Abstract

We investigate whether improvements in street-food safety can be achieved by providing information to vendors in the form of a training. Among randomly assigned vendors in Kolkata, India, we find large improvements in knowledge and awareness but little change in observed behavior. We provide two main explanations for these findings. First, information acquisition by itself does not make it significantly easier for vendors to provide customers with safer food options. Second, although consumers have a positive willingness to pay for perceived hygiene, they struggle to distinguish between safe and contaminated food. We recommend policies targeting supply-side constraints and consumers' awareness.

I. Introduction

THE Food and Agricultural Organization (FAO) and the World Health Organization (WHO) regularly report that the consumption of unsafe street food affects the health status of millions of people all over the world (WHO, 1996, 2006; FAO/WHO, 2003). Since street food accounts for a significant proportion of the daily food consumption of 2.5 billion consumers (FAO, 2007), establishing a more reliable and safe food supply chain remains one of the major public health concerns of many governments (Bhowmik, 2012). Among the several factors that contribute to this status quo, street vending is an occupation that requires little start-up capital and therefore draws a large section of the urban poor who lack awareness and adequate training in safe food-handling methods (NCEUS, 2007; World Bank, 2013). As a result, one of the key messages for governments and relevant stakeholders by the FAO and WHO is that food handlers' knowledge and awareness about food safety and hygiene is a necessary pillar for setting up national strategies to control food-related hazards, particularly in urban areas (WHO, 2006; FAO, 2009a, 2009b). Food safety training and certificates are common in developed countries but less so in many developing countries. Moreover, among the range of interventions, they are possibly the cheapest, but evidence on whether and why they might work is still limited.

In this paper, we aim to fill this gap by designing a training aimed at improving vendors' awareness and reducing street-food safety hazards. The simple hypothesis underlying such an intervention is that information acquisition could enable vendors to offer consumers a wider array of desirable products and services by removing the knowledge barrier. We verify that this is appropriate in our context by demonstrating both that vendors have low awareness and high perceived costs of providing hygienic services at baseline and that consumers have a positive willingness to pay for food that is perceived as more hygienic. We implement the program by randomly allocating eligibility to the training to 681 street-food vendors in Kolkata, India. We show that the program increases vendors' awareness, but it is not sufficient to improve observed food safety behaviors. We provide evidence that this is likely due to the training not sufficiently reducing the difficulty of providing these services, as well as consumers struggling to detect whether street food is contaminated. We suggest that information to vendors is not the key constraint in this context and that policymakers should focus on mitigating the supply-side constraints (Duflo, Galiani, & Mobarak, 2012), as well as improving food safety awareness among consumers.

Our project was undertaken from March 2015 to July 2016, with the logistic support of InnoAid, a Denmark-based nongovernmental organization (NGO).1 We collected a sample of street-food vendors, spread out in different areas of the city, and grouped them into 74 blocks of roughly ten vendors each.2 We randomized vendors across three conditions: control, training, and training with promotional material, allowing them to advertise to customers their participation in the program. The training consisted of three workshops and three follow-ups, organized over ten consecutive weeks, and had two main objectives: first, to enhance vendors' awareness of health risks related to food vending, and second, to develop their capacity to make the necessary improvements in food safety practices.

At the beginning and the end of the study, we collected data on vendors' socioeconomic characteristics, business practices, and awareness about food safety and hygiene. Additionally, we monitored vendors' actual behavior between the baseline and the endline surveys for six data points on each vendor (once every eight weeks). The direct monitoring of vendors' daily behavior at their kiosks allows us to have a more comprehensive and accurate evaluation of their practices without relying exclusively on self-reported information. Moreover, at each point of the data collection and training, the take-up rate was always above 70%. This also makes our study novel in terms of the degree of accuracy of the data collected on vendors.

Our design allows us to identify the effect of the program in a simple first-difference framework, comparing outcomes for vendors in treatment and control groups.3 The parameter we identify is the effect of increased awareness on vendors' behavior such that it is not confounded by the fear of punishment. This is because the eligibility for the training was randomly allocated, the participation was voluntary, and there was no mandated penalty for noncompliance to food-safety practices. This would not be the case, for instance, if one were evaluating a government-designed food hygiene monitoring program. In this case, the vendors' observed behavior could be driven by both improved awareness and the threat of sanctions in case of noncompliance, with the two being difficult to disentangle.4 In our setup, any effect of the training provides evidence toward a potential scope of self-regulation by the vendors.5 This is informative to a policymaker, because it would make the case for the efficacy of such an intervention due to its low cost of implementation. Indeed, fines and sanctions may provide stronger incentives for vendors to comply with higher food-safety practices, but it would likely be a much more expensive intervention, especially in a weak and complex institutional environment.6

The first set of results is twofold. We find a substantial effect of the training on vendors' awareness, knowledge of food safety hazards, and self-reported safety behaviors. Overall, treated vendors' awareness improves by 7.1 percentage points relative to the control group. We also see a significant improvement in terms of self-reported actions that vendors claim to have engaged in, by about 10.8 percentage points.7 However, we fail to discern a significant impact of the training on actual food safety behavior as measured by the external monitors.

Given the complexity of the sector, there are several reasons that might explain why behavioral improvements in food safety are difficult to undertake. First, vendors may lack the necessary incentives to produce safer foods because even if they changed their behavior, they know the improved quality would be difficult to observe in practice, and hence they would not be rewarded by the consumers. This would be the case if, for example, food safety improvements were not visible to customers, as in the case of some food storage and preservation processes. Since one treatment arm is given free tools to advertise participation in the program, we are able to test directly whether food safety improvements are prevented because vendors fear that they might go unnoticed by consumers. We do not find a significantly different impact for this group of vendors compared to the basic treatment.8

Second, the information acquired from the training might not actually have made it easier for the vendors to provide consumers with a wider array of hygienic products and services. This could be the case because, besides lacking awareness about food safety, vendors continue to be constrained by the lack of several other costly infrastructural facilities—for example, clean water, waste disposal, or electricity. We validate this hypothesis by collecting direct information on individual perceived cost or difficulty of engaging in safer practices (before and after intervention). Vendors have a high perceived cost at baseline, and though the training modestly decreases it, the reduction is not significant. We elicit further information from vendors about the primary problems they face, which corroborates the idea that infrastructural, institutional, or monetary problems are highly prevalent.9

Third, consumers in this sector may not have a sufficiently high demand for hygienic food. We investigate this hypothesis in a follow-up study, conducted between June and August 2019, on 1,480 customers of street food. This study allows us to speak to the factors that determine their consumption choices and to directly elicit their willingness to pay for alternative food options. We find that consumers' stated preferences indicate that they do in fact value safer food and have a positive willingness to pay for it. However, almost 70% of the sample reports that they find it impossible or extremely difficult to detect whether food is contaminated. Moreover, consumers also take into consideration several other attributes, such as taste, price, and location of street food, while making their purchase decisions. For instance, as many as 63% of consumers regard something other than hygiene as the key determinant of their decision. Additionally, compared to their preferences for safer food, they are less willing to pay when vendors simply display a signal of external endorsement (unaccompanied by an actual improvement in their practices). Not surprisingly, our data on vendors show that they do not exhibit increases in their revenue and profits, given the lack of actual enhancements in their food safety standards. The identical picture that emerges from our vendor and consumer samples lends further credence to our results.

