TY - JOUR
AU1 - Dalton, Patricio S
AU2 - Rüschenpöhler, Julius
AU3 - Uras, Burak
AU4 - Zia, Bilal
AB - Abstract Business practices and performance vary widely across businesses within the same sector. A key outstanding question is why profitable practices do not readily diffuse. We conduct a field experiment among urban retailers in Indonesia to study whether alleviating informational and behavioral frictions can facilitate such diffusion in a cost-effective manner. Through quantitative and qualitative fieldwork, we curate a handbook that associates locally relevant practices with performance, and provides idiosyncratic implementation guidance informed by exemplary local retailers. We complement this handbook with two light-touch interventions to facilitate behavior change. A subset of retailers is invited to a documentary movie screening featuring the paths to success of exemplary peers. Another subset is offered two 30-minute personal visits by a local facilitator. A third group is offered both. Eighteen months later, we find significant impacts on practice adoption when the handbook is coupled with the two behavioral nudges, and up to a 35% increase in profits and 16.7% increase in sales. These findings suggest both informational and behavioral constraints are at play. The types of practices adopted map the performance improvements to efficiency gains rather than other channels. A simple cost–benefit analysis shows such locally relevant knowledge can be codified and scaled successfully at relatively low cost. Teaching Slides A set of Teaching Slides to accompany this article are available online as Supplementary Data. 1. Introduction A large body of empirical evidence on firms shows that there is remarkable heterogeneity in the adoption of business practices within economic sectors (e.g. Bloom and Van Reenen 2007; Syverson 2011; Gibbons and Henderson 2013). Importantly, such heterogeneity is strongly associated with firm-level productivity differences (de Mel, McKenzie, and Woodruff 2009; McKenzie and Woodruff 2017; Bloom et al. 2019). These two stylized facts imply that knowledge on locally profitable business practices does exist, yet it does not diffuse naturally. Understanding these constraints as well as factors and processes that can facilitate knowledge diffusion is therefore an important and outstanding research question. This paper addresses this area of research through a field experiment with small-scale urban retailers in Indonesia. We hypothesize that small-scale firms in developing countries likely face multiple and potentially overlapping constraints to the adoption of profitable business practices. Specifically, we characterize and distinguish two types of constraints, informational and behavioral, and implement a field experiment to study whether alleviating these constraints can facilitate the diffusion of business practices in a cost-effective manner.1 Addressing informational constraints is challenging given the muted impacts of standard business training programs found in the literature that ostensibly are designed to target informational asymmetries (Quinn and Woodruff 2019; McKenzie 2020). For this reason, we take a fundamentally different approach than standard business training. Instead of providing set-courses, we harness the existing local heterogeneity observed in practices and profits and curate a handbook that associates practical and locally relevant best practices with performance. This handbook identifies and prioritizes particular types of information that are most relevant for performance improvements in the local setting, and provides idiosyncratic implementation guidance informed by exemplary local retailers. We complement the handbook with two light-touch interventions to address potentially complementary behavioral constraints. A subset of retailers is invited to a movie screening featuring the paths to success of exemplary peers. Another subset is offered two 30-minute in-person counseling visits by a trained facilitator. A third group is offered both. These two behavioral nudges, the movie and counseling, are grounded in the hypothesis that information alone, even if curated, may not be sufficient to facilitate sustained behavior change. The movie targets potential aspirations constraints (Dalton, Ghosal, and Mani 2016), while the in-person counseling targets constraints related to inattention (Steiner, Stewart, and Matéjka 2017), status-quo bias (Kahneman, Knetsch, and Thaler 1991), and procrastination (Duflo, Kremer, and Robinson 2011). The combined treatment of both the movie and counseling offers potentially complementary benefits. Our experiment design consists of 1,301 urban retail shop owners randomized into 4 treatment groups of 260 retailers and a Control group (N = 261). All 1,040 treated retailers are offered a free copy of the handbook. While retailers assigned to the Handbook group receive the handbook alone, those assigned to the Movie group additionally receive an invitation to the movie screening, and those assigned to the Counseling group additionally receive an invitation to the counseling visits. Finally, retailers assigned to the All Three group receive the handbook and the two invitations. The analysis presented in this paper is based on pooled intent-to-treat (ITT) estimates from 2 follow-up surveys, 6 and 18 months after the intervention. First, we find underwhelming evidence for the effectiveness of the handbook alone and find no significant improvements in the adoption of business practices. By contrast, we find strong and statistically significant impacts when the handbook is complemented with the behavioral interventions. We find that adding either the movie, counseling, or both result in sizable and statistically significant improvements over Control on an aggregate score of all practices covered in the handbook. Specifically, our results show a 12.5% improvement (0.22 s.d.) for Movie, a 10.1% improvement (0.18 s.d.) for Counseling, and a 16.0% improvement (0.29 s.d.) for All Three. These coefficients also reflect statistically significant improvements over providing the handbook alone. Unpacking the aggregate effects into individual practice components, we find statistically significant improvements in all five dimensions of practices for the All Three group: record keeping (27.5%), planning (16.9%), stocking up (11.3%), marketing (24.4%), and joint decision-making (21.9%). These results are robust to multiple hypothesis corrections for false discovery. Second, on business performance, we find these practice effects translate to improvements in profits and sales. Specifically, we find significant and economically meaningful impacts for firms assigned to Counseling and All Three, with an increase in profits of 35% (0.27 s.d.) and 21% (0.17 s.d.), respectively, compared to Control. Likewise, the analysis reveals statistically significant improvements in sales, representing a 17% (0.15 s.d.) improvement for Counseling and a 16% (0.14 s.d.) increase for All Three. Businesses assigned to Movie also increase their profits and sales with respect to Control, but the effects are not statistically significant at conventional levels, partly due to low take-up of the movie. Following the null effects on practices, we find no significant effects on profits and sales of offering the handbook alone. In fact, the three groups with behavioral add-ons significantly outperform the Handbook group on both profits and sales. Next, we turn to mechanisms of impact to understand the pathways for achieving these performance gains. We find no significant increase in the number of customers, total expenses, shop size, or number of employees for any of the treated businesses. Instead, we find improvements in business operational efficiency due to adoption of business practices. Our analysis shows that stocking up and marketing are the two most important channels that mediate the impact on sales and profits. Unpacking the improvements in these practices, we find significant improvements in stocking responsiveness to product profitability, negotiating lower prices with suppliers, consulting with former customers to understand changes in market demand, and offering discounts to loyal customers to maintain customer continuity. Collectively, these practices point to improvements in business efficiency—a mechanism that implies that higher sales can be achieved without an equivalent and corresponding increase in expenses. As a placebo test, we find no significant improvements in business practices that were not covered in the handbook. Overall, these results suggest both informational and behavioral constraints are jointly binding and that efficiency gains can be achieved by curating and distributing existing and locally relevant knowledge. As hypothesized, we find that information alone—even when curated—is not enough to facilitate behavior change. The movie helps move the needle on performance outcomes we measure, but ultimately the changes are not statistically significant. We also do not detect any significant changes in the growth aspirations of retailers, which suggests that the movie was not effective in this regard. Third, the success of the treatments including counseling suggests that information constraints were present and a nudge in the form of short in-person visits was sufficient to help retailers internalize the information in the handbook. Inattention, status-quo, and procrastination are likely important channels explaining the success of the counseling nudge, but it is not possible to disentangle among these with the data available in our study. Finally, a simple cost–benefit analysis shows that the highest increase in monthly profits in our study is more than double the per capita cost of the program. Hence, the study points to a highly cost-effective method for achieving efficiency gains. This paper contributes to at least three literatures. The first is an emerging literature on scalability of programs to facilitate firm upgrading. As McKenzie (2020) highlights in a recent review paper, the biggest challenge for policymakers relates to scaling up good quality business development services to the vast numbers of small businesses that operate in developing countries. Even the best classroom-based business training programs have limited scalability due to cost, infrastructure, and instructor availability constraints. In contrast, our study shows that combining and scaling the handbook with behavioral add-ons can be achieved at relatively low cost, without the need for quality instructors or infrastructure such as classrooms or technical media. As such, our approach offers an alternative and novel avenue for low cost scalability of locally relevant business practices.2 Second, our findings are closely related to the literature on firm upgrading and “PPDs among SSEs” [“Persistent Performance Differences among Seemingly Similar Enterprises”] in a fundamental way (e.g. Gibbons and Henderson (2013)). The motivation of our experimental design is based on the empirical fact that profitable practices exist within the market, yet they do not diffuse without outside intervention. As pointed out by Verhoogen (2020), a key factor preventing diffusion of knowledge is that much of the knowledge needed to perform well is tacit and cannot be simply purchased on the open market.3 Many organizational capabilities “need to be worked out in the practice of producing,” while Gibbons (2010) highlights that they need to be “homegrown” (pg. 30). Similarly, Arrow (1969) emphasizes the importance of decoding tacit knowledge to facilitate the diffusion of innovation, arguing that “different communication channels have different costs (or, equivalently, different capacities), where these costs include the ability of the sender to ‘code’ the information and the recipient to ‘de-code’ it” (pg. 33). To illustrate the problem of transferring organizational knowledge, Winter (2006) uses the analogy of baking a cake from a recipe: “knowing how to bake a cake is clearly not the same thing as knowing how to bring together in one place all the ingredients for a cake” (cited in Gibbons and Henderson (2013), pg. 38). These are precisely the types of frictions that we address in our study. Finally, our study is related to the recent literature on mobilizing peer learning to stimulate business growth (Hardy Morgan and McCasland 2016; Brooks, Donovan, and Johnson 2018; Cai and Szeidl 2018; Fafchamps and Quinn 2018; Lafortune, Riutort, and Tessada 2018), though there are key differences. In our design, local knowledge is identified, processed, aggregated, and diffused anonymously by us who act as intermediaries. By contrast, in Cai and Szeidl (2018), Fafchamps and Quinn (2018), and Brooks, Donovan, and Johnson (2018), the local knowledge is shared directly and personally by the firm owners who are encouraged to meet up regularly.4 This not only makes the interventions more costly, but also has the disadvantage that firm owners may choose whom they want to share their best practices with and to what extent. Cai and Szeidl (2018) show that successful business owners share relevant business knowledge with their peers when they are not competitors. Similarly, the findings of Hardy Morgan and McCasland (2016) suggest that competition is an important barrier to the diffusion of technological know-how. In Lafortune, Riutort, and Tessada (2018), it is former alumni of a formal business training program who share their experiences with peers after having attended the program. In their set-up, the mentors can also choose the extent and type of information they want to share with the current trainees. Given that competition can hinder the diffusion of knowledge and information as established in these studies, our research design is robust to such competitive effects as it does not involve direct interaction between peers; in fact, we intermediate these transfers of knowledge through the handbook. Moreover, most of the practices covered in the handbook for which we identify positive and significant adoption effects relate to gains in efficiency, and as such are non-rivalrous—that is the adoption of a practice by one retailer does not come at the expense of another retailer’s performance. This difference may partly explain why we capture larger treatment effects both on the adoption of practices and on business performance. Further, since retailers do not engage in direct contact with their peers, our design allows us to measure the pure effect of information and its different dissemination channels without needing to disentangle the effect of information alone from the effect of contact with peers per-se, which is a usual concern in the peer-to-peer learning literature. Finally, we also differentiate from this literature in that the focus of our study is not manufacturing firms but small retail shops. We believe highlighting the sector is helpful for scalability and external validity as these types of firms are ubiquitous in developing countries and lessons from our study have potentially wide relevance and applicability. 