, D. Kennedy, E. Dunn, Whitney Wallace, Stephanie Meadows, A. Cafiero (2001)
Results of a Community Pharmacy‐Based Breast Cancer Risk‐Assessment and Education Program
Pharmacotherapy: The Journal of Human Pharmacology and Drug Therapy, 21
R. Ready, J. Whitehead, G. Blomquist (1995)
Contingent Valuation When Respondents Are Ambivalent
Journal of Environmental Economics and Management, 29
(1999)
surveys: is the dichotomous choice approach incentive-compatible? American Economic Review
J. Sample, R. Warland (1973)
Attitude and Prediction of Behavior
Social Forces, 51
A. Alberini, K. Boyle, Michael Welsh (2003)
Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty
Journal of Environmental Economics and Management, 45
B. Bluml, James McKenney, M. Cziraky (2000)
Pharmaceutical care services and results in project ImPACT: hyperlipidemia.
Journal of the American Pharmaceutical Association, 40 2
P. Champ, R. Bishop (2001)
Donation Payment Mechanisms and Contingent Valuation: An Empirical Study of Hypothetical Bias
Environmental and Resource Economics, 19
Ariely (2003)
???Coherent arbitrariness??: stable demand curves without stable preferences?
Quarterly Journal of Economics, 118
(1987)
citation_publisher=Cambridge University Press, ; Laboratory Experimentation in Economics: Six Points of View
Colin Camerer, R. Hogarth (1999)
The Effects of Financial Incentives in Experiments: A Review and Capital-Labor-Production Framework
Journal of Risk and Uncertainty, 19
(1942)
citation_publisher=University of Chicago Press, ; Studies in Mathematical Economics and Econometrics in Memory of Henry Schultz
M. Johannesson, B. Liljas, P. Johansson (1998)
An experimental comparison of dichotomous choice contingent valuation questions and real purchase decisions
Applied Economics, 30
R. Cummings, G. Harrison, E. Rutström (1995)
Homegrown Values and Hypothetical Surveys: Is the Dichotomous Choice Approach Incentive-Compatible?
The American Economic Review, 85
Don Glines (1983)
Resources on the Future.
Educational Leadership, 41
I. Ajzen, T. Brown, Franklin Carvajal (2004)
Explaining the Discrepancy between Intentions and Actions: The Case of Hypothetical Bias in Contingent Valuation
Personality and Social Psychology Bulletin, 30
(1989)
Environmental and Resource Economics, vol
R. Cummings, Laura Taylor (1999)
Unbiased Value Estimates for Environmental Goods: A Cheap Talk Design for the Contingent Valuation Method
The American Economic Review, 89
W. Viscusi, W. Evans (1990)
Utility Functions That Depend on Health Status: Estimates and Economic Implications
The American Economic Review, 80
R. Fazio, M. Zanna (1978)
Attitudinal qualities relating to the strength of the attitude-behavior relationship☆
Journal of Experimental Social Psychology, 14
P. Champ, R. Bishop, T. Brown, Daniel Mccollum (1997)
Using Donation Mechanisms to Value Nonuse Benefits from Public Goods
Journal of Environmental Economics and Management, 33
(2001)
Contingent valuation, resources and environmental in International Encyclopedia of Social and Behavioral Sciences
(2001)
?Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards?
American Economic Review, 91
ÔThe theory of planned behaviorÕ
C. Plott, V. Smith (2008)
Handbook of Experimental Economics Results
Richard Carson (1997)
Contingent Valuation: Theoretical Advances and Empirical Tests Since the NOAA Panel
American Journal of Agricultural Economics, 79
(1996)
?Contingent valuation and revealed preference methodologies: comparing the estimates for quasi?public goods?
Land Economics, 72
P. Diamond, J. Hausman (1994)
Contingent Valuation: Is Some Number Better than No Number?
Journal of Economic Perspectives, 8
S. Fujii, T. Gärling (2003)
Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis
Transportation Research Part A-policy and Practice, 37
L. Posey (2003)
Proving that pharmaceutical care makes a difference in community pharmacy.
Journal of the American Pharmaceutical Association, 43 2
James Murphy, Thomas Stevens, Darryl Weatherhead (2004)
Is Cheap Talk Effective at Eliminating Hypothetical Bias in a Provision Point Mechanism?
Environmental and Resource Economics, 30
(1994)
American Economic Association
R. Bishop, T. Heberlein (1979)
Measuring Values of Extramarket Goods: Are Indirect Measures Biased?
American Journal of Agricultural Economics, 61
(1942)
ÔThe empirical derivation of indifference functionsÕ
(1989)
citation_publisher=Resources for the Future, ; Using Surveys to Value Public Goods: The Contingent Valuation Method
K. Blumenschein, M. Johannesson, G. Blomquist, B. Liljas, R. O'Conor (1998)
Experimental Results on Expressed Certainty and Hypothetical Bias in Contingent Valuation
Southern Economic Journal, 65
R. Thaler (1987)
Laboratory experimentation in economics: The psychology of choice and the assumptions of economics
M. Pauly (1994)
Universal health insurance in the Clinton plan: coverage as a tax-financed public good.
The journal of economic perspectives : a journal of the American Economic Association, 8 3
(1994)
?The contingent valuation debate: why economists should care?
Journal of Economic Perspectives, 8

Publisher: Oxford University Press
Copyright: © The Author(s). Journal compilation © Royal Economic Society
ISSN: 0013-0133
eISSN: 1468-0297
DOI: 10.1111/j.1468-0297.2007.02106.x
Publisher site: See Article on Publisher Site

Abstract

Abstract Concern exists that hypothetical willingness to pay questions overestimate real willingness to pay. In a field experiment, we compare two methods of removing hypothetical bias, a cheap talk approach and a certainty approach, with real purchases. We find evidence of hypothetical bias for unadulterated contingent valuation. Contingent valuation with certainty statements removes the hypothetical bias, but the cheap talk approach has no significant impact. Our findings suggest that willingness to pay can be accurately estimated by adding a simple follow‐up question about the certainty of responses and that cheap talk is not a generally effective approach. Information about the willingness to pay for non‐market goods is crucial for understanding the welfare implications of different policies. Ideally we would like to rely on actual choices for inferring willingness to pay. But for many public programmes, especially those in the environmental and health area, revealed preference information is limited. This limitation can arise from the lack of markets or third‐party financing. Researchers have therefore tried to estimate willingness to pay based on stated preferences that are hypothetical choices. This methodology, known as contingent valuation, typically involves a survey or experiment in which there is a detailed description of the good being offered, a description of how the good would be provided, a method for eliciting preferences for the good, follow up debriefing questions and questions about socioeconomic characteristics (Portney, 1994; Carson, 2001). Contingent valuation is still somewhat controversial in economics (Diamond and Hausman, 1994; Carson et al., 2001; Ariely et al., 2003). At the heart of this controversy is the extent to which hypothetical choices correspond to real economic choices.1Carson et al. (1996) find that values from contingent valuation and revealed preference studies match fairly well, yet strong counterevidence exists. For example, laboratory experiments by Cummings et al. (1995, 1997) comparing real and hypothetical willingness to pay suggest that hypothetical responses sometimes substantially overestimate willingness to pay. Meta‐analysis by List and Gallet (2001) and reviews by Harrison (2006) and Harrison and Rutström (forthcoming) also suggest overestimation. This overstatement problem is known as hypothetical bias. The evidence on hypothetical bias has stimulated research into various techniques for removing this bias. Two basic approaches have evolved. Cummings and Taylor (1999) introduced a cheap talk approach. With this approach a cheap talk script precedes the elicitation of willingness to pay. Cheap talk scripts include an explicit discussion about hypothetical bias. Subjects are told what hypothetical bias is, that it is a common problem in hypothetical valuation questions, and why it might occur. Furthermore, subjects are asked to adjust for hypothetical bias in responding to the willingness to pay question. After the subject has read or heard the cheap talk script he/she responds to the willingness to pay question. Cummings and Taylor provided evidence of the effectiveness of the cheap talk approach in experimental referenda about donations to public goods. Subsequent studies on the cheap talk approach have found mixed results. Brown et al. (2003), using a similar design as Cummings and Taylor, found that the effectiveness of the cheap talk approach varied with the price level. In a second‐price auction for sports cards, List (2001) found that the cheap talk approach removed the hypothetical bias for nondealers, but not for dealers. Ajzen et al. (2004) found that a cheap talk corrective entreaty eliminated hypothetical bias on a referendum to contribute to a college scholarship fund. In a study by Murphy et al. (2005) on a voluntary contribution mechanism, the cheap talk approach did not fully remove the hypothetical bias. Aadland and Caplan (2006) develop a general cheap talk script that is short and neutral with regard to the direction of hypothetical bias and find that it exacerbates, rather than mitigates, the bias. Concurrently a second approach to mitigating hypothetical bias was developed. This approach is based on respondent certainty about willingness to pay. Two different versions of this approach have been used. In the first version a scale is used to assess the degree of certainty in hypothetical willingness to pay responses. Such a certainty scale was first used by Champ et al. (1997), who compared hypothetical dichotomous choice questions about donating a specified amount to a public good with actual donations to the public good. They assessed the certainty of the hypothetical donation responses on a 1–10 scale from very uncertain to very certain. They found that hypothetical donations significantly exceeded real donations but that there was no significant difference when only subjects who were very certain of their yes response were counted as real yes responses. Similar results using a certainty scale were also found by Ethier et al. (2000), Champ and Bishop (2001), Poe et al. (2002) and Vossler et al. (2003).2 A drawback of using a scale to assess certainty is that it is necessary to estimate the cut‐off level of certainty at which a hypothetical decision corresponds to a real decision. In the second version of the certainty approach, hypothetical willingness to pay responses are divided based on a follow‐up