All in all, our findings imply that training is not sufficient to foster substantial changes in food safety, suggesting that information alone to vendors is not the key constraint in this sector. What is perhaps required is a more active role played by the local authorities, both in terms of providing easier access to infrastructure as well as imposing regulations to mandate safer practices. While our data validate that there is a demand for safer street food, increased awareness among consumers would enable them to better distinguish between desirable food options and contaminated or unhygienic alternatives and generate further incentives for vendors to improve. Indeed, what emerges from our analysis is that the market currently suffers from welfare losses stemming from vendors not being able to capitalize on higher profits and prices that consumers would be willing to pay in exchange for a perceived improvement in food hygiene.

Our study also has a strong temporal relevance in this context, because the Indian government is currently undertaking a process of formalization of street vendors. Indeed, the Parliament of India enacted legislation to regulate street vendors in public areas, passing a National Act for Urban Street Vendors (“Protection of Livelihood and Regulation of Street Vending”) in March 2014. The aim of the act is to regulate the unorganized street vending sector, providing specific zones to the street vendors, a license to run their business, and recommendations concerning health and hygienic standards (NPUSV, 2009). The implementation of this policy has so far been limited to a few urban areas, and, as of 2019, Kolkata was not among them yet. Therefore, much of the business still remains informal throughout India.10 Our results can inform policymakers in India as well as similar countries in Southeast Asia, who are attempting to undertake a similar formalization of a hitherto unregulated sector, a key component of which is bolstering the quality of the good produced.

The remainder of this paper is organized as follows. In section II, we review the relevant literature and highlight our contributions. In section III, we describe in detail the context of our study and the design of the experiment. In section IV, we present our data. We describe our empirical strategy and report our main results in section V. We then detail the findings from our choice experiments and consumer survey in section VI, and conclude in section VII.

II. Literature Review and Contribution

The paper contributes to a flourishing literature investigating the role of trainings offered to microentrepreneurs in developing countries. While they mostly focus on improving business practices to foster profitability and employment,11 our main goal is to reduce food safety hazards through informational training. In this regard, we relate to information experiments aimed at improving health outcomes. Previous studies randomized information on the importance of wearing eyeglasses (Ma et al., 2013), how to avoid intestinal worm infections (Meredith et al., 2013) or prevent HIV (Dupas, 2011; Duflo, Dupas, & Kremer, 2015), and how to access family planning sessions (Ashraf, Field, & Lee, 2014). In a recent review of this literature, Dupas and Miguel (2017) state that “information is necessary but often not sufficient to generate take-up” (41). Similar conclusions can be drawn from our experiment. The only previous study that we are aware of that deals with retailers is Banerjee et al. (2015), who provide information and financial incentives (higher markups) to retailers selling a specific type of salt aimed at reducing anemia. Unlike that paper, we do not rely on direct financial incentives because they are difficult to implement in our context. Additionally, while much of this literature follows only the entrepreneurs comprising the supply side of the market, we combine it with direct evidence from the consumers they cater to, thereby making our study more comprehensive.

Our study also contributes to broadening our understanding of whether training aimed at improving food safety among street vendors might change their behaviors. Although they are often recommended by the FAO and the WHO (FAO/WHO, 2003; WHO, 2006; FAO, 2009b, 2013), we lack conclusive evidence on their effectiveness. For example, Soon, Baines, and Seaman (2012) run a meta-analysis on several food-safety training evaluations and find a positive effect of the training on both food safety knowledge and self-reported practices. Along the same line, Medeiros et al. (2011) reach a similar conclusion in a review of different food safety training methods. However, both reviews have several limitations. First, they include food handlers from very different contexts, such as restaurants or canteen employees, who may face different constraints from those of street vendors. Second, food safety practices are evaluated through self-reporting only.12 A third limitation is the small sample size of such studies, which, in the case of Soon et al. (2012), is on average 47. Finally, more recent studies report null findings, in which self-reported behaviors are not affected by the training (da Cunha, Stedefeldt, & de Rosso, 2014; da Cunha et al., 2015).

With respect to the above cited literature, our work differs in several dimensions: (i) we base our inferences on a larger sample of vendors and statistical power; (ii) we do not only rely on self-reported behaviors, as we also provide a novel approach to assess and measure food safety standards, based on a multitude of information collected on multiple observations, which could be replicated in similar contexts; (iii) we collect precise information about each vendor's network on the street to account for possible spillover effects and spread of information contained in the workshops; (iv) our intervention allows measuring how lasting the informational shock provided by the training is; and (v) our consumer survey allows us to comment directly on the demand aspect of this market.

III. Background, Motivation, and Intervention

A. Street-Food, Health Hazards, and Professional Training

Street vending is the activity of selling goods and services in the streets without having a permanent built-up structure (Wongtada, 2014). This is an important and growing source of employment in developing countries, especially for the urban poor, as it does not require specialized skills or a large start-up capital (FAO, 2013). Street vending is mostly conducted in the informal sector, as local authorities are often unable (or unwilling) to regulate this sector (Bhowmik, 2012). On their side, vendors are often criticized for providing low-quality goods, creating congestion and safety risks (Bromley, 2000). This precarious position and highly uncertain environment make them ideal victims for harassment such as evictions, confiscation of merchandise, petty offenses, and extortion (Bhowmik, 2012). Among street vendors, food sellers represent the most visible group as they provide affordable and nutritional food to millions of urban consumers every day, especially to low- and middle-income consumers with limited time and means for shopping and cooking (FAO, 2007).