2. Research Design 2.1. Hypotheses Our research is founded on the hypothesis that small businesses face multiple and likely overlapping constraints to the adoption of business practices. This hypothesis offers a potential explanation for the stylized heterogeneity observed in business practices across firms of similar size, which is present even in developed economies such as the United States (Bloom et al. 2019). We design and implement a cost-effective and scalable experimental design that identifies the role of informational and behavioral constraints on the adoption of business practices and ultimately on business performance among small-scale firms. A handbook of business practices forms the basis of our design. The handbook aims to address informational constraints associated with the adoption decision. A key advance in our study is that the handbook identifies and prioritizes particular lines of information that are most relevant for performance improvements in the local setting. In addition, the handbook clarifies common misconceptions about business practices and offers practical tips on adoption. A good analogy for the handbook, following the argument by Winter (2006), is that it provides a curated list of the best local recipes for domestic chefs rather than a comprehensive and universal cookbook. For these reasons, we expect business owners to find it easy to read and follow, which in turn allows us to focus on low-cost and scalable behavioral nudges. The behavioral nudges of the study are derived from the hypothesis that information alone, even if curated, may not be sufficient to facilitate sustained behavior change. The literature identifies complementary constraints related to inattention (Steiner, Stewart, and Matéjka 2017), aspirations (Dalton, Ghosal, and Mani 2016), status-quo bias (Kahneman, Knetsch, and Thaler 1991), and procrastination (Duflo, Kremer, and Robinson 2011) that can individually or collectively block behavior change among informed decision-makers. With this motivation in mind, our study design complements the handbook with two behavioral add-ons that seek to relax these additional constraints. The movie targets the aspirations constraint, and is based on the notion that the handbook may not be helpful if business owners do not aspire to grow. The format of the movie used in our study with successful peers implementing practices is motivated by recent evidence that such media can influence aspirations and behavior (Bernard et al. 2014; La Ferrara 2019). Acting as role models, these peers, along with their unique success stories and hands-on advice, may serve as an exemplar and therefore foster growth by opening up aspiration windows and facilitating the adoption of successful practices. The in-person counseling targets potential constraints related to inattention, status-quo bias, and procrastination over information alone. Moreover, the psychology literature has long advocated the benefits of human interaction in counseling over inanimate information sources such as computer messages or printed materials (King et al. 2007). Based on this evidence, we consider the short facilitator visits as a nudge that can help internalize the lessons of the handbook. Finally, the combined treatment of both the movie and counseling coupled with the handbook (All Three) is designed to alleviate the information as well as both types of behavioral constraints discussed above. Moreover, this study arm seeks to identify possible complementarities across treatment arms. We acknowledge that both the movie and counseling can influence behavior through different and possibly complementary channels than those hypothesised above. The movie may serve as an endorsement of the handbook by successful peers, or it may further facilitate reading of the handbook as it constitutes an accessible means of communicating key ideas. Likewise, the counseling intervention may facilitate learning on top (or instead) of acting as a nudge, since it offers (a) additional tailoring to individuals’ specific circumstances, (b) one-to-one interaction on topic areas that are elicited prior to each session, and (c) trained facilitators who themselves come from similar backgrounds as the retailers. Although we are not able to separate these channels experimentally, in Section 5.2 we analyze available survey evidence to try and discern whether any of these channels are activated in our study sample. 2.2. Study Setting This study is based in Jakarta, Indonesia. With a population of 10.1 million in inner Jakarta and an urban area of around 30 million (“DKI Jakarta”), Jakarta is the largest city in South-East Asia, and the capital and economic center of Indonesia. In 2015, the city generated a nominal GDP of almost one-sixth of the total nominal GDP of Indonesia (Statistics Indonesia 2016). Our sample consists of traditional retail shops, locally called Warung or Toko Kelontong. Most of the shop owners in our sample are situated in residential areas or adjacent to “wet markets” for meat, fish, and vegetables.5 Our baseline analysis shows a spread of 24 different product categories offered by shop owners and the average shop owner selling a variety of them. Of the sample, 71% lists tobacco and cigarettes among their top-3 most sold products and 50% list this category as their top selling product. Out of the shops that do not sell tobacco and cigarettes; rice, gas, and petrol, and soft drinks are the main products on sale. Less than 5% of shops list fruits and vegetables or meat and fish as their top-3 main selling products. Of the sample, 9% list eggs in their top-3, but only 2.5% list it as their top product. Small retail businesses make up a large fraction of all micro and small enterprises (MSEs) in Indonesia: about 22% of all employees in MSEs work in retail and hospitality which makes it the second largest sector after agriculture (Indonesian Ministry of Cooperatives and SMEs Indonesia 2011). The sectoral choice of our study is helpful for scalability and external validity as these types of firms are ubiquitous in developing countries. 2.3. Sampling The city of Jakarta comprises 144 districts (“Kelurahan”), which include the urban area of Jakarta proper as well as some agglomerations in the wider Jabodetabek metropolitan area. We restricted the sampling base for this study to the 112 districts in South, East, and West urban Jakarta.6 As a first step, 29 of the 112 districts were randomly selected to be included in the study.7 Across the 29 selected districts, we first conducted a listing exercise of small retail shop owners that met the following three inclusion criteria: (1) the shop is at least 4m2 in size; (2) the shop offers at least two different product categories; and (3) the shop is at least 30 meters away from other shops in the listing.8 In addition, we excluded movable establishments, franchise businesses of larger retail chains, and shops located in densely populated marketplaces. These listing criteria were chosen to ensure the sample would consist of business owners with an established store, who sold a variety of products, and where spillovers were minimized by design. Out of the 2,042 businesses listed through this method, we randomly selected 1,301 to be part of this study. 2.4. Experimental Design and Timeline The 1,301 shops in the sample were randomly divided into a handbook treatment group (N = 1,040) and a pure control group (N = 261), stratified by district, gender, shop size (below 6m2, between 6 and 10m2, or above 10m2), and a composite score of business practices (above and below the median). Among handbook recipients, we implemented two additional and orthogonal treatments: the movie screening and the two counseling visits. Online Appendix Figure M.2(a) maps the spatial distribution of our study sample across Jakarta and Online Appendix Figure M.2(b) illustrates the spatial distribution in one sample district. Overall, the study design consists of four experimental treatment arms of 260 firms each:9 handbook only (Handbook group), handbook and an invitation to the movie screening (Movie group), handbook and an invitation to two counseling visits (Counseling group), and all three interventions (All Three group).10 The timing was as follows.11 In January 2016, we sampled the 2,042 businesses. In March and April 2016, we administered the baseline survey and registered the trial of the study at the American Economic Association’s Randomized-Controlled-Trial Registry website.12 Interventions took place in October and November 2016 and were followed by a midline survey held in April and May 2017 and then an endline survey held in May 2018. 2.4.1. Handbook The content of the handbook was developed through primary qualitative and quantitative assessment of best practices among local business peers. First, prior to the start of quantitative fieldwork, we conducted detailed qualitative interviews with a sample of 102 shop owners chosen from two comparable urban markets outside our study area. The goal of these interviews was to understand the most common and successful business practices from the perspective of local business owners, and to familiarize ourselves with the various implementation processes and constraints to adoption. With this objective in mind, we asked a number of open-ended questions in the format of a conversation, and responses were recorded and later transcribed.13 The analysis of these interviews allowed us to identify the following five categories of locally relevant business practices: record keeping, financial planning, stocking up, marketing, and joint decision-making. In the next stage, these categories directly fed into the quantitative baseline survey, from which we were able to associate the contribution of within-category individual practices to profits, sales, and number of customers. Specifically, for each category we estimated linear regressions of profits, sales, and number of customers on individual practices and a set of firm-level controls. Online Appendix Table I.2 shows the (top) practices and the baseline regression coefficients. Column (1) presents the number of specifications in which the variables had a significant coefficient and columns (2) and (3) present the coefficients of the regressions with sales and profits as dependent variables, respectively. These regression coefficients were used in the handbook to illustrate a quantitative association of a particular practice with sales and profits. The findings from these quantitative assessments were then matched with the qualitative fieldwork to arrive at the final set of locally relevant best practices to be included in the handbook. With this information and the help of a local NGO, we developed a handbook of best practices implemented by peers. The handbook consists of five chapters: keeping business records, calculating profits, making stock-up decisions, attracting customers, and cooperation in business decisions, in this order. Although there is no chapter exclusively on financial planning, practices such as reviewing financial performance, analyzing where there are areas for improvement, and comparing sales with targets are natural consequences of implementing the best practices suggested in the handbook. Online Appendix Table I.1 lists all the practices measured in this formative stage of the study, and Online Appendix K presents a summary of the information on beliefs, reasons to adopt practices, step-by-step implementation guidance, and tips presented in each of the five chapters of the handbook.14 Each chapter of the handbook is structured as follows. First, it confronts common false beliefs and misconceptions held by shop owners about the usefulness of implementing the different practices. These misconceptions are based on statements that were heard repeatedly during the qualitative interviews. For example, one common thought by shop owners was that keeping records is difficult for individuals without higher education, or that it is complicated. The handbook emphasizes that this is not the case, and that we have observed shop owners from different educational backgrounds keeping records. Likewise, to confront the belief that keeping records is complicated, the handbook provides simple step-by-step guidance on how to make record keeping an easy routine to follow, also learned from peers. Second, it presents arguments for why it is important to implement the practices, providing locally relevant evidence-based reasons. To this end, the handbook uses regression coefficients associating top practices with profits and sales using data from the quantitative baseline survey. This information is presented in simple layman terms.15 As a concrete example, consider the marketing practice of consulting former customers, which is a binary variable equal to 1 if the retailer consulted with former customers on why they stopped buying from this shop. As Online Appendix Table I.2 shows, the baseline analysis returns coefficients of 0.24 on sales and 0.23 on profits. On page 7 of handbook, this information is conveyed as follows: “retailers who decide to get back to former customers to see the reasons why they quit buying at their shop have monthly sales 24% higher than the sales of businesses whose owners just let it go. Their monthly profits are also 23% higher.” In addition, the handbook provides an approximation to the monetary value of implementing each practice, both for a “typical” size shop and for “larger” shops. The following is an illustrative example: “From the survey we know that shops that use discounts to attract new and retain loyal customers have monthly sales that are 40% higher than the sales of shops which do not give discounts. Also, their monthly profits are 29% higher.” And this is the monetary value: “For a typical shop with IDR 15 million in monthly sales, shops that offered discounts earned IDR 21 million in sales. For a bigger shop with sales of IDR 30 million per month, it would mean IDR 42 million in monthly sales.” Finally, the handbook suggests other reasons to implement the practices. For instance, it suggests that profit-calculation “is useful to plan finances and save up money” and record keeping is necessary to “compare your sales with a benchmark you have,” or to “be better able to save up and withstand unexpected events’, or “unless you keep proper records, there is no good way of knowing how much cash you have on your hands or how much to save. And without savings, unexpected events can hit you and your family hard.” Third, the handbook provides clear, step-by-step implementation guidance illustrated by idiosyncratic practical examples taken from exemplary shop owners. For example, record keeping involves nine specific steps, stocking up comprises seven steps, and marketing involves three steps. The following is an example that we used in the handbook to introduce the practice of joint decision making: “Imagine you are in the back of the shop making a list of items to stock up while your spouse is serving customers in front. You should ask your spouse which items are selling well. That way, you will know better what exactly to stock up on and by how much.” Fourth, the handbook offers several tips to facilitate the adoption of practices that we learned during the qualitative interviews with shop owners. For example, it explains how to deal with the uncertainty of electricity bills in Jakarta, and why switching to a local voucher system works well for a lot of firms. Regarding marketing practices, one tip is to take advantage of the fact that suppliers may want to leave a new product in the shop without charging for it right away. For instance, in the handbook we state: “Remember that the suppliers also want to find out how well the new product sells in the market.” As another example, in the handbook, we refer to the following tip to induce joint decision-making: “by including your co-worker or spouse in the process of making decisions about the shop, chances are they will become more involved and thus be of even greater use.” For record keeping, the handbook recommends using different colors for household expenses, family “loans,” monthly expenses (e.g. bills and stock up), and brackets for stalled payments. For stocking up, the handbook suggests using tallies. To finalize the description of the handbook, it is important to highlight that curating local knowledge in this manner is quite scalable. We worked with an initial sample of 102 firms (less than 10% of our study sample), and gathered rich qualitative data from them. The variance within this sample was very informative for the purposes of the handbook.16 2.4.2. Documentary Movie The movie aims to inspire and motivate the adoption of business practices with the help of locally selected successful business peers. The motivation for this intervention is grounded in research that suggests that exposure to positive role models can facilitate behavior change (Chong and La Ferrara 2009; La Ferrara, Chong, and Duryea 2012; Berg and Zia 2017; Riley 2017) and ignite aspirations and forward-looking behavior (Bernard et al. 2014). The retailers who participated in the movie were selected from the initial pool of 102 qualitative survey participants. Based on these interviews, we shortlisted nine retailers who were ostensibly successful, had an organized shop and books, and employed the largest number of business practices in our study. These characteristics qualified the selected business owners as potential role models who could be inspirational for their peers and alleviate aspirational constraints. In conjunction with the production of the handbook, we conducted further in-depth interviews with the selected owners about their personal business trajectory and about business practices and implementation advice they regarded as crucial to achieve growth. We also wanted diversity in gender, age, and ethnicity to appeal to the different business owners in our sample. This heterogeneity is important since similarity cues based on these factors have been shown to facilitate social learning besides cues of success, competence, skill, and knowledge (see e.g. Henrich and Gil-White 2001; Efferson, Lalive, and Fehr 2008; McElreath et al. 2008; Corriveau and Harris 2009; Rendell et al. 2011; Chudek et al. 2013). This procedure resulted in a final set of five shop owners representing the local frontier of best practices in each domain of business operations. Moreover, these shop owners regularly employed the practices in the same way featured in the handbook and agreed to describe their implementation methods and paths to success in a documentary movie. The movie featuring these five successful business owners was filmed on shop sites and edited in post-production by a professional media company hired by us. We were involved at each stage of implementation, including