The World Bank Economic Review, 38(3), 2024, 580–597 https://doi.org10.1093/wber/lhad038 Article Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Heuristics on Call: The Impact of Mobile-Phone-Based Business-Management Advice Shawn Cole, Mukta Joshi, and Antoinette Schoar Abstract There is growing evidence that business training for micro-entrepreneurs can be effective. However, in-person training can be expensive and imposes costs on the target beneficiaries. This paper presents the results of a two-site randomized evaluation of a light-touch, mobile-phone-based business-training service for micro- entrepreneurs in India and the Philippines. The results show that the training had a statistically significant im- pact on the adoption of improved business practices, with an increase of 0.06 to 0.12 standard deviation points when considering a binary indicator of business practices. The study finds no evidence of impacts on business sales or profits, though the confidence intervals are wide enough to include meaningful effect sizes (positive or negative). These results suggest that mobile-phone-based training can be a cost-effective and scalable way to impart business skills to micro-entrepreneurs. JEL classification: C93, D91, L26, M53, O12 Keywords: Business training, rules of thumb, digital, micro-entrepreneurs, India, Philippines 1. Introduction There are approximately 420–510 million micro, small, and medium enterprises (MSMEs) around the world (International Finance Corporation 2013). Micro, small, and medium enterprises account for about 90 percent of businesses and over 50 percent of employment worldwide (International Council for Small Business 2019) and employ a majority of the population in many low-income countries (International Finance Corporation 2013). A majority of MSME entrepreneurs do not receive training or support to help them manage the financial complexity of a small enterprise. Yet it is known that in-person training programs based on simple business-management heuristics or “rules of thumb” can significantly improve Shawn Cole is the John G. McLean Professor of Business Administration at the Harvard Business School, Cambridge, USA. His email is scole@hbs.edu. Mukta Joshi (corresponding author) is a Principal Behavioral Designer at ideas42, New York, USA. Her email is mjoshi@ideas42.org. Antoinette Schoar is the Stewart C. Myers-Horn Family Professor of Finance at MIT Sloan School of Management at Cambridge, USA. Her email is aschoar@mit.edu. The research for this article was financed by Development Innovation Ventures (DIV) and Consultative Group to Assist the Poor (CGAP). The authors also acknowledge financial support from the Division of Faculty Development and Research at Harvard Business School. The authors thank Marina Dimova for her contributions to content development and Yuting Wang and Anshul Maudar for their excellent research assistance. The authors are grateful to Janalakshmi and Negros Women for Tomorrow Foundation (NWTF) for their partnership on this work. Lastly, the authors thank the editor David McKenzie and three anonymous referees for their constructive comments. A supplementary online appendix for this article can be found at The World Bank Economic Review website. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com The World Bank Economic Review 581 micro-entrepreneurs’ business practices and firm revenue in low-income country settings (Drexler, Fischer, and Schoar 2014). This study involves a two-site field experiment whose primary purpose is to examine whether financial- Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 heuristics training—delivered via mobile phone technology—can affect the management practices of micro-entrepreneurs and improve firm outcomes. The mobile-phone-based delivery of this training seeks to overcome key barriers that may hamper the efficacy and reach of “traditional” business-training courses. First, such courses are costly: per-pupil direct cost estimates often range from $20 to above $750, and these estimates do not factor in the value of the entrepreneurs’ time (McKenzie 2020). These programs are often subsidized by governments (van Lieshout and Mehtha 2017) and few, if any, charge tuition to beneficiaries, suggesting a low willingness to pay. Nevertheless, developing low-cost, convenient ways to reach MSME owners may be particularly important, as a recent meta-analysis finds such programs can have modest positive effects (McKenzie and Woodruff 2013; McKenzie 2020). In contrast to traditional training, distilling information into actionable, simple rules of thumb may lower adoption barriers and improve financial management even in the absence of complete understanding of accounting or business planning (Feldman 2003; Maddox et al. 2008). Financial heuristics or rules of thumb which lighten the cognitive burden of learning useful financial-management skills may therefore be more effective for many micro-entrepreneurs in low-income countries who face frequent scarcity of money and time (Mullainathan and Shafir 2013). Drexler, Fischer, and Schoar (2014), in a field experiment in the Dominican Republic, demonstrate that in-person classroom training of this sort can improve micro- entrepreneurs’ management practices and revenues. Arráiz, Bhanot, and Calero (2019) use a randomized controlled trial to test the effects of a traditional training program relative to the effects of a tailor- made heuristics-based program for micro-entrepreneurs in Ecuador and find statistically and economically meaningful incremental effect sizes on sales and profit of the tailored-heuristics training. Building on Drexler, Fischer, and Schoar (2014), this intervention delivers and evaluates a similar financial-heuristics training via lower-touch mobile-phone delivery. A key advantage of mobile delivery is that programs can be scaled quickly, at low cost, with high fidelity (a recorded message is identical whether sent to one thousand or one million people, while scaling traditional “train the trainers” programs could require significant human resources and management). Partnering with microfinance institutions (MFIs) in a two-site field experiment in the Philippines and India, financial-heuristics training content was de- livered to micro-entrepreneurs via weekly audio messages over their mobile phones. Combining various recommended practices into a binary indicator of improved practices, this study finds that the low-touch intervention increases adoption of recommended practices (on a binary scale) by 0.01 to 0.019 (0.06 to 0.12 standard deviation points), roughly 20–40 percent of the effect size found in McKenzie (2020)’s meta-analysis of higher-cost and higher-touch traditional classroom training programs. While evidence of change in business practices is seen, effects on sales and profits are substantially noisier with point estimates of effects statistically indistinguishable from zero. This paper contributes to two broad strands of related literature. The first is on managerial skill, firm productivity, and economic growth. Managerial skill can contribute to firm productivity (Bruhn, Schoar, and Karlan 2010). Some businesses in low-income-country settings employ poor business practices (Bloom et al. 2012) and these practices hamper productivity (Hsieh and Klenow 2009). In a randomized controlled trial with Peruvian group lending clients, Karlan and Valdivia (2011) find little or no evidence of any marginal effect of an add-on business-training module on revenues and profits. McKenzie (2020) provides a comprehensive review of the training literature. While some individual studies find that customized management consulting advice can improve management practices and firm outcomes (Bloom and Van Reenen 2010; Bruhn, Karlan, and Schoar 2018; dalla Pellegrina et al. 2021), others are underpowered to detect meaningful effects on sales and profits. McKenzie (2020)’s meta-analysis finds positive effects on profits and sales of the order of magnitude of 5 to 10 percentage points. 582 Cole, Joshi, and Schoar Relative to the literature, this study finds that a much lower-touch intervention—mobile-phone-based heuristic financial advice—delivered over a longer period of time, in the preferred specification, can have 20–40 percent of the effect of these much higher-touch interventions. Additionally, the intervention is Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 longer lasting and less affected by the attendance and attrition problems other studies have faced.1 It is, also, in sites that few papers in this literature have reported on.2 The work presented in this paper is perhaps most closely related to Acimovic et al. (2020), which reports on a field experiment conducted by a mobile network operator with a goal of encouraging its independent sales agents to improve their inventory management. Agents who received both in-person training and explicit recommendations improved their performance. Lastly, this paper contributes to the literature on digital service delivery in low-income countries. Cole and Fernando (2020) find that agricultural advice delivered via a phone-based platform improves farmer decision making, and a meta-analysis by Fabregas, Kremer, and Schilbach (2019) suggests that such ap- proaches are effective at increasing agricultural yield in a variety of contexts. Aker, Ksoll, and Lybbert (2012) find that basic skills can be taught via mobile phone. The present study extends the application of mobile-phone-based delivery by showing that a mobile-phone-based heuristics training can in fact change entrepreneurs’ business-management behaviors. 