However, street vendors are also considered a threat to public health, as street food is a main determinant of food-borne diseases. According to WHO, the overall burden of such diseases is substantial: “An estimated 600 million—almost 1 in 10 people in the world—fall ill after eating contaminated food and 420,000 die every year, resulting in the loss of 33 million healthy life years.”13 Evidence on the links between streetfood and food-borne diseases is provided by microbiological studies, which also highlight that food contamination could be reduced by avoiding common food safety hazards (Desenclos et al., 1991; Flint et al., 2005; Bhaskar et al., 2004; Chumber, Kaushik, & Savy, 2007; Ghosh et al., 2007; Greig et al., 2007; Chan & Chan, 2008; Sousa, 2008; Choudhury et al., 2011; Rane, 2011; Feglo & Sakyi, 2012; among many others).14

Moreover, related studies show that street-food vendors often lack food safety knowledge, supporting the idea that lack of food safety awareness is an important determinant of food safety hazards and food-borne illnesses.15 The FAO and the WHO, in collaboration with microbiologists as well as food experts based at national food safety authorities, have acknowledged the public health literature on this topic and recommended professional training targeted at food vendors (WHO, 2008 and FAO, 2009a). For example, in 2003, a joint FAO/WHO commission published the Codex Alimentarius (FAO/WHO, 2003), which advocates training on food safety, which is “fundamentally important to any food hygiene system” (20). More recent reports further recognize the strategic importance of the street-food sector in providing accessible and nutritious food, and recommend food-safety training to curb the health threat due to poor food safety knowledge (WHO, 2006; FAO, 2009b, 2013).16

B. Motivation for Our Intervention

Professional trainings are recognized as a necessary pillar to address the problem of food-related hazards. However, our knowledge on whether and why they might work is still limited. We posit that such trainings might be an effective tool to bring about food safety improvements if information acquisition can result in higher profits for vendors. This is because the information imparted at the trainings leads to a broadening of their choice set in terms of the products and services that they can offer to their customers. For example, product features that the vendor did not previously know were valuable (e.g., clean utensils, uncontaminated water) are now introduced into their set of possibilities.17 Conditional on consumers valuing these features, adopting such practices could increase vendors' profits by improving their sales and revenues or enabling them to sell their fare at a higher price.18

Hence, the motivation behind our intervention is that vendors could be currently facing a loss of business due to their ignorance of food safety hazards. This hypothesis finds support in some of the reports by the WHO. For instance, WHO (2006) states that in street-food “markets, loss of business will result from unclean and poorly managed facilities and unhygienic food handling practices. All these losses are more regrettable because they are preventable with basic investments in training and infrastructure” (6). The vendors in our sample themselves seemed to be aligned with the idea that our program could boost their business. Our baseline data showed that the majority of them (53%) expected that professional training related to hygiene practices would increase their profits.

We further establish why such a training intervention was appropriate in our context by bringing empirical evidence to bear on both the supply and demand sides of our market. First, in the absence of such a program, acquiring knowledge about food safety is costly and unlikely for several reasons: (i) street vendors have low levels of education (e.g., 29% of vendors in our sample have no formal education at all, and only 11% completed secondary school), which limits their capability for independent knowledge acquisition on food safety; (ii) they operate in an informal sector, and hence they rarely deal with official institutional bodies that could in principle provide guidance and support on food safety matters; and (iii) there are no self-help groups in this market that might help in spreading such information (e.g., only 1.8% of vendors in our sample were part of any such group or association). Moreover, at the baseline, we directly collected data on vendors' perceived cost or difficulty of engaging in desirable practices. A large share of vendors reported a high perceived cost of providing clean drinking water (60%), maintaining clean surroundings at the stall (48%), keeping hands clean (41%), having a clean cooking area (42%), and providing disposable utensils and cutlery (68%). Such costs could stem from both a lack of relevant knowledge or awareness, as well as inadequate infrastructural and other external support. Our informational intervention was designed to alleviate the former, providing them with the capability to make necessary improvements by removing the information barrier.

Second, we supplement the above with direct evidence from the demand side that consumers do in fact value such products and services. Based on a discrete choice experiment that we conducted on customers of street food as part of a follow-up survey on consumers, we show that the stated preferences of consumers in this market indicate a positive marginal willingness to pay for food that is perceived as more hygienic.19

C. Design of the Experiment

Our study is based in Kolkata, the third largest urban agglomeration in India, with a population of about 14 million, and the main business, commercial, and financial hub of eastern India. The most recent estimates suggest that there are 300,000 street vendors in greater Kolkata, and approximately half of them are street-food vendors (NCEUS, 2009), selling rich, diverse, and affordable food to millions of customers every day. Like other urban areas, Kolkata is not exempt from public health issues generated by street-food vending.20

Our experiment consisted of three initial phases: identification, mobilization, and randomization of vendors. These were followed by the survey collection, workshops, monitoring, and evaluation. During the identification phase, we selected suitable areas for the training based on the presence of a significant cluster of street-food vendors. This process led to the selection of ten areas distributed across the entire city (see figure A1 in appendix A1). These neighborhoods differ in terms of socioeconomic background, including different types of consumers. For instance, some vendors are located in business districts, while others are close to schools or public transportation.

The mobilization phase proceeded in the following way. Since street vending is an informal activity and vendors are generally organized in informal groups, we had to first engage local vendor leaders, informal representatives of vendors on the street and often affiliated to vendor unions. Our team made arrangements with several leaders throughout the city who represented more than 2,000 vendors. Based on their location, we clustered them in 200 blocks of roughly 10 vendors each.21 From these blocks, we randomly selected 100 blocks that constituted our initial sample: 26 of them were used as pilot to evaluate the workshops, and the remaining 74 constituted our preferred sample for the analysis.22

This phase was also aimed at guaranteeing a satisfactory take-up rate. To ensure that, our team visited the selected areas to explain to the eligible vendors the benefits of attending a series of workshops focused on food safety. Table A1 in appendix A1 provides summary statistics of the most important elements of our experimental design. Overall, we included 681 vendors in the study at the baseline; of these, 230 vendors (25 blocks) were assigned to the control group (C), 210 vendors (23 blocks) were assigned to a first treatment group (T1), eligible for the training and the follow-up sessions, and 241 vendors (26 blocks) were assigned to a second treatment group (T2), eligible to receive promotional material endorsing their participation in the program, on top of the training and follow-up. As we discuss below, the purpose of including the second treatment group was to examine whether vendors could be incentivized to alter their practices only if they were assured of being able to communicate such improvements to their customers.

Within each area, we randomly allocated vendors to treatment and control groups so as to ensure that all areas were represented among all three types of vendors. The treatment was assigned at the block level; that is, all vendors in the same block were eligible to attend the workshops together. As one can see from table A1, all three workshops were attended by at least 70% of the eligible vendors.23 Moreover, we were also successful at keeping the take-up rate of the monitorings and the endline survey high. One possible concern from the point of view of the identification is that vendors working in blocks next to each other, but assigned to different treatments, could lead to confounding spillover effects in their behavior, which would be difficult to control. We address this issue by collecting precise information on each vendor's network on the street, as we explain in detail in section IV.