script development, test runs, filming, and post production. The end product was a 25 minute movie.17 The movie was publicly screened in each of the 29 districts at a local school or other public space. All screening locations were central and accessible to all invited businesses. In order to increase attendance, shop owners were offered IDR100,000 (US|${\$}$| 4.68 PPP) as a show-up fee and to compensate them for their time and transport expenses. In addition, we offered two alternative screening dates in each district and sent individual text message reminders the day prior to each screening. Each screening was followed by a facilitation session by a trained counselor who clarified any doubts and answered questions from the audience. The screening ended with a short feedback survey and payment of the show-up fee. 2.4.3. Counseling For the counseling treatment, we trained a set of local facilitators based on the content of the handbook. These individuals were required to have a bachelor’s degree in a related field and some experience interacting with businesses similar to those in our study. All facilitators were fieldwork enumerators who expressed interest, had the experience of conducting business interviews, but were not involved as enumerators in the survey-stage of this study. The training of the enumerators was conducted over three days and included classroom lectures, role play exercises, and pilot visits to retail businesses in districts external to the study. The 20 facilitators trained through this process were then randomly assigned to businesses in our study and were supervised by senior field staff. The objective of the counseling treatment was to reduce procrastination and inattention associated with reading and applying the content of the handbook. We aimed to achieve this objective by implementing the following protocol. The facilitator first confirmed the identity of the business owner and then asked which aspects of the handbook needed clarification. Based on the owner’s response, the facilitator chose one of three options. First, if the entrepreneur had started implementing a practice but encountered problems along the way, the facilitator would document the issues and start giving standardized implementation advice. Second, if the entrepreneur had not started implementing any practice but had made progress reading the handbook, the facilitator would document any issues with the material and then give implementation advice. Once all issues were dealt with, the facilitator would encourage the entrepreneur to go through the rest of the chapter and follow the written guidance. Third, if the entrepreneur had not even started reading the handbook, the facilitator would elicit their priorities among the practices and start introducing the chapter corresponding to the most relevant practice. Each shop visit lasted approximately 30 minutes. At the end of the first visit, the shop owners were asked to establish goals for the implementation of a practice covered during the visit and for the study of selected material. A second visit was scheduled two weeks after the first and at the convenience of the shop owners. This second visit followed the same protocol as the first with the difference that the starting point was determined by the work left from the first session and the shop owner’s priorities elicited during that visit. There are several features that make the counseling sessions special. First, the sessions were provided on-site rather than in an outside venue, which allowed shop owners to tend to their business matters with only minor interruption. Second, the counseling sessions involved direct one-to-one interaction on topic areas that were elicited prior to each session. As a result, although the sessions were only 30 minutes each, they were intense in engagement. Third, the facilitators themselves came from similar socioeconomic backgrounds as the shop owners, which was a deliberate choice to facilitate closer interaction. 3. Data Firm-level data were collected at baseline, mid-line (6 months after the intervention), and end-line (18 months after the intervention). The baseline assessed business owner background characteristics, business characteristics, and the use of business practices.18 3.1. Summary Statistics, Attrition, and Compliance Summary statistics from the baseline survey are presented in Table 1. Columns (1) and (2) provide mean and standard deviation values for the total sample of 1,301 businesses, respectively, while column (3) presents p-values for equality of means tests across all experimental groups (treatment and control). All p-values are large (well above 0.10). We also ran a multinomial logit specification regressing assignment to treatment on baseline characteristics and checked for joint orthogonality with a chi-square test. The p-value of this test is 0.819, which also suggests that the randomization was successful. Table 1. Summary statistics and tests of randomization. . (1) . (2) . (3) . . . . F-test of joint equality of . . . Standard . means across all . Variable . Sample mean . deviation . treatments (p-value) . Respondent is male 0.29 0.96 Respondent age 45.27 11.31 1.00 Respondent years of education 9.39 3.78 0.66 Respondent risk preference score (0–1 scale) 0.37 0.21 0.64 Respondent time preference score (0–1 scale) 0.52 0.23 0.92 Shop size in square meters 13.22 12.34 0.63 Age of firm 13.60 11.79 0.72 Firm has tax ID 0.19 0.37 Total number of employees 2.00 1.22 0.58 Total number of customers on a normal day 49.33 43.32 0.29 Sales last month (USD PPP) 4786.16 4853.48 0.63 Estimated profits last month (USD PPP) 907.73 1185.09 0.71 Obtained business loan in last 12 months 0.16 0.52 Aggregate score of practices covered in handbook (0–1 scale) 0.36 0.18 0.83 Aggregate score for practices not covered in handbook (0–1 scale) 0.18 0.13 0.59 Aggregate score of enumerator shop evaluation (0–1 scale) 0.63 0.24 0.52 Omnibus chi-square test of joint orthogonality from multinomial logit (p-value): 0.819 . (1) . (2) . (3) . . . . F-test of joint equality of . . . Standard . means across all . Variable . Sample mean . deviation . treatments (p-value) . Respondent is male 0.29 0.96 Respondent age 45.27 11.31 1.00 Respondent years of education 9.39 3.78 0.66 Respondent risk preference score (0–1 scale) 0.37 0.21 0.64 Respondent time preference score (0–1 scale) 0.52 0.23 0.92 Shop size in square meters 13.22 12.34 0.63 Age of firm 13.60 11.79 0.72 Firm has tax ID 0.19 0.37 Total number of employees 2.00 1.22 0.58 Total number of customers on a normal day 49.33 43.32 0.29 Sales last month (USD PPP) 4786.16 4853.48 0.63 Estimated profits last month (USD PPP) 907.73 1185.09 0.71 Obtained business loan in last 12 months 0.16 0.52 Aggregate score of practices covered in handbook (0–1 scale) 0.36 0.18 0.83 Aggregate score for practices not covered in handbook (0–1 scale) 0.18 0.13 0.59 Aggregate score of enumerator shop evaluation (0–1 scale) 0.63 0.24 0.52 Omnibus chi-square test of joint orthogonality from multinomial logit (p-value): 0.819 Notes: This table presents summary statistics for the baseline survey data. Columns (1) and (2) present mean and standard deviations for the full sample. Column (3) presents p-values for equality of means tests across all experimental groups (treatment and control). The bottom of the table presents the p-value from a chi-square test of joint orthogonality of all variables reported in the table using a multinomial logit specification. Open in new tab Table 1. Summary statistics and tests of randomization. . (1) . (2) . (3) . . . . F-test of joint equality of . . . Standard . means across all . Variable . Sample mean . deviation . treatments (p-value) . Respondent is male 0.29 0.96 Respondent age 45.27 11.31 1.00 Respondent years of education 9.39 3.78 0.66 Respondent risk preference score (0–1 scale) 0.37 0.21 0.64 Respondent time preference score (0–1 scale) 0.52 0.23 0.92 Shop size in square meters 13.22 12.34 0.63 Age of firm 13.60 11.79 0.72 Firm has tax ID 0.19 0.37 Total number of employees 2.00 1.22 0.58 Total number of customers on a normal day 49.33 43.32 0.29 Sales last month (USD PPP) 4786.16 4853.48 0.63 Estimated profits last month (USD PPP) 907.73 1185.09 0.71 Obtained business loan in last 12 months 0.16 0.52 Aggregate score of practices covered in handbook (0–1 scale) 0.36 0.18 0.83 Aggregate score for practices not covered in handbook (0–1 scale) 0.18 0.13 0.59 Aggregate score of enumerator shop evaluation (0–1 scale) 0.63 0.24 0.52 Omnibus chi-square test of joint orthogonality from multinomial logit (p-value): 0.819 . (1) . (2) . (3) . . . . F-test of joint equality of . . . Standard . means across all . Variable . Sample mean . deviation . treatments (p-value) . Respondent is male 0.29 0.96 Respondent age 45.27 11.31 1.00 Respondent years of education 9.39 3.78 0.66 Respondent risk preference score (0–1 scale) 0.37 0.21 0.64 Respondent time preference score (0–1 scale) 0.52 0.23 0.92 Shop size in square meters 13.22 12.34 0.63 Age of firm 13.60 11.79 0.72 Firm has tax ID 0.19 0.37 Total number of employees 2.00 1.22 0.58 Total number of customers on a normal day 49.33 43.32 0.29 Sales last month (USD PPP) 4786.16 4853.48 0.63 Estimated profits last month (USD PPP) 907.73 1185.09 0.71 Obtained business loan in last 12 months 0.16 0.52 Aggregate score of practices covered in handbook (0–1 scale) 0.36 0.18 0.83 Aggregate score for practices not covered in handbook (0–1 scale) 0.18 0.13 0.59 Aggregate score of enumerator shop evaluation (0–1 scale) 0.63 0.24 0.52 Omnibus chi-square test of joint orthogonality from multinomial logit (p-value): 0.819 Notes: This table presents summary statistics for the baseline survey data. Columns (1) and (2) present mean and standard deviations for the full sample. Column (3) presents p-values for equality of means tests across all experimental groups (treatment and control). The bottom of the table presents the p-value from a chi-square test of joint orthogonality of all variables reported in the table using a multinomial logit specification. Open in new tab According to the statistics in Table 1, shop owners in our sample are mostly female (71%) and are 45 years old on average. Educational backgrounds are mixed, with a mean educational attainment of 9 years of schooling. The average entrepreneur is risk averse, and is neither patient nor impatient.19 The average business has a size of about 13 sq. mts, has been in operation for 13 years, employs 2 workers, receives about 50 customers per day, and has monthly sales of US|${\$}$| 4,786 PPP and monthly profits of US|${\$}$|908 PPP (rounded).20 Only 19% of the businesses report having a tax ID and 16% obtained a loan in the last 12 months. The average adoption rate of business practices included in the handbook is 36% at baseline, compared to 18% for practices not covered in the handbook. Finally, the normalized index for shop appearance as judged by the enumerators is 0.63, where 0 indicates the least tidy and 1 the most. This empirical observation implies that the shops were considered to be in relatively good condition. This sample of retailers is similar on many observable characteristics to other samples compiled by McKenzie and Woodruff (2017) in Bangladesh, Kenya, Mexico, Ghana, Nigeria, Sri Lanka, and Chile. Moreover, MSEs like the ones in our sample constitute the majority of small firms in the developing world. According to Hsieh and Olken (2014), “about 90 % of firms in Mexico employ less than 10 workers. In India and Indonesia, the fraction of firms with less than 10 workers is almost visually indistinguishable from 100 %” (p. 93). Similarly, McKenzie (2017) reports 99.6 % of firms in Nigeria have fewer than ten workers. Online Appendix B presents the analysis on survey attrition. As Online Appendix Table B.1 shows, attrition was very low. We were able to reach 92% of the sample at midline and 81% of the sample at endline, and the small attrition rates are not correlated with treatment status. Nevertheless, Online Appendix Table B.2 presents Lee bounds for the main treatment effects in this paper and finds statistically significant impacts even at the lower bound. Online Appendix C discusses treatment compliance and attendee evaluation, summarizes statistics collected during the counseling visits, and presents analysis for selection into treatment. Online Appendix Table C.1 shows that out of the 520 shop owners invited to the movie screening, only 52% showed up at the venue for the film screening session. The take-up of the counseling treatment was higher—out of the 520 shop owners offered personalized counseling sessions, 77% received the counseling once and 68% received it twice. Lastly, we do not observe consistent discernable patterns of selection on observables (Online Appendix Table C.3). 4. Main Results 4.1. Estimation Strategy We present pooled ITT estimates using the following ANCOVA regression specification: $$\begin{align} {Y}_{(2,3)i} = \alpha + \sum \limits _{m=1}^4 \beta _m \text{T}_{mi} + \gamma \text{X}_{1i} + \delta {V} + \theta M + \zeta \text{Y}_{1i} + \epsilon _i, \end{align}$$(1) where Y(2,3)i is the stacked outcome for business i at midline t = 2 and endline t = 3.21T is a dummy variable equal to one if business i was assigned to a particular treatment group, while m = 1 to 4 represent the four types of interventions in the study. Since the randomization was done after stratifying by gender, shop size (micro, small, or mid-sized), and a median split of a business practice composite score, we follow Bruhn and McKenzie (2009) and include the strata dummies represented by the vector X. V represents district fixed effects, while M is a dummy variable for the midline survey round. Y1i is the baseline value of the outcome of interest. Standard errors are clustered at the business level.22 To correct for multiple hypothesis testing we adopt the method of Benjamini, Krieger, and Yekutieli (2006) as outlined in Anderson (2008) to calculate sharpened false discovery rate (FDR) q-values.23 This method applies the (Benjamini and Hochberg 1995) (henceforth BH) correction in two stages where the first stage uses the standard BH procedure to estimate the number of true null hypotheses and the second stage uses these estimates and re-applies the BH procedure to achieve sharpened FDR q-values. This procedure provides better power than the standard BH method (Anderson 2008). Given the multiple treatment arms and multiple outcomes in our study, we balance the issue of power against the likelihood of false discovery by adopting the following approach. First, we group outcomes into domains, where each domain encompasses a particular aspect of the business. Second, we apply FDR corrections simultaneously for all tests performed within each domain. We specify five such domains: business practices, business performance, business expenditure, mechanisms, and personal attributes of the business owner. The business practices domain includes aggregate indices for record keeping, planning, stocking up, marketing, and joint decision-making. We also present an overall aggregate index of practices and compare it against alternative aggregation techniques such as principal component analysis and lasso in Online Appendix D. The business performance domain includes profits and sales, which are the main performance outcomes of a firm. We also construct an overall performance z-score measure, where we first compute z-scores for profits and sales and then take the average of the two z-scores. The expenditure domain includes total business expenses as well as disaggregated individual expense categories. The mechanism domain includes shop size, total number of employees, and total number of customers, that is outcomes that inform on the mechanism of treatment impact. As with the performance domain, we construct an overall mechanism z-score measure. Finally, the personal domain includes business owner outcomes of satisfaction with life, satisfaction with finances, systematic and intuitive working style, as well as an index of standardized z-scores for business aspirations.24 4.2. Business Practices We first analyze primary treatment impacts on business practices that were covered in the handbook. In Table 2, we report results for the five dimensions of business practices and the aggregate of all practices with sharpened FDR q-values reported in italics.25 The table shows that providing the handbook alone leads to no significant improvement in the adoption of business practices in any category. The point estimates are also very small and close to zero. In contrast, we find strong and statistically significant impacts when this pure information treatment is combined with the behavioral add-ons. Table 2. Business practices domain. . (1) . (2) . (3) . (4) . (5) . (6) . . Aggregate of all practices . Record keeping . Planning . Stocking up . Marketing . Joint decision making . Assigned handbook 0.008 0.025 0.028 −0.007 −0.010 0.011 (0.013) (0.017) (0.022) (0.019) (0.022) (0.027) 0.157 0.127 0.167 0.290 0.290 0.290 Assigned handbook & movie 0.042*** 0.058*** 0.044* 0.038* 0.042 0.039 (0.013) (0.018) (0.021) (0.020) (0.023) (0.028) 0.002 0.009 0.071 0.086 0.103 0.142 Assigned handbook & counseling 0.034*** 0.065*** 0.034 0.011 0.039 0.037 (0.013) (0.018) (0.021) (0.019) (0.023) (0.027) 0.006 0.005 0.111 0.290 0.111 0.152 Assigned all three 0.054*** 0.054*** 0.068*** 0.053** 0.061** 0.059* (0.013) (0.017) (0.022) (0.019) (0.024) (0.027) 0.001 0.009 0.009 0.017 0.026 0.070 R-squared 0.308 0.205 0.193 0.187 0.152 0.120 N 2205 2205 2204 2205 2205 2205 Dependent variable mean in control group 0.337 0.196 0.402 0.471 0.250 0.269 Dependent variable SD in control group 0.189 0.252 0.310 0.270 0.320 0.420 F-test (p-value): Book = book & movie 0.008 0.063 0.464 0.013 0.024 0.305 F-test (p-value): Book = book & counseling 0.040 0.026 0.769 0.307 0.032 0.346 F-test (p-value): Book = all three 0.000 0.104 0.078 0.001 0.002 0.082 F-test (p-value): Book & movie = book & counseling 0.539 0.701 0.629 0.144 0.885 0.928 F-test (p-value): Book & movie = all three 0.363 0.826 0.276 0.421 0.451 0.488 . (1) . (2) . (3) . (4) . (5) . (6) . . Aggregate of all practices . Record keeping . Planning . Stocking up . Marketing . Joint decision making . Assigned handbook 0.008 0.025 0.028 −0.007 −0.010 0.011 (0.013) (0.017) (0.022) (0.019) (0.022) (0.027) 0.157 0.127 0.167 0.290 0.290 0.290 Assigned handbook & movie 0.042*** 0.058*** 0.044* 0.038* 0.042 0.039 (0.013) (0.018) (0.021) (0.020) (0.023) (0.028) 0.002 0.009 0.071 0.086 0.103 0.142 Assigned handbook & counseling 0.034*** 0.065*** 0.034 0.011 0.039 0.037 (0.013) (0.018) (0.021) (0.019) (0.023) (0.027) 0.006 0.005 0.111 0.290 0.111 0.152 Assigned all three 0.054*** 0.054*** 0.068*** 0.053** 0.061** 0.059* (0.013) (0.017) (0.022) (0.019) (0.024) (0.027) 0.001 0.009 0.009 0.017 0.026 0.070 R-squared 0.308 0.205 0.193 0.187 0.152 0.120 N 2205 2205 2204 2205 2205 2205 Dependent variable mean in control group 0.337 0.196 0.402 0.471 0.250 0.269 Dependent variable SD in control group 0.189 0.252 0.310 0.270 0.320 0.420 F-test (p-value): Book = book & movie 0.008 0.063 0.464 0.013 0.024 0.305 F-test (p-value): Book = book & counseling 0.040 0.026 0.769 0.307 0.032 0.346 F-test (p-value): Book = all three 0.000 0.104 0.078 0.001 0.002 0.082 F-test (p-value): Book & movie = book & counseling 0.539 0.701 0.629 0.144 0.885 0.928 F-test (p-value): Book & movie = all three 0.363 0.826 0.276 0.421 0.451 0.488 Notes: This table presents regression analysis for the business practices domain, including the aggregate of all practices (column (1)) and sub-practice indices of record keeping (column (2)), planning (column (3)), stocking-up (column (4)), marketing (column (5)), and joint decision making (column (6)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(6), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Statistically significant q-values are highlighted by: * (10% significance level), ** (5% significance level), and *** (1% significance level). Open in new tab Table 2. Business practices domain. . (1) . (2) . (3) . (4) . (5) . (6) . . Aggregate of all practices . Record keeping . Planning . Stocking up . Marketing . Joint decision making . Assigned handbook 0.008 0.025 0.028 −0.007 −0.010 0.011 (0.013) (0.017) (0.022) (0.019) (0.022) (0.027) 0.157 0.127 0.167 0.290 0.290 0.290 Assigned handbook & movie 0.042*** 0.058*** 0.044* 0.038* 0.042 0.039 (0.013) (0.018) (0.021) (0.020) (0.023) (0.028) 0.002 0.009 0.071 0.086 0.103 0.142 Assigned handbook & counseling 0.034*** 0.065*** 0.034 0.011 0.039 0.037 (0.013) (0.018) (0.021) (0.019) (0.023) (0.027) 0.006 0.005 0.111 0.290 0.111 0.152 Assigned all three 0.054*** 0.054*** 0.068*** 0.053** 0.061** 0.059* (0.013) (0.017) (0.022) (0.019) (0.024) (0.027) 0.001 0.009 0.009 0.017 0.026 0.070 R-squared 0.308 0.205 0.193 0.187 0.152 0.120 N 2205 2205 2204 2205 2205 2205 Dependent variable mean in control group 0.337 0.196 0.402 0.471 0.250 0.269 Dependent variable SD in control group 0.189 0.252 0.310 0.270 0.320 0.420 F-test (p-value): Book = book & movie 0.008 0.063 0.464 0.013 0.024 0.305 F-test (p-value): Book = book & counseling 0.040 0.026 0.769 0.307 0.032 0.346 F-test (p-value): Book = all three 0.000 0.104 0.078 0.001 0.002 0.082 F-test (p-value): Book & movie = book & counseling 0.539 0.701 0.629 0.144 0.885 0.928 F-test (p-value): Book & movie = all three 0.363 0.826 0.276 0.421 0.451 0.488 . (1) . (2) . (3) . (4) . (5) . (6) . . Aggregate of all practices . Record keeping . Planning . Stocking up . Marketing . Joint decision making . Assigned handbook 0.008 0.025 0.028 −0.007 −0.010 0.011 (0.013) (0.017) (0.022) (0.019) (0.022) (0.027) 0.157 0.127 0.167 0.290 0.290 0.290 Assigned handbook & movie 0.042*** 0.058*** 0.044* 0.038* 0.042 0.039 (0.013) (0.018) (0.021) (0.020) (0.023) (0.028) 0.002 0.009 0.071 0.086 0.103 0.142 Assigned handbook & counseling 0.034*** 0.065*** 0.034 0.011 0.039 0.037 (0.013) (0.018) (0.021) (0.019) (0.023) (0.027) 0.006 0.005 0.111 0.290 0.111 0.152 Assigned all three 0.054*** 0.054*** 0.068*** 0.053** 0.061** 0.059* (0.013) (0.017) (0.022) (0.019) (0.024) (0.027) 0.001 0.009 0.009 0.017 0.026 0.070 R-squared 0.308 0.205 0.193 0.187 0.152 0.120 N 2205 2205 2204 2205 2205 2205 Dependent variable mean in control group 0.337 0.196 0.402 0.471 0.250 0.269 Dependent variable SD in control group 0.189 0.252 0.310 0.270 0.320 0.420 F-test (p-value): Book = book & movie 0.008 0.063 0.464 0.013 0.024 0.305 F-test (p-value): Book = book & counseling 0.040 0.026 0.769 0.307 0.032 0.346 F-test (p-value): Book = all three 0.000 0.104 0.078 0.001 0.002 0.082 F-test (p-value): Book & movie = book & counseling 0.539 0.701 0.629 0.144 0.885 0.928 F-test (p-value): Book & movie = all three 0.363 0.826 0.276 0.421 0.451 0.488 Notes: This table presents regression analysis for the business practices domain, including the aggregate of all practices (column (1)) and sub-practice indices of record keeping (column (2)), planning (column (3)), stocking-up (column (4)), marketing (column (5)), and joint decision making (column (6)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(6), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Statistically significant q-values are highlighted by: * (10% significance level), ** (5% significance level), and *** (1% significance level). Open in new tab First, we find that adding either the movie, counseling, or both results in statistically significant improvements in the aggregate of all practices. Compared to an adoption rate of 33.7% in Control, our results show a 12.5% improvement (0.22 s.d.) for Movie, a 10.1% improvement (0.18 s.d.) for Counseling, and a 16.0% improvement (0.29 s.d.) for All Three. All three coefficients are statistically significant at the 1% level. Furthermore, these coefficients are statistically larger than providing the Handbook alone, as represented by significant F-test p-values at the bottom of column (1). Second, disaggregated analysis in columns (2)–(6) shows that even with FDR corrections for all tests performed across practice categories and treatment arms, we find that the All Three intervention significantly stimulates the adoption of business practices on all dimensions of the handbook. Specifically, the results show statistically significant impacts on record keeping (27.6%), planning (16.9%), stocking up (11.3%), marketing (24.4%), and joint decision-making (21.9%). We also analyze the distributional impact on business practices by plotting and comparing the kernel densities of the business practices aggregate for each treatment group. Figure 1 plots the Probability Density Functions (PDFs) for the baseline and endline surveys. We then perform Kolmogorov–Smirnov tests of equality of distributions for all treatment and control arms at baseline and endline. For baseline, we cannot reject equality of distributions for any combination, which is by design as the assignment to treatments was random. For endline, we can reject the equality of distribution null between the All Three treatment group and the Control at the 5% significance level. Online Appendix Table D.7 presents quantile regression analysis for each decile of the business practice distribution and finds statistically significant effects in the All Three treatment group at every decile except the highest. These results show that the treatment was effective in improving practices among all businesses, even those at the lower tail of the business practices distribution. Figure 1. Open in new tabDownload slide PDFs of business practices aggregate score. This figure plots the PDF of the business practices aggregate score at baseline and endline. The figure overlays the PDF of each treatment group and the control group. A Kolmogorov–Smirnov test of equality of distributions cannot be rejected for any combination at baseline. At endline, equality between the “All Three” treatment group and the control group is rejected at the 5% significance level. Figure 1. Open in new tabDownload slide PDFs of business practices aggregate score. This figure plots the PDF of the business practices aggregate score at baseline and endline. The figure overlays the PDF of each treatment group and the control group. A Kolmogorov–Smirnov test of equality of distributions cannot be rejected for any combination at baseline. At endline, equality between the “All Three” treatment group and the control group is rejected at the 5% significance level. Finally, as a placebo test, we measure impacts on unrelated business practices. Specifically Online Appendix Table H.2 repeats the analysis for practices that were not covered in the handbook, and finds no treatment impacts on either the composite or any of the individual practices. This finding implies that the practices that we intended to treat indeed exhibited a change in the treatment groups while practices unrelated to our intervention were not affected. 4.3. Business Performance: Profits and Sales Next we turn to performance impacts of the handbook and behavioral add-ons in Table 3. Column (1) presents results for the average z-score of profits and sales, column (2) for the estimated monthly profits, and column (3) for the monthly sales. Both profits and sales are winsorized on both tails at the 5th and 95th percentiles. Online Appendix E explains how the profits measure is computed and shows that the reported impacts are robust to alternative profit measures. Table 3. Business performance domain. . (1) . (2) . (3) . . Performance domain Z-score . Estimated profits last month (win 5%) . Sales last month (win 5%) . Assigned handbook −0.038 −90.194 −384.190 (0.029) (78.342) (312.447) 0.153 0.162 0.162 Assigned handbook & movie 0.030 113.842 365.202 (0.031) (86.829) (336.910) 0.200 0.162 0.162 Assigned handbook & counseling 0.100** 309.980*** 835.637** (0.033) (89.633) (373.159) 0.011 0.005 0.047 Assigned all three 0.074** 190.267** 803.081** (0.032) (84.814) (357.874) 0.034 0.047 0.047 R-squared 0.399 0.180 0.493 N 2197 2172 2197 Dependent variable mean in control group −0.123 894.544 4998.923 Dependent variable SD in control group 0.481 1127.783 5623.257 F-test (p-value): Book = book & movie 0.026 0.018 0.017 F-test (p-value): Book = book & counseling 0.000 0.000 0.000 F-test (p-value): Book = all three 0.000 0.001 0.000 F-test (p-value): Book & movie = book & counseling 0.037 0.043 0.203 F-test (p-value): Book & movie = all three 0.182 0.409 0.221 . (1) . (2) . (3) . . Performance domain Z-score . Estimated profits last month (win 5%) . Sales last month (win 5%) . Assigned handbook −0.038 −90.194 −384.190 (0.029) (78.342) (312.447) 0.153 0.162 0.162 Assigned handbook & movie 0.030 113.842 365.202 (0.031) (86.829) (336.910) 0.200 0.162 0.162 Assigned handbook & counseling 0.100** 309.980*** 835.637** (0.033) (89.633) (373.159) 0.011 0.005 0.047 Assigned all three 0.074** 190.267** 803.081** (0.032) (84.814) (357.874) 0.034 0.047 0.047 R-squared 0.399 0.180 0.493 N 2197 2172 2197 Dependent variable mean in control group −0.123 894.544 4998.923 Dependent variable SD in control group 0.481 1127.783 5623.257 F-test (p-value): Book = book & movie 0.026 0.018 0.017 F-test (p-value): Book = book & counseling 0.000 0.000 0.000 F-test (p-value): Book = all three 0.000 0.001 0.000 F-test (p-value): Book & movie = book & counseling 0.037 0.043 0.203 F-test (p-value): Book & movie = all three 0.182 0.409 0.221 Notes: This table presents regression analysis for the business performance domain, including the average z-score of profits and sales (column (1)) as well as the individual outcomes of estimated profits (column (2)) and sales (column (3)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(3), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Statistically significant q-values are highlighted by: ** (5% significance level), and *** (1% significance level). Open in new tab Table 3. Business performance domain. . (1) . (2) . (3) . . Performance domain Z-score . Estimated profits last month (win 5%) . Sales last month (win 5%) . Assigned handbook −0.038 −90.194 −384.190 (0.029) (78.342) (312.447) 0.153 0.162 0.162 Assigned handbook & movie 0.030 113.842 365.202 (0.031) (86.829) (336.910) 0.200 0.162 0.162 Assigned handbook & counseling 0.100** 309.980*** 835.637** (0.033) (89.633) (373.159) 0.011 0.005 0.047 Assigned all three 0.074** 190.267** 803.081** (0.032) (84.814) (357.874) 0.034 0.047 0.047 R-squared 0.399 0.180 0.493 N 2197 2172 2197 Dependent variable mean in control group −0.123 894.544 4998.923 Dependent variable SD in control group 0.481 1127.783 5623.257 F-test (p-value): Book = book & movie 0.026 0.018 0.017 F-test (p-value): Book = book & counseling 0.000 0.000 0.000 F-test (p-value): Book = all three 0.000 0.001 0.000 F-test (p-value): Book & movie = book & counseling 0.037 0.043 0.203 F-test (p-value): Book & movie = all three 0.182 0.409 0.221 . (1) . (2) . (3) . . Performance domain Z-score . Estimated profits last month (win 5%) . Sales last month (win 5%) . Assigned handbook −0.038 −90.194 −384.190 (0.029) (78.342) (312.447) 0.153 0.162 0.162 Assigned handbook & movie 0.030 113.842 365.202 (0.031) (86.829) (336.910) 0.200 0.162 0.162 Assigned handbook & counseling 0.100** 309.980*** 835.637** (0.033) (89.633) (373.159) 0.011 0.005 0.047 Assigned all three 0.074** 190.267** 803.081** (0.032) (84.814) (357.874) 0.034 0.047 0.047 R-squared 0.399 0.180 0.493 N 2197 2172 2197 Dependent variable mean in control group −0.123 894.544 4998.923 Dependent variable SD in control group 0.481 1127.783 5623.257 F-test (p-value): Book = book & movie 0.026 0.018 0.017 F-test (p-value): Book = book & counseling 0.000 0.000 0.000 F-test (p-value): Book = all three 0.000 0.001 0.000 F-test (p-value): Book & movie = book & counseling 0.037 0.043 0.203 F-test (p-value): Book & movie = all three 0.182 0.409 0.221 Notes: This table presents regression analysis for the business performance domain, including the average z-score of profits and sales (column (1)) as well as the individual outcomes of estimated profits (column (2)) and sales (column (3)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(3), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Statistically significant q-values are highlighted by: ** (5% significance level), and *** (1% significance level). Open in new tab Table 3 shows statistically and economically significant treatment effects on the three performance measures in the Counseling and All Three groups. Compared to Control, businesses assigned to Counseling improve profits by 35%. This effect is statistically significant at the 1% level and represents a 0.27 standard deviation increase with respect to Control. Businesses in Counseling also increase their monthly sales by 17% (0.15 standard deviations) and the effect is significant at the 5% level. Similarly, businesses assigned to All Three improve profits by 21% and sales by 16% (0.17 and 0.14 standard deviation improvement) compared to Control, effects that are statistically significant at the 5% level. In monetary terms, these effects are sizable, and in the Counseling group translate to USD PPP 310 more profits on a monthly basis and a USD PPP 835 increase in monthly sales. In All Three the outcome effects imply USD PPP 190 increase in monthly profits and USD PPP 803 increase in monthly sales. Table 3 also illustrates that businesses assigned to the Movie improve profits by 13% (0.10 standard deviations) over the Control group, but this effect is not statistically significant at conventional levels. The lack of statistical significance is in part due to the relatively low take-up of the movie. The impact of Counseling and All Three on profits is quite large, especially considering that firms only received one hour of counseling in total. To put this effect size into context, other studies that show positive impact on profits find comparable size effects, but utilize extensive “treatment hours” (see Online Appendix Table A.14 for a comparison with other studies). Specifically, with respect to the profit impact, de Mel, Mckenzie, and Woodruff (2012) report an increase by 43% after 49 to 63 hours of training; Calderón, Cunha, and De Giorgi (2013) find a 24% increase after 48 hours of training; Lafortune, Riutort, and Tessada (2018) find a 31% increase following 48–56 hours of role model training; and Cai and Szeidl (2018) find a 35% increase after 144 hours of owners–managers meetings. Table 3 also shows that the handbook alone does not have any significant impact on profits or sales; in fact, the F-test p-values in the bottom half of the table show that the Movie, Counseling, and All Three significantly outperform the Handbook on all performance outcomes. These findings suggest that the curated information provided in the handbook is internalized only after it is accompanied with behavioral add-ons in the form of counseling or both counseling and the movie. Finally, we analyze the distributional impact on profits and sales by plotting the PDFs of estimated profits and sales at baseline and endline (see Figure 2). We also perform Kolmogorov–Smirnov tests of equality of distributions for all treatment and control arms at baseline and endline. While we cannot reject equality of distributions at baseline, we reject the equality of distribution null at endline between the All Three treatment group and Control at the 5% significance level for estimated profits and 1% level for sales. In addition, for estimated profits at endline, equality of distributions is rejected between: (a) All Three and Handbook at the 5% level; (b) All Three and Movie at the 10% level; (c) and Counseling and Handbook at the 10% level. For sales at endline, equality of distributions is rejected between: (a) All Three and Handbook at the 5% level; (b) Counseling and the Control at the 1% level; (c) and Counseling and Handbook at the 10% level. Figure 2. Open in new tabDownload slide PDFs of profits and sales. This figure plots the PDFs of estimated profits and sales at baseline and endline. For both outcomes, the figure overlays the PDF of each treatment group and the control group. A Kolmogorov–Smirnov test of equality of distributions cannot be rejected for any combination at baseline for either estimated profits or sales. At endline, equality between the “All Three” treatment group and the control group is rejected at the 5% significance level for estimated profits and 1% level for sales. In addition, for estimated profits at endline, equality of distributions is rejected between: (a) “All Three” and “Book Only” at the 5% level; (b) “All Three” and “Book & Movie” at the 10% level; (c) and “Book & Counseling” and “Book Only” at the 10% level. For sales at endline, equality of distributions is rejected between: (a) “All Three” and “Book Only” at the 5% level; (b) “Book & Counseling” and the control group at the 1% level; (c) and “Book & Counseling” and “Book Only” at the 10% level. Figure 2. Open in new tabDownload slide PDFs of profits and sales. This figure plots the PDFs of estimated profits and sales at baseline and endline. For both outcomes, the figure overlays the PDF of each treatment group and the control group. A Kolmogorov–Smirnov test of equality of distributions cannot be rejected for any combination at baseline for either estimated profits or sales. At endline, equality between the “All Three” treatment group and the control group is rejected at the 5% significance level for estimated profits and 1% level for sales. In addition, for estimated profits at endline, equality of distributions is rejected between: (a) “All Three” and “Book Only” at the 5% level; (b) “All Three” and “Book & Movie” at the 10% level; (c) and “Book & Counseling” and “Book Only” at the 10% level. For sales at endline, equality of distributions is rejected between: (a) “All Three” and “Book Only” at the 5% level; (b) “Book & Counseling” and the control group at the 1% level; (c) and “Book & Counseling” and “Book Only” at the 10% level. Finally, to benchmark the findings of our study, we compare the standard deviation effect sizes on profits and sales with what other studies in the literature have found. As Online Appendix Table J.1 shows, most studies in this literature do not report control group standard deviations, so this comparison has limited scope. There is also variation in how profits and sales are estimated in the literature as measuring business performance still remains a challenge. Nevertheless, the two studies for which these data are available are Anderson, Chandy, and Zia (2018) and Lafortune, Riutort, and Tessada (2018). Following 80 hours of intensive business training on a pre-screened sample of aspiring entrepreneurs, the former study finds a 0.3 SD increase in profits and a 0.3 SD increase in sales. Following 48–56 hours of role model based training and individual technical assistance, the latter study finds up to a 0.17 SD improvement in profits and up to a 0.19 SD improvement in sales. In comparison, we find a 0.15 SD improvement in profits and 0.14 SD improvement in sales for All Three. The handbook involved no beneficiary interaction, the movie was 25 minutes long, and the two counseling sessions were 30 minutes each. Notwithstanding the substantial difference in hours of involvement with beneficiaries, the effect sizes we find are still consistent and comparable with what has been identified in the literature. 4.4. Business Expenses Table 4 presents regressions results for total expenses in the last month in column (1) and for individual expenses (columns (2)–(6)). While the regression analysis does not show statistically significant impacts, the results in Table 4 do show higher point estimates for precisely the treatment groups that have corresponding and significant increases in sales. The coefficients are positive and directionally consistent, but the standard errors are large and therefore the coefficients are not statistically significant at conventional levels. Nevertheless, column (1) of Table 4 shows that total expenses in All Three are significantly larger than Handbook, with a p-value of 0.09 in an equal coefficients F-test. Decomposing this effect into individual components, Table 4 also shows that stocking up expenses account for more than 95% of total business expenses when comparing the control group means across expenditure categories. Table 4. Business expenditure domain. . (1) . (2) . (3) . (4) . (5) . (6) . . Total expenses last month (win 5%) . Stocking up (win 5%) . Wages (win 5%) . Rent (win 5%) . Electricity (win 5%) . Transport (win 5%) . Assigned handbook −110.180 −97.514 −2.385 −6.854 1.957 0.348 (253.384) (249.141) (1.593) (5.005) (3.508) (2.199) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & movie 154.728 128.388 −1.293 −1.342 3.266 0.074 (283.544) (278.493) (1.905) (4.893) (3.397) (2.231) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 174.898 188.687 2.784 −1.271 5.955* −0.048 (294.527) (284.571) (2.120) (4.986) (3.394) (2.112) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned all three 334.029 321.822 −1.620 −3.085 −0.685 −0.342 (290.281) (287.664) (1.601) (4.763) (3.373) (2.156) 1.000 1.000 1.000 1.000 1.000 1.000 R-squared 0.532 0.535 0.314 0.628 0.285 0.223 N 2188 2183 2187 2187 2172 2180 Dependent variable mean in control group 4287.789 4093.664 7.072 51.573 66.853 30.601 Dependent variable SD in control group 4811.178 4702.590 30.335 90.935 50.646 29.975 F-test (p-value): Book = book & movie 0.303 0.367 0.582 0.220 0.717 0.901 F-test (p-value): Book = book & counseling 0.294 0.271 0.017 0.222 0.275 0.849 F-test (p-value): Book = all three 0.097 0.111 0.646 0.380 0.467 0.743 F-test (p-value): Book & movie = book & counseling 0.946 0.833 0.093 0.987 0.454 0.954 F-test (p-value): Book & movie = all three 0.542 0.504 0.868 0.678 0.264 0.846 . (1) . (2) . (3) . (4) . (5) . (6) . . Total expenses last month (win 5%) . Stocking up (win 5%) . Wages (win 5%) . Rent (win 5%) . Electricity (win 5%) . Transport (win 5%) . Assigned handbook −110.180 −97.514 −2.385 −6.854 1.957 0.348 (253.384) (249.141) (1.593) (5.005) (3.508) (2.199) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & movie 154.728 128.388 −1.293 −1.342 3.266 0.074 (283.544) (278.493) (1.905) (4.893) (3.397) (2.231) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 174.898 188.687 2.784 −1.271 5.955* −0.048 (294.527) (284.571) (2.120) (4.986) (3.394) (2.112) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned all three 334.029 321.822 −1.620 −3.085 −0.685 −0.342 (290.281) (287.664) (1.601) (4.763) (3.373) (2.156) 1.000 1.000 1.000 1.000 1.000 1.000 R-squared 0.532 0.535 0.314 0.628 0.285 0.223 N 2188 2183 2187 2187 2172 2180 Dependent variable mean in control group 4287.789 4093.664 7.072 51.573 66.853 30.601 Dependent variable SD in control group 4811.178 4702.590 30.335 90.935 50.646 29.975 F-test (p-value): Book = book & movie 0.303 0.367 0.582 0.220 0.717 0.901 F-test (p-value): Book = book & counseling 0.294 0.271 0.017 0.222 0.275 0.849 F-test (p-value): Book = all three 0.097 0.111 0.646 0.380 0.467 0.743 F-test (p-value): Book & movie = book & counseling 0.946 0.833 0.093 0.987 0.454 0.954 F-test (p-value): Book & movie = all three 0.542 0.504 0.868 0.678 0.264 0.846 Notes: This table presents regression analysis for the business expenditure domain, including the total expenses (column (1)) as well as the individual expense categories of stocking up (column (2)), wages (column (3)), rent (column (4)), electricity (column (5)), and transport (column (6)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(6), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Open in new tab Table 4. Business expenditure domain. . (1) . (2) . (3) . (4) . (5) . (6) . . Total expenses last month (win 5%) . Stocking up (win 5%) . Wages (win 5%) . Rent (win 5%) . Electricity (win 5%) . Transport (win 5%) . Assigned handbook −110.180 −97.514 −2.385 −6.854 1.957 0.348 (253.384) (249.141) (1.593) (5.005) (3.508) (2.199) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & movie 154.728 128.388 −1.293 −1.342 3.266 0.074 (283.544) (278.493) (1.905) (4.893) (3.397) (2.231) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 174.898 188.687 2.784 −1.271 5.955* −0.048 (294.527) (284.571) (2.120) (4.986) (3.394) (2.112) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned all three 334.029 321.822 −1.620 −3.085 −0.685 −0.342 (290.281) (287.664) (1.601) (4.763) (3.373) (2.156) 1.000 1.000 1.000 1.000 1.000 1.000 R-squared 0.532 0.535 0.314 0.628 0.285 0.223 N 2188 2183 2187 2187 2172 2180 Dependent variable mean in control group 4287.789 4093.664 7.072 51.573 66.853 30.601 Dependent variable SD in control group 4811.178 4702.590 30.335 90.935 50.646 29.975 F-test (p-value): Book = book & movie 0.303 0.367 0.582 0.220 0.717 0.901 F-test (p-value): Book = book & counseling 0.294 0.271 0.017 0.222 0.275 0.849 F-test (p-value): Book = all three 0.097 0.111 0.646 0.380 0.467 0.743 F-test (p-value): Book & movie = book & counseling 0.946 0.833 0.093 0.987 0.454 0.954 F-test (p-value): Book & movie = all three 0.542 0.504 0.868 0.678 0.264 0.846 . (1) . (2) . (3) . (4) . (5) . (6) . . Total expenses last month (win 5%) . Stocking up (win 5%) . Wages (win 5%) . Rent (win 5%) . Electricity (win 5%) . Transport (win 5%) . Assigned handbook −110.180 −97.514 −2.385 −6.854 1.957 0.348 (253.384) (249.141) (1.593) (5.005) (3.508) (2.199) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & movie 154.728 128.388 −1.293 −1.342 3.266 0.074 (283.544) (278.493) (1.905) (4.893) (3.397) (2.231) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 174.898 188.687 2.784 −1.271 5.955* −0.048 (294.527) (284.571) (2.120) (4.986) (3.394) (2.112) 1.000 1.000 1.000 1.000 1.000 1.000 Assigned all three 334.029 321.822 −1.620 −3.085 −0.685 −0.342 (290.281) (287.664) (1.601) (4.763) (3.373) (2.156) 1.000 1.000 1.000 1.000 1.000 1.000 R-squared 0.532 0.535 0.314 0.628 0.285 0.223 N 2188 2183 2187 2187 2172 2180 Dependent variable mean in control group 4287.789 4093.664 7.072 51.573 66.853 30.601 Dependent variable SD in control group 4811.178 4702.590 30.335 90.935 50.646 29.975 F-test (p-value): Book = book & movie 0.303 0.367 0.582 0.220 0.717 0.901 F-test (p-value): Book = book & counseling 0.294 0.271 0.017 0.222 0.275 0.849 F-test (p-value): Book = all three 0.097 0.111 0.646 0.380 0.467 0.743 F-test (p-value): Book & movie = book & counseling 0.946 0.833 0.093 0.987 0.454 0.954 F-test (p-value): Book & movie = all three 0.542 0.504 0.868 0.678 0.264 0.846 Notes: This table presents regression analysis for the business expenditure domain, including the total expenses (column (1)) as well as the individual expense categories of stocking up (column (2)), wages (column (3)), rent (column (4)), electricity (column (5)), and transport (column (6)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(6), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Open in new tab 4.5. Business Size, Employees, and Customers Table 5 presents regression analysis for the mechanism domain, which relates to shop size (column (2)), number of employees (column (3)), and number of customers (column (4)), as well as the average z-score of these three outcomes (column (1)). Although the coefficients for corresponding treatments with behavioral add-ons are positive, they are not statistically significant for any outcome. These results imply that the increase in sales is not due to new customers or business expansion. In Section 5, we explore the mechanisms of impact and tag these improvements as efficiency gains. Table 5. Mechanisms domain. . (1) . (2) . (3) . (4) . . Mechanism domain z-score . Shop size in square meters . Total number of employees . Total number of customers . Assigned handbook −0.007 −0.129 0.013 −0.351 (0.038) (0.666) (0.086) (2.452) 1.000 1.000 1.000 1.000 Assigned handbook & movie 0.035 0.132 0.076 1.077 (0.039) (0.676) (0.083) (2.212) 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.066 0.934 0.130 0.635 (0.041) (0.716) (0.079) (2.421) 0.764 1.000 1.000 1.000 Assigned all three 0.021 0.388 0.010 1.456 (0.038) (0.635) (0.084) (2.414) 1.000 1.000 1.000 1.000 R-squared 0.365 0.357 0.227 0.336 N 2205 2204 2205 2203 Dependent variable mean in control group −0.049 12.856 1.954 40.232 Dependent variable SD in control group 0.551 9.235 1.150 35.578 F-test (p-value): Book = book & movie 0.297 0.702 0.468 0.573 F-test (p-value): Book = book & counseling 0.084 0.135 0.167 0.724 F-test (p-value): Book = all three 0.465 0.414 0.976 0.511 F-test (p-value): Book & movie = book & counseling 0.475 0.258 0.513 0.862 F-test (p-value): Book & movie = all three 0.732 0.685 0.442 0.877 . (1) . (2) . (3) . (4) . . Mechanism domain z-score . Shop size in square meters . Total number of employees . Total number of customers . Assigned handbook −0.007 −0.129 0.013 −0.351 (0.038) (0.666) (0.086) (2.452) 1.000 1.000 1.000 1.000 Assigned handbook & movie 0.035 0.132 0.076 1.077 (0.039) (0.676) (0.083) (2.212) 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.066 0.934 0.130 0.635 (0.041) (0.716) (0.079) (2.421) 0.764 1.000 1.000 1.000 Assigned all three 0.021 0.388 0.010 1.456 (0.038) (0.635) (0.084) (2.414) 1.000 1.000 1.000 1.000 R-squared 0.365 0.357 0.227 0.336 N 2205 2204 2205 2203 Dependent variable mean in control group −0.049 12.856 1.954 40.232 Dependent variable SD in control group 0.551 9.235 1.150 35.578 F-test (p-value): Book = book & movie 0.297 0.702 0.468 0.573 F-test (p-value): Book = book & counseling 0.084 0.135 0.167 0.724 F-test (p-value): Book = all three 0.465 0.414 0.976 0.511 F-test (p-value): Book & movie = book & counseling 0.475 0.258 0.513 0.862 F-test (p-value): Book & movie = all three 0.732 0.685 0.442 0.877 Notes: This table presents regression analysis for the mechanisms domain, including the average z-score of shop size, employees, and customers (column (1)); as well as the individual outcomes of shop size (column (2)), number of employees (column (3)), and number of customers (column (4)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(4), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Open in new tab Table 5. Mechanisms domain. . (1) . (2) . (3) . (4) . . Mechanism domain z-score . Shop size in square meters . Total number of employees . Total number of customers . Assigned handbook −0.007 −0.129 0.013 −0.351 (0.038) (0.666) (0.086) (2.452) 1.000 1.000 1.000 1.000 Assigned handbook & movie 0.035 0.132 0.076 1.077 (0.039) (0.676) (0.083) (2.212) 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.066 0.934 0.130 0.635 (0.041) (0.716) (0.079) (2.421) 0.764 1.000 1.000 1.000 Assigned all three 0.021 0.388 0.010 1.456 (0.038) (0.635) (0.084) (2.414) 1.000 1.000 1.000 1.000 R-squared 0.365 0.357 0.227 0.336 N 2205 2204 2205 2203 Dependent variable mean in control group −0.049 12.856 1.954 40.232 Dependent variable SD in control group 0.551 9.235 1.150 35.578 F-test (p-value): Book = book & movie 0.297 0.702 0.468 0.573 F-test (p-value): Book = book & counseling 0.084 0.135 0.167 0.724 F-test (p-value): Book = all three 0.465 0.414 0.976 0.511 F-test (p-value): Book & movie = book & counseling 0.475 0.258 0.513 0.862 F-test (p-value): Book & movie = all three 0.732 0.685 0.442 0.877 . (1) . (2) . (3) . (4) . . Mechanism domain z-score . Shop size in square meters . Total number of employees . Total number of customers . Assigned handbook −0.007 −0.129 0.013 −0.351 (0.038) (0.666) (0.086) (2.452) 1.000 1.000 1.000 1.000 Assigned handbook & movie 0.035 0.132 0.076 1.077 (0.039) (0.676) (0.083) (2.212) 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.066 0.934 0.130 0.635 (0.041) (0.716) (0.079) (2.421) 0.764 1.000 1.000 1.000 Assigned all three 0.021 0.388 0.010 1.456 (0.038) (0.635) (0.084) (2.414) 1.000 1.000 1.000 1.000 R-squared 0.365 0.357 0.227 0.336 N 2205 2204 2205 2203 Dependent variable mean in control group −0.049 12.856 1.954 40.232 Dependent variable SD in control group 0.551 9.235 1.150 35.578 F-test (p-value): Book = book & movie 0.297 0.702 0.468 0.573 F-test (p-value): Book = book & counseling 0.084 0.135 0.167 0.724 F-test (p-value): Book = all three 0.465 0.414 0.976 0.511 F-test (p-value): Book & movie = book & counseling 0.475 0.258 0.513 0.862 F-test (p-value): Book & movie = all three 0.732 0.685 0.442 0.877 Notes: This table presents regression analysis for the mechanisms domain, including the average z-score of shop size, employees, and customers (column (1)); as well as the individual outcomes of shop size (column (2)), number of employees (column (3)), and number of customers (column (4)). The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy, baseline value of the outcome variable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. For column (1), FDR corrections are performed across treatment arms. For columns (2)–(4), FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes). Open in new tab 4.6. Business Owner Personal Attributes Before turning to mechanisms of impact, Table 6 presents results on business owners’ personal attributes. In particular, we analyze effects on satisfaction with life (column (1)) and finances (column (2)), on a measure of systemic (column (3)) and intuitive (column (4)) working style, and on an aspirations z-score index (column (5)).26 Table 6. Personal attributes domain. . (1) . (2) . (3) . (4) . (5) . . Satisfied with life (0–1 scale) . Satisfied with finances (0–1 scale) . Systematic working style (0–1 scale) . Intuitive working style (0–1 scale) . Aspirations z-score index . Assigned handbook 0.031 0.022 −0.034 −0.015 −0.045 (0.021) (0.021) (0.016) (0.015) (0.025) 1.000 1.000 0.409 1.000 0.819 Assigned handbook & movie 0.000 0.022 −0.009 0.000 −0.003 (0.021) (0.021) (0.015) (0.015) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.016 0.020 −0.002 −0.024 0.024 (0.020) (0.020) (0.015) (0.016) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned all three 0.015 0.045 −0.019 −0.004 0.002 (0.019) (0.020) (0.015) (0.016) (0.026) 1.000 0.409 1.000 1.000 1.000 R-squared 0.040 0.037 0.158 0.101 0.357 N 1019 1018 1181 1181 2205 Dependent variable mean in control group 0.699 0.635 0.799 0.618 −0.091 Dependent variable SD in control group 0.212 0.211 0.182 0.171 0.379 F-test (p-value): Book = book & movie 0.152 0.969 0.091 0.309 0.107 F-test (p-value): Book = book & counseling 0.474 0.920 0.039 0.573 0.009 F-test (p-value): Book = all three 0.414 0.256 0.307 0.477 0.074 F-test (p-value): Book & movie = book & counseling 0.455 0.887 0.656 0.133 0.322 F-test (p-value): Book & movie = all three 0.464 0.271 0.469 0.781 0.877 . (1) . (2) . (3) . (4) . (5) . . Satisfied with life (0–1 scale) . Satisfied with finances (0–1 scale) . Systematic working style (0–1 scale) . Intuitive working style (0–1 scale) . Aspirations z-score index . Assigned handbook 0.031 0.022 −0.034 −0.015 −0.045 (0.021) (0.021) (0.016) (0.015) (0.025) 1.000 1.000 0.409 1.000 0.819 Assigned handbook & movie 0.000 0.022 −0.009 0.000 −0.003 (0.021) (0.021) (0.015) (0.015) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.016 0.020 −0.002 −0.024 0.024 (0.020) (0.020) (0.015) (0.016) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned all three 0.015 0.045 −0.019 −0.004 0.002 (0.019) (0.020) (0.015) (0.016) (0.026) 1.000 0.409 1.000 1.000 1.000 R-squared 0.040 0.037 0.158 0.101 0.357 N 1019 1018 1181 1181 2205 Dependent variable mean in control group 0.699 0.635 0.799 0.618 −0.091 Dependent variable SD in control group 0.212 0.211 0.182 0.171 0.379 F-test (p-value): Book = book & movie 0.152 0.969 0.091 0.309 0.107 F-test (p-value): Book = book & counseling 0.474 0.920 0.039 0.573 0.009 F-test (p-value): Book = all three 0.414 0.256 0.307 0.477 0.074 F-test (p-value): Book & movie = book & counseling 0.455 0.887 0.656 0.133 0.322 F-test (p-value): Book & movie = all three 0.464 0.271 0.469 0.781 0.877 Notes: This table presents regression analysis for the personal attributes domain, including a measure of the business owner’s satisfaction with life (column (1)), a measure of satisfaction with finances (column (2)), a measure of systemic working style (column (3)), a measure of intuitive working style (column (4)), and an aspirations z-score index (column (5)). The aspirations index is the average z-score of 12-month aspirations for shop size, employment, customers, and sales. Columns (1) and (2) outcomes are only available for the endline; and columns (3) and (4) outcomes are only available for the baseline and midline. The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy where applicable, baseline value of the outcome variable where applicable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes) reported in the table. Open in new tab Table 6. Personal attributes domain. . (1) . (2) . (3) . (4) . (5) . . Satisfied with life (0–1 scale) . Satisfied with finances (0–1 scale) . Systematic working style (0–1 scale) . Intuitive working style (0–1 scale) . Aspirations z-score index . Assigned handbook 0.031 0.022 −0.034 −0.015 −0.045 (0.021) (0.021) (0.016) (0.015) (0.025) 1.000 1.000 0.409 1.000 0.819 Assigned handbook & movie 0.000 0.022 −0.009 0.000 −0.003 (0.021) (0.021) (0.015) (0.015) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.016 0.020 −0.002 −0.024 0.024 (0.020) (0.020) (0.015) (0.016) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned all three 0.015 0.045 −0.019 −0.004 0.002 (0.019) (0.020) (0.015) (0.016) (0.026) 1.000 0.409 1.000 1.000 1.000 R-squared 0.040 0.037 0.158 0.101 0.357 N 1019 1018 1181 1181 2205 Dependent variable mean in control group 0.699 0.635 0.799 0.618 −0.091 Dependent variable SD in control group 0.212 0.211 0.182 0.171 0.379 F-test (p-value): Book = book & movie 0.152 0.969 0.091 0.309 0.107 F-test (p-value): Book = book & counseling 0.474 0.920 0.039 0.573 0.009 F-test (p-value): Book = all three 0.414 0.256 0.307 0.477 0.074 F-test (p-value): Book & movie = book & counseling 0.455 0.887 0.656 0.133 0.322 F-test (p-value): Book & movie = all three 0.464 0.271 0.469 0.781 0.877 . (1) . (2) . (3) . (4) . (5) . . Satisfied with life (0–1 scale) . Satisfied with finances (0–1 scale) . Systematic working style (0–1 scale) . Intuitive working style (0–1 scale) . Aspirations z-score index . Assigned handbook 0.031 0.022 −0.034 −0.015 −0.045 (0.021) (0.021) (0.016) (0.015) (0.025) 1.000 1.000 0.409 1.000 0.819 Assigned handbook & movie 0.000 0.022 −0.009 0.000 −0.003 (0.021) (0.021) (0.015) (0.015) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned handbook & counseling 0.016 0.020 −0.002 −0.024 0.024 (0.020) (0.020) (0.015) (0.016) (0.026) 1.000 1.000 1.000 1.000 1.000 Assigned all three 0.015 0.045 −0.019 −0.004 0.002 (0.019) (0.020) (0.015) (0.016) (0.026) 1.000 0.409 1.000 1.000 1.000 R-squared 0.040 0.037 0.158 0.101 0.357 N 1019 1018 1181 1181 2205 Dependent variable mean in control group 0.699 0.635 0.799 0.618 −0.091 Dependent variable SD in control group 0.212 0.211 0.182 0.171 0.379 F-test (p-value): Book = book & movie 0.152 0.969 0.091 0.309 0.107 F-test (p-value): Book = book & counseling 0.474 0.920 0.039 0.573 0.009 F-test (p-value): Book = all three 0.414 0.256 0.307 0.477 0.074 F-test (p-value): Book & movie = book & counseling 0.455 0.887 0.656 0.133 0.322 F-test (p-value): Book & movie = all three 0.464 0.271 0.469 0.781 0.877 Notes: This table presents regression analysis for the personal attributes domain, including a measure of the business owner’s satisfaction with life (column (1)), a measure of satisfaction with finances (column (2)), a measure of systemic working style (column (3)), a measure of intuitive working style (column (4)), and an aspirations z-score index (column (5)). The aspirations index is the average z-score of 12-month aspirations for shop size, employment, customers, and sales. Columns (1) and (2) outcomes are only available for the endline; and columns (3) and (4) outcomes are only available for the baseline and midline. The sample is pooled across two follow-up survey rounds, a midline at 6 months and endline at 18 months after the intervention. All regression specifications control for a midline survey round dummy where applicable, baseline value of the outcome variable where applicable, as well as stratification controls. Standard errors in parentheses are clustered at the retail shop level. Sharpened q-values that correct for false discovery rates using the method of Benjamini et al. (2006) are reported in italics. FDR corrections are simultaneously performed across all tests (i.e. # of treatment arms * # of outcomes) reported in the table. Open in new tab Our hypotheses regarding the impact of the interventions on these personal attributes were as follows. Regarding satisfaction with life, we left it as an open empirical question, since it is not clear that improvements in sales and profits would necessarily imply greater satisfaction, as they may have come along with other subjective costs. On cognitive styles, our hypothesis was that the entrepreneur would think more systematically than intuitively after improving business practices. Finally, because the treatments (especially the Movie) had an aspirational component, our hypothesis was that treated retailers would raise their aspirations. However, as Table 6 shows, we do not detect significant treatment effects on any personal outcome, which suggests that the channel practice adoption and performance improvement is purely through gains in knowledge rather than aspirations, working style, or subjective well-being. 5. Mechanisms We now turn to the mechanisms behind the increase in profits and sales identified in the previous section. The experiment design of this paper implies a strong causal link between improved business practices and better performance outcomes. This link is especially merited given the handbook and its accompanying behavioral add-ons are precisely based on the business practices for which we identify treatment effects. In this section, we delve more deeply into the details of these practices and how they relate to gains in efficiency rather than other mechanisms. We also investigate possible alternative explanations for the performance improvements. 5.1. Measurement Concerns 5.1.1. Efficiency Gains or Just Better Record-Keeping? In this subsection, we unpack the individual practices underlying the aggregate indices to shed light on the mechanisms of impact. This exercise also helps address the concern that the performance gains we identify are simply the result of shop owners being able to record sales and profits better without any real improvements in the business. First, it is important to note that there is no effect on profits for the Movie treatment (Table 3) even though record keeping improves in that treatment arm. Second, while record keeping practices do improve as a result of the interventions, shop owners improve other key handbook practices as well. In particular, they adopt practices that are bound to make the business more efficient. For instance, businesses assigned to All Three “adjust stocks based on product profitability” and “negotiate lower prices with suppliers” (Online Appendix Table D.4); “consult with former customers” and “offer discounts” (Online Appendix Table D.5); “make joint decisions” and “draft agreements to make joint decisions” with a business partner (Online Appendix Table D.6). These practices are expected to either lower the average unit cost of the shop owner (vis-a-vis suppliers) or lower frictions when reaching customers and improve business efficiency. Additionally, we observe a substantial positive impact on practices related to financial planning (Online Appendix Table D.3), such as “reviewing financial performance to identify channels of improvement,” “making anticipated budget for upcoming costs,” and “comparing target vs actual sales.” Combined, these practices strongly suggest gains in efficiency as a primary mechanism for improvement in profits and sales. This efficiency mechanism implies that improvements in sales can be accommodated without an equivalent increase in expenses. The improvements in individual practices can be traced directly to the content of the handbook, where, for example, the chapter on stocking up does not focus on “stocking up more,” rather it focuses on “stocking up efficiently” based on set schedules and restocking targets. Specifically, page 40 of the handbook encourages business owners to stock up based on the demand of customers on the best sold products. Also, pages 41–46 aim to improve the business owners’ awareness and management of shop inventory. In particular, the handbook provides step-by-step guidance on how to keep track of inventory and avoid supply chain inefficiencies. Second, we apply formal causal mediation analysis to investigate the proportion of the treatment effects on sales, profits, and expenses in All Three that can be attributed to each dimension of practices. Specifically, we follow the decomposition framework outlined in Carpena and Zia (2020) and motivated by Imai et al. (2011), which separates the average treatment effect on sales, profits, and expenses into an Average Causal Mediation Effect (ACME) and an Average Direct Effect (ADE). The ACME isolates the impact of a particular intermediary channel (e.g. marketing practices), while the ADE represents all other pathways. Empirically, these effects are estimated using coefficients from two regressions: one for the effect of the treatment on the mediator (i.e. practice score); and the other for the effect of the mediator on the outcome conditional on the treatment. The product of these two coefficients, the ACME, captures the portion of the average treatment effect that can be attributed to the mediating practice.27 Online Appendix Table F.1 presents the ACME analysis for profits (panel A), sales (panel B), and expenses (panel C) in All Three. With a focus on relative rather than absolute coefficients, the results suggest that record keeping is in fact not the most important mediator of performance, rather stocking up and marketing practices contribute the most to sales and expenses and where the mediation effects are largest and statistically significant. Likewise, for profits, the mediation effects are strongest and statistically significant for marketing practices. 5.1.2. Real Adoption of Practices or Self-Reporting Bias? A possible concern with the analysis in this paper is related to social desirability bias or experimenter demand effects. Specifically, the concern relates to treated shop owners simply misreporting higher adoption rates, profits, and sales. First, we would like to note that this is not the first paper facing such a concern; in fact, previous research has specifically tested for self-reporting bias in field experiments and does not find much evidence. For example, Bruhn, Karlan, and Schoar (2018) compare survey data with administrative data on employment levels and wages and do not find a positive treatment effect with survey data that then disappears with administrative data. In addition, the authors test whether treated firms were more likely to (a) provide alternative contacts on the survey, and (b) report sales on the follow-up survey, which should be the case if they wanted to please the interviewer. They find no statistically significant differences in either measure across the treatment and control groups, and interpret these findings as indication that self-reporting bias is likely absent. Similarly, Cai and Szeidl (2018) provide evidence on the lack of experimenter demand effect in businesses surveys by comparing self-reported sales with actual book value of sales and finding no statistical difference. It is worth noting that both Bruhn, Karlan, and Schoar (2018) and Cai and Szeidl (2018) involve interventions which, compared to ours, are much more intense, expensive, and lengthy, which would make their studies more prone to experimenter demand effects. Second, we have more specific reasons to argue that self-reporting bias is unlikely in our setting as we do not explicitly link the survey to the intervention as in other studies. In addition, if the results were influenced by experimenter demand effects, we would likely observe significant effects for the handbook treatment as well, which we do not. Also, such misreporting would affect all business practices and not just the ones included in the handbook. As a direct test, we run placebo regressions to assess treatment effects on practices that were not mentioned in the handbook. Online Appendix Table H.2 confirms that our treatments did not have any significant impact on these placebo practices, which supports the argument that the impacts on business practice adoption are legitimate effects. Finally, we use data on “objective” measures gathered by the survey team. At the end of the baseline and endline surveys, enumerators were instructed to report their own views on shop appearance. For instance, they were asked to record whether the shop appeared clean, well stocked, and whether prices were clearly marked. While these measures do not directly indicate or confirm the use of specific business practices, they do provide a rough indication of whether the shop is better “organized.” As Online Appendix Table H.4 shows, we do see improvements on some of these measures in shop owners assigned to Movie and Counseling. Even though shop owners assigned to All Three do not show significant changes in these measures, the sign of the coefficients go in the right direction. 