question with two degrees of certainty. An early form of this version was used by Johannesson et al. (1998) who divided hypothetical yes responses to buy a box of chocolates into ‘fairly sure’ and ‘absolutely sure’ yes responses. They found that the percentage of absolutely sure yes responses was lower than the proportion of real yes responses and the approach provided a conservative estimate of willingness to pay.3 This certainty question was subsequently adjusted by Blumenschein et al. (1998), who divided hypothetical yes responses into ‘probably sure’ and ‘definitely sure’ responses. Only yes responses about which subjects were ‘definitely sure’ were treated as yes responses. Yes responses about which subjects were only ‘probably sure’ were treated the same as no responses. This approach has been effective in removing hypothetical bias in both laboratory and field experiments (Blumenschein et al., 1998, 2001). Furthermore, by studying individuals who first respond to a hypothetical willingness to pay question and subsequently face a real purchase decision, it has been demonstrated that the degree of certainty in hypothetical yes responses is a very strong predictor of whether a hypothetical yes response corresponds to a real yes response (Blumenschein et al., 1998; Johannesson et al., 1999). In the current study we provide new evidence from a field experiment about the effectiveness of the cheap talk approach and the certainty approach in calibrating hypothetical willingness to pay responses. The purpose of the field experiment is to compare: real payments with stated hypothetical willingness to pay, real payments with stated hypothetical willingness to pay elicited with a cheap talk entreaty, and real payments with stated hypothetical willingness to pay adjusted based on follow up certainty statements. The main contribution of the article is the direct comparison of the cheap talk and certainty approaches in a field experiment with real payments at several prices. For a field experiment, we have a relatively large sample of more than 260 individuals, and in parametric analysis we control for more than 20 personal characteristics including income, health, health behaviour and time costs. The field experiment is carried out under the same conditions as a high quality contingent valuation study. Face‐to‐face interviews are used, and the subjects value a non‐trivial good that is not available on the market. In an environment of social concern for obesity and diabetes we chose a diabetes management programme as the good. It is primarily a private good, but spillovers exist from third party financing of health care. The subjects in the experiment are divided into three treatments groups: real, stated preference after cheap talk, and stated preference with certainty follow up. Subjects receive a real or hypothetical offer to purchase a real, but previously unavailable, service. By varying the price across subjects we are able to estimate the aggregate demand curve (Bishop and Heberlein, 1979). Our findings suggest that the cheap talk approach is not a generally effective method of removing hypothetical bias, but that it is possible to accurately estimate willingness to pay by adding a simple follow‐up question about the certainty of responses. 1. Experimental Design The experiment involved three different treatments referred to as the ‘real group’, ‘the hypothetical group’ and the ‘cheap talk group’. Subjects in the real group were actually given the opportunity to purchase a pharmacist‐provided diabetes management programme. Subjects in the hypothetical group received a dichotomous choice contingent valuation question about purchasing the same diabetes management programme followed by a question about the certainty of their hypothetical responses (with the categories: probably sure/definitely sure). When subjects answered the willingness to pay question they did not know that they would subsequently be asked about the certainty of their willingness to pay responses. Subjects in the cheap talk group received a dichotomous choice contingent valuation question preceded by a cheap talk script. In all three groups the price was varied between $15, $40 and $80. Each subject received only one price offer. The questionnaires were pilot‐tested in a focus group of diabetics in Lexington, Kentucky prior to the study. The focus group was also used to determine the prices used in the experiment. The data collection was carried out as ‘face‐to‐face’ interviews, and the same trained interviewer carried out all the interviews.4 The study had approval from the University of Kentucky Medical Institutional Review Board. Details of the experiment are further described below. 1.1. The Good The good used in our experiment was a pharmacist provided, type‐2 diabetes management programme. This programme was designed to assist type‐2 diabetics in attaining optimal management of their diabetes, thereby enhancing their life quality and decreasing their utilisation of expensive health care services such as Emergency Department visits and hospitalisations. The profession of pharmacy is slowly progressing towards a ‘patient care’ focus of practice rather than the historical ‘drug dispensing’ focus. Pharmacists in a variety of practice settings are developing new clinical services and disease management programmes to help patients achieve desired health outcomes. These services range from blood lipid management and diabetes education to smoking cessation programmes and cancer risk assessment (Bluml et al., 2000; Rodriguez de Bittner and Haines, 1997; Kennedy and Small, 2002; Giles et al., 2001). Although new pharmacist‐provided patient care and disease management programmes are being developed, these services are rarely provided in the community pharmacies where most people encounter pharmacists (Posey, 2003). Thus, it is very unlikely that the general public has a pre‐conceived market price for such services. Furthermore, in the current health care environment in the US, pharmacist‐provided disease management is rarely included as a benefit in health plans. The scope of a comprehensive disease management programme, such as the one in this experiment, exceeds the ethical obligations and expectations for required pharmacist service provision; therefore, it is not unethical for a pharmacist to deny the programme to individuals who do not wish to pay for the service. 1.2. Subject Recruitment Subjects were recruited from nine pharmacies in the state of Kentucky in the US. Three pharmacies were included in each of the experimental groups. The three pharmacies in the real group agreed to deliver the diabetes management programme to any subject that purchased the service at the price offered. All of the pharmacists in the real group had received extensive training from the American Pharmacy Services Corporation Foundation for Education and Research on providing the diabetes management programme prior to the implementation of this study. Only one price was used at each pharmacy, as the pharmacists were unwilling to charge patients within the same pharmacy different prices. The pharmacists identified their type‐2 diabetes patients who were age 18 or older and who had received a prescription for a type‐2 diabetes medication in the past 6 months. Potential subjects were contacted by phone. After confirming the diagnosis of diabetes they were asked if they would participate in a scientific study that involved an interview of approximately 15–20 minutes. Individuals who agreed to participate in the study were given a mutually convenient appointment time for the interview, which was carried out in the pharmacy. As compensation, each subject received $25 upon completion of the survey. The interviews took place between May 1 and July 23, 2003. In total 267 patients were interviewed (90 in the real group, 91 in the hypothetical group and 86 in the cheap talk group). 1.3. The Questionnaires Subjects were first given a questionnaire with background questions to fill in. In all experimental groups, a written description of the pharmacist‐provided diabetes management programme was then given to the subject. (The description of the diabetes management programme is given in Appendix 1.) The interviewer read the programme description to the subject while the subject read along on his/her own copy. The interviewer responded to any questions the subject had regarding the service. Next, the interviewer gave the subject a written copy of the survey. The interviewer read the valuation/purchase question (including the cheap talk script in that treatment) to the subject and the subject marked his/her response on the survey form. Subsequently in the hypothetical group the interviewer also read the certainty follow up question to the subject, and again, the subject marked his/her response on the survey form.5 1.3.1. Real Group Subjects in the real group were given the opportunity to purchase the diabetes management programme at their pharmacy. After receiving a description of the diabetes management programme the following question was posed:6 ‘You are now being offered the opportunity to purchase the diabetes disease management service that was just described to you. All of the services that were described to you would be provided for one flat rate. If you choose to purchase the service, you will have to use some of your household income to pay for it here and now with cash, cheque or credit card. Will you buy this service here and now at a price of $40? Please circle your answer below.’ 1.3.2. Hypothetical Group Subjects in the hypothetical group received a hypothetical dichotomous choice contingent valuation question. The question consisted of a description of the diabetes management programme after which the following question was posed: ‘Assume that you are being offered the opportunity to purchase the diabetes disease management service that was just described to you. All of the services that were described to you would be provided for one flat rate. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. Would you buy this service here and now at a price of $40? Please circle your answer below.’ The dichotomous choice contingent valuation question was followed by a question in which the subjects were asked if they were ‘probably sure’ or ‘definitely sure’ of their yes (no) answer. This question appeared on the page following the willingness to pay question and was phrased in the following way: ‘If you answered YES (NO), are you ‘probably sure’ or ‘definitely sure’ that you would (not) buy the diabetes management service here and now at a price of $40? Please circle your answer below.’ 