In order to be able to discern some of the incentives underlying the behavior of the vendors, we differentiated the two treatment arms by providing additional promotional materials to a subsample of vendors.24 Specifically, they received posters, napkins, and certificates endorsing their participation in the program, and thereby signaling to customers their efforts to cope with food safety issues. Panel A of figure A2 in appendix A1 shows a translated version (from Bengali) of the posters displayed by vendors at their kiosk, and panel B shows a sample of the napkins provided by our partners on the right. We designed the treatment together with InnoAid marketing experts and our local partners. Our objective was to maximize the probability that the intervention for this group would not go unnoticed by the customers. Our first goal was to increase the saliency around quality, motivated by evidence that although consumers declare to be concerned about food safety (WHO, 2008; FAO, 2013), they might not necessarily recognize food safety hazards and, in turn, food safety improvements. Therefore, the purpose of the promotional material was to explicitly signal vendors' engagement with this problem. Second, we aimed at solving a signaling problem: customers might not trust vendors when they simply claim to provide cleaner food, and vendors do not provide it. Conversely, customers might trust more third-party certification (InnoAid promotional material), making vendors more willing to improve hygienic standards.

D. Training and Workshops

At the beginning of the program, we administered a baseline survey to all vendors in order to obtain socioeconomic and demographic information, as well as to have a clear picture of their business practices and food safety awareness before the intervention. Consistent with the literature outlined in section IIIB, in our baseline survey, we also find that only a minority of vendors were able to mention possible causes of food contamination. Table A2 in appendix A1 reports summary statistics from a set of variables preliminarily investigating vendors' food-safety knowledge.25 First, we asked whether they knew about the ongoing debate on the National Act to formalize street vending. Only 11% of them reported awareness of it. Second, we asked vendors to list any possible sources of food contamination: only a minority were able to do it. They most mentioned bacteria (22%), while only 10% of vendors mentioned water contamination. This preliminary evidence also suggests that the level of food-safety awareness was similarly limited across treatment groups, because there are no significant differences across groups.26

The vendors assigned to treatment were then eligible to attend three workshops of about 1.5 hours each and a series of three follow-ups at the vendor's kiosk. The workshops took place between October 2015 and May 2016 in areas close to the vendors' workplace.27 This was decided in order to facilitate the identification of attending vendors and ensure a homogeneous environment across vendors' workshops.

Each meeting took place approximately every two or three weeks and focused on specific topics. Workshop 1 focused on providing information on the Street Vendors Act 2014, aiming at increasing vendors' awareness of their rights and the formalization process. This workshop also focused on the importance of cooperation among vendors to achieve common goals. Workshop 2 included a wide range of information about food-safety hazards related to the vendors' daily behaviors. Workshop 3 focused on how to change daily behavior to reduce food-safety hazards. For instance, it stressed the importance of clean hands, vendors' personal hygiene, use of containers for water, covering food, availability of paper plates and cups, and waste management. The follow-ups consisted of one-on-one meetings at the vendor's kiosk, where a trainer could observe the vendor and suggest personalized advice to improve his food-safety practices. More information on each workshop is provided in the online appendix.

The workshop phase was accompanied by four rounds of monitoring, where data collectors observed and recorded the food safety practices adopted by the vendors (this means that, together with the baseline and endline surveys, we collected six data points on behavior). This information was collected by observing the vendors at their stalls, instead of asking them questions, so as to minimize reporting errors. Several precautions were taken to ensure that the monitoring captured the vendors' actual food-handling practices and that the vendors were not altering their behavior for the survey. (i) The data collectors for the monitoring rounds were not involved with the actual training, so that they were unknown to the vendors. (ii) The vendors had no prior knowledge of when their stall would be monitored (they were obviously aware of being monitored at some point). (iii) The monitoring was undertaken by two teams of two data collectors visiting each block together; they were approaching from either side of the street, so that it would be difficult for vendors to be alerted and behave differently on seeing the interviewers on either side of the street. (iv) For additional accuracy, the data collectors took pictures of the stalls, which were then cross-checked with the data.

At the end of the program, we collected an endline survey that repeated the monitoring of the vendors' food-handling practices and collected general information on their business practices. The endline also included detailed questions targeted at understanding whether the vendors had learned about safer food handling methods or engaged in any of the practices advocated by the training workshops.

IV. Data

A. Independent Variables and Balancing

One of the novelties of our study is the degree of accuracy of the data collected about street vendors' information and practices. Table 1 provides socioeconomic and demographic information of the vendors in each of the treatment arm. The last column of the table reports the p-value for the means of the two treatment groups and the control group being equal. The treatment and control blocks are statistically similar in terms of most baseline characteristics. Around 86% of vendors in our sample are male, which is not surprising given the heavily skewed gender ratio in this sector (panel A). Vendors' average age is about 41, and they have a remarkably long experience in this sector (19 years). Moreover, 72% of them have some education and approximately 70% of the vendors are associated with a union.

Table 1.