5.2. Unpacking the Counseling Channel: Which Behavioral Constraints are Likely Binding? The strong and sustained effects of Counseling and All Three in our study suggest that constraints like procrastination, status quo, and/or inattention may be important impediments to practice and performance improvement. At face value, this is not surprising as people have sticky habits which are difficult to influence (Kahneman, Knetsch, and Thaler 1991), and changing practices that have been implemented for years (sometimes through generations) require time investment with uncertain returns (Verhoogen 2020). Furthermore, distrust regarding the veracity of information can further restrict behavior change. The handbook coupled with the counseling in either Counseling or All Three addresses each of these constraints. It gives retailers a nudge to overcome their procrastination. It also provides credible information from successful peers so the information comes across as accessible and achievable. It links practices to business performance, so the uncertainty about the benefits of investing in changing a habit is reduced. Overall, the handbook with counseling alleviates both information and behavioral constraints. While it is difficult to isolate how precisely the counseling channel operates, we have some data on the facilitation visits that can shed light on mechanisms. First, we have data on facilitator gender. Since facilitators were randomly assigned to beneficiaries, we have exogenous variation in whether a beneficiary received counseling from a facilitator of the same gender (low social distance) or opposite gender (high social distance). While an imperfect measure of social distance, we believe gender match does carry some weight in a conservative Muslim society such as Indonesia, where it is likely that the level of comfort and acceptance would be higher with matched genders. We use this variation to assess whether there are differences in practice adoption for the Counseling and All Three treatment groups based on whether there was a gender match between the facilitator and beneficiary. Online Appendix Table H.1 presents these results and finds no statistical difference based on gender match, in fact, the point estimates are almost identical. With the earlier stated caveat on a limited social distance measure, this finding suggests that matching demographics between facilitators and beneficiaries was not a main determinant of impact. Next, as part of the first shop visit, the facilitators recorded whether the beneficiary had started reading the handbook (extensive margin) and whether the beneficiary expressed difficulty in comprehending the material (intensive margin). These data in Online Appendix Table C.2 show that 38% of beneficiaries visited had not yet started reading the handbook, and of the 62% who had started only 6% expressed difficulties in reading comprehension. Furthermore, among the 38% who had not yet started reading, only 1% stated reading comprehension as a binding constraint. In addition, we do not find heterogeneous impacts by education levels in Online Appendix Table H.5. While not conclusive, collectively these findings suggest that inattention, status-quo, and/or procrastination are likely to be important channels for why many beneficiaries did not even start reading the handbook even though reading comprehension is not a reported constraint. On the intensive margin, it is very difficult to distinguish from the alternative channel that facilitators could tailor the handbook to specific individual needs. We acknowledge that we can only claim that the counseling related arms operate through some combination of these channels. Dissecting individual components in an experimental setting beyond what we can do with observable data remains an interesting and important question for future research. 5.3. Do Treated Businesses Improve Performance at the Expense of the Control Group? Another concern relates to whether the gains in sales and profits are achieved at the expense of the control group. In order to allay this concern, we present two key pieces of evidence. First, a simple comparison of sales and profits for the control group between baseline and endline shows that trends in both these variables are flat: USD PPP 4,454 and USD PPP 890, respectively, at baseline and USD PPP 4,999 and USD PPP 895, respectively, at endline. Neither of these performance variables are statistically different from baseline to endline. Given that sales and profits could have changed in the absence of the treatment for macroeconomic or other external reasons, this evidence suggests a flat rather than an increasing or declining growth trajectory. Second, to detect geographic spillovers in the spirit of Miguel and Kremer (2004), we use GPS data to directly measure linear distance from each shop in the control group to its nearest shop in the treatment group. If treated shops were improving performance at the expense of others, then control group shops closer to a treated shop would have worse performance outcomes than those located further away. Moreover, we would expect a positive and significant coefficient on the “distance to nearest treated shop” variable in a specification restricted to the control group. Online Appendix Table H.3 presents a simple regression of profits and sales on this distance measure. Although treated shops are randomly assigned and hence location is exogenous, we nevertheless include a control for “distance to nearest control shop” in these regressions to account for market density. The coefficients for both profits and sales are in fact negative and not statistically significant, which suggests that the gains in performance were not achieved at the expense of the control group.28 6. Conclusion This paper shows that it is possible to stimulate the efficiency and growth of small firms by harnessing and curating locally relevant business knowledge. Our findings show that the delivery mechanism of this information is critical—pure information alone, even when carefully curated to cater to the local context, does not have an impact on the adoption of business practices or on performance, but combining it with inexpensive behavioral nudges results in sizable and significant improvements on both dimensions. These effects persist up to 18 months after the interventions, which is indicative of their long-term durability. On scalability, a back-of-the-envelope tabulation reveals the per-firm cost of the Handbook intervention was approximately US|${\$}$|100, with the Movie and Counseling costing an additional US|${\$}$|25 each.29 In contrast, the benefits were up to US|${\$}$|310 per month in profits, along with a high adoption rate of efficient business practices. This simple comparison shows that the interventions were highly cost-effective with rich potential for scaling up. The cost effectiveness and scalability are further supported by the local relevance and ease of implementation of our study design. Moreover, the interventions studied in this paper do not require extensive training programs, long business consulting or peer-to-peer sessions, or other heavy time demands from participants. Indeed, other methods studied in the literature that involve face-to-face meetings cost substantially more money, demand higher engagement, and require highly skilled instructors. Finally, in light of the existing literature on small business growth in developing economies, the overarching insight of this paper is that we are neither in a world where only first principles matter, because otherwise curated information alone would be very effective, nor are we in a world where the keys to business growth are so idiosyncratic that professional consultants or other forms of deeper engagement are required to unlock success. Moreover, our findings suggest that locally useful knowledge is context dependent but not atomistic, and our methodology offers a simple, effective, and inexpensive blueprint for harnessing its potential. Acknowledgments We thank the Abdul Latif Jameel Poverty Action Lab (J-PAL) for hosting our study, in particular Ni Luh Putu Satyaning Pradnya Paramita, Dwitri Amalia, Raisa Annisa, Lukman Edwindra, and Gabriel Halm for excellent research assistance. We also thank participants at the 8th Wageningen-Lucerne-Tilburg Development Economics workshop, IPA 8th SME Working Group Meeting at MIT Sloan, 9th Tilburg-Wageningen Development Economics Workshop, ENTER-Jamboree 2018 in Toulouse School of Economics, UCPH Field Days Conference at U. of Copenhagen, NBER 2018 Entrepreneurship Working Group in Boston, SEU 2018 at U. de Montevideo, Innovation Growth Lab 2019 Global Conference in Berlin, the Empirical Management Conference 2019 at The World Bank, The Econometric Society World Congress 2020 at Bocconi, and Economics Seminars at EBRD, Radbout Nijmegen, UC Louvain, Stanford, Tilburg, U. Torcuato Di Tella, U. of Glasgow, U. of Gottingen, U. of Florence, Tinbergen Institute, Erasmus University, and the Development Economics Network Berlin. Special thanks to Imran Rasul (Editor), four anonymous referees, Thorsten Beck, Nick Bloom, Erwin Bulte, Michelle Brock, Esther Duflo, David McKenzie, Karen Macours, Rossa O’Keeffe-O’Donovan, Menno Pradhan, Sarada, Antoinette Schoar, Daan van Soest, Ben Vollaard, and Chris Woodruff for helpful discussions and comments. This paper was produced under the framework of the “Enabling Innovation and Productivity Growth in Low Income Countries (EIP-LIC/PO5639)” project, funded by the Department for International Development (DFID) of the United Kingdom and implemented by Tilburg University. Additional funding was received from the World Bank Strategic Research Program (SRP) and the Tilburg Economics Department. Research on the ground was conducted in collaboration with J-PAL South-East Asia, SurveyMETER and Micra. AEA RCT Registry ID: 0001175. Notes The editor in charge of this paper was Imran Rasul. Footnotes 1. The term “behavioral constraints” encompasses psychological biases (e.g. status-quo, present bias, and aspirations) and cognitive shortcuts (e.g. inattention) that can hinder behavior change. 2. For literature on market based solutions see Maffioli, McKenzie, and Ubfal (2020) and Anderson and McKenzie (2020); for online or edutainment delivery see Barsoum et al. (2018) and Bjorvatn et al. (2020); and for firm targeting see Anderson, Chandy, and Zia (2018) and McKenzie and Sansone (2019). 3. The idea of “tacit knowledge” is not new and dates back to Polanyi (1958), who describes it as the psychological and social construct of overall knowledge. 4. The information that experienced mentors typically share in Brooks, Donovan, and Johnson (2018) does not necessarily concern managerial practices of the local context but mostly relate to current information about local market conditions. 5. Online Appendix Figure M.1 shows pictures of two typical shops in our sample, representative in both size and appearance. 6. We excluded all 32 villages of North Jakarta due to a MSE training program being run concurrently in that area by a local retail chain. 7. We initially selected 30 districts, however, in one of these districts only five businesses were identified and they differed markedly from the remaining sample. Hence, they were dropped from the sample. 8. Within a market, we picked a starting point (i.e. the first shop in the sample) at random. From here, we made sure that all subsequent shops in the sample were at least 30 meters away from the first shop and from each other. This protocol ensured that we have a minimum of 30 meters distance between shops throughout the study sample. 9. Online Appendix Figure M.3 tabulates the various treatment arms of this study 10. Among the 1,040 handbooks, 520 had the economic returns to the adoption of each business practice described as gains and 520 had them described as losses. However, as shown in Online Appendix G and Online Appendix Table G.1, the effects of the framing on the main outcomes of this paper are statistically indistinguishable except for the Movie group. For this reason and to maximize statistical power, we focus on pooled estimates in this paper. 11. See Online Appendix Figure M.4 for a detailed timeline. 12. https://www.socialscienceregistry.org/trials/1175. Online Appendix L provides a report clarifying any departure of the research analysis in this paper from what was pre-registered in the AEA RCT Registry. 13. Respondents were informed that the interviews would be used to develop a program to help small businesses and to advance research in Indonesia. 14. The full handbook can be found here. 15. To make sure retailers understand these messages, we developed the handbook after several pilots with retailers of similar backgrounds in Jakarta, and they found this type of information comprehensible. Moreover, the average retailer in our sample has 9 years of formal education, which implies that they possess basic reading and mathematical comprehension. 16. In Online Appendix A, we provide a detailed comparison of the handbook vis-a-vis different approaches to business training found in the literature. 17. The movie can be accessed here. 18. Online Appendix Table I.1 shows the list of the business practices measured in the surveys. 19. Risk attitude is measured with the answer to the question: “Some people usually avoid taking any risk, others are generally fully prepared to take risks. Please imagine a yard stick from 0 to 10. This time 0 means you usually avoid taking any risk and 10 means you are generally fully prepared to take risks.” Time preferences are measured in a similar way: “Some people usually want to have things now rather than later, others are generally willing to wait a long time. Now, please imagine a yardstick from 0 to 10. 0 means you usually want things now rather than later and 10 means you are generally willing to wait.” 20. The profit and sales figures are winsorized on both tails, at the 5th and 95th percentile. Profits are calculated by aggregating all costs up and then calculating sales minus total costs. This applies some of the suggestions on how to calculate more accurate profits from Anderson, Lazicky, and Zia (2019). Online Appendix E explains the estimated profits measure used in this paper, and for robustness presents regression results using alternative measures. 21. Online Appendix Table H.6 reports regression results on practice adoption, profits, and sales after 6 and 18 months using the main ANCOVA specification in equation (1). Given that we do not see differential effects across waves, we present pooled estimates. 22. We did consider Treatment-on-Treated (TOT) analysis owing to the low take-up rates especially for the movie. The TOT coefficients do scale up by the inverse of compliance, but standard errors also increase so the p-values are not different or significant for TOT. 23. The FDR is the expected proportion of rejections that are type 1 errors (i.e. false rejections). 24. The aspirations index is an average of z-scores for 12-month respondent aspirations for shop size, employment, customers, and sales. Given that the satisfaction questions are only asked in the endline and the cognitive style questions are only asked in the baseline and midline, this domain does not have an overall z-score measure. 25. Online Appendix D presents regressions results for individual business practices in Online Appendix Tables D.2–D.6. 