1.3.3. Cheap Talk Group Subjects in the cheap talk group received a hypothetical dichotomous choice contingent valuation question preceded by a cheap talk script. The cheap talk script explained hypothetical bias and asked subjects to state what they would really do. We modelled our cheap talk script after the script used in the Cummings and Taylor (1999) study, with some adaptation due to the differences in the valuation question (a dichotomous purchase question versus a dichotomous referendum question) and the good (a predominately private, health good versus a public, environmental good). Cummings and Taylor (1999) used two versions of the cheap talk script; one with numbers to illustrate hypothetical bias and one without specific numbers. These two scripts yielded statistically indistinguishable results in their study. We used the script without numbers, as it is more generally applicable.7 Such cheap talk has reduced hypothetical bias in some previous studies. (The full cheap talk script is given in Appendix 2.) The following excerpts from the script illustrate the approach: ‘…in a recent study, a group of patients were asked if they were willing to purchase a disease management service that is similar to the diabetes disease management service that I have just described to you. Payment was hypothetical for these patients, as it will be for you. No one had to pay money if they said they would buy the disease management service. Another similar group of patients also participated in this study. These patients were offered the opportunity to actually purchase the disease management service at the same price. If patients in this second group agreed to purchase the programme they really did have to pay money. On average, more patients said ‘yes’ when the purchase question was hypothetical than when it was real. We call this ‘hypothetical bias’. ‘Hypothetical bias’ is the difference that we continually see in the way people respond to hypothetical situations as compared to real situations… How can we get people to think about their response to a hypothetical purchase question and respond as if it was a real purchase decision, where if they agree to the purchase they will really have to pay the price? How do we get them to think about what it means to really pay money, if in fact they really aren't going to have to do it? Let me tell you why I think we continually see this hypothetical bias, why people behave differently in a hypothetical purchase situation than they do when the purchase situation is real. I think that when we respond to hypothetical purchase questions, we give some thought to what we might do, but we know we can always change our minds especially if we don't want to buy. But, when the purchase offer is real, and we would actually have to spend our money if we say yes, we think a different way. If I were in your shoes… I would think about how I feel about spending my money this way. When I got ready to answer the question, I would ask myself: if this were a real offer to purchase the diabetes service, and I had to pay $40 if I said yes: do I really want to spend my money this way? If I really did, I would say yes; if I did not, I would say no… In any case, I ask you to respond just exactly as you would respond if you were really going to face the consequences of your response: which is to pay money if you say yes. Please keep this in mind when responding to the question.’ After reading the cheap talk script the subjects responded to the dichotomous choice contingent valuation question. As in the hypothetical group, the dichotomous choice contingent valuation question was followed (on a separate page) by a question in which the subjects were asked if they were ‘probably sure’ or ‘definitely sure’ of their yes (no) answer. 1.4. Hypotheses and Tests To test the null hypothesis of no hypothetical bias, the percentage of yes responses is compared between the experimental groups. Three comparisons are carried out: hypothetical versus real, hypothetical definitely sure versus real, and cheap talk versus real. The first comparison tests for the presence of hypothetical bias in a standard, dichotomous choice willingness to pay question and the other two comparisons test if the certainty approach and/or the cheap talk approach removes the hypothetical bias. A contingency table Pearson chi‐square test is used to compare the percentage of yes responses between the groups (D'Agostino et al., 1988). While this test is informative, the drawback of the non‐parametric test is that it does not control for any differences in personal characteristics between the groups. Therefore, we also carry out logistic regression analysis to test if the probability of a yes response differs between the groups, controlling for the background variables collected in the study.8 The price of the diabetes management programme is included in the regression analysis in order to derive the demand curve and estimate the mean willingness to pay. The mean willingness to pay is also estimated non‐parametrically using the method developed by Kriström (1990). In the regression analysis we control for a host of personal characteristics collected in the study. To test and control for an income effect we include annual household income, household size and an indicator for wealth (if the subject owns his or her residence).9 To control for differences in health and health behaviour we include the following variables: previous participation in a diabetes management programme, membership in a diabetes support group, time since diabetes was diagnosed, perceived diabetes severity (mild, moderate, or severe), if the subject has suffered from any of the following complications from diabetes: cardiovascular disease, renal disease, vision problems or neuropathies; if any family member has suffered from complications of diabetes, smoking, body mass index, an indicator for whether the subject knows his/her haemoglobin A1c level, and perceived general health status (excellent, very good, good, fair, poor).10 We also control for the following socioeconomic characteristics: age, gender, years of education and ethnic background. Finally, we control for an indicator of the time cost of participating in the diabetes management programme (the travel time to the pharmacy).11 2. Results 2.1. Background Characteristics Table 1 provides the background characteristics for the three experimental groups.12 There are great similarities among the experimental groups in terms of background characteristics such as household income, household size, home ownership, age and ethnic background. In only three cases is the difference between groups significant at the 5% level. We control for differences in the regression analysis. Table 1 Background Characteristics (standard deviations in parentheses) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) Open in new tab Table 1 Background Characteristics (standard deviations in parentheses) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) . Real group (n = 90) . Hypothetical group (n = 91) . Cheap talk group (n = 86) . p‐value real vs hypothetical difference . p‐value real vs cheap talk difference . Income and wealth: Annual household income ($1,000) 31.34 30.58 31.26 0.867 0.988 (32.98) (26.54) (27.38) Household size 2.43 2.54 2.14 0.597 0.089 (1.29) (1.38) (0.97) Owns residence (%) 81.11 83.52 73.26 0.671 0.214 Health and health behaviour: Previous participation in disease management (%) 10.00 8.79 8.14 0.780 0.668 Member of diabetes support group (%) 3.33 2.20 4.65 0.641 0.655 Time with diabetes (years) 8.49 7.99 10.16 0.625 0.160 (6.95) (6.72) (8.61) Diabetes severity: 0.616 0.701 Mild (%) 33.33 27.47 34.88 Moderate (%) 53.33 60.44 55.81 Severe (%) 13.33 12.09 9.30 Cardiovascular disease (%) 84.44 81.32 83.72 0.577 0.896 Renal disease (%) 13.33 3.30 8.14 0.014 0.267 Vision problems (%) 41.11 34.07 39.53 0.328 0.831 Neuropathies (%) 57.78 59.34 50.00 0.831 0.301 Complications of diabetes in family (%) 57.78 53.85 56.98 0.594 0.914 Smoking (%) 30.00 18.68 33.72 0.076 0.596 Body mass index 34.01 31.91 33.82 0.064 0.882 (8.62) (6.23) (7.97) Know their haemoglobin A1c level (%) 18.89 19.78 19.77 0.879 0.883 General health: Excellent (%) 3.33 0.00 1.16 0.018 0.390 Very good (%) 12.22 13.19 13.95 Good (%) 28.89 36.26 38.37 Fair (%) 31.11 41.76 31.40 Poor (%) 24.44 8.79 15.12 Socioeconomics: Age (years) 56.71 59.98 59.04 0.093 0.253 (13.00) (13.05) (13.88) Women (%) 60.00 68.13 69.77 0.254 0.175 Education (years) 11.04 11.80 12.09 0.138 0.028 (3.24) (3.59) (3.09) Ethnic background white (%) 90.00 92.31 90.70 0.584 0.876 Time cost: Travel time to pharmacy (minutes) 11.87 13.63 12.88 0.177 0.465 (8.99) (8.41) (9.20) Open in new tab 2.2. Experimental Results The experimental results are shown in Table 2. In the real group 45% of the subjects bought the diabetes management programme at a price of $15, 23% bought the programme at a price of $40 and 10% of the subjects bought the programme at a price of $80. If there is hypothetical bias, then these percentages will be higher in the hypothetical group. We find this is clearly the case; the percentage of yes responses is higher at all prices in the hypothetical group. Overall the percentage of ‘buyers’ is about twice as high in the hypothetical group (45% versus 26%), and this difference is highly significant. Table 2 Number (%) of Yes Responses in the Experimental Groups Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 *p‐value of the difference compared to the yes responses in the real group. Open in new tab Table 2 Number (%) of Yes Responses in the Experimental Groups Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 Price . Real group Number (%) . Hypothetical group: All yes responses . Cheap talk group . Hypothetical group: Definitely sure yes responses . Number (%) . p‐value* . Number (%) . p‐value* . Number (%) . p‐value* . $15 13/29 (45) 22/31 (71) 0.040 17/29 (59) 0.293 11/31 (35) 0.460 $40 7/30 (23) 14/34 (41) 0.129 11/30 (37) 0.260 11/34 (32) 0.423 $80 3/31 (10) 5/26 (19) 0.301 11/27 (41) 0.006 0/26 (0) 0.103 All 23/90 (26) 41/91 (45) 0.006 39/86 (45) 0.006 22/91 (24) 0.830 *p‐value of the difference compared to the yes responses in the real group. Open in new tab The overall percentage of yes responses in the cheap talk group (45%) is the same as in the hypothetical group; the hypothetical bias is substantial in the cheap talk group also. The certainty approach seems to work better in terms of reducing hypothetical bias. The overall percentage of definitely sure yes responses is close to the percentage of real yes responses (24% versus 26%), and we cannot reject the null hypothesis of no difference in this case. The results of the parametric tests using logistic regression analysis are shown in Table 3.13 In the regressions we include dummy variables for the experimental groups. We run two regression equations for the two interpretations of yes responses in the hypothetical group. The first one considers all yes responses to be yes responses in the hypothetical group. The second one considers only the definitely sure yes responses to be yes responses, i.e., the probably sure yes responses are coded as ‘no’. In the first regression, the dummy variable for the hypothetical group is highly significant, and the marginal effect is 23.4 percentage points.14 Also the dummy variable for the cheap talk group is highly significant, with a marginal effect of 24.9 percentage points. In the second regression, the one that considers only definitely sure yes responses to be yes responses, the hypothetical dummy variable is close to zero and not significant. These results imply that the follow up certainty approach is effective in removing the hypothetical bias. Table 3 Results of Logistic Regression Analysis of the Effect of Experimental Group on the Probability of a Yes Response . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab Table 3 Results of Logistic Regression Analysis of the Effect of Experimental Group on the Probability of a Yes Response . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 . All yes responses in hypothetical group (Equation 1) . Definitely sure yes responses in hypothetical group (Equation 2) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −2.545 1.933 0.188 −1.464 2.040 0.473 Hypothetical group 1.046 0.400 0.009 0.234 −0.103 0.416 0.804 −0.019 Cheap talk group 1.114 0.405 0.006 0.249 1.215 0.411 0.003 0.229 Price −0.033 0.007 <0.001 −0.007 −0.033 0.007 <0.001 −0.006 Household income 0.016 0.006 0.014 0.004 0.017 0.007 0.012 0.003 Household size −0.324 0.156 0.038 −0.073 −0.378 0.169 0.025 −0.071 Owns residence 1.032 0.425 0.015 0.231 1.181 0.459 0.010 0.223 Disease management 0.699 0.516 0.176 0.156 0.795 0.540 0.141 0.150 Diabetes support group 2.047 0.924 0.027 0.458 1.745 0.835 0.037 0.329 Time with diabetes −0.002 0.022 0.928 −0.005 −0.008 0.023 0.732 −0.001 Moderate diabetes* 0.463 0.359 0.197 0.104 0.368 0.372 0.322 0.069 Severe diabetes* 0.861 0.564 0.127 0.193 0.707 0.592 0.232 0.134 Cardiovascular disease −0.013 0.443 0.976 −0.003 −0.082 0.462 0.858 −0.016 Renal disease −1.102 0.668 0.099 −0.246 −0.826 0.679 0.224 −0.156 Vision problems 0.087 0.352 0.804 0.020 −0.089 0.364 0.808 −0.017 Neuropathies 0.196 0.358 0.584 0.044 0.619 0.376 0.100 0.117 Complications of diabetes in family −0.221 0.318 0.486 −0.049 −0.264 0.327 0.419 −0.050 Smoking −0.055 0.379 0.885 −0.012 −0.050 0.398 0.899 −0.009 Body mass index 0.032 0.022 0.146 0.007 0.024 0.022 0.275 0.005 Know haemoglobin A1c 0.887 0.402 0.027 0.198 0.876 0.404 0.030 0.165 Good health† 1.303 0.517 0.012 0.291 0.898 0.514 0.080 0.170 Fair health† 0.773 0.529 0.144 0.173 0.457 0.527 0.385 0.086 Poor health† 1.094 0.665 0.100 0.245 0.585 0.674 0.385 0.110 Age −0.001 0.015 0.959 −0.0002 −0.002 0.016 0.902 −0.0004 Woman 0.204 0.335 0.543 0.046 0.081 0.347 0.816 0.015 Education −0.022 0.059 0.705 −0.005 −0.053 0.063 0.403 −0.010 Ethnic: white −0.454 0.560 0.417 −0.102 −0.362 0.589 0.539 −0.068 Travel time 0.014 0.019 0.466 0.003 0.001 0.020 0.961 0.002 Number of obs. 267 267 Chi‐square 77.518 <0.001 70.319 <0.001 Log‐likelihood −139.281 −131.112 McFadden pseudo‐R2 0.218 0.211 % individual prediction 76.030 76.030 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab We can see also that there is a highly significant effect of price in the regression equations, consistent with a downward sloping demand curve. The price coefficient measures the slope of the demand curve. It has been argued that the slopes will differ for hypothetical and real willingness to pay data, due to a higher variance in the hypothetical willingness to pay (Haab et al., 1999). In our data there is a tendency for the hypothetical responses to be closer to the real responses for higher prices than for lower prices whereas with cheap talk it is the other way round. These tendencies would suggest that the slopes of the demand curves may differ between the three experimental groups. Therefore, we tested adding interaction terms between the price and the experimental group dummy variables, allowing the slope of the aggregate demand function to vary between the three experimental groups. The interaction terms are, however, not significant at the 10% level (neither individually nor jointly), and we cannot reject the null hypothesis of equal slopes between the three experimental groups (for either equation 1 or equation 2 in Table 3).15 There is also a significant effect of income and the indicator for wealth. Higher income and wealth shift the demand curve outwards, consistent with economic theory. The negative effect of household size on demand might also reflect an income effect (as the income per household member decreases with household size). Membership in a diabetes support group is also associated with a significantly higher demand for the diabetes management programme, as is knowledge about the haemoglobin A1c level. These variables might reflect a greater concern for the impact of diabetes. There is also a tendency for an effect of the perceived general health status, with a higher demand for subjects with a worse overall health status. Patients with a worse health status may derive greater health gains from participating in the diabetes management programme. As suggested by the work of Viscusi and Evans (1990), a worse health status may also decrease the marginal utility of income, which will increase the willingness to pay for a given health gain. The follow‐up question about the certainty of yes/no responses was also included in the cheap talk group. The overall percentage of definitely sure yes responses is 30% in the cheap talk group, which according to a chi‐square test is not significantly different from the percentage of real yes responses (26%; p = 0.489) or the percentage of definitely sure yes responses in the hypothetical group (24%; p = 0.365). We also re‐estimated the second regression equation in Table 3 counting only definitely sure yes as yes responses in the cheap talk group. The cheap talk group dummy variable is not significant in that regression equation (coefficient = 0.406; p = 0.322; marginal effect = 0.068). We also tested pooling the definitely sure yes responses in the hypothetical group and the cheap talk group, including a dummy variable for this joint sample (comparing definitely sure yes responses for the pooled hypothetical and cheap talk groups with the real yes responses). This dummy variable is not significant (coefficient = 0.108; p = 0.760; marginal effect = 0.018). The overall percentage of definitely sure yes responses in the pooled sample is 27%, which according to a chi‐square test is not significantly different from the percentage of real yes responses (26%; p = 0.785). 2.3. Aggregate Demand Curves and Willingness To Pay In Figures 1 and 2 we show the non‐parametric and parametric aggregate demand curves. The parametric demand curves are based on the logistic regression equations in Table 3 and are estimated at the mean of the covariates. The Figures confirm the previous results. The hypothetical and cheap talk demand curves are relatively close and both overestimate the actual demand. In contrast to cheap talk, the certainty calibration through recoding leads to a demand curve that is close to the real demand curve. Fig. 2. Open in new tabDownload slide Parametric Demand Curves Fig. 2. Open in new tabDownload slide Parametric Demand Curves Fig. 1. Open in new tabDownload slide Non‐parametric Demand Curves Fig. 1. Open in new tabDownload slide Non‐parametric Demand Curves The mean willingness to pay can be estimated as the area below the demand curves, and the estimated means are shown in Table 4.16 Using the parametric method the mean real willingness to pay is $22 and the mean hypothetical willingness to pay is $42. The cheap talk approach yields a mean willingness to pay of $44, whereas the certainty approach yields a mean willingness to pay of $20. The non‐parametric and the parametric methods yield similar estimates of the mean willingness to pay especially for the real and definitely sure hypothetical willingness to pay. The difference of $2 between the real and definitely sure willingness to pay, for both the non‐parametric and parametric methods, is not statistically different from zero. Table 4 Mean Willingness to Pay in the Experimental Groups ($) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) *p‐value of the difference compared to the mean willingness to pay in the real group and 95% confidence interval of the difference. Open in new tab Table 4 Mean Willingness to Pay in the Experimental Groups ($) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) . Real group Mean (SE) . Hypothetical group: all yes responses . Cheap talk group . Hypothetical group: definitely sure yes responses . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Mean (SE) . p‐value (95% confidence interval for difference)* . Non‐parametric method 21.85 36.74 0.005 36.38 0.010 20.27 0.759 (3.73) (3.84) (4.40–25.38) (4.21) (3.50–25.56) (3.55) (−11.67–8.51) Parametric method 22.11 42.36 0.010 43.90 0.007 19.77 0.715 (4.71) (6.34) (4.77–35.73) (6.56) (5.96–37.62) (4.34) (−14.89–10.21) *p‐value of the difference compared to the mean willingness to pay in the real group and 95% confidence interval of the difference. Open in new tab 2.4. Sensitivity Analysis As an additional test of whether or not the definitely sure yes responses differ from the real yes responses, we compare the effects of all covariates between the real group and the definitely‐sure hypothetical group. This is done by adding interaction terms between the hypothetical group dummy variable and all the other variables; equivalent to running separate regression functions for the real group and the definitely sure hypothetical group.17 With a likelihood ratio test we test if the hypothetical dummy variable and the interaction terms improve the significance of the model. We cannot reject the null hypothesis of no significant difference between the regression models at the 10% level.18 Our results suggest that probably sure and definitely sure yes responses are quite different (i.e. probably sure yes responses correspond to real ‘no’ responses). To investigate this further, we estimate two separate regressions on the hypothetical group data. The first regression (n = 69) counts probably sure yes responses as yes and excludes the definitely sure yes responses, and the second regression (n = 72) counts definitely sure yes responses as yes and excludes the probably sure yes responses. If the two categories of yes responses are identical the estimated coefficients and explanatory power should be similar for the two regressions. The results of these two regressions, shown in Table 5, differ substantially. In the regression on probably sure yes responses, the McFadden pseudo‐R2 is 0.233, no variable is significant at the 5% level, and the regression equation is far from significant (chi‐square 18.895 (23 df); p = 0.707). In the regression on definitely sure yes responses the McFadden pseudo‐R2 is 0.584 and the regression equation is highly significant (chi‐square = 51.804 (23 df); p = 0.001). Both the price variable and the wealth indicator are also highly significant with the expected signs. Household income has the expected sign but is not quite significant (p = 0.187).19 Table 5 Results of Logistic Regression Analysis Comparing the Probably Sure and Definitely Sure Yes Responses in the Hypothetical Group . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab Table 5 Results of Logistic Regression Analysis Comparing the Probably Sure and Definitely Sure Yes Responses in the Hypothetical Group . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 . Probably sure yes responses = 1 (definitely sure yes responses excluded) . Definitely sure yes responses = 1 (probably sure yes responses excluded) . β . SE . p‐value . Marginal effect . β . SE . p‐value . Marginal effect . Constant −10.018 6.708 0.135 2.147 8.016 0.789 Price −0.022 0.016 0.150 −0.003 −0.155 0.050 0.002 −0.004 Household income 0.013 0.026 0.620 0.002 0.058 0.044 0.187 0.002 Household size 0.038 0.431 0.929 0.006 −0.367 0.671 0.585 −0.010 Owns residence −0.352 1.260 0.780 −0.054 8.453 3.233 0.009 0.241 Disease management 0.115 1.230 0.926 0.018 2.512 2.670 0.347 0.071 Time with diabetes 0.035 0.075 0.644 0.005 −0.345 0.163 0.034 −0.010 Moderate diabetes* 1.434 1.063 0.177 0.222 2.991 2.090 0.152 0.085 Severe diabetes* 1.731 1.424 0.224 0.268 8.813 4.229 0.037 0.251 Cardiovascular disease 1.148 1.160 0.322 0.178 −0.521 1.979 0.792 −0.015 Vision problems −0.341 0.937 0.716 −0.053 −2.883 2.022 0.154 −0.082 Neuropathies −1.876 1.015 0.067 −0.290 2.328 1.739 0.181 0.066 Complications of diabetes in family 1.187 0.951 0.212 0.184 0.504 1.370 0.713 0.014 Smoking −0.209 1.055 0.843 −0.032 1.815 1.928 0.346 0.052 Body mass index 0.072 0.071 0.310 0.011 −0.041 0.096 0.667 −0.001 Know haemoglobin A1c 0.423 1.138 0.710 0.065 −0.327 1.462 0.823 −0.009 Good health† 2.890 1.811 0.111 0.447 1.810 2.204 0.412 0.052 Fair health† 2.205 1.826 0.227 0.341 −0.661 2.440 0.786 −0.019 Poor health† 3.248 2.289 0.156 0.503 −3.533 4.164 0.396 −0.101 Age 0.019 0.046 0.676 0.003 0.035 0.060 0.563 0.001 Woman 1.090 1.095 0.319 0.169 −0.738 1.127 0.512 −0.021 Education 0.091 0.135 0.499 0.014 −0.499 0.291 0.086 −0.014 Ethnic: white −0.518 1.281 0.686 −0.080 0.941 2.531 0.710 0.027 Travel time 0.050 0.406 0.218 0.008 −0.211 0.130 0.103 −0.006 Number of obs. 69 72 Chi‐square 18.895 0.707 51.804 0.001 Log‐likelihood −31.160 −18.414 McFadden pseudo‐R2 0.233 0.584 % individual prediction 79.710 84.722 *Baseline category: mild diabetes; †Baseline category: Excellent or very good general health. Open in new tab In the experimental results in Table 2 the percentage of yes responses at the $80 price in the cheap talk group seems high; i.e. substantially higher than in the hypothetical group and somewhat higher than at the $40 price in the cheap talk group. We therefore investigated if the background characteristics differed in the $80 cheap talk group compared to the rest of the sample. There is a significant difference at the 5% level for only two of the over twenty background characteristics. The number of subjects that had previously participated in a diabetes management programme is higher in the cheap talk group at the $80 price (22% versus 8%) and the number of patients with vision problems is higher in this group (59% versus 36%). Both of these variables are controlled for in the regression analysis. In the regression analysis, the vision problem variable is far from significant and has a point estimate that is close to zero. The variable for previous participation in a diabetes management programme has a sizeable positive point estimate but is not quite significant. Six of the 27 subjects at the $80 price in the cheap talk group had previously participated in a diabetes management programme and four of them said yes in the hypothetical willingness to pay question. This may be one explanation for the high rate of yes responses in the cheap talk group at the $80 price. As an extra sensitivity analysis we therefore re‐estimated our results excluding all subjects that had previously participated in a diabetes management programme (24 subjects). After this exclusion the fraction of yes responses in the cheap talk group is 61% at $15, 37% at $40 and 33% at $80, and the overall fraction of yes responses is still significantly higher than in the real group (44% in the cheap talk group and 23% in the real group; p = 0.005).20 The effect of the cheap talk group is also still highly significant in the regression analysis (p = 0.012 in the first regression equation in Table 3 and p = 0.006 in the second regression equation in Table 3). 3. Discussion and Concluding Remarks Our field experiment yields two important findings for eliciting willingness to pay using contingent valuation. First, hypothetical bias was removed with the follow‐up certainty question; this approach yielded dichotomous choice responses that were indistinguishable from the real decisions. Second, the cheap talk approach was not effective in removing hypothetical bias. We discuss each of these findings in turn below. Prior evidence on the cheap talk approach is mixed. Cummings and Taylor (1999) found that the cheap talk approach was effective in removing hypothetical bias in experimental referendums about donations to public goods. However, in the recent studies by List (2001), Brown et al. (2003) and Murphy et al. (2005), the cheap talk approach was effective only in specific sub‐groups. The diabetics in our study have had diabetes for an average of about nine years. To the extent that the diabetics in our study are informed and experienced, like the card dealers in List's experiment, our finding that hypothetical bias in stated preferences for the diabetes management programme is not mitigated by cheap talk is similar to List's finding. In terms of validity, we think our findings are important evidence. Our field experiment is carried out in a similar fashion as an actual contingent valuation study. Face‐to‐face interviews (which are often recommended) are used and subjects value a non‐trivial good for which there is no available market price on which to anchor. The consistent effects of price and income also support the validity of our results. The follow up certainty approach used in this study yielded estimates of willingness to pay that matched real payments. These results are encouraging and they are consistent with previous evidence from both a laboratory experiment on a private good and a field experiment on a health care good (Blumenschein et al., 1998, 2001). This follow up certainty approach is also a very practical approach and appealing against Occam's razor. It only entails adding a simple follow‐up question with two degrees of certainty to a dichotomous choice contingent valuation study. Why do certainty statements that follow contingent valuation reduce hypothetical bias? One straightforward explanation is that definitely yes more closely resembles the response necessary to make a purchase in real market situations. The manner in which the offer is made requires a decision that must be made immediately. Only respondents who are ready to produce the cash, cheque, or credit/debit card get to purchase. If probably yes signals some interest but not enough to make the payment when offered the opportunity, then probably yes is tantamount to no. A sales representative will not let the person have the product unless payment is made. Another explanation of how certainty statements produce a good match between statements and actual purchase behaviour draws on social psychology. Sample and Warland (1973) view attitude as a predisposition to behaviour. They hypothesise and find that certainty of attitude is a moderator variable that produces better measurement of attitude and intention and better prediction of behaviour. Certainty is used to partition people into more homogenous groups. Attitude is a major predictor of both intentions and behaviour for the high‐certainty group. Fazio and Zanna (1978a,b) and Raden (1985) also explore the link between attitude and behaviour and state that individuals who hold attitudes with greater certainty and more confidence behave more consistently. More recently Ajzen (1991) offers a theory of planned behaviour in which attitudes, subjective norms and perceived behavioural control predict intentions and, in turn, intentions predict behaviour. Intentions capture motivational forces. The stronger the intention, the stronger is the link to behaviour.21 The Fujii and Gärling (2003) application of attitude theory to improve the accuracy of stated preferences for travel modes provides a clearer link yet between stated willingness to pay, certainty statements and behaviour. Their assessment is that the single most important insight from attitude theory is that behavioural intention, the commitment to act, is the best predictor of behaviour. Stated preference methods are designed to elicit stable preferences that reflect a well‐behaved utility function. Stated preference is interpreted as a behavioural intention. Fujii and Gärling's view is that preference is composed of two parts, core preference determined by the utility function and contingent preference that may depend on factors such as framing, response mode and anchoring. Studies that do not eliminate contingent preference embedded in stated preference data will produce results that do not match well with behaviour. They expect that weak intention cause errors in predicting behaviour. Their before‐ and after‐panel survey of use of a new subway line in Kyoto, Japan in 1997–8 shows that strength of intention to ride the new line (yes instead of yes to some degree) was a factor that substantially increased the match between intention and behaviour. Our findings that follow up certainty statements permit reclassification of yes responses so that stated willingness to pay and purchase behaviour match well is consistent with the view from social psychology that strength of attitude and intention improve the link between them and behaviour. Our findings are consistent too with the view that only strong, certain stated purchase intentions match well with the typical market purchase. Cheap talk appears to be effective in mitigating hypothetical bias in some applications and ineffective in others. Our evidence suggests that certainty statements will be more effective in removing hypothetical bias but further research is warranted. Field experiments on goods with varying degrees of publicness and different subject populations would be valuable in establishing the certainty approach as a consistently reliable way to improve contingent valuation and determining if follow up questions with two degrees of certainty is always the best form of the certainty approach. Footnotes 1 " Wallis and Friedman (1942) criticised the use of hypothetical choices in experiments and it has been debated ever since (Thaler, 1987; Kagel and Roth, 1995; Camerer and Hogarth, 1999). In experimental economics it is standard practice to provide monetary incentives, whereas the bulk of experimental work in psychology is based on hypothetical choices. For an overview of the methodological debate between psychologists and economists, see Hertwig and Ortmann (2001). 