2. Setting: Training and Intervention This study was conducted with micro-entrepreneurs in India and the Philippines. Micro, small, and medium enterprises employ a substantial share of the national labor force in both countries: MSMEs are 99 percent of all registered enterprises in the Philippines (Congressional Policy and Budget Research Department, Congress of the Philippines 2020) and employ approximately 63 million workers in India (Ministry of Micro and Medium Enterprises, Government of India 2019). Additionally, both countries have high mobile phone penetration (International Telecommunication Union 2019). In both countries, the mobile-phone-based heuristics training was rolled out to a subset of MSME clients of two trusted local MFIs. In the Philippines, the team partnered with Negros Women for To- morrow Foundation (NWTF), while in India, Janalakshmi was the partner MFI. In collaboration with these partners, extensive qualitative interviews with micro-entrepreneurs were conducted to understand the bottlenecks their businesses faced, the kinds of training that would be useful, and different delivery alternatives that could be explored. Prior to this study, a robust pilot was run in India (with Janalakshmi and the Institute for Financial Management and Research LEAD) which differed materially in execu- tion, but was similar in spirit. That pilot found no effects on behavior or business outcomes. Refinements to that pilot—in content, delivery method, technology partners, etc.—in conjunction with participating micro-entrepreneurs’ interviews, informed the structure of this intervention. The training curriculum was developed by a partner non-profit, ideas42, that oversaw the overall execution and project management of the field experiment. They did so in concert with extensive field interviews with partner institutions and micro-entrepreneurs and learnings from the pilot study. 1 Many traditional training programs deliver training over multiple days, typically a week, and require participants to travel to a training site. In this setting, participants do not need to travel to obtain training, contributing to a low training-attrition rate. In addition, it can be conjectured that the borrowing relationships between participants and partner microfinance institutions (through whom this training was delivered) may have also led to the low levels of attrition. 2 A close review of the literature found no studies set in the Philippines, and only one set in India, which evaluates the impact of business training on micro-entrepreneurs. dalla Pellegrina et al. (2021) find management training offered by a microfinance organization in India has a positive effect on financial management skills. Bloom et al. (2012) was conducted in India but involves an intense management consulting program. The World Bank Economic Review 583 Treated entrepreneurs received a 30-minute in-person orientation to the training that they would re- ceive over the subsequent 20+ weeks via pre-recorded weekly interactive voice response (IVR) messages on their mobile devices. The in-person orientation sessions were held in group settings with 25–30 treated Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 entrepreneurs in each session.3 The orientation session had three objectives. First, they introduced the training program to the treated MFI clients and set their expectations on what to expect over the coming weeks. Second, it ensured the MFI clients knew how to use their mobile phones to receive incoming IVR calls.4 Finally, the in-person orientation introduced clients to the concept of “cash separation,” one of the four key pillars of the training curriculum.5 Clients were introduced to the first module at the brief in- person orientation session where they were given two handouts in their local language. One summarized how to access the training service via their phones and the second was a visual aid describing the concept of cash separation between household and business. The training curriculum was built upon Drexler, Fischer, and Schoar (2014) and contained simple rules of thumb on business management organized into four modules: (1) Cash separation (profit calculation): Presented micro-entrepreneurs with simple action steps around how to separate business and household cash, and to pay themselves a fixed weekly salary, in order to better monitor their own business’s profitability. (2) Customer credit: Provided simple rules of thumb on when and when not to offer credit to customers. (3) Inventory management: Presented simple action steps on how to manage inventory of a retail business. (4) Supplier management: Provided action steps on selecting reliable suppliers that offer the best price and product quality. Exemplary messages of the specific heuristics taught are provided in supplementary online appendices S1A and S1B. The core of the intervention was 3- to 4-minute-long audio messages that delivered the training content each week via IVR calls to treated micro-entrepreneurs. Entrepreneurs were called at their preferred times, which they had specified at the in-person orientation. Entrepreneurs could also access a missed call service by ringing the training number to receive a free call back to access the training content from the current week and the previous week. The training was offered free of charge to participants. The training was delivered in a soap-opera format to bolster engagement. The lead character (female voice actor) in the series played the part of a successful small business owner with years of experience who offered the micro-entrepreneurs practical tips on business management that she had learned over the years from running her own business. The messages were delivered to the treated entrepreneur in their local language weekly for a total of 21 weeks in the Philippines6 (between August 2016 and January 2017) 3 Clients who were unable to attend the in-person training were oriented individually. 4 The IVR service was based on outgoing “push” calls, which automatically called MFI clients and played a recorded message when MFI clients answered the phone. Treated clients could also dial the training phone number and leave a missed call in order to trigger a call back to listen to the previous two weeks’ messages. These outbound weekly calls were free for treated MFI clients. The decision to use voice calls meant that all mobile phone types were compatible with the service (most of the micro-entrepreneurs in the study had basic phones, and not smart phones), and sidestepped concerns about literacy with text-based training. 5 The MFI partners felt in-person orientation was necessary due to the novel nature of mobile-phone service delivery. The overall treatment can be thought of as a bundle of financial heuristics delivered via mobile phone and this brief in-person orientation. The majority of treated clients received the orientation: 95 percent in the Philippines and 92 percent in India. 6 In the Philippines, 21 training messages were delivered (vs. 22 in India) because a message on availing cash discounts from suppliers was not relevant for the Filipino entrepreneurs as they did not buy their stock on credit. Indian entrepreneurs, on the other hand, often bought goods for sale on credit and as a result received an additional message in the supplier management module that taught them to consider paying for inventory in full at the time of purchase, and asking for a discount for paying in full to increase their profit margin. 584 Cole, Joshi, and Schoar and 22 weeks in India (between August 2016 and March 2017).7 The per participant cost of training delivery (airtime only) in India was $2.04 for messages in Hindi (total of 73 minutes) and $2.38 for messages in Kannada (total of 85 minutes). In the Philippines, where airtime charges are much higher, the Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 per participant cost of training delivery was $14.99 (total of 81 minutes).8 In this study, microfinance institutions, which have been extensively studied, were partnered with. Banerjee, Karlan, and Zinman (2015), as well as Meager (2022), summarize the evidence of the causal effects of expansion of microcredit, and, importantly, find heterogeneous effects, with limited impact on sales or profits for the median borrower, but large gains for borrowers at the upper tail. This study seeks to measure the incremental benefit of advisory services on top of credit. 3. Experimental Design The two-site experiment was conducted in the Philippines and in India. The experimental design in each setting is described below. The field experiment in the Philippines ran from March 2016 to June 2017. The sample was drawn from active group loan clients of the partner MFI: NWTF. To be eligible for the experiment, clients had to speak Hiligaynon, manage a retail business, and have access to a mobile phone. The baseline data- collection exercise was conducted from March to June 2016. It consisted of in-person interviews with eligible clients to gather detailed information on their demographics, business ownership, financial and managerial practices, and business outcomes. A total of 2,096 clients were interviewed. Clients were then randomly assigned into a control arm (1,030 clients) and a treatment arm (1,066 clients). As the clients were beneficiaries of group loans, randomization was carried out at the group level to account for the possibility of spillovers; 676 groups were randomly assigned to treatment and 675 into control. Stratified group randomization by the number of members in each group (which ranged from one to five) was done to ensure an equal number of treatment and control clients for each stratum of group size. As described earlier, clients in the treatment group received an in-person orientation. The orientation sessions were conducted by NWTF staff and held at weekly MFI group meeting sites where group mem- bers met on a weekly basis as part of their group lending program with NWTF. About 78 percent of treatment clients attended the group orientation sessions. For those clients who were unable to attend the initial large group session, make-up orientation was held by NWTF staff in a smaller group or individual settings. Roughly 5 percent of treatment clients did not receive any orientation. Treatment group clients received weekly heuristic training messages for a total of 21 weeks between August 2016 and January 2017. Endline in-person surveys were conducted between April and June 2017. A total of 1,898 clients were surveyed at the endline out of the baseline sample of 2,096. The field experiment in India ran from March 2016 to July 2017. The sample was drawn from indi- vidual loan clients of partner MFI Janalakshmi. To be eligible for being part of the sample, clients had to speak Kannada, Hindi, or Urdu, and—just as in the Philippines—manage a retail business and have access to a mobile phone. The study began with a total sample of 2,407 clients in Bangalore (henceforth referred to as Wave-1). The sample was increased by 1,442 clients (Wave-2) midway through the experi- ment in order to increase the study’s statistical power. The Wave-2 sample was drawn based on the same eligibility criteria and included clients from Bangalore, Mysore, Davangere, Gulbarga, Indore, and Delhi. Baseline data collection for Wave-1 clients was conducted between March and May of 2016. However, 7 The India sample was augmented after the start of the experiment in a manner explained in the Experimental Design section . 8 This per participant cost includes the airtime used for listening to all training messages in entirety. Actual costs incurred are in fact lower, as the service only charges for airtime used in a call. Because content from a successful intervention could quite easily be scaled to reach hundreds of thousands (or even millions) of entrepreneurs, focus is put on marginal costs, rather than the content development cost. The primary content development costs were staff time of ideas42. The World Bank Economic Review 585 a baseline survey for Wave-2 clients was not conducted. In India, randomization was conducted at the individual level. For Wave-1 clients, randomization was stratified by MFI branch in order to ensure an equal number of treatment and control clients at each branch. Wave-1 randomization was conducted after Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 baseline data collection. Wave-2 randomization was also conducted at the individual level, but stratified by region and language to ensure an equal proportion of treatment and control clients in each region and for each language. The in-person, group orientation sessions were held at Janalakshmi’s local offices. Only 22 percent of clients attended these sessions. Non-attendees were followed up individually for a make-up orientation conducted on an individual basis by the research team.9 About 70 percent of treatment clients received individual make-up orientation sessions. Heuristic training messages were sent out for a total of 22 weeks in India. Wave-1 clients received messages from August to December 2016 and were interviewed for endline data collection between April and June 2017. Wave-2 clients received messages from October 2016 to March 2017. Their endline surveys were conducted between June and July 2017. A total of 3,318 clients were surveyed at the endline out of the baseline sample of 3,849. 4. Data and Estimation Strategy 4.1. Data The primary data sources of this study are the in-person interviews conducted at baseline and endline with the study participants. The baseline and endline surveys in both sites were conducted in private, one-on-one interviews between enumerators and the entrepreneurs, who were informed that the financial institution would not have access to their individual data from the surveys. For impact evaluation, self- reported measures, including those on firm productivity, which are imperfect (de Mel, McKenzie, and Woodruff 2009) but allow for comparability to a number of other studies, are relied on. These survey responses were augmented with administrative data on the study participants from the partner MFIs. Pick-up and listenership data from the IVR platform provider were also collected to gauge engagement of the treated clients with the training content. Tables 1A and 1B report baseline summary statistics and balance tests for the two samples. In the Philippines, the typical study participant is, on average, a 45-year-old female entrepreneur, who in 72 percent of cases has a high-school diploma. In the Philippines, 53 percent of the sample entrepreneurs report that their current business is their primary source of income. The sample in India is somewhat different. The typical study participant is, on average, a 38-year-old female entrepreneur. Only 35 percent of the Indian sample has a high-school diploma and 49 percent report that their business is their primary source of income. The second panel reports a snapshot of the business practices that sampled entrepreneurs adopt at the baseline. In the Philippines, 68 percent report they separate business and household cash, 77 percent calculate profits, 78 percent give credit for no more than seven days, 77 percent keep business records, 70 percent keep records of customer credit, 68 percent record important information of customer credit, and 76 percent take full advantage of cash discounts offered by suppliers. These baseline adoption measures are similar to those in Drexler, Fischer, and Schoar (2014) in the Dominican Republic wherein 74 percent of sampled entrepreneurs reported separating business and personal cash, 66 percent kept reports of their accounts, and 81 percent formally calculated their revenues. Baseline adoption of recommended business practices is considerably lower in India, as table 1B il- lustrates. Of the sampled entrepreneurs at baseline (Wave-1), only 35 percent separate household and business cash, 81 percent calculate profits, 28 percent give credit to customers for a week, and only 40 percent keep business records. Unlike in the Philippines, only 42 percent keep records of customer credit, 9 Innovations for Poverty Action (IPA) and Institute for Financial Management and Research (IFMR-LEAD) oversaw the implementation of the experiment in the Philippines and India respectively. 586 Cole, Joshi, and Schoar Table 1A. Summary Statistics and Balance-Philippines. (1) (2) (3) (2)-(3) Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Total Sample Control Treatment Pairwise t-test Variable N Mean/(SD) N Mean/(SD) N Mean/(SD) N P-value A. Client Characteristics Age 2096 44.688 1030 44.734 1066 44.644 2096 0.852 (11.075) (11.184) (10.973) Female 2096 0.998 1030 0.999 1066 0.996 2096 0.192 (0.049) (0.031) (0.061) Education High School or Above 2096 0.722 1030 0.731 1066 0.714 2096 0.380 (0.448) (0.444) (0.452) Business Type Business is Primary Source of Income 2096 0.528 1030 0.520 1066 0.535 2096 0.512 (0.499) (0.500) (0.499) Retail: Food 2096 0.524 1030 0.533 1066 0.516 2096 0.434 (0.500) (0.499) (0.500) B. Business Practices Do Separate Business & Household Cash 2096 0.683 1030 0.676 1066 0.689 2096 0.499 (0.466) (0.468) (0.463) Do Pay Salary to Self 2096 0.152 1030 0.141 1066 0.162 2096 0.170 (0.359) (0.348) (0.369) Do Calculate Profits 2096 0.773 1030 0.763 1066 0.782 2096 0.293 (0.419) (0.425) (0.413) Give customers credit for at most 7 days 1793 0.779 876 0.759 917 0.797 1793 0.053∗ (0.415) (0.428) (0.402) Do nothing when customers do not pay credit 2096 0.102 1030 0.100 1066 0.104 2096 0.755 (0.303) (0.300) (0.306) Keep Business Records 2096 0.771 1030 0.773 1066 0.770 2096 0.885 (0.420) (0.419) (0.421) Keep Customer Credit Records 1802 0.701 881 0.720 921 0.684 1802 0.099∗ (0.458) (0.449) (0.465) Record Important Customer Credit Information 1802 0.680 881 0.699 921 0.662 1802 0.093∗ (0.466) (0.459) (0.473) Determine stock based on a good strategy 2096 0.239 1030 0.232 1066 0.246 2096 0.461 (0.427) (0.422) (0.431) Never Visit competitors to check prices/quality 2096 0.758 1030 0.775 1066 0.742 2096 0.080∗ (0.428) (0.418) (0.438) Never Talk to customers to understand needs 2096 0.532 1030 0.541 1066 0.523 2096 0.427 (0.499) (0.499) (0.500) Never do supplier quality comparison 2096 0.388 1030 0.396 1066 0.380 2096 0.447 (0.487) (0.489) (0.486) Never Negotiated terms with suppliers 2096 0.504 1030 0.514 1066 0.495 2096 0.403 (0.500) (0.500) (0.500) Took full advantage of cash discount 592 0.762 288 0.757 304 0.766 592 0.787 (0.426) (0.430) (0.424) C. Business Performance Sales-Regular Week (Winsorized at 1%) 1930 6093.881 944 6287.659 986 5908.357 1930 0.293 (7917.268) (8647.279) (7148.402) Profits-Regular Week (Winsorized at 1%) 1961 2343.967 961 2297.904 1000 2388.233 1961 0.459 (2702.364) (2612.478) (2786.603) Source: Primary data collected by research team Notes: This table presents summary statistics based on baseline survey data. Standard deviations (column 2, 3, 4) of variables and p-values (column 5) appear in parentheses. ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗∗∗ at the 1%-level The World Bank Economic Review 587 Table 1B. Summary Statistics and Balance-India. (1) (2) (3) (2)-(3) Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Total Sample Control Treatment Pairwise t-test Variable N Mean/(SD) N Mean/(SD) N Mean/(SD) N P-value A. Client Characteristics Age 2407 38.371 1204 38.605 1203 38.138 2407 0.171 (8.356) (8.381) (8.328) Female 2407 0.795 1204 0.802 1203 0.788 2407 0.385 (0.404) (0.398) (0.409) Education High School or Above 2407 0.354 1204 0.354 1203 0.353 2407 0.978 (0.478) (0.478) (0.478) Business Type Business is Primary Source of Income 2407 0.492 1204 0.478 1203 0.506 2407 0.160 (0.500) (0.500) (0.500) Retail: Food 2402 0.271 1201 0.266 1201 0.275 2402 0.646 (0.444) (0.442) (0.447) B. Business Practices Do Separate Business & Household Cash 2407 0.351 1204 0.363 1203 0.340 2407 0.238 (0.478) (0.481) (0.474) Do Pay Salary to Self 2407 0.045 1204 0.042 1203 0.048 2407 0.429 (0.207) (0.200) (0.214) Do Calculate Profits 2407 0.818 1204 0.830 1203 0.805 2407 0.124 (0.386) (0.376) (0.396) Give customers credit for at most 7 days 1325 0.278 660 0.253 665 0.304 1325 0.039∗∗ (0.448) (0.435) (0.460) Do nothing when customers do not pay credit 1324 0.073 659 0.070 665 0.077 1324 0.631 (0.261) (0.255) (0.266) Keep Business Records 2407 0.396 1204 0.393 1203 0.398 2407 0.790 (0.489) (0.489) (0.490) Keep Customer Credit Records 1327 0.417 662 0.421 665 0.412 1327 0.728 (0.493) (0.494) (0.493) Record Important Customer Credit Information 1327 0.101 662 0.110 665 0.092 1327 0.263 (0.301) (0.313) (0.289) Determine stock based on a good strategy 2407 0.160 1204 0.165 1203 0.155 2407 0.475 (0.367) (0.372) (0.362) Never Visit competitors to check prices/quality 2407 0.568 1204 0.566 1203 0.569 2407 0.916 (0.496) (0.496) (0.495) Never Talk to customers to understand needs 2407 0.639 1204 0.639 1203 0.639 2407 0.978 (0.480) (0.481) (0.480) Never do supplier quality comparison 2407 0.360 1204 0.369 1203 0.352 2407 0.381 (0.480) (0.483) (0.478) Never Negotiated terms with suppliers 2407 0.262 1204 0.262 1203 0.262 2407 0.990 (0.440) (0.440) (0.440) Took full advantage of cash discount 739 0.388 364 0.382 375 0.395 739 0.722 (0.488) (0.487) (0.489) C. Business Performance Sales-Regular Week (Winsorized at 1%) 2399 14114.973 1199 13702.085 1200 14527.517 2399 0.174 (14870.715) (14358.239) (15360.613) Profits-Regular Week (Winsorized at 1%) 2389 5232.516 1194 5128.308 1195 5336.636 2389 0.290 (4809.680) (4710.809) (4906.242) Source: Primary data collected by research team Notes: This table presents summary statistics based on baseline survey data. Standard deviations (column 2, 3, 4) of variables and p-values (column 5) appear in parentheses. ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗∗∗ at the 1%-level 588 Cole, Joshi, and Schoar 10 percent note important customer credit details, and 39 percent take advantage of cash discounts offered by suppliers. Average sales for sampled entrepreneurs are about 6,094 pesos ($128) in the Philippines and 14,115 Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 rupees ($210) in India. Profits in a regular week are, on average, 2,344 pesos ($49) in the Philippines and about 5,233 rupees ($78) in India. In Drexler, Fischer, and Schoar (2014) reported sales in an aver- age week were on 6,399 Dominican pesos ($181).10 An important strength of this study is that it tests whether similar business training can affect business practices in three different settings—starting with the Dominican Republic and then in India and the Philippines. In comparing the two study sites, it is worth noting that the baseline adoption of recommended prac- tices is substantially higher in the Philippines. This may be due to the higher level of education of the Filipino entrepreneurs in the sample (72 percent high school or better versus 35 percent) since education level and business practices are positively correlated (correlation not reported). Additionally, the market structures likely vary substantially across the two sites. Listenership rates in India and the Philippines are reported in tables 2A and 2B. These rates provide a snapshot of participant engagement with the program. The average pickup rate across the two countries was 76 percent. Listenership rate conditional on pickup across the two countries was 84 percent. Table 2A. Pickup & Listening Rates-Philippines. Pick-up Rate (%) Listening Rate Regardless Listening Rate Conditional of Pick-up (%) on Pick-up (%) Module (1) (2) (3) Cash Separation 78% 67% 89% Customer Credit 72% 61% 89% Inventory Management 70% 60% 90% Supplier Management 68% 56% 85% Overall 73% 62% 92% Source: Primary data collected by research team Notes: This table presents the pick up rates and listening rates for each section of financial heuristics training curriculum. Pick up rate = number of pickups / total number of calls. Listing rate regardless of pick up = listenership / total duration. Listening rate conditional on pick up = listenership / total duration if call picked up. Table 2B. Pickup & Listening Rates-India. Pick-up Rate (%) Listening Rate Regardless Listening Rate Conditional of Pick-up (%) on Pick-up (%) Module (1) (2) (3) Cash Separation 83% 52% 67% Customer Credit 81% 50% 66% Inventory Management 78% 46% 65% Supplier Management 72% 44% 68% Overall 79% 48% 76% Source: Primary data collected by research team Notes: This table presents the pick up rates and listening rates for each section of financial heuristics training curriculum. Pick up rate = number of pickups / total number of calls. Listing rate regardless of pick up = listenership / total duration. Listening rate conditional on pick up = listenership / total duration if call picked up. 10 Exchange rate in 2016: India: USD 1 = INR 67.18; the Philippines: USD 1 = PHP 47.49. The intervention in Drexler, Fischer, and Schoar (2014) was implemented in the Dominican Republic between March and May 2007 at which time the exchange rate was roughly 1 USD = 35.29 Dominican pesos. Exchange rates sourced from https://www.exchangerates. org.uk/. The World Bank Economic Review 589 4.2. Estimation Strategy The primary purpose of this paper was to examine whether mobile-phone-based business training could impact business practices and firm productivity. As described earlier, this study was inspired by Drexler, Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Fischer, and Schoar (2014), and thus focuses on the outcomes identified in that paper, namely, business practices, sales, and profits.11 Since treatment was randomly assigned, the differences estimator, specified in the model below, provides unbiased estimates of the target estimand—the training’s average treatment effect: yE B i = α + β1 Ti + β2 yi + γ Wi + εi , where yE i is the endline outcome of interest, Ti is a treatment dummy, Wi is a vector of controls, and yi B 12 is the baseline measure of the outcome variable and is included where available. The outcomes of interest are business practices, sales, and profits. The business-practice measures col- lected at endline were enumerated on a three-point scale, with 1 being the least desired and 3 being the most desired outcome in reference to the practices taught in the training. For instance, it was asked how often entrepreneurs contact customers whose credit is due. An entrepreneur answering “None of the time” would get a score of 1, “Some of the time” would get a score of 2, and “Often/all of the time” would get a score of 3. Much of the literature reports business outcomes as binary practices, rather than the ranges used in this paper. For example, de Mel et al. (2009) state that “for every 20 practices that business train- ing attempts to teach firms to do, on average firms invited to training only implement one additional practice.” To provide some (admittedly imperfect) comparability, for each business practice, this paper converts the three-value range into a binary variable.13 ,14 Both results are presented in table 3, but the binary index measure is discussed for comparability to other papers. In terms of productivity measures, “regular week” sales and profits are considered as outcomes of interest. These are reported in levels winsorized at the 1 percent level as well as in logs.15 For both sales and profits, respondents were asked three questions at endline. The first two included enumeration of sales/profits on the previous day and an assessment of whether the previous day’s profits were “good,” “bad,” or “regular.” This was followed by respondents being asked to report sales/profits in a typical week.16 Under this baseline model, standard errors are clustered at the group level in the Philippines given the weekly group meetings in that setting. Heteroskedasticity robust standard errors are reported in India where randomization was done at the individual level. 11 A pre-analysis plan was not created. 12 Covariates in the Philippines include location, age of business, own a mobile phone indicator, primary source of income indicator, education level, business type, and variables used for stratification: number of clients in each group. Covariates in India include wave dummy, time of survey, gender, age of business, own a mobile phone indicator, primary source of income indicator, education level, business type, and variables used for stratification: branch and language. 13 This is done by assigning the low value 0, the high value 1, and the intermediate value to zero if there are more lows than highs for that item, or to 1 if there are more highs than lows. This, in effect, passes the three-point scale through an above/below median filter to convert it into a binary outcome. 14 The relevant business practices include separating business and household cash, paying a fixed weekly salary to self, calculating profits, giving credits for no more than seven days, calling the customers whose credit is due, keeping business and credit records, buying more of the most popular products and less of the least popular products, visiting competitors to check out price, talking to customers to check out need, introducing new products, comparing price and quality of various suppliers, negotiating prices and terms with suppliers, and taking advantage of cash discount from suppliers. 15 It can be noted regarding the log transformation that no entrepreneurs in the sample reported null or negative profits for a “regular week.” 16 The precise wording of the sales questions were “(1) What were the sales in your business yesterday? (Sales from primary business only); (2) How would you classify yesterday in terms of sales?; (3) Can you tell us what the average sales per week are in your business?” Respondents were asked the same question for profits. 590 Cole, Joshi, and Schoar Table 3. Intent to Treat Analysis. Philippines India Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Dependent Variables Control Mean Treatment Effect Control Mean Treatment Effect (1) (2) (3) (4) Business Practice Index (1-3) 1.911 .037∗∗∗ 1.773 .021∗∗ [.282] (.01) [.28] (.01) .006 .027 N 1897 3311 Business Practice Index (0-1) .47 .019∗∗ .365 .01∗ [.155] (.01) [.157] (.01) .01 .058 N 1897 3311 Regular Week Sales-Winsorized at 1% 6918.448 −483.22 11153.823 686.093 [9520.558] (378.07) [13151.163] (447.34) .201 .125 N 1733 3311 Regular Week Sales-Log Transformed 8.272 −.049 8.782 .065 [1.084] (.05) [1.21] (.04) .288 .104 N 1733 3311 Regular Week Profits-Winsorized at 1% 2210.957 −91.64 4973.52 −11.976 [2512.83] (106.96) [5012.34] (167.88) .392 .943 N 1773 3306 Regular Week Profits-Log Transformed 7.189 −.017 8.066 −.007 [1.23] (.06) [1.16] (.04) .765 .86 N 1773 3306 Source: Primary data collected by research team Notes: This table presents the impact of training on business practices and performance for the experiments conducted in India and the Philippines. Control means are presented in columns 1 and 3 and standard deviations in square brackets. Each coefficient reported in columns 2 and 4 is from the regression for each outcome variable on the treatment variable. Covariates include time of survey (India), gender, age of business, own a cellphone indicator, primary source of income indicator, education level, business type, and variables used for stratification and language. Heteroskedasticity-robust standard errors in parentheses. ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗∗∗ at the 1%-level. The analysis also tests for heterogenous treatment effects along four dimensions of heterogeneity, namely, the entrepreneur’s level of education, age, business size, and baseline adoption of recommended business practices. This is done by running the following model on the full sample: yE B i = α + β1 Ti + β2 Xi + β3 Ti ∗ Xi + β4 yi + γ Wi + εi , (1) where Xi is a dummy that is turned on when the entrepreneur has an above median measure of the relevant axis of heterogeneity being tested. For instance, for age, Xi = 1 if the entrepreneur is as old as or older than the median age of the sample and zero otherwise. In this model, β ˆ 3 is the estimate of interest. To correct for multiple hypothesis testing in the heterogeneity analysis, we calculate and report Anderson (2008) sharpened q-values. 5. Results 5.1. Uptake and Engagement Take-up and engagement with the training content is presented in tables 2A and 2B. In India, the mean (median) number of calls answered was 16.7 (19), while in the Philippines it was 15.3 (18). The uptake The World Bank Economic Review 591 of the program, as measured by pick-up rates, was high, with around three-quarters of calls picked up across all four training modules in both sites. Moreover, participants were engaged with the training as listenership rates were above 60 percent in the Philippines and about 48 percent in India.17 Participants Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 were more engaged with the first two modules, which covered cash separation and customer credit, com- pared to the later modules on inventory and supplier management. Additionally, around 32 percent of treated beneficiaries in the Philippines used the missed call service at least once and about 40 percent did so in India. As part of the endline data collection, participants’ feedback on the training was also collected. About 78 percent of the training participants in the Philippines and 62 percent of the training participants in India reported that they were likely to recommend this training program to their family, friends, and other business owners like them. Tables 4A and 4B report predictors of engagement with the training content. Pick-up and listenership rates increase with age in the Philippines but not in India. Owning a mobile phone (as opposed to having access via some other means, say a family member, for instance) is the strongest predictor of engage- ment with the training calls in both contexts. In the Philippines, education is negatively correlated with engagement. Engagement does not appear to vary materially by other covariates. 5.2. Impact on Practices and Productivity The main experimental findings are reported in table 3. Financial-heuristics training delivered via mobile phones increases the adoption of recommended business practices by 0.01 (India) to 0.02 (Philippines) on the unweighted average binary business-practice index. These results are significant at the 10 percent and 5 percent levels respectively. The magnitude of this change in business practices is 0.06–0.12 standard deviation points, an effect size of 20–40 percent of the magnitude of the effect of traditional training programs suggested by McKenzie (2020)’s meta-analysis. The results are of comparable magnitude but more statistically significant when expressed on the three-point range over which initial survey responses were enumerated. No statistically significant changes are found in firm productivity. Point estimates are positive for rev- enues in the Indian sample and mildly negative in the Philippines. Point estimates for profits are near zero for both samples as well. For both sales and profits in both sites, none of the productivity measures are statistically distinguishable from zero at the 10 percent confidence level. Attrition is relatively low in both the Philippines (treatment (9.1 percent) and control (9.8 percent)) and India (treatment (13.88 percent) and control (13.71 percent)). In both settings the attrition rate is statistically indistinguishable across treatment and control groups. Evidence of differential attrition is examined by testing whether there is a statistically significant difference in baseline variables between the treatment and control observations in the attrited samples. However, no statistically significant difference is found (supplementary online tables S2A and S2B). Why might not the same positive, statistically significant effects on sales and profits in this sample as Drexler, Fischer, and Schoar (2014) report in the Dominican Republic be observed? A number of possibilities are explored. One possibility is that mobile-phone-based intervention is simply weaker: while the barriers to attendance were lower, the total “contact time” in this intervention was much lower.18 Another possible explanation is that generalized heuristics are not optimized for specific entrepreneurs and as such will not drive growth in sales and profits in some specific settings for some entrepreneurs. 17 An individual’s listenership rate is calculated as the total number of minutes listened to by a client (across all messages) divided by the total number of minutes of content the participant would have heard had they listened to all messages in entirety. If a client never answered a call, then their listenership rate would be zero. 18 An alternative possibility is that entrepreneurs in the Dominican Republic respond more to business training; unfor- tunately, conducting this study with a sufficient number of entrepreneurs in the Dominican Republic would have been prohibitively expensive. 592 Cole, Joshi, and Schoar Table 4A. Predictors of Engagement-Philippines. Dependent Variables Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Pick Up Rate (%) Listenership- Regardless of Pickup(%) Predictors (1) (2) (3) (4) Age of Respondent .255∗∗∗ .248∗∗ .243∗∗ .217∗∗ (.09) (.1) (.1) (.11) Urban 3.836 4.161 4.343∗ 3.649 (2.4) (2.63) (2.52) (2.74) Age of Business −.05 −.047 .037 .085 (.13) (.14) (.14) (.14) Business is Primary Source of Income −2.208 −1.519 −3.581∗ −2.779 (1.88) (2.03) (1.93) (2.08) Own a Cellphone 11.145∗∗∗ 10.075∗∗∗ 13.495∗∗∗ 13.391∗∗∗ (3.09) (3.23) (3.05) (3.19) Less than 5th class −16.281∗∗∗ −15.403∗∗ −13.988∗∗ −12.742∗∗ (5.94) (6.02) (5.76) (5.85) Above 5th class −11.317∗∗∗ −11.2∗∗∗ −11.914∗∗∗ −11.354∗∗∗ (2.75) (2.97) (2.84) (3.08) Completed High School −5.347∗∗ −4.697∗ −5.908∗∗ −5.185∗∗ (2.24) (2.44) (2.39) (2.6) Sari Sari Store 1.648 5.002 5.605 9.218 (4.61) (5.9) (5.22) (6.47) Food Retail −2.063 .445 1.364 4.312 (4.56) (5.82) (5.21) (6.43) Baseline Practice Score −.697 −.573 (3.64) (3.64) Log-Baseline Regular Week Sales −1.664 −1.607 (1.17) (1.22) Log-Baseline Regular Week Profits −.52 −.315 (.91) (.99) Constant 59.382∗∗∗ 75.553∗∗∗ 43.664∗∗∗ 56.915∗∗∗ (6.85) (12.45) (7.54) (13.1) N 1066 947 1066 947 Source: Primary data collected by research team Notes: This table presents the predictors of engagement in training. Pickup rate (%) and listenership rate (%) are regressed on characteristics of participants. Urban takes value 1 for urban particiapnts and 0 for rural participants. The fourth level of education is graduate/post graduate, which is omitted due to multicollinearity. The third type of business is non-food retail, which is omitted due to multicolinearity. Standard errors, clustered at the group-level, in parentheses. Business practice score, ranging from 1 to 3, is a scaled score such that higher score indicate better business practices. ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗ ∗ ∗ at the 1%-level. Participants in India were affected by a demonetization policy that may have negatively impacted business. Moreover, there is in fact very little systematic evidence isolating the specific business practices that drive the most sales and profit growth. Like most of the literature, this paper evaluates a bundled program, based on the best curriculum that could be devised for this study. But perhaps the suggestion of credit limits for certain customers in some contexts, for example, might have limited sales. To summarize the main finding, the primary purpose of this study was to examine whether mobile- phone-based training could affect business practices. While evidence of change in business practices is seen, a shortcoming of this paper’s approach is that there is limited information on financial performance. While the business practices are relatively precisely measured, financial outcomes, such as sales and profits, are unfortunately substantially noisier. Even with a sample of 3,849 entrepreneurs in India and 2,096 in the Philippines, when we consider regular week profits winsorized at the 1 percent level, the standard The World Bank Economic Review 593 Table 4B. Predictors of Engagement-India. Dependent Variables Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Pick Up Rate (%) Listenership- Regardless of Pickup(%) Predictors (1) (2) (3) (4) Age of Respondent .035 .03 .142 .148 (.1) (.1) (.13) (.13) Female −2.691 −2.61 1.984 2.31 (1.86) (1.97) (2.6) (2.68) Age of Business −.125 −.146 −.095 −.084 (.11) (.12) (.15) (.16) Own a Cellphone 53.488∗∗∗ 52.778∗∗∗ 33.522∗∗∗ 32.723∗∗∗ (3.25) (3.12) (6.79) (6.96) Less than 5th class 7.917∗∗ 8.054∗∗ .594 1.407 (3.94) (4.04) (5.04) (5.1) Above 5th class 3.736 4.117 .37 1.194 (3.97) (4.07) (4.96) (5.01) Completed High School 3.409 3.767 1.311 1.964 (4.02) (4.11) (5.02) (5.07) Shop −3.502∗ −3.846∗ −4.144∗ −3.957 (1.96) (2.04) (2.49) (2.54) Food Retail −2.147 −2.604 −2.392 −2.187 (2.07) (2.13) (2.8) (2.86) Baseline Practice Score −2.539 4.831 (3.23) (3.23) Log-Baseline Regular Week Sales .735 .688 (1.2) (1.25) Log-Baseline Regular Week Profits −.362 −.215 (.99) (1.09) Constant 27.934∗∗∗ 29.256∗∗ 21.216∗∗ 7.595 (7.1) (11.83) (10.49) (15.36) N 979 972 979 972 Source: Primary data collected by research team Notes: This table presents the predictors of engagement in training. Pickup rate (%) and listenership rate (%) are regressed on characteristics of participants, controlling for wave dummy and language. The fourth level of education is graduate/postgraduate, which is omitted due to multicollinearity. The third type of business is non-food retail, which is omitted due to multicollinearity. Heteroscedasticity-robust standard errors in parentheses. Business practice score, ranging from 1 to 3, is a scaled score such that higher score indicates better business practices. ∗ Denotes significance at 10%-level, ∗∗ at the 5%-level, and ∗ ∗ ∗ at the 1%-level. error is approximately 5 percent of the sample mean in the Philippines and about 3.4 percent in India, which prevents the ruling out of meaningful economic effects. Like most evaluations in this literature, only one program is evaluated, and therefore the paper is unable to answer important questions such as whether training combined with credit is more effective than training alone, or whether the identity of the provider of information affects take-up. These are important design questions, and it can be noted that a digitally designed and delivered service, such as the one evaluated here, may be particularly well suited to investigate these questions, through for example a series of A/B tests with a large population. 5.3. Heterogeneity Analysis There is no reason to believe that treatment effects must be homogeneous, and the differential findings in the Dominican Republic, India, and the Philippines suggest it is worth exploring treatment heterogeneity. Tables 5 and 6 report heterogeneous effects, as estimated by equation (1), along the dimensions expected to matter most. 594 Cole, Joshi, and Schoar Table 5. Heterogeneous Impact of Training. PHILIPPINES Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Level of Education Age of Entrepreneur Outcome Variables Treatment Low Treatment ∗ Low Treatment Old Treatment ∗ Old Business Practices Index .0305∗ −.0157 .0201 .0631∗∗∗ .0426∗∗ −.0532∗∗ Standard Error (.02) (.02) (.03) (.02) (.02) (.03) P-Value .053 .445 .49 .001 .028 .046 Sharpened q-value .451 .147 N 1897 1897 1897 1897 1897 1897 Regular Week Sales-Winsorized at 1% −358.9 120.7 −466.6 −418.7 658.9 −139.5 Standard Error (445.88) (609.08) (780.28) (534.91) (633.02) (753.96) P-Value .421 .843 .55 .434 .298 .853 Sharpened q-value 1 1 N 1733 1733 1733 1733 1733 1733 Regular Week Profits-Winsorized at 1% −153 −212.4 198.9 −229.5∗ 231.4 257.2 Standard Error (127.36) (165.25) (229.33) (136.74) (173.59) (212.17) P-Value .23 .199 .386 .093 .183 .226 Sharpened q-value 1 .825 N 1773 1773 1773 1773 1773 1773 PHILIPPINES Size of Business Baseline Business Practices Outcome Variables Treatment Small Treatment ∗Small Treatment Low Score Treatment ∗ Low Business Practices Index .0147 −.0327∗ .047∗ .0293 −.0071 .0132 Standard Error (.02) (.02) (.03) (.02) (.03) (.03) P-Value .418 .073 .064 .136 .777 .621 Sharpened q-value .147 .451 N 1897 1897 1897 1897 1897 1897 Regular Week Sales-Winsorized at 1% −1149.9∗ −905.1 1332.5∗ −204.4 959.4∗ −556.1 Standard Error (645.71) (623.06) (719.6) (527) (580.04) (746.36) P-Value .075 .147 .064 .698 .098 .456 Sharpened q-value .289 1 N 1733 1733 1733 1733 1733 1733 Regular Week Profits-Winsorized at 1% −319.9∗ −740.8∗∗∗ 480.6∗∗ −85.3 173.6 −24.9 Standard Error (169.31) (146.54) (204.46) (148.72) (151.81) (209.36) P-Value .059 0 .019 .567 .253 .905 Sharpened q-value .18 1 N 1773 1773 1773 1773 1773 1773 Source: Primary data collected by research team Notes: This table presents heterogeneous treatment effect using interaction terms. Endline business outcome are regressed on treatment dummy, subgroup variable, and the interaction between treatment dummy and subgroup variable, controlling for covariates. Covariates include age of business, own a cellphone indicator, primary source of income indicator, education level, business type, and variables used for stratification. Heteroskedasticity-robust standard errors in parentheses. Sharpened q-values that correct for multiple hypothesis testing are also presented ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗∗∗ at the 1%-level the 5%-level, and ∗∗∗ at the 1%-level Four dimensions of heterogeneity are focused on, namely, the entrepreneur’s level of education, age, business size, and baseline adoption of recommended business practices. Differences along the education dimension help to assess whether the effectiveness of training depends on the baseline level of human capital where we predict that micro-entrepreneurs with lower levels of educational attainment might be able to understand and apply the practices equally as well as micro-entrepreneurs with higher levels of The World Bank Economic Review 595 Table 6. Heterogeneous Impact of Training. INDIA Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Level of Education Age of Entrepreneur Outcome Variables Treatment Low Treatment ∗ Low Treatment Old Treatment ∗ Old Business Practices Index .0161 −.0586∗∗∗ .0051 .0278 .0203 −.0159 Standard Error (.02) (.02) (.03) (.02) (.02) (.02) P-Value .428 .001 .84 .111 .237 .509 Sharpened q-value 1 1 N 2034 2034 2034 2034 2034 2034 Regular Week Sales-Winsorized at 1% 249.7 −276.8 −164.4 −179.5 218.7 611.6 Standard Error (937.07) (795.9) (1137.75) (743.28) (764.65) (1059.9) P-Value .79 .728 .885 .809 .775 .564 Sharpened q-value 1 1 N 2027 2027 2027 2027 2027 2027 Regular Week Profits-Winsorized at 1% −184.2 −691.2∗∗ 197.1 −119.6 −81.7 119.4 Standard Error (392.52) (345.42) (478.08) (320.88) (320.24) (441.02) P-Value .639 .046 .68 .71 .799 .787 Sharpened q-value 1 1 N 2016 2016 2016 2016 2016 2016 INDIA Size of Business Baseline Business Practices Outcome Variables Treatment Treatment ∗ Small Small Treatment Low Score Treatment ∗ Low Business Practices Index .026 −.0289 −.011 −.002 −.032 .042∗ Standard Error (.02) (.02) (.02) (.02) (.02) (.02) P-Value .171 .103 .654 .914 .176 .082 Sharpened q-value 1 .489 N 2034 2034 2034 2034 2034 2034 Regular Week Sales-Winsorized at 1% −244.4 −195.3 649.3 −72.6 −428.6 425.3 Standard Error (1038.29) (1021.25) (1188.12) (766.08) (820.14) (1073.87) P-Value .814 .848 .585 .925 .601 .692 Sharpened q-value 1 1 N 2027 2027 2027 2027 2027 2027 Regular Week Profits-Winsorized at 1% −117.1 101.4 266.6 281.9 −635.7 −1196.2∗∗∗ Standard Error (417.11) (387.11) (478.2) (324.3) (328.9) (441.27) P-Value .779 .002 .832 .411 .392 .15 Sharpened q-value 1 .489 N 2016 2016 2016 2016 2016 2016 Source: Primary data collected by research team Notes: This table presents heterogeneous treatment effect using interaction terms. Endline business outcome are regressed on treatment dummy, subgroup variable, and the interaction between treatment dummy and subgroup variable, controlling for covariates. Covariates include age of business, own a cellphone indicator, primary source of income indicator, education level, business type, and variables used for stratification. Heteroskedasticity-robust standard errors in parentheses. Sharpened q-values that correct for multiple hypothesis testing are also presented ∗ Denotes significance at 10%-level, ∗ ∗ at the 5%-level, and ∗∗∗ at the 1%-level the 5%-level, and ∗∗∗ at the 1%-level educational attainment, given that the rule of thumb training is designed to be relatively easy to implement regardless of one’s educational background. The hypothesis that older individuals may be “set in their ways” is analyzed by measuring treatment heterogeneity by age. Testing along business size (baseline regular week sales used as a proxy) helps examine whether there are differential treatment effects for small versus large businesses. Micro-entrepreneurs with larger busi- 596 Cole, Joshi, and Schoar nesses may be less willing to change practices that have enabled them to reap larger sales. Additionally, micro-entrepreneurs with lower sales may have a greater incentive to learn and adopt more effective practices to increase their sales. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Finally, treatment heterogeneity by baseline business-practice adoption is also tested. It is hypothesized that micro-entrepreneurs with lower baseline adoption of recommended financial practices may have larger room for improvement and thereby see greater gains from the training. As tables 5 illustrates, there is evidence of heterogeneous treatment effects in the Philippines. Specif- ically, the treatment is seen to be twice as effective among young entrepreneurs compared to older en- trepreneurs, and to be substantially more effective among small businesses. The results in the Philippines suggest the training is more effective at changing the behavior of younger entrepreneurs who are likely less set in their business practices. This is consistent with the effect size being larger for small businesses as well, as they might have greater flexibility to change and adopt new practices. In India, no evidence of heterogeneity along the four dimensions is found. See table 6. 6. Conclusion This paper presents the results of a two-site randomized experiment assessing the impact of mobile-phone- based financial-heuristics training on micro-entrepreneur’s business practices and firm outcomes. The training intervention was taken up by most entrepreneurs and the majority engaged with the content. It is found that the training led to a significant improvement in business practices among the treated micro-entrepreneurs. The effect size estimate on improved business practices ranges between 0.06 and 0.12 standard deviation points of the practice adoption index. In the wider literature that focuses on much higher-touch interventions, effect sizes of the order of magnitude of 20–40 percent are found. The paper extends the work in the training and firm productivity literature. The focus on a mobile phone-based intervention allows for greater scalability due to lower implementation costs. The modest effect sizes might be more beneficial on a cost-benefit basis since the marginal cost of extending this training to other entrepreneurs is negligible. Conflict of Interest Authors have no financial conflict of interest with regard to this research. Shawn Cole is on the board of Precision Development, which provides mobile phone based agricultural advice to smallholder farmers. Antoinette Schoar is a co-founder and one of the board of directors of the non-profit organization ideas42. Data Availability Statement The underlying data and code necessary to recreate the tables in this paper are available on the Abdul Latif Jameel Poverty Action Lab Dataverse. https://dataverse.harvard.edu/dataset.xhtml?persistentId=doi: 10.7910/DVN/WF284V. References Acimovic, J., C. Parker, D. Drake, and K. Balasubramanian, 2020. “Show or Tell? Improving Inventory Support for Agent-Based Businesses at the Base of the Pyramid.”. Manufacturing & Service Operations Management. 24(1): https://doi.org/10.1287/msom.2020.0922. Aker, J. C., C. Ksoll, and T. J. Lybbert, 2012. “Can Mobile Phones Improve Learning? Evidence from a Field Experiment in Niger.” American Economic Journal: Applied Economics 4(4): 94–120. Anderson, M. L., 2008. “Multiple Inference and Gender Differences in the Effects of Early Intervention: A Reevalua- tion of the Abecedarian, Perry Preschool, and Early Training Projects.” Journal of the American Statistical Associ- ation 103(484): 1481–95. The World Bank Economic Review 597 Arráiz, I., S. P. Bhanot, and C. Calero, 2019. “Less Is More: Experimental Evidence on Heuristic-Based Business Training in Ecuador.” Inter-American Development Bank, Development through the Private Sector Series. TN 18. Banerjee, A., D. Karlan, and J. Zinman, 2015. “Six Randomized Evaluations of Microcredit: Introduction and Further Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Steps.” American Economic Journal: Applied Economics 7(1): 1–21. Bloom, N., B. Eifert, A. Mahajan, D. McKenzie, and J. Roberts, 2012. “Does Management Matter? Evidence from India.” Quarterly Journal of Economics 128(1): 1–51. Bloom, N., and J. Van Reenen, 2010. “Why Do Management Practices Differ across Firms and Countries?” Journal of Economic Perspectives 24(1): 203–24. Bruhn, M., D. Karlan, and A. Schoar, 2018. “The Impact of Consulting Services on Small and Medium Enterprises: Evidence from a Randomized Trial in Mexico.” Journal of Political Economy 126(2): 635–87. Bruhn, M., A. Schoar, and D. Karlan, 2010. “What Capital Is Missing in Developing Countries?” American Economic Review 100(2): 629–33. Cole, S. A., and A. N. Fernando, 2020. “‘Mobile’izing Agricultural Advice Technology Adoption Diffusion and Sus- tainability.” Economic Journal 131(633): 192–219. Government of the Philippines. Congressional Policy and Budget Research Department. House of Representatives. 2020. "Facts in Figures: MSMEs in the Philippines." The Philippines. dalla Pellegrina, L., M. Skully, T. Walker, P. Wanke, and M. Wijesiri, 2021. “Does Ownership Structure Affect Firm Performance? Evidence of Indian Bank Efficiency before and after the Global Financial Crisis.” American Economic Journal: Applied Economics 29(3): 1842–67. de Mel, S., D. J. McKenzie, and C. Woodruff, 2009. “Measuring Microenterprise Profits: Must We Ask How the Sausage Is Made?” Journal of Development Economics 88(1): 19–31. Drexler, A., G. Fischer, and A. Schoar, 2014. “Keeping It Simple: Financial Literacy and Rules of Thumb.” American Economic Journal: Applied Economics 6(2): 1–31. Fabregas, R., M. Kremer, and F. Schilbach, 2019. “Realizing the Potential of Digital Development: The Case of Agri- cultural Advice.” Science 366(6471): eaay3038. Feldman, J., 2003. “The Simplicity Principle in Human Concept Learning.” Current Directions in Psychological Sci- ence 12(6): 227–32. Hsieh, C.-T., and P. J. Klenow, 2009. “Misallocation and Manufacturing TFP in China and India.” Quarterly Journal of Economics 124(4): 1403–48. International Council for Small Business. 2019. “Annual Global Micro, Small and Medium-Sized Enterprises Report.” Washington DC, USA. International Finance Corporation. 2013. “Small and Medium Enterprise Finance: New Findings, Trends and G- 20/Global Partnership for Financial Inclusion Progress.” Washington DC, USA. International Telecommunication Union. 2019. “World Telecommunication/ICT Indicators Database.”(23rd Edition). Geneva, Switzerland. Karlan, D., and M. Valdivia, 2011. “Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions.” Review of Economics and Statistics 93(2): 510–27. Maddox, W. T., C. L. Bradley, B. D. Glass, and J. V. Filoteo, 2008. “When More Is Less: Feedback Effects in Perceptual Category Learning.” Cognition 108(2): 578–89. McKenzie, D., 2020. “Small Business Training to Improve Management Practices in Developing Coun- tries: Reassessing the Evidence for ‘Training Doesn’t Work’.” Policy Research Working Paper Series. 9408. Washington DC, USA. McKenzie, D., and C. Woodruff, 2013. “What Are We Learning from Business Training and Entrepreneurship Evalu- ations around the Developing World?” World Bank Research Observer 29(1): 48–82. Meager, R., 2022. “Aggregating Distributional Treatment Effects: A Bayesian Hierarchical Analysis of the Microcredit Literature.” American Economic Review 112(6): 1818–47. Government of India. Ministry of Micro and Medium Enterprises. 2019. “Annual Report 2019-20.” New Delhi, India. Mullainathan, S., and E. Shafir, 2013. Scarcity: Why Having Too Little Means So Much(1st ed.). New York: Times Books, Henry Holt and Company. ISBN 9780805091239. van Lieshout, S., and P. Mehtha, 2017. “The Next 15 Million Start and Improve Your Business Global Tracer Study 2011-15.” The International Labour Organization. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Supplementary Online Appendix Heuristics on Call: The Impact of Mobile-Phone-Based Business-Management Advice Shawn Cole, Mukta Joshi, and Antoinette Schoar S1A. Module 1: Cash Separation—Message 2: Two Physical Locations to Separate Cash [Philippines] Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Lesson: Find two locations to keep business and household cash separate. Episode structure Script Standard episode intro Hello! And welcome back to the Project Dungannon business training program. This is Tita Jo again. Introduce the customer credit topic A small business like my sari-sari store should not need a complex system just to determine its weekly profit, right? Then what should we do to track our weekly profits? I will help you by teaching you Cash Separation. Today, I’ll teach you the first of the three steps of Cash Separation—how to keep your business and household cash separate. The problem One of the first problems I encountered as a business owner was how to keep track of my weekly profits. I had income from the sari-sari store, my husband’s salary, and a sideline viand selling business, but I also had expenses for both business and family—and they always mixed. I always had a hard time knowing which is which. For example, when I needed to buy Gasul for my viand selling business, I used to get the money from the day’s sales of the sari-sari store. And when my daughter asked for money for her school project, I also got it from the sari-sari store money. As you can see, I was mixing up all my expenses, and didn’t know how much my sari-sari store was making. How do I know then how much I can take from the business to spend for the home? The solution The good news is, I have discovered a very simple way to keep business and household cash and expenses separate. It is called Cash Separation. You just need to have two separate places to keep the money for the business and household. All you need to do is find 2 separate places to keep your business and household money. You may use whatever is convenient for you—a drawer, a box, a garapon. For me, what worked best was a belt bag for the sari-sari store cash and a drawer in the aparador for the household money. Keep your business cash handy in your business, so you can do all business transactions from it. And remember not to mix them up—put all business income only in the business location and pay all business expenses out of the same business cash location. Same with the household cash. Separating your business and household cash will be useful regardless if your family has only one source of income or earns income from multiple sources. Call to action Now it’s time to act: To start separating your cash, in the next two days go and find your two separate locations for your business and household cash. Choose one location for the business cash and one for the household cash. Next week, I will teach you how to use your two cash locations, so you can track how well your business is doing. Closing statement Again, this is Tita Jo saying “Thank You” for listening! If you’d like to hear this message again, please give me a missed call at <0239XXXXX> any time. And remember, keep listening and keep prospering. Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 S1B. Module 2: Customer Credit-Message 1: 7-day Credit Rule [India] Episode structure Script Standard episode intro Hello and welcome back to the Janalakshmi business training program. This is Sangeetha again. Introduce the topic In the next few weeks, I want to share with you a few new tips to help you manage your customer credit better. For business owners like us, a key to business success is treating our customers right. Customer credit is often an important element of our relationship with our customers. But there is such a thing as giving too much credit to your customers—I’ve experienced it in my own business. When you give credit to your customers you are not getting money into your business. So you might not have the money you need to pay your business loan or buy more supplies for your business. That is why I want to share with you some simple tricks for when and how to give credit to your customers in a way that doesn’t hurt your business. Introduce the topic Today I will share with you a tip I use to make sure that I do not have too much credit outstanding. When I started with my kirana store, I struggled to find a way to limit how much credit to give to my customers. For example, I had a relative, Karthik, who asked me to buy milk and yogurt on credit every week. I felt bad to say no, but week after week, he did not pay me back. His credit grew bigger every week. The problem I know we all have had such problems. But remember, customers that do not pay you back on time hurt your business. The money they owe you could make it hard for you to pay your own business expenses. The solution I have a simple solution for you: Only sell goods on credit to customers who can promise to pay you back in the next 7 days. At first, it will be hard to ask your trusted customers to pay back in 7 days. But if you explain politely to them that credit for longer than 7 days hurts your business, they will understand. Tell them that if you get paid after 7 days, it makes it harder for you to pay your business expenses. This will get your business in debt, and your business and family will suffer. Call to action Time to act: Starting tomorrow, ask each of your customers who wants to buy on credit when they would pay you back. Give credit only to those customers who confirm that they will pay you back in the next 7 days. Explain to your customers that you cannot give them credit for longer than 7 days, as this would hurt your business. In the next weeks, I will teach you two more tricks to help you limit how much credit you give to customers. Closing statement This is Sangeetha! If you’d like to hear this message again, please give me a missed call at 0804XXXXXXX. And remember, keep listening and keep prospering. Table S2A. Attrited Sample Summary Statistics and Balance—Philippines (1) (2) (3) (2) − (3) Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Total sample Attrited control Attrited treatment Pairwise t-test Variable N Mean/(SD) N Mean/(SD) N Mean/(SD) N p-value A. Client characteristics Age 198 42.040 101 42.089 97 41.990 198 0.950 (11.177) (11.123) (11.291) Female 198 1.000 101 1.000 97 1.000 .n .n (0.000) (0.000) (0.000) Education High school or above 198 0.773 101 0.782 97 0.763 198 0.748 (0.420) (0.415) (0.428) Business type Business is primary source of income 198 0.545 101 0.554 97 0.536 198 0.796 (0.499) (0.500) (0.501) Retail: Food 198 0.576 101 0.614 97 0.536 198 0.271 (0.495) (0.489) (0.501) B. Business practices Do separate business & household cash 198 0.687 101 0.644 97 0.732 198 0.182 (0.465) (0.481) (0.445) Do pay salary to self 198 0.157 101 0.139 97 0.175 198 0.481 (0.364) (0.347) (0.382) Do calculate profits 198 0.773 101 0.743 97 0.804 198 0.304 (0.420) (0.439) (0.399) Give customers credit for at most 7 days 165 0.812 82 0.817 83 0.807 165 0.872 (0.392) (0.389) (0.397) Do nothing when customers do not pay credit 198 0.066 101 0.079 97 0.052 198 0.435 (0.248) (0.271) (0.222) Keep business records 198 0.717 101 0.733 97 0.701 198 0.623 (0.452) (0.445) (0.460) Keep customer credit records 165 0.661 82 0.695 83 0.627 165 0.355 (0.475) (0.463) (0.487) Determine stock based on a good strategy 198 0.207 101 0.168 97 0.247 198 0.171 (0.406) (0.376) (0.434) Never visit competitors to check prices/quality 198 0.788 101 0.792 97 0.784 198 0.883 (0.410) (0.408) (0.414) Never talk to customers to understand needs 198 0.545 101 0.545 97 0.546 198 0.979 (0.499) (0.500) (0.500) Never do supplier quality comparison 198 0.414 101 0.446 97 0.381 198 0.363 (0.494) (0.500) (0.488) Never negotiated terms with suppliers 198 0.470 101 0.485 97 0.454 198 0.659 (0.500) (0.502) (0.500) Took full advantage of cash discount 50 0.740 25 0.720 25 0.760 50 0.753 (0.443) (0.458) (0.436) C. Business performance Sales—Regular week (winsorized at 1%) 179 3,944.050 89 4,063.876 90 3,825.556 179 0.713 (4,317.047) (4,958.475) (3,596.528) Profits—Regular week (winsorized at 1%) 185 1,793.405 95 1,634.947 90 1,960.667 185 0.314 (2,194.321) (2,506.625) (1,806.503) Source: Primary data collected by research team. Note: This table presents summary statistics based on baseline survey data. Standard deviations (columns 2, 3, 4) of variables and p-values (column 5) appear in parentheses. ∗ Denotes significance at the 10 percent level, ∗∗ at the 5 percent level, and ∗∗∗ at the 1 percent level. Table S2B. Attrited Summary Statistics and Balance—India (1) (2) (3) (2) − (3) Downloaded from https://academic.oup.com/wber/advance-article/doi/10.1093/wber/lhad038/7612583 by LEGVP Law Library user on 01 August 2024 Total sample Attrited control Attrited treatment Pairwise t-test Variable N Mean/(SD) N Mean/(SD) N Mean/(SD) N p-value A. Client characteristics Age 531 37.245 264 37.292 267 37.199 531 0.896 (8.176) (8.643) (7.703) Female 531 0.864 264 0.864 267 0.865 531 0.959 (0.343) (0.344) (0.342) Education High school or above 531 0.347 264 0.367 267 0.326 531 0.315 (0.476) (0.483) (0.470) Business type Business is primary source of income 531 0.337 264 0.322 267 0.352 531 0.464 (0.473) (0.468) (0.479) Retail: Food 531 0.260 264 0.284 267 0.236 531 0.207 (0.439) (0.452) (0.425) B. Business practices Do separate business & household cash 531 0.243 264 0.254 267 0.232 531 0.563 (0.429) (0.436) (0.423) Do pay salary to self 531 0.038 264 0.030 267 0.045 531 0.377 (0.191) (0.172) (0.208) Do calculate profits 531 0.582 264 0.580 267 0.584 531 0.912 (0.494) (0.495) (0.494) Give customers credit for at most 7 days 193 0.285 98 0.255 95 0.316 193 0.353 (0.453) (0.438) (0.467) Do nothing when customers do not pay credit 193 0.047 98 0.031 95 0.063 193 0.286 (0.211) (0.173) (0.245) Keep business records 365 0.411 180 0.444 185 0.378 365 0.201 (0.493) (0.498) (0.486) Keep customer credit records 193 0.435 98 0.469 95 0.400 193 0.334 (0.497) (0.502) (0.492) Determine stock based on a good strategy 531 0.109 264 0.114 267 0.105 531 0.747 (0.312) (0.318) (0.307) Never visit competitors to check prices/quality 531 0.386 264 0.367 267 0.404 531 0.381 (0.487) (0.483) (0.492) Never talk to customers to understand needs 531 0.446 264 0.447 267 0.446 531 0.976 (0.498) (0.498) (0.498) Never do supplier quality comparison 531 0.232 264 0.250 267 0.213 531 0.320 (0.422) (0.434) (0.411) Never negotiated terms with suppliers 531 0.183 264 0.189 267 0.176 531 0.691 (0.387) (0.393) (0.382) Took full advantage of cash discount 115 0.461 56 0.482 59 0.441 115 0.659 (0.501) (0.504) (0.501) C. Business performance Sales—Regular week (winsorized at 1%) 364 13,651.099 179 13,262.570 185 14,027.027 364 0.618 (14,613.778) (14,519.390) (14,734.123) Profits—Regular week (winsorized at 1%) 363 5,267.218 179 5,074.022 184 5,455.163 363 0.465 (4,962.215) (4,561.282) (5,329.171) Source: Primary data collected by research team Note: This table presents summary statistics based on baseline survey data. Standard deviations (columns 2, 3, 4) of variables and p-values (column 5) appear in parentheses. ∗ Denotes significance at the 10 percent level, ∗∗ at the 5 percent level, and ∗∗∗ at the 1 percent level. C The Author(s) 2024. Published by Oxford University Press on behalf of the International Bank for Reconstruction and Development / THE WORLD BANK. All rights reserved. For permissions, please e-mail: journals.permissions@oup.com