26. We use the following two questions from the World Values Survey to measure subjective well-being: On a scale from 1 (“very dissatisfied”) to 10 (“very satisfied”) (1) “How satisfied are you with the financial situation of your household?” and (2) “How satisfied are you with your life at this point?” (see, Inglehart et al. 2014). To measure cognitive style, we use a ten-item questionnaire proposed by Sagiv et al. (2010). This includes five statements measuring a systematic approach (e.g. “Before I do something important, I plan carefully.”) and five statements measuring an intuitive approach (e.g. “I often follow my instincts.”) to working and thinking. The aspirations index is the average z-score of 12-month aspirations for shop size, employment, customers, and sales. These aspirations measures are closely related to those used in the literature (e.g. (Bernard et al. 2014)), although we acknowledge that aspirations are difficult to measure precisely. Columns (1) and (2) outcomes are only available for the endline; and columns (3) and (4) outcomes are only available for the baseline and midline. 27. As highlighted in Carpena and Zia (2020) and Imai, Keele, and Yamamoto (2010), there are two important assumptions needed to identify the ACME and ADE. First is that treatment assignment is independent of the outcomes, which is satisfied in our study because treatment status is randomly assigned. Second, conditional on treatment status, the mediator is also independent of the outcomes. For further discussion on the merits of mediation analysis performed in this paper, see Online Appendix F. 28. Note that we do not have data on all firms in the study area so cannot rule out general equilibrium effects elsewhere. As in most field studies of this type, we acknowledge that our results represent a partial equilibrium. 29. These cost estimates do not include our own research time, but they include the cost of qualitative and quantitative interviews, handbook production, printing, and distribution. The Movie and Counseling cost estimates additionally include production expenses, facilitator training and salaries, equipment and location rental, and respondent travel and meal compensation. Further, embedded within the cost of interviews and salaries are costs associated with RA time as well as overhead charged by our implementation partner (JPAL) for research services. These include several rounds of piloting and fixed costs associated with getting a new research project off the ground. When scaled, these costs are likely to go down. References Anderson Michael L. ( 2008 ). “Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevaluation of the Abecedarian, Perry Preschool, and Early Training Projects .” Journal of the American Statistical Association , 103 , 1481 – 1495 . Google Scholar Crossref Search ADS WorldCat Anderson Stephen , Chandy Rajesh, Zia Bilal ( 2018 ). “Pathways to Profits: Identifying Separate Channels of Small Firm Growth through Business Training .” Management Science , 64 , 1 – 25 . Google Scholar OpenURL Placeholder Text WorldCat Anderson Stephen , Lazicky Christy, Zia Bilal ( 2019 ). “Measuring the Unmeasured: Combining Technology and Behavioral Insights to Improve Measurement of Business Outcomes.” World Bank Policy Research Working Paper No. 8836 . Anderson Stephen , McKenzie David ( 2020 ). “Improving Business Practices and the Boundary of the Entrepreneur: A Randomized Experiment Comparing Training, Consulting, Insourcing and Outsourcing.” Policy Research Working Paper No. 9502 . World Bank , Washington, DC . Arrow Kenneth ( 1969 ). “Classificatory Notes on the Production and Transmission of Technological Knowledge .” American Economic Review , 59 ( 2 ), 29 – 35 . Google Scholar OpenURL Placeholder Text WorldCat Barsoum Ghada , Crepon Bruno, Gardiner Drew, Michel Bastien, Pariente William ( 2018 ). “Evaluating the Impact of Entrepreneurship Edutainment in Egypt: An experimental approach with Peer Effects.” Working paper . Benjamini Yoav , Hochberg Yosef ( 1995 ). “Controlling the False Discovery Rate: A Practical and Powerful Approach to Multiple Testing .” Journal of the Royal Statistical Society. Series B (Methodological) , 57 , 289 – 300 . Google Scholar Crossref Search ADS WorldCat Benjamini Yoav , Krieger Abba M., Yekutieli Daniel ( 2006 ). “Adaptive Linear Step-up Procedures that Control the False Discovery Rate .” Biometrika , 93 , 491 – 507 . Google Scholar Crossref Search ADS WorldCat Berg Gunhild , Zia Bilal ( 2017 ). “Harnessing Emotional Connections to Improve Financial Decisions: Evaluating the Impact of Financial Education in Mainstream Media.” Journal of the European Economic Association , 15 1025 – 1055 . Google Scholar Crossref Search ADS WorldCat Bernard Tanguy , Dercon Stefan, Orkin Kate, Taffesse Alemayehu S. ( 2014 ). “The Future in Mind: Aspirations and Forward-Looking Behaviour in Rural Ethiopia.” CEPR Discussion Paper , 10224 . Bjorvatn Kjetil , Cappelen Alexander W., Helgesson Sekei Linda, Sørensen Erik and Tungodden Bertil ( 2020 ). “Teaching through Television: Experimental Evidence on Entrepreneurship Education in Tanzania .” Management Science , 66 , 2308 – 2325 . Google Scholar Crossref Search ADS WorldCat Bloom Nicholas , Brynjolfsson Erik, Foster Lucia, Jarmin Ron, Patnaik Megha, Saporta-Eksten Itay, Van Reenen John ( 2019 ). “What Drives Differences in Management Practices?” American Economic Review , 109 ( 5 ), 1648 – 83 . Google Scholar Crossref Search ADS WorldCat Bloom Nicholas , Van Reenen John ( 2007 ). “Measuring and Explaining Management Practices across Firms and Countries .” The Quarterly Journal of Economics , 122 , 1351 – 1408 . Google Scholar Crossref Search ADS WorldCat Brooks Wyatt , Donovan Kevin, Johnson Terence R. ( 2018 ). “Mentors or Teachers? Microenterprise Training in Kenya .” American Economic Journal: Applied Economics , 10 , 196 – 221 . Google Scholar OpenURL Placeholder Text WorldCat Bruhn Miriam , Karlan Dean, Schoar Antoinette ( 2018 ). “The Impact of Consulting Services on Small and Medium Enterprises: Evidence from a Randomized Trial in Mexico .” Journal of Political Economy , 126 , 635 – 687 . Google Scholar Crossref Search ADS WorldCat Bruhn Miriam , McKenzie David ( 2009 ). “In Pursuit of Balance: Randomisation in Practice in Development Field Experiments .” American Economic Journal: Applied Economics , 1 , 200 – 232 . Google Scholar OpenURL Placeholder Text WorldCat Cai Jing , Szeidl Adam ( 2018 ). “Interfirm Relationships and Business Performance .” Quarterly Journal of Economics , 7 , 125 – 161 . Google Scholar OpenURL Placeholder Text WorldCat Calderón Gabriela , Cunha Jesse M., De Giorgi Giacomo ( 2013 ). “Business Literacy and Development: Evidence from a Randomized Controlled Trial in Rural Mexico.” Barcelona GSE Working Paper Series , 742 . Carpena Fenella , Zia Bilal ( 2020 ). “The Causal Mechanism of Financial Education: Evidence from Mediation Analysis .” Journal of Economic Behavior and Organization , 177 , 143 – 184 . Google Scholar Crossref Search ADS WorldCat Chong Alberto , La Ferrara Eliana ( 2009 ). “Television and Divorce: Evidence from Brazilian Novelas .” Journal of the European Economic Association , 7 , 458 – 468 . Google Scholar Crossref Search ADS WorldCat Chudek Maciej , Brosseau-Liard Patricia, Birch Susan, Henrich Joseph ( 2013 ). “Culture-Gene Coevolutionary Theory and Children’s Selective Social Learning.” In Navigating the Social World: What Infants, Children, and Other Species Can Teach Us , edited by Banaji Mahzarin, Gelman Susan. Oxford University Press , pp. 1 – 10 . Google Scholar Crossref Search ADS Google Preview WorldCat COPAC Corriveau Kathleen H. , Harris Paul L. ( 2009 ). “Choosing Your Informant: Weighing Familiarity and Recent Accuracy .” Developmental Science , 12 , 426 – 437 . Google Scholar Crossref Search ADS PubMed WorldCat Dalton Patricio S. , Ghosal Sayantan, Mani Anandi ( 2016 ). “Poverty and Aspirations Failure .” Economic Journal , 126 , 165 – 188 . Google Scholar Crossref Search ADS WorldCat de Mel Suresh , McKenzie David, Woodruff Christopher ( 2009 ). “Measuring Microenterprise Profits: Must we Ask How the Sausage is Made?” Journal of Development Economics , 88 , 19 – 31 . Google Scholar Crossref Search ADS WorldCat de Mel Suresh , Mckenzie David, Woodruff Christopher ( 2012 ). “One-Time Transfers of Cash or Capital Have Long-Lasting Effects on Microenterprises in Sri Lanka .” Science , 335 , 962 – 966 . Google Scholar Crossref Search ADS PubMed WorldCat Duflo Esther , Kremer Michael, Robinson Jonathan ( 2011 ). “Nudging Farmers to Use Fertilizer: Theory and Experimental Evidence from Kenya .” American Economic Review , 101 ( 6 ), 2350 – 90 . Google Scholar Crossref Search ADS WorldCat Efferson Charles , Lalive Rafael, Fehr Ernst ( 2008 ). “The Coevolution of Cultural Groups and Ingroup Favoritism .” Science , 321 , 1844 – 1849 . Google Scholar Crossref Search ADS PubMed WorldCat Fafchamps Marcel , Quinn Simon ( 2018 ). “Networks and Manufacturing Firms in Africa: Results from a Randomized Field Experiment .” The World Bank Economic Review , 32 , 656 – 675 . Google Scholar OpenURL Placeholder Text WorldCat Gibbons Robert ( 2010 ). “Inside Organizations: Pricing, Politics, and Path Dependence .” Annual Review of Economics , 2 , 337 – 365 . Google Scholar Crossref Search ADS WorldCat Gibbons Robert , Henderson Rebecca ( 2013 ). What Do Managers Do? Exploring Persistent Performance Differences among Seemingly Similar Enterprises , chap. 17 , pp. 680 – 731 . The Handbook of Organizational Economics . Hardy Morgan , McCasland James ( 2016 ). “It Takes Two: Experimental Evidence on the Determinants of Technology Diffusion.” Working Paper . Henrich Joseph , Gil-White Francisco J. ( 2001 ). “The Evolution of Prestige: Freely Conferred Deference as a Mechanism for Enhancing the Benefits of Cultural Transmission .” Evolution and Human Behavior , 22 , 165 – 196 . Google Scholar Crossref Search ADS PubMed WorldCat Hsieh Chang Tai , Olken Benjamin A. ( 2014 ). “The Missing ‘Missing Middle’ .” Journal of Economic Perspectives , 28 ( 3 ), 89 – 108 . Google Scholar Crossref Search ADS WorldCat Imai Kosuke , Keele Luke, Tingley Dustin, Yamamoto Teppei ( 2011 ). “Unpacking the Black Box of Causality: Learning about Causal Mechanisms from Experimental and Observational Studies .” American Political Science Review , 105 , 765 – 789 . Google Scholar Crossref Search ADS WorldCat Imai Kosuke , Keele Luke, Yamamoto Teppei ( 2010 ). “Identification, Inference and Sensitivity Analysis for Causal Mediation Effects .” Statistical Science , 25 , 51 – 71 . Google Scholar Crossref Search ADS WorldCat Indonesian Ministry of Cooperatives and SMEs Indonesia ( 2011 ). “Statistik Usaha Mikro, Kecil dan Menengah.” Technical Report , Kementerian Koperasi dan UMKM . Inglehart R. , Haerpfer C., Moreno A., Welzel C., Kizilova K., Diez-Medrano J., Lagos M., Norris P., Ponarin E., Puranen B. (eds.) ( 2014 ). World Values Survey: Round Six . JD Systems Institute , Madrid , www.worldvaluessurvey.org/WVSDocumentationWV6.jsp . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Kahneman Daniel , Knetsch Jack L., Thaler Richard H. ( 1991 ). “Anomalies: The Endowment Effect, Loss Aversion, and Status Quo Bias .” Journal of Economic Perspectives , 5 ( 1 ), 193 – 206 . Google Scholar Crossref Search ADS WorldCat King Abby C. , Friedman Robert, Marcus Bess, Castro Cynthia, Napolitano Melissa, Ahn David, Baker Lawrence ( 2007 ). “Ongoing Physical Activity Advice by Humans versus Computers: the Community Health Advice by Telephone (CHAT) Trial .” Health Psychology: Official Journal of the Division of Health Psychology, American Psychological Association , 26 , 718 – 27 . Google Scholar Crossref Search ADS PubMed WorldCat La Ferrara Eliana ( 2019 ). “Presidential Address: Aspirations, Social Norms, and Development .” Journal of the European Economic Association , 17 , 1687 – 1722 . Google Scholar Crossref Search ADS WorldCat La Ferrara Eliana , Alberto Chong, Duryea Suzanne ( 2012 ). “Soap Operas and Fertility: Evidence from Brazil .” American Economic Journal: Applied Economics , 4 , 1 – 31 . Google Scholar OpenURL Placeholder Text WorldCat Lafortune Jeanne , Riutort Julio, Tessada Jose ( 2018 ). “Role Models or Individual Consulting: The Impact of Personalizing Micro-entrepreneurship Training .” American Economic Journal: Applied Economics , 10 , 222 – 245 . Google Scholar OpenURL Placeholder Text WorldCat Maffioli Alessandro , McKenzie David, Ubfal Diego ( 2020 ). “Estimating the Demand for Business Training: Evidence from Jamaica.” Policy Research Working Paper No. 9415 . World Bank , Washington, DC . McElreath Richard , Bell Adrian V., Efferson Charles, Lubell Mark, Richerson Peter J., Waring Timothy ( 2008 ). “Beyond Existence and Aiming Outside the Laboratory: Estimating Frequency-dependent and Pay-off-biased Social Learning Strategies .” Philosophical Transactions of the Royal Society B: Biological Sciences , 363 , 3515 – 3528 . Google Scholar Crossref Search ADS WorldCat McKenzie David ( 2017 ). “Identifying and Spurring High-Growth Entrepreneurship: Experimental Evidence from a Business Plan Competition .” American Economic Review , 107 ( 8 ), 2278 – 2307 . Google Scholar Crossref Search ADS WorldCat McKenzie David ( 2020 ). “Small Business Training to Improve Management Practices in Developing Countries: Reassessing the Evidence for ‘Training Doesn’t Work’.” Policy Research Working Paper No. WPS 9408 , Washington, D.C. : World Bank Group . McKenzie David , Sansone Dario ( 2019 ). “Predicting Entrepreneurial Success is Hard: Evidence from A Business Plan Competition in Nigeria .” Journal of Development Economics , 141 , 102369 . Google Scholar Crossref Search ADS WorldCat McKenzie David , Woodruff Christopher ( 2017 ). “Business Practices in Small Firms in Developing Countries .” Management Science , 63 , 2967 – 2981 . Google Scholar Crossref Search ADS WorldCat Miguel Edward , Kremer Michael ( 2004 ). “Worms: Identifying Impacts on Education and Health in the Presence of Treatment Externalities .” Econometrica , 72 , 159 – 217 . Google Scholar Crossref Search ADS WorldCat Polanyi Michael ( 1958 ). Personal Knowledge: Towards a Post-Critical Philosophy . University of Chicago Press . Google Scholar Google Preview OpenURL Placeholder Text WorldCat COPAC Quinn Simon , Woodruff Christopher ( 2019 ). “Experiments and Entrepreneurship in Developing Countries .” Annual Review of Economics , 11 , 225 – 248 . Google Scholar Crossref Search ADS WorldCat Rendell Luke E. , Fogarty Laurel, Hoppitt William J. E., Morgan Thomas J. H., Webster Mike M., Laland Kevin N. ( 2011 ). “Cognitive Culture: Theoretical and Empirical Insights into Social Learning Strategies .” Trends in Cognitive Sciences , 15 , 68 – 76 . Google Scholar Crossref Search ADS PubMed WorldCat Riley Emma ( 2017 ). “Increasing Students’ Aspirations: The Impact of Queen of Katwe on Students’ Educational Attainment.” CSAE Working Paper Series , 2017 – 13 . Sagiv Lilach , Arieli Sharon, Goldenberg Jacob, Goldschmidt Ayalla ( 2010 ). “Structure and Freedom in Creativity: The Interplay Between Externally Imposed Structure and Personal Cognitive Style .” Journal of Organizational Behavior , 31 , 1086 – 1110 . Google Scholar Crossref Search ADS WorldCat Statistics Indonesia ( 2016 ). “Statistical Yearbook of Indonesia 2016.” Techical Report , Badan Pusat Statistik . Steiner Jakub , Stewart Colin, Matéjka Filip ( 2017 ). “Rational Inattention Dynamics: Inertia and Delay in Decision-Making .” Econometrica , 85 , 521 – 553 . Google Scholar Crossref Search ADS WorldCat Syverson Chad ( 2011 ). “What Determines Productivity?” Journal of Economic Literature , 49 , 326 – 65 . Google Scholar Crossref Search ADS WorldCat Verhoogen Eric ( 2020 ). “Firm-Level Upgrading in Developing Countries.” CDEP-CGEG Working Paper no. 83, March 2020 . OpenURL Placeholder Text WorldCat Winter Sidney G. ( 2006 ). “Toward a neo-Schumpeterian theory of the firm .” Industrial and Corporate Change , 15 , 125 – 141 . Google Scholar Crossref Search ADS WorldCat © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association. This is an Open Access article distributed under the terms of the Creative Commons Attribution Non-Commercial License (https://creativecommons.org/licenses/by-nc/4.0/), which permits non-commercial re-use, distribution, and reproduction in any medium, provided the original work is properly cited. For commercial re-use, please contact journals.permissions@oup.com © The Author(s) 2021. Published by Oxford University Press on behalf of European Economic Association.
TI - Curating Local Knowledge: Experimental Evidence from Small Retailers in Indonesia
JF - Journal of the European Economic Association
DO - 10.1093/jeea/jvab007
DA - 2021-10-14
UR - https://www.deepdyve.com/lp/oxford-university-press/curating-local-knowledge-experimental-evidence-from-small-retailers-in-E0LCnObmtW
SP - 2622
EP - 2657
VL - 19
IS - 5
DP - DeepDyve
ER -