2 " A potentially important variant of this approach incorporates different degrees of certainty directly into the contingent valuation questions and adjusts the estimated willingness to pay (Johannesson et al., 1993; Ready et al., 1995, 2001; Alberini et al., 2003; Evans et al., 2003; Vossler and Poe, 2005). In the current study we choose to focus on dichotomous choice contingent valuation and a follow up certainty question with two degrees of certainty in order to isolate the effect of follow up certainty calibration. 3 " This study was also carried out in Swedish, making the exact translation of the certainty categories difficult. 4 " List et al. (2004) found that subject anonymity can affect willingness to pay for a public good using a referendum format. They argue that subjects may vote yes to publicly advertise their own goodwill. However, any social approval should be less important with the diabetes management programme used in our study. 5 " The follow up certainty question was asked after the willingness to pay question in the cheap talk group also. All data and questionnaires are available from the first author upon request. 6 " The price was varied between $15, $40 and $80 but below we illustrate the questions with a $40 price. 7 " A problem with using actual numbers to illustrate hypothetical bias is that the size of hypothetical bias (and thus numbers) may vary across studies and it therefore becomes unclear which numbers to use. There is also a risk that numbers will lead to anchoring if prices are given. 8 " Using a probit model instead of a logit model yielded similar results and does not change the reported conclusions. 9 " For household income the subject could either fill in the exact amount or mark one of the following categories: <$5,000, $5,000–10,000, $10,001–20,000, $20,001–30,000, $30,001–50,000, $50,001–100,000, $100,001–150,000, >$150,000. A continuous income measure was constructed by setting the income for each subject to the midpoint of the interval ($175,000 was used for the highest income category). 10 " Haemoglobin A1c (HbA1c), also referred to as glycosylated haemoglobin, is a useful indicator of how well the blood glucose level has been controlled in the recent past and is routinely used to monitor diabetic patients; complications of diabetes can be delayed or prevented if the HbA1c level can be kept close to 7%. 11 " There were missing data for household income for four subjects and for body mass index for two subjects. In the regression analysis we imputed the mean value of household income and body mass index in the sample for these subjects to avoid losing any observations. Excluding the six observations with missing data from the regression analysis leads to similar results, and does not change the conclusions reported below. 12 " For continuous variables a two‐sided t‐test was used to test for statistical differences between the groups. For the categorical variables a contingency table Pearson chi‐square test was used. 13 " The continuous variables (price, time with diabetes, body mass index, age, education, household size, household income, and travel time) are included without any transformation in Table 3. We also tested a logarithmic transformation for these variables, and that yielded similar results. 14 " The marginal effects are evaluated at the mean of the other covariates. 15 " The chi‐square value of the likelihood‐ratio test of the joint significance of the two interaction coefficients is 3.36 (2 df) for the first regression equation in Table 3 and 3.97 (2 df) for the second regression equation in Table 3. The critical value at the 10% level is 4.61. 16 " In the estimation of mean willingness to pay with the non‐parametric method it was assumed that the maximum willingness to pay was equal to the highest price ($80) used in the study, and that the proportion of subjects with zero willingness to pay was equal to the proportion of ‘no’ responses at the lowest price used in the study ($15). The variance of mean willingness to pay was estimated based on 2,000 bootstrap replications using the method of Tambour and Zethraeus (1998). The estimation of mean willingness to pay in the parametric method was based on estimating the area below the demand curve using the formula: −(1/β) ln(1 + eα), where β is the price coefficient in the logistic regression equation and α is the constant in the logistic regression equation (with the effect of all other covariates added to the constant). This formula restricts willingness to pay to be nonnegative as is appropriate for a private good that does not have to be consumed. See Johansson (1995, p. 113) for details. 17 " The interaction terms are added to the second regression model in Table 3 (‘Definitely sure yes responses in hypothetical group’) but with the cheap talk group excluded. 18 " The chi‐square value of the test is 32.11 (26 df) and the critical value at the 10% level is 35.56. 19 " These regressions include the same variables as the regression results in Table 3 (except the experimental group dummy variables), with the exception that two variables (diabetes support group and renal disease) are excluded due to a lack of variation in the dependent variable for one of the dummy variable categories. 20 " The p‐values comparing the cheap talk group and the real group at the individual prices are now 0.102 ($15), 0.234 ($40) and 0.052 ($80). 21 " Mitchell and Carson (1989, Chapter 8) offer an insightful discussion of contingent valuation in the context of attitude‐intention‐behaviour relationships studied in social psychology and marketing. Stated willingness to pay is interpreted as an intention to actually pay. They offer three factors that will increase the predictive power of intention: correspondence, proximity and familiarity. They do not discuss increasing the predictive power of stated preferences through the use of stated certainty. References Aadland , D. and Caplan , A.J. (2006). ‘Cheap talk revisited: new evidence from CVM’ , Journal of Economic Behavior and Organisation , vol. 60 ( 4 ) (August), pp. 562 – 78 . Crossref Search ADS WorldCat Ajzen , Icek. ( 1991 ). ‘The theory of planned behavior’ , Organizational Behavior and Human Decision Processes , vol. 50 ( 2 ) (December), pp. 179 – 211 . Google Scholar Crossref Search ADS WorldCat Ajzen , I. , Brown , T.C. and Carvajal , F. ( 2004 ). ‘Explaining the discrepancy between intentions and actions: the case of hypothetical bias in contingent valuation’ , Personality and Social Psychology Bulletin , vol. 30 ( 9 ) (September), pp. 1108 – 21 . Google Scholar Crossref Search ADS PubMed WorldCat Alberini , A. , Boyle , K. and Welsh , M. ( 2003 ). ‘Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty’ , Journal of Environmental Economics and Management , vol. 45 ( 1 ) (January), pp. 40 – 62 . Google Scholar Crossref Search ADS WorldCat Ariely , D. , Loewenstein , G. and Prelec , D. ( 2003 ). ‘‘‘Coherent arbitrariness’’: stable demand curves without stable preferences’ , Quarterly Journal of Economics , vol. 118 ( 1 ) (February), pp. 73 – 105 . Google Scholar Crossref Search ADS WorldCat Bishop , R.C. and Heberlein , T.A. ( 1979 ). ‘Measuring values of extramarket goods: are indirect measures biased?’ , American Journal of Agricultural Economics , vol. 61 ( 5 ) (December), pp. 926 – 30 . Google Scholar Crossref Search ADS WorldCat Blumenschein , K. , Johannesson , M., Blomquist , G.C., Liljas , B. and O'Conor R.M. ( 1998 ). ‘Experimental results on expressed certainty and hypothetical bias in contingent valuation’ , Southern Economic Journal , vol. 65 ( 1 ) (July), pp. 169 – 77 . Google Scholar Crossref Search ADS WorldCat Blumenschein , K. , Johannesson , M., Yokoyama , K.K. and Freeman , P.R. ( 2001 ). ‘Hypothetical versus real willingness to pay in the health care sector: results from a field experiment’ , Journal of Health Economics , vol. 20 ( 3 ) (May), pp. 441 – 57 . Google Scholar Crossref Search ADS PubMed WorldCat Bluml , B.M. , McKenney , J.M., and Cziraky , M.J. ( 2000 ). ‘Pharmaceutical care services and results in Project ImPACT: Hyperlipidemia’ , Journal of the American Pharmaceutical Association , vol. 40 ( 2 ) (March – April), pp. 157 – 65 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Brown , T.C. , Ajzen , I. and Hrubes , D. ( 2003 ). ‘Further test of entreaties to avoid hypothetical bias in referendum contingent valuation’ , Journal of Environmental Economics and Management , vol. 46 ( 2 ) (September), pp. 353 – 61 . Google Scholar Crossref Search ADS WorldCat Camerer , C.F. and Hogarth , R.M. ( 1999 ). ‘The effects of financial incentives in experiments: a review and capital‐labor‐production framework’ , Journal of Risk and Uncertainty , vol. 19 ( 1 ) (December), pp. 7 – 42 . Google Scholar Crossref Search ADS WorldCat Carson , R. T. , ( 2001 ). ‘Contingent valuation, resources and environmental’, in ( N. J. Smelser and P.B. Baltes, eds.), International Encyclopedia of Social and Behavioral Sciences , vol. 19 , pp. 13272 – 5 , Amsterdam: Elsevier Science . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC 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 , vol. 72 ( 1 ) (February), pp. 80 – 99 . Google Scholar Crossref Search ADS WorldCat Carson , R.T. , Flores , N.E. and Meade , N.F. ( 2001 ). ‘Contingent valuation: controversies and evidence’ , Environmental and Resource Economics , vol. 19 ( 2 ) (June), pp. 173 – 210 . Google Scholar Crossref Search ADS WorldCat Champ , P.A. and Bishop , R.C. ( 2001 ). ‘Donation payment mechanisms and contingent valuation: an empirical study of hypothetical bias’ , Environmental and Resource Economics , vol. 19 ( 4 ) (August), pp. 383 – 402 . Google Scholar Crossref Search ADS WorldCat Champ , P.A. , Bishop , R.C., Brown , T.C. and McCollum , D.W. ( 1997 ). ‘Using donation mechanisms to value nonuse benefits from public goods’ , Journal of Environmental Economics and Management , vol. 33 ( 2 ) (June), pp. 151 – 62 . Google Scholar Crossref Search ADS WorldCat Cummings , R.G. , Elliott , S., Harrison , G.W. and Murphy , J. ( 1997 ). ‘Are hypothetical referenda incentive compatible?’ , Journal of Political Economy , vol. 105 ( 3 ) (June), pp. 609 – 21 . Google Scholar Crossref Search ADS WorldCat Cummings , R.G. , Harrison , G.W. and Rutström , E.E. ( 1995 ). ‘Homegrown values and hypothetical surveys: is the dichotomous choice approach incentive‐compatible?’ , American Economic Review , vol. 85 ( 1 ) (March), pp. 260 – 6 . OpenURL Placeholder Text WorldCat 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 , vol. 89 ( 3 ) (June), pp. 649 – 65 . Google Scholar Crossref Search ADS WorldCat D'Agostino , R.B. , Chase , W. and Belanger , B. ( 1988 ). ‘The appropriateness of some common procedures for testing the equality of two independent binomial populations’ , American Statistician , vol. 42 ( 3 ) (August), pp. 198 – 202 . OpenURL Placeholder Text WorldCat Diamond , P.A. and Hausman , J.A. ( 1994 ). ‘Contingent valuation: is some number better than no number?’ , Journal of Economic Perspectives , vol. 8 ( 4 ) (Fall), pp. 45 – 64 . Google Scholar Crossref Search ADS PubMed WorldCat Ethier , R.G. , Poe , G.L., Schulze , W.D. and Clark , J. ( 2000 ). ‘A comparison of hypothetical phone and mail contingent valuation responses for green‐pricing electricity programs’ , Land Economics , vol. 76 ( 1 ) (February), pp. 54 – 67 . Google Scholar Crossref Search ADS WorldCat Evans , M.F. , Flores , N.E. and Boyle , K.J. ( 2003 ). ‘Multiple‐bounded uncertainty choice data as probabilistic intentions’ , Land Economics , vol. 79 ( 4 ) (November), pp. 549 – 60 . Google Scholar Crossref Search ADS WorldCat Fazio , R.H. and Zanna , M.P. ( 1978a ). ‘Attitudinal qualities relating to the strength of the attitude‐behavior relationship’ , Journal of Experimental Social Psychology , vol. 14 ( 4 ) (July), pp. 398 – 408 . Google Scholar Crossref Search ADS WorldCat Fazio , R.H. and Zanna , M.P. ( 1978b ). ‘On the predictive validity of attitudes: the roles of direct experience and confidence’ , Journal of Personality , vol. 46 ( 2 ) (June), pp. 228 – 43 . Google Scholar Crossref Search ADS WorldCat Fujii , S. and Gärling , T. ( 2003 ). ‘Application of attitude theory for improved predictive accuracy of stated preference methods in travel demand analysis’ , Transportation Research Part A , vol. 37 ( 4 ) (May), pp. 389 – 402 . OpenURL Placeholder Text WorldCat Giles , J.T. , Kennedy , D.T., Dunn , E.C., Wallace W.L., Meadows , S.L. and Cafiero , A.C. ( 2001 ). ‘Results of a community pharmacy‐based breast cancer risk‐assessment and education program’ , Pharmacotherapy , vol. 21 ( 2 ) (February), pp. 243 – 53 . Google Scholar Crossref Search ADS PubMed WorldCat Haab , T.C. , Huang , J.‐C. and Whitehead , J.C. ( 1999 ). ‘Are hypothetical referenda incentive compatible? A comment’ , Journal of Political Economy , vol. 107 ( 1 ) (February), pp. 186 – 96 . Google Scholar Crossref Search ADS WorldCat Harrison , G.W. ( 2006 ). ‘Experimental evidence on alternative environmental valuation methods’ , Environmental and Resource Economics , vol. 34 ( 1 ) (May), pp. 125 – 62 . Google Scholar Crossref Search ADS WorldCat Harrison , G.W. and Rutström , E.E. (forthcoming). ‘Experimental evidence on the existence of hypothetical bias in value elicitation methods’, in ( C. Plott and V. L. Smith, eds.), Handbook of Experimental Economics Results , New York: Elsevier Science . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Hertwig , R. and Ortmann , A. ( 2001 ). ‘Experimental practices in economics: a methodological challenge for psychologists?’ , Behavioral and Brain Sciences , vol. 24 ( 3 ) (June), pp. 383 – 403 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Johannesson , M. , Blomquist , G.C., Blumenschein , K., Johansson , P‐O., Liljas , B. and O'Conor , R.M. ( 1999 ). ‘Calibrating hypothetical willingness to pay responses’ , Journal of Risk and Uncertainty , vol. 18 ( 1 ) (April), pp. 21 – 32 . Google Scholar Crossref Search ADS WorldCat Johannesson , M. , Johansson , P‐O., Kriström , B. and Gerdtham , U‐G. ( 1993 ). ‘Willingness to pay for antihypertensive therapy: further results’ , Journal of Health Economics , vol. 12 ( 1 ) (April), pp. 95 – 108 . Google Scholar Crossref Search ADS PubMed WorldCat Johannesson , M. , Liljas , B. and Johansson , P‐O. ( 1998 ). ‘An experimental comparison of dichotomous choice contingent valuation questions and real purchase decisions’ , Applied Economics , vol. 30 ( 5 ) (May), pp. 643 – 7 . Google Scholar Crossref Search ADS WorldCat Johansson , P‐O. ( 1995 ). Evaluating Health Risks: An Economic Approach , Cambridge: Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kagel , J.H. and Roth , A.E. (eds.). ( 1995 ). The Handbook of Experimental Economics , Princeton, New Jersey: Princeton University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kennedy , D.T. and Small , R.E. ( 2002 ). ‘Development and implementation of a smoking cessation clinic in community pharmacy practice’ , Journal of the American Pharmaceutical Association , vol. 42 ( 1 ) (January – February), pp. 83 – 92 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Kriström , B. ( 1990 ). ‘A non‐parametric approach to the estimation of welfare measures in discrete response valuation studies’ , Land Economics , vol. 66 ( 2 ) (May), pp. 135 – 9 . Google Scholar Crossref Search ADS WorldCat List , J.A. ( 2001 ). ‘Do explicit warnings eliminate the hypothetical bias in elicitation procedures? Evidence from field auctions for sportscards’ , American Economic Review , vol. 91 ( 5 ) (December), pp. 1498 – 507 . Google Scholar Crossref Search ADS WorldCat List , J.A. , Berrens , R.P., Bohara , A.K. and Kerkvliet , J. ( 2004 ). ‘Examining the role of social isolation on stated preferences’ , American Economic Review , vol. 94 ( 3 ) (June), pp. 741 – 52 . Google Scholar Crossref Search ADS WorldCat List , J.A. and Gallet , C.A. ( 2001 ). ‘What experimental protocol influence disparities between actual and hypothetical stated values?’ , Environmental and Resource Economics , vol. 20 ( 3 ) (November), pp. 241 – 54 . Google Scholar Crossref Search ADS WorldCat Mitchell , R.C. and Carson , R.T. ( 1989 ). Using Surveys to Value Public Goods: The Contingent Valuation Method , Washington, DC: Resources for the Future . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Murphy , J.J. , Stevens , T. and Weatherhead , D. ( 2005 ). ‘Is cheap talk effective at eliminating hypothetical bias in a provision point mechanism?’ , Environmental and Resource Economics , vol. 30 ( 3 ) (March), pp. 327 – 43 . Google Scholar Crossref Search ADS WorldCat Poe , G.L. , Clark , J.E., Rondeau , D. and Schulze , W.D. ( 2002 ). ‘Provision point mechanisms and field validity tests of contingent valuation’ , Environmental and Resource Economics , vol. 23 ( 1 ) (September), pp. 105 – 31 . Google Scholar Crossref Search ADS WorldCat Portney , P.R. ( 1994 ). ‘The contingent valuation debate: why economists should care’ , Journal of Economic Perspectives , vol. 8 ( 4 ) (Fall), pp. 3 – 18 . Google Scholar Crossref Search ADS PubMed WorldCat Posey , L.M. ( 2003 ). ‘Proving that pharmaceutical care makes a difference in community pharmacy’ , Journal of the American Pharmaceutical Association , vol. 43 ( 2 ) (March‐April), pp. 136 – 9 . Google Scholar PubMed OpenURL Placeholder Text WorldCat Raden , D. ( 1985 ). ‘Strength‐related attitude dimensions’ , Social Psychology Quarterly , vol. 48 ( 4 ) (December), pp. 312 – 30 . Google Scholar Crossref Search ADS WorldCat Ready , R.C. , Navrud , S. and Dubourg , W.R. ( 2001 ). ‘How do respondents with uncertain willingness to pay answer contingent valuation questions?’ , Land Economics , vol. 77 ( 3 ) (August), pp. 315 – 26 . Google Scholar Crossref Search ADS WorldCat Ready , R.C. , Whitehead , J.C. and Blomquist , G.C. ( 1995 ). ‘Contingent valuation when respondents are ambivalent’ , Journal of Environmental Economics and Management , vol. 29 ( 2 ) (September), pp. 181 – 96 . Google Scholar Crossref Search ADS WorldCat Rodriguez de Bittner , M. and Haines , S.T. ( 1997 ). ‘Pharmacy‐based diabetes management: a practical approach’ , Journal of the American Pharmaceutical Association , vol. 37 ( 4 ) (July‐August), pp. 443 – 55 . OpenURL Placeholder Text WorldCat Sample , J. and Warland , R. ( 1973 ). ‘Attitude and prediction of behavior’ , Social Forces , vol. 51 ( 3 ) (March), pp. 292 – 304 . Google Scholar Crossref Search ADS WorldCat Tambour , T. and Zethraeus , N. ( 1998 ). ‘Nonparametric willingness‐to‐pay measures and confidence statements’ , Medical Decision Making , vol. 18 ( 3 ) (July – September), pp. 330 – 6 . Google Scholar Crossref Search ADS PubMed WorldCat Thaler , R. ( 1987 ). ‘The psychology of choice and the assumptions of economics’, in ( A. E. Roth, ed.), Laboratory Experimentation in Economics: Six Points of View , pp. 99 – 130 , Cambridge: Cambridge University Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Viscusi , W.K. and Evans , W.N. ( 1990 ). ‘Utility functions that depend on health status: estimates and economic implications’ , American Economic Review , vol. 80 ( 3 ) (June), pp. 353 – 74 . OpenURL Placeholder Text WorldCat Vossler , C.A. , Ethier , R.G., Poe , G.L., and Welsh , M.P. ( 2003 ). ‘Payment certainty in discrete choice contingent valuation responses: results from a field validity test’ , Southern Economic Journal , vol. 69 ( 4 ) (April), pp. 886 – 902 . Google Scholar Crossref Search ADS WorldCat Vossler , C.A. and Poe , G.L. ( 2005 ). ‘Analysis of contingent valuation data with multiple bids and response options allowing respondents to express uncertainty: a comment’ , Journal of Environmental Economics and Management , vol. 49 ( 1 ) (January), pp. 197 – 200 . Google Scholar Crossref Search ADS WorldCat Wallis , W.A. and Friedman , M. ( 1942 ). ‘The empirical derivation of indifference functions’, in ( O. Lange, F. McIntyre and T.O. Yntema, eds.), Studies in Mathematical Economics and Econometrics in Memory of Henry Schultz , pp. 175 – 89 , Chicago: University of Chicago Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Appendices Appendix 1: Description of the Diabetes Management Programme Pharmacist‐Provided Diabetes Management Programme Description Controlling diabetes is extremely important for your health. When diabetes is not properly controlled it can lead to complications such as nerve problems, heart problems, sexual problems, kidney failure, blindness, and amputations. These complications usually occur in people who have had uncontrolled diabetes for many years. Many