Preintervention: Summary Statistics and Balancing

TrainingTraining w/SignalControlFull samplep-value difference
(T1)(T2)(C)T1 = CT2 = CT1 = T2
 Panel A. Socioeconomic and demographic information 
Male 0.85 0.87 0.87 0.86 0.72 0.96 0.77 
Age 40.32 41.41 41.01 40.94 0.60 0.75 0.45 
Years of vending 19.21 19.20 19.97 19.47 0.60 0.64 0.99 
Some education 0.73 0.72 0.71 0.72 0.37 0.63 0.68 
Member of a union 0.70 0.66 0.74 0.70 0.63 0.38 0.70 
 Panel B. Business information 
Cooked food 0.71 0.68 0.77 0.72 0.23 0.14 0.71 
Light snacks 0.59 0.51 0.57 0.55 0.74 0.45 0.33 
Heavy snacks 0.13 0.12 0.15 0.14 0.59 0.51 0.92 
Meals 0.19 0.23 0.19 0.20 0.94 0.49 0.49 
Drinks 0.11 0.07 0.12 0.10 0.72 0.17 0.25 
Fruits and vegetables 0.06 0.09 0.05 0.07 0.62 0.37 0.58 
Sweets 0.05 0.08 0.06 0.06 0.75 0.57 0.39 
Other 0.01 0.01 0.04 0.02 0.35 0.24 0.54 
Revenue 1,294 1,463 1,253 1,337 0.81 0.45 0.53 
Expenditure 954 1,125 928 1,003 0.84 0.42 0.47 
Profits 339 338 326 334 0.76 0.77 0.97 
Hours worked 11 11 11 11 0.98 0.73 0.77 
 Panel C. Daily problems faced 
Bribes to police 0.37 0.30 0.37 0.35 0.79 0.41 0.27 
Daily problems faced are:        
Lack of electricity connection 0.44 0.49 0.51 0.48 0.35 0.67 0.54 
Shortage of capital 0.34 0.32 0.34 0.33 0.94 0.76 0.79 
Competition from vendors 0.20 0.19 0.12 0.17 0.18 0.26 0.90 
Competition from other 0.06 0.03 0.03 0.04 0.24 0.89 0.37 
Lack of toilets 0.48 0.61 0.55 0.55 0.48 0.43 0.16 
Lack of drinkable water 0.45 0.58 0.52 0.52 0.44 0.45 0.14 
Extortion and bribes from police 0.18 0.11 0.16 0.15 0.79 0.41 0.27 
 Panel D. Network (number of vendors known) 
Own block 7.56 7.77 7.76 7.70 0.68 0.99 0.66 
Neighboring blocks 4.65 4.64 4.84 4.71 0.79 0.77 0.99 
 Panel E. Observed behavior 
Facilities index 0.48 0.44 0.46 0.46 0.77 0.19 0.12 
Food handling index 0.64 0.60 0.61 0.62 0.44 0.67 0.21 
Costumers index 0.69 0.68 0.69 0.69 0.92 0.89 0.76 
Total index 0.60 0.56 0.57 0.58 0.55 0.33 0.11 
Observations 210 241 230 681    
TrainingTraining w/SignalControlFull samplep-value difference
(T1)(T2)(C)T1 = CT2 = CT1 = T2
 Panel A. Socioeconomic and demographic information 
Male 0.85 0.87 0.87 0.86 0.72 0.96 0.77 
Age 40.32 41.41 41.01 40.94 0.60 0.75 0.45 
Years of vending 19.21 19.20 19.97 19.47 0.60 0.64 0.99 
Some education 0.73 0.72 0.71 0.72 0.37 0.63 0.68 
Member of a union 0.70 0.66 0.74 0.70 0.63 0.38 0.70 
 Panel B. Business information 
Cooked food 0.71 0.68 0.77 0.72 0.23 0.14 0.71 
Light snacks 0.59 0.51 0.57 0.55 0.74 0.45 0.33 
Heavy snacks 0.13 0.12 0.15 0.14 0.59 0.51 0.92 
Meals 0.19 0.23 0.19 0.20 0.94 0.49 0.49 
Drinks 0.11 0.07 0.12 0.10 0.72 0.17 0.25 
Fruits and vegetables 0.06 0.09 0.05 0.07 0.62 0.37 0.58 
Sweets 0.05 0.08 0.06 0.06 0.75 0.57 0.39 
Other 0.01 0.01 0.04 0.02 0.35 0.24 0.54 
Revenue 1,294 1,463 1,253 1,337 0.81 0.45 0.53 
Expenditure 954 1,125 928 1,003 0.84 0.42 0.47 
Profits 339 338 326 334 0.76 0.77 0.97 
Hours worked 11 11 11 11 0.98 0.73 0.77 
 Panel C. Daily problems faced 
Bribes to police 0.37 0.30 0.37 0.35 0.79 0.41 0.27 
Daily problems faced are:        
Lack of electricity connection 0.44 0.49 0.51 0.48 0.35 0.67 0.54 
Shortage of capital 0.34 0.32 0.34 0.33 0.94 0.76 0.79 
Competition from vendors 0.20 0.19 0.12 0.17 0.18 0.26 0.90 
Competition from other 0.06 0.03 0.03 0.04 0.24 0.89 0.37 
Lack of toilets 0.48 0.61 0.55 0.55 0.48 0.43 0.16 
Lack of drinkable water 0.45 0.58 0.52 0.52 0.44 0.45 0.14 
Extortion and bribes from police 0.18 0.11 0.16 0.15 0.79 0.41 0.27 
 Panel D. Network (number of vendors known) 
Own block 7.56 7.77 7.76 7.70 0.68 0.99 0.66 
Neighboring blocks 4.65 4.64 4.84 4.71 0.79 0.77 0.99 
 Panel E. Observed behavior 
Facilities index 0.48 0.44 0.46 0.46 0.77 0.19 0.12 
Food handling index 0.64 0.60 0.61 0.62 0.44 0.67 0.21 
Costumers index 0.69 0.68 0.69 0.69 0.92 0.89 0.76 
Total index 0.60 0.56 0.57 0.58 0.55 0.33 0.11 
Observations 210 241 230 681    

The table provides socioeconomic and demographic information (panel A), business information (panel B), daily problems faced by vendors (panel C), network information (panel D), and observed behavioral outcomes (panel E) in each treatment arm. Each column for T1, T2, C, and Total reports the mean value. P-values are based on standard errors clustered by block.

In panel B, we can see that roughly 72% of vendors in our sample sell food that is cooked at the stall. The most represented category are vendors selling light snacks28 (55%), followed by vendors selling meals (20%), heavy snacks (14%), drinks (10%), fruits and vegetables (7%), and sweets (6%). Vendors profit on average 334 rupees a day, well above the poverty threshold line (35 rupees in 2015). Panel C shows that vendors are concerned with a wide set of problems, such as lack of basic infrastructure (electricity, clean water, and toilets), extortion, and capital constraints. In section VB, we discuss to what extent this set of obstacles might affect vendors' behaviors.

We also collected data on the network for each vendor. Specifically, we showed each of them a picture of the other vendors on the block, as well as the vendors in the two neighboring blocks, and asked (i) whether they knew this vendor well, (ii) whether this vendor was a family member, and (iii) how often they interacted during the week. This gives us crucial information on the possible flow of information. As we can see in table 1, panel D, on average, vendors from all the treatment arms know a similar number of people from their own blocks (7.7 vendors), as well as their neighboring blocks (4.7 vendors). Panel E provides information about the balance for the variables on observed behavior.

B. Outcome Variables

We now turn to the first inspection of our outcomes of interest as measured at the endline, after the intervention. This gives us a first overview of the effects, which are then measured with precision in section V. The variables are divided into three groups: (i) awareness, capturing to what extent vendors are aware of food safety matters; (ii) claimed behavior, capturing vendors' self-reported actions taken to improve their behavior with respect to food safety; and (iii) observed behavior, capturing vendors' behavior as observed by the data collectors. While the data on the latter were collected by observing vendors in all six survey rounds, data on the first two categories were collected only at the endline. In what follows, we discuss each category of outcome variables in detail.

First, regarding awareness, the endline survey asked vendors what, according to them constitutes personal hygiene and food hygiene, what some common contaminants of food are, and the reasons for using clean utensils.29 Based on these, we compute a proxy of awareness that takes value 1 if the vendor mentioned any point at all and 0 otherwise.30 It is worth noting that these outcome variables are based on whether the vendor can clearly enumerate relevant hygienic practices. Vendors who know the answers but do not say it are observationally equivalent to those who are not aware. Thus, part of what we are identifying is whether the training makes different aspects of hygienic practices more salient, easier to think about systematically, articulate, and enumerate for the vendors. Panel A of table 2 reports the proportion of vendors who can provide any answer for each of the three treatment arms. We see that this proportion is consistently higher for the two treatment arms compared to the control.

Table 2.