studies have shown keeping your blood sugar and haemoglobin A1c at recommended levels can delay or even prevent these complications. You can significantly improve your health, feel better and prevent these complications by maintaining an appropriate diet and exercise plan and taking your diabetes medicines properly. Your pharmacist can provide education and motivation to help you understand and control your diabetes. This diabetes programme was developed by pharmacists at the American Pharmacy Services Corporation. The programme incorporates all of the latest recommendations for optimal diabetes management from the American Diabetes Association and the Centers for Disease Control and Prevention. The programme will allow you and your pharmacist to work together as a team to reach your desired blood sugar and haemoglobin A1c levels through appropriate medication use, monitoring, and lifestyle changes. This programme will last three months and will consist of three formal appointments with your pharmacist. With your permission, your pharmacist will measure your blood sugar, haemoglobin A1c, and conduct other tests to assess your overall health at the beginning and end of the programme. This will allow your pharmacist to make recommendations to you and your physician to maximise the effectiveness of your diabetes therapy. You will have the opportunity to discuss any concerns or problems you have during any of the scheduled visits. The first visit with your pharmacist will last between 45 minutes and one hour. During this visit, your pharmacist will take a medication history and discuss with you the medicines you take for diabetes. You will take a short survey to assess symptoms associated with diabetes and you will complete a short questionnaire about the relationship of haemoglobin A1c to diabetes control. Your pharmacist will then discuss the results of these surveys with you. Your pharmacist will review the relationship of diabetes control to symptoms and blood sugar testing, the proper way to take your diabetes medicine, and target levels for blood sugar and haemoglobin A1c to prevent complications in the future. In addition your pharmacist will measure your blood sugar, haemoglobin A1c, blood pressure and weight. These measures will help your pharmacist assess the progress that you make during the three month programme. The second visit will last between 25 and 35 minutes. Your pharmacist will again ask you to complete a short survey to assess symptoms associated with diabetes. Then, your pharmacist will provide you with information about eating a healthy diet, maintaining regular exercise, and weight control if you are overweight. In addition, your pharmacist will measure your blood sugar and weight, discuss any problems you have encountered since the last visit, and answer any questions you may have. The third and final visit will last between 30 and 45 minutes. You will complete a short survey assessing your diabetes symptoms and control. Your pharmacist will measure your blood sugar, haemoglobin A1c, weight, and blood pressure. Your pharmacist will discuss the progress you have made since the first visit and will review potential complications associated with uncontrolled diabetes and provide tips to help you maintain the progress you have made during the programme. You will have the opportunity to discuss any problems you have encountered since the last visit and ask questions. At the end of the third month, you may choose to continue the diabetes management programme, and you can ask your pharmacist to schedule new appointments with you to best suit your individual needs. Appendix 2: The Cheap Talk Script (with a $40 price). Assume that you are being offered the opportunity to purchase the diabetes disease management service that was just described to you. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. I am going to ask you: ‘Would you buy this service at a price of $40.’ But, before you answer the question, I want to describe a problem that we have in studies like this one. This is a hypothetical purchase question – not a real one. If you say that you would purchase the diabetes service, you will not actually purchase the service or pay any money at the end of our interview. But, I would like for you to respond to the question as though your response would involve a real cash payment. And that is the problem. In most studies of this kind, folks seem to have a hard time doing this. They respond differently to a hypothetical purchase question, where they do not really have to pay money, than they do in a real purchase situation where they actually will have to pay money. For example, in a recent study, a group of patients were asked if they were willing to purchase a disease management service that is similar to the diabetes disease management service that I have just described to you. Payment was hypothetical for these patients, as it will be for you. No one had to pay money if they said they would buy the disease management service. Another similar group of patients also participated in this study. These patients were offered the opportunity to actually purchase the disease management service at the same price. If patients in this second group agreed to purchase the programme they really did have to pay money. On average, more patients said ‘yes’ when the purchase question was hypothetical than when it was real. We call this ‘hypothetical bias’. ‘Hypothetical bias’ is the difference that we continually see in the way people respond to hypothetical situations as compared to real situations – people seem to respond differently to purchase questions when they really don't have to pay money as a result of their response. In the real offer to purchase the disease management service, where people knew they would have to pay money if they said ‘yes’, fewer said ‘yes’ than when payment was hypothetical and people knew they would not pay anything if they said ‘yes’. How can we get people to think about their response to a hypothetical purchase question and respond as if it was a real purchase decision, where if they agree to the purchase they will really have to pay the price? How do we get them to think about what it means to really pay money, if in fact they really aren't going to have to do it? Let me tell you why I think we continually see this hypothetical bias, why people behave differently in a hypothetical purchase situation than they do when the purchase situation is real. I think that when we respond to hypothetical purchase questions, we give some thought to what we might do, but we know we can always change our minds especially if we don't want to buy. But, when the purchase offer is real, and we would actually have to spend our money if we say yes, we think a different way. When we are faced with the possibility of having to spend money, we think about our options: if I spend money on this, that is money I don't have to spend on other things. If I spend money on a diabetes management service, that is money I don't have to spend on groceries, go to a movie, or perhaps spend on some other way of improving my diabetes. So when the payment is real we answer in a way that takes into account the limited amount of money we have. We answer realising that we just don't have enough money to do everything we might like to do. This is just my opinion, of course, but it is what I think may be going on in hypothetical purchase questions. In any case, the only way that we know to get people like you to respond to our hypothetical purchase question just like you would respond if the purchase offer was real is to simply ask you: when you reply to the hypothetical purchase question below, please do the following: Think about what you are replying to. If this were real and you said yes, you would actually have to pay $40 right now – do you really want the diabetes management service enough that you would be willing to spend the money? If I were in your shoes, and I were asked whether or not I would purchase the diabetes management service that was just described at a price of $40, I would think about how I feel about spending my money this way. When I got ready to answer the question, I would ask myself: if this were a real offer to purchase the diabetes service, and I had to pay $40 if I said yes: do I really want to spend my money this way? If I really did, I would say yes; if I did not, I would say no – I wouldn't throw my money around. That is just my opinion, of course. You must do whatever you want to do. In any case, I ask you to respond just exactly as you would respond if you were really going to face the consequences of your response: which is to pay money if you say yes. Please keep this in mind when responding to the question. So, assume that you are being offered the opportunity to purchase the diabetes disease management service that was described to you. Assume that if you choose to purchase the service, you would have to use some of your household income to pay here and now with cash, cheque or credit card. Would you buy this service here and now at a price of $ 40? Please circle your answer below. Yes No Author notes " An earlier draft of this article was presented at the iHEA 5th World Congress held July 10–13, 2005 in Barcelona. We are grateful to Jason Shogren and Laura Taylor for detailed suggestions on the design of our experiment and the cheap talk script. For comments, we thank Richard Bishop, Patricia Champ, Tommy Gärling, Glenn Harrison, John List, John Whitehead, Kip Viscusi, seminar participants at the University of Connecticut and Appalachian State University, Editor Leonardo Felli, and three anonymous referees. We also thank Niklas Zethraeus for research assistance and Ronald Langley for suggestions about certainty statements. Finally, we are indebted to the American Pharmacy Services Corporation and the following pharmacists who participated in this experiment: Matt Cull, Drane Stephens, Aaron McIntosh, Kathy Salyer, Steve and Alicia Dawson, Gary Hamm, Alyson and Leon Claywell, Jason Wallace and Melodie Hawkins. This research was supported by Cooperative Agreement Number E11/CCE421825 from the Centers for Disease Control and Prevention (CDC). Views and contents are solely the responsibility of the authors and do not necessarily represent the official views of CDC. Magnus Johannesson received financial support from the Swedish Research Council. © The Author(s). Journal compilation © Royal Economic Society

Journal

The Economic Journal – Oxford University Press

Published: Jan 1, 2008

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

Eliciting Willingness to Pay Without Bias: Evidence from a Field Experiment

References (80)

Abstract

Journal

Recommended Articles

There are no references for this article.