Postintervention: Summary of Main Outcome Variables

TrainingTraining w/SignalControlFull sample
(T1)(T2)(C)
 Panel A. Food safety awareness 
Food hygiene 0.70 0.67 0.58 0.65 
Personal hygiene 0.80 0.77 0.75 0.77 
Contamination 0.73 0.72 0.69 0.71 
Utensils 0.84 0.86 0.77 0.82 
 Panel B. Claimed Behavior 
Discussions 0.24 0.21 0.05 0.17 
Get information 0.16 0.16 0.07 0.13 
Improvements 0.47 0.43 0.27 0.39 
Investments 0.09 0.05 0.06 0.07 
Discuss NA 0.12 0.12 0.03 0.09 
Information on NA 0.11 0.08 0.06 0.08 
Unions 0.04 0.05 0.04 0.04 
Meetings 0.10 0.12 0.04 0.09 
 Panel C. Observed Behavior 
Facilities index 0.54 0.50 0.51 0.52 
Food handling index 0.80 0.80 0.83 0.81 
Customers index 0.64 0.60 0.62 0.62 
Total index 0.64 0.60 0.64 0.63 
Observations 176 192 193 561 
TrainingTraining w/SignalControlFull sample
(T1)(T2)(C)
 Panel A. Food safety awareness 
Food hygiene 0.70 0.67 0.58 0.65 
Personal hygiene 0.80 0.77 0.75 0.77 
Contamination 0.73 0.72 0.69 0.71 
Utensils 0.84 0.86 0.77 0.82 
 Panel B. Claimed Behavior 
Discussions 0.24 0.21 0.05 0.17 
Get information 0.16 0.16 0.07 0.13 
Improvements 0.47 0.43 0.27 0.39 
Investments 0.09 0.05 0.06 0.07 
Discuss NA 0.12 0.12 0.03 0.09 
Information on NA 0.11 0.08 0.06 0.08 
Unions 0.04 0.05 0.04 0.04 
Meetings 0.10 0.12 0.04 0.09 
 Panel C. Observed Behavior 
Facilities index 0.54 0.50 0.51 0.52 
Food handling index 0.80 0.80 0.83 0.81 
Customers index 0.64 0.60 0.62 0.62 
Total index 0.64 0.60 0.64 0.63 
Observations 176 192 193 561 

The table provides a first inspection of our outputs of interest as measured at the endline, after the intervention. This gives us a first overview of the effects, which are then measured with precision in section V. The variables are divided into three groups: Panel A reports measures of awareness, capturing to what extent vendors are aware of food safety matters. Panel B reports measures of claimed behavior, capturing vendors' self-reported actions taken to improve their behavior with respect to food safety. Panel C reports observed behavior, capturing vendors' behavior as observed by the data collectors.

Second, regarding claimed behavior, the endline questionnaire included a set of questions directly inquiring whether the vendor had engaged in any of the following activities over the previous six months: (i) discussing food safety with other vendors, (ii) trying to obtain more information on food safety, (iii) trying to improve the hygiene and sanitation levels at their stalls, (iv) buying new utensils or investing in community solutions such as a bigger waste bin for multiple vendors on that street, (v) talking about the National Act with other vendors, (vi) trying to find out more information about the National Act, (vii) discussing it with their unions, and (viii) interacting with other vendors about food safety improvements. Vendors could either agree or disagree that they had engaged in each of these actions. Based on their answers, we define a dummy variable corresponding to each action, taking the value 1 if the vendor had participated in that activity and 0 otherwise. This set of measures was based on questionnaires used by public health scholars to evaluate self-reported food safety behaviors (Soon et al., 2012; da Cunha et al., 2014; Clayton et al., 2002). The proportion of vendors engaging in each of these activities is reported in panel B of table 2. Again, we see that this proportion is, on average, much higher for the treated vendors compared to the control group. For instance, as many as 24% of the vendors in the first treatment group and 21% of vendors in the second treatment group report that they engaged in discussions with other vendors about safe food handling, while this proportion was only 5% for the control vendors.

Third, regarding observed behavior, the variables are defined on the basis of the information collected on observed vendors' food safety practices and can be grouped into three broad categories.31 The first category is facilities, and it is applicable to all vendors. We assigned vendors scores based on (i) whether they used an apron and a proper trash can, (ii) whether there was safe drinking water, (iii) whether the vendor was using a clean cloth to wipe his hands, and (iv) whether there was food debris or waste on the floor of the stall or in the cooking area. We then calculated the percentage of the total score that the vendor received and obtained the final “Facilities index.” For instance, a vendor exhibiting perfect behavior on all dimensions would get a score of 1, corresponding to 100%, on this index. The index for the second category, handling, is similarly constructed and based on scores assigned for (i) whether the ingredients are covered and separated from the cooked food, (ii) whether the cooked food is covered, and (iii) whether tongs, spatulas, or other implements are used to cook and to serve (as opposed to touching the food with bare hands). This index is relevant only for vendors who are actually cooking food at the stall, not for vendors who sell only precooked or prepackaged snacks, cut fruits, beverages, and so on. The corresponding index for the third category, customer, is calculated based on scores for (i) whether there is food debris on the tables, chairs, or benches, (ii) whether the vendor uses disposable plates, cups, and cutlery, and (iii) whether the nondisposable utensils (usually made of steel) are washed with soap. This last index is applicable only for vendors who have accommodations for customers to sit and eat at the stall. Finally, the index Total is the average of the three indices and is representative of the general (audited) observed vendors' behavior or hygiene practices of the street-food vending sector in Kolkata. The online appendix includes further details on how each of these indices was constructed.

As shown in panel C of table 2, vendors displayed an average score of 52% for the facilities index, around 81% for the food-handling index, around 62% for the customer service index, and 63% for the total index. In contrast to the other two sets of output variables, there seems to be no marked difference between the treatment arms in terms of these observed indices at the time of the endline.32 We analyze this in the next section.

V. Empirical Strategy and Results

A. Specifications

We now empirically estimate the effects of the intervention. Our identification strategy exploits the random assignment of blocks of vendors to the different treatment groups, to recover the effect of being assigned to the training on each of the three categories of outcome variables.

Our empirical specification is
yit=α+βTi+Xiγ+δt+ɛit,
(1)
where yit is the outcome variable for vendor i in period t; Ti is a dummy variable indicating the treatment status of vendor i; Xit is a vector of several controls such as the vendor's age, gender, and years of experience, the area of the city where the stall is located, whether the stall sells cooked food, and the number of pilot blocks within a 1 km radius; and δt is a period fixed effect. In this specification, as well as in those that follows, standard errors are robust and clustered at the block level. The randomization of treatment implies that the coefficient β yields an unbiased effect of the intervention. It is worth noting here that we consider the intention-to-treat (ITT) estimate as the policy-relevant parameter for our purpose, since the broad question that we are interested in is to identify the effect of this intervention were it to be replicated exactly on a larger scale.33
In order to separately identify the effects of the two different treatments and to compare their effects, we estimate an identical specification with a dummy for each treatment as follows,
yit=α+β1T1,i+β2T2,i+Xiγ+δt+ɛit,
(2)
where T1,i refers to the treatment group with only the training, and T2,i refers to the treatment group with both training and signal.
The difference specification is our preferred specification: the treatment being randomized, a simple comparison of treatment and control group vendors, does not suffer from endogeneity due to differences in pretreatment trends. The staggered nature of the training and the multiple rounds of monitoring further enables us to estimate a difference-in-difference specification, exploiting the difference in timing of the training administered to different blocks as follows,
yit=α+β1T*Afterit+β2Ti+Xiγ+δt+ɛit,
(3)
where Afterit is a dummy variable for whether the data were collected after the workshop. Our coefficient of interest here is β1, which measures the difference in pre- and posttraining outcomes for the treatment groups, compared to the difference for the control group, and gives a more precise estimate of the training effect.34 The results from this latter specification are qualitatively identical and are reported in appendix A3.

B. Main Results

In this section we report the results we obtained for each outcome variable. First, table 3 reports the coefficients for the effects of the training program on vendors' awareness. Panel A outlines the effects of belonging to any treatment group: β in equation (1). As one can see, during the endline survey, vendors assigned to the training were on average more likely to recall a relevant issue about food hygiene (8.2 percentage points, or 8.2/58 = 14.1% of mean control group), personal hygiene (2.6 p.p.), contamination of food (8 p.p.), and cleanliness of utensils (9.7 p.p.), as compared to the control group. Panel B shows the effects of the two treatment groups separately: β1 and β2 from equation (2). Notice that even though the vendors in T2 were provided with the means of signaling to customers their participation in the training program, we do not see a larger improvement as compared to T1.

Table 3.

Awareness about Food Safety: Difference Specification

(1)(2)(3)(4)
Food HygienePersonal HygieneContaminationUtensils
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.082*** 0.026* 0.080** 0.097*** 
 (0.030) (0.013) (0.040) (0.035) 
Observations 561 561 561 561 
R-squared 0.571 0.782 0.405 0.299 
Mean Dep. Var. (C) 0.580 0.746 0.694 0.767 
 Panel B. Separate Treatments 
Training (T1) 0.075** 0.025 0.072* 0.094** 
 (0.033) (0.018) (0.043) (0.039) 
Training w/Signal (T2) 0.088** 0.027 0.085* 0.099** 
 (0.035) (0.018) (0.045) (0.039) 
p-value (T1 = T2) 0.70 0.95 0.73 0.90 
Observations 561 561 561 561 
R-squared 0.571 0.782 0.406 0.299 
Mean Dep. Var. (C) 0.580 0.746 0.694 0.767 
(1)(2)(3)(4)
Food HygienePersonal HygieneContaminationUtensils
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.082*** 0.026* 0.080** 0.097*** 
 (0.030) (0.013) (0.040) (0.035) 
Observations 561 561 561 561 
R-squared 0.571 0.782 0.405 0.299 
Mean Dep. Var. (C) 0.580 0.746 0.694 0.767 
 Panel B. Separate Treatments 
Training (T1) 0.075** 0.025 0.072* 0.094** 
 (0.033) (0.018) (0.043) (0.039) 
Training w/Signal (T2) 0.088** 0.027 0.085* 0.099** 
 (0.035) (0.018) (0.045) (0.039) 
p-value (T1 = T2) 0.70 0.95 0.73 0.90 
Observations 561 561 561 561 
R-squared 0.571 0.782 0.406 0.299 
Mean Dep. Var. (C) 0.580 0.746 0.694 0.767 

The table reports the coefficients for the effects of the training program on vendors' awareness. Panel A reports the effects of belonging to any treatment group: β in equation (1). Panel B shows the effects of the two treatment groups separately: β1 and β2 from equation (2). All regressions control for vendor's age, gender, education level, a binary indicator for stall ownership, years vending, area of vending, a binary indicator for food cooked at the stall, and the type of food, and include interviewer fixed effects. Robust standard errors, clustered at the block level, are in parentheses. These regressions are estimated only for the endline sample. *p<0.10, **p<0.05, and ***p<0.01.

Second, table 4 reports identical coefficients for claimed behavior, or actions that vendors claim to have taken, based on the endline data. According to Panel A vendors assigned to treatment are significantly more likely to have taken part in most of the activities they were asked about. Coefficients range from as large as 23 p.p. more for having tried to improve sanitation levels at their stalls, to a modest 6 p.p. more for trying to obtain more information on the National Act. Similar to the results on awareness, Panel B shows that the effects for the two treatment arms are statistically indistinguishable from each other for all outcomes, with the exception of investing in stall equipment, for which vendors in the first treatment group in fact claim to have done more than those in the second group.

Table 4.

Claimed Behaviors: Difference Specification

(1)(2)(3)(4)(5)(6)(7)(8)
DiscussionsGet infoImprovementsInvestmentsDiscuss NAInfo on NAUnionsMeetings
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.204*** 0.132*** 0.230*** 0.028 0.120*** 0.060** 0.002 0.090*** 
 (0.037) (0.038) (0.052) (0.023) (0.032) (0.029) (0.020) (0.030) 
Observations 518 519 517 517 518 516 517 518 
R-squared 0.219 0.188 0.466 0.261 0.215 0.179 0.124 0.158 
Mean Dep. Var. (C) 0.049 0.068 0.270 0.062 0.031 0.062 0.043 0.037 
 Panel B. Separate Treatments 
Training (T1) 0.225*** 0.142*** 0.213*** 0.075*** 0.135*** 0.077** −0.007 0.086** 
 (0.047) (0.048) (0.065) (0.025) (0.038) (0.038) (0.025) (0.035) 
Training w/Signal (T2) 0.187*** 0.124*** 0.244*** −0.008 0.108*** 0.047 0.008 0.094** 
 (0.044) (0.044) (0.054) (0.027) (0.037) (0.033) (0.025) (0.037) 
p-value (T1 = T2) 0.49 0.72 0.58 0.00 0.46 0.44 0.60 0.84 
Observations 518 519 517 517 518 516 517 518 
R-squared 0.220 0.188 0.467 0.276 0.216 0.181 0.125 0.158 
Mean Dep. Var. (C) 0.049 0.068 0.270 0.062 0.031 0.062 0.043 0.037 
(1)(2)(3)(4)(5)(6)(7)(8)
DiscussionsGet infoImprovementsInvestmentsDiscuss NAInfo on NAUnionsMeetings
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.204*** 0.132*** 0.230*** 0.028 0.120*** 0.060** 0.002 0.090*** 
 (0.037) (0.038) (0.052) (0.023) (0.032) (0.029) (0.020) (0.030) 
Observations 518 519 517 517 518 516 517 518 
R-squared 0.219 0.188 0.466 0.261 0.215 0.179 0.124 0.158 
Mean Dep. Var. (C) 0.049 0.068 0.270 0.062 0.031 0.062 0.043 0.037 
 Panel B. Separate Treatments 
Training (T1) 0.225*** 0.142*** 0.213*** 0.075*** 0.135*** 0.077** −0.007 0.086** 
 (0.047) (0.048) (0.065) (0.025) (0.038) (0.038) (0.025) (0.035) 
Training w/Signal (T2) 0.187*** 0.124*** 0.244*** −0.008 0.108*** 0.047 0.008 0.094** 
 (0.044) (0.044) (0.054) (0.027) (0.037) (0.033) (0.025) (0.037) 
p-value (T1 = T2) 0.49 0.72 0.58 0.00 0.46 0.44 0.60 0.84 
Observations 518 519 517 517 518 516 517 518 
R-squared 0.220 0.188 0.467 0.276 0.216 0.181 0.125 0.158 
Mean Dep. Var. (C) 0.049 0.068 0.270 0.062 0.031 0.062 0.043 0.037 

The table reports the coefficients for the effects of the training program on vendors' claimed behavior. Panel A reports the effects of belonging to any treatment group: β in equation (1). Panel B shows the effects of the two treatment groups separately: β1 and β2 from equation (2). All regressions control for vendors' age, gender, education level, a binary indicator for stall ownership, years vending, area of vending, a binary indicator for food cooked at the stall, and the type of food, and include interviewer fixed effects. Robust standard errors, clustered at the block level, are in parentheses. These regressions are estimated only for the endline sample. *p<0.10, **p<0.05, and ***p<0.01.

Third, regarding observed behavior, we estimate specification (1) on all six rounds of actual behavioral data as observed by the data collectors. Table 5 presents these results. Panel A reports the effect of the training on each of the behavior indices, pooling the treatments together. We do not find a significant difference in the observed behavior of the vendors who were eligible to attend the training compared to those who were not. In particular, this is also the case for disaggregated individual outcomes, which do not require any investments or monetary outlay and where information alone should have been the primary constraint, such as keeping the cooking area and the stall clean, storing raw ingredients separately from cooked food, and keeping ingredients and cooked food covered (results in table A4 in the appendix). The only individual outcome for which we observe an improvement in the treatment vendors compared to the control vendors is in wearing an apron, which in fact requires the purchase of one; a possible explanation is that wearing an apron is a behavior that vendors know would definitely be perceived by the consumers, while the same may not be true for some of the other outcomes.

Table 5.

Observed Behavior: Difference Specification

(1)(2)(3)(4)
Facilities IndexFood handling IndexCustomer care IndexTotal Index
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.013 -0.000 -0.000 0.003 
 (0.013) (0.014) (0.016) (0.012) 
Observations 3,478 2,648 3,201 3,478 
R-squared 0.174 0.189 0.354 0.227 
Mean Dep. Var. (C) 0.447 0.694 0.689 0.581 
 Panel B. Separate Treatments 
Training (T1) 0.026* 0.011 -0.004 0.016 
 (0.014) (0.015) (0.018) (0.012) 
Training w/Signal (T2) 0.003 -0.010 0.003 -0.007 
 (0.016) (0.016) (0.018) (0.015) 
p-value (T1 = T2) 0.12 0.09 0.68 0.09 
Observations 3,478 2,648 3,201 3,478 
R-squared 0.175 0.190 0.354 0.229 
Mean Dep. Var. (C) 0.447 0.694 0.689 0.581 
(1)(2)(3)(4)
Facilities IndexFood handling IndexCustomer care IndexTotal Index
 Panel A. Pooled Treatment 
Training (T1 or T2) 0.013 -0.000 -0.000 0.003 
 (0.013) (0.014) (0.016) (0.012) 
Observations 3,478 2,648 3,201 3,478 
R-squared 0.174 0.189 0.354 0.227 
Mean Dep. Var. (C) 0.447 0.694 0.689 0.581 
 Panel B. Separate Treatments 
Training (T1) 0.026* 0.011 -0.004 0.016 
 (0.014) (0.015) (0.018) (0.012) 
Training w/Signal (T2) 0.003 -0.010 0.003 -0.007 
 (0.016) (0.016) (0.018) (0.015) 
p-value (T1 = T2) 0.12 0.09 0.68 0.09 
Observations 3,478 2,648 3,201 3,478 
R-squared 0.175 0.190 0.354 0.229 
Mean Dep. Var. (C) 0.447 0.694 0.689 0.581 

The table reports the coefficients for the effects of the training program on vendors' observed behavior. Panel A reports the effects of belonging to any treatment group: β in equation (1). Panel B shows the effects of the two treatment groups separately: β1 and β2 from equation (2). All regressions control for vendor's age, gender, education level, a binary indicator for stall ownership, years vending, area of vending, a binary indicator for food cooked at the stall, and the type of food, and include interviewer fixed effects. Robust standard errors, clustered at the block level, are in parentheses. These regressions are estimated using all six survey rounds. *p<0.10, **p<0.05, and ***p<0.01.

Panel B further reports the results for the two treatments separately. As one can see, contrary to expectations, the estimated effect is in fact marginally higher for the T1 group. We see a modest positive effect of the training on the facilities index (2.6 p.p., or 5.8% of the mean control group). Moreover, the point estimates for the second treatment group are sometimes negative (though not significant). One may argue that there seems to be a substitution between effort and advertising: when vendors are provided with promotional material, they may feel that they already signaled their quality to consumers, and therefore make less of an effort to prove through their food handling that their product is safe. We show later that such a substitution possibly contributed to vendors not earning greater revenues or profits despite signaling their engagement with food safety.

Overall, there is a very small effect of the training on behavior, despite the absence of any direct monetary incentives or repercussions. However, since this is only modest in magnitude, it does not allow us to conclude that a cheap intervention such as ours can alone sustain a marked improvement in food safety.

We run several robustness checks to further explore the effects of the training on actual behavior. First, in order to explore the possible heterogeneity over time, we estimate the difference specification, further splitting the vendors by the amount of time elapsed since each the start of the program. Appendix table A8 reports these results. We do not observe any clear pattern over time.35 Second, in order to explore the possible heterogeneity of the training with respect to the type of business run, we restrict the analysis to the group of vendors who might benefit the most from the workshops: vendors selling cooked food on the street. The results are qualitatively similar to the ones reported in the text and are available on request. Third, we use the information on the networks that the vendors operate in to explore whether the effects of the training are exacerbated by peer effects. From the location of each block, we know which two blocks are its closest neighbors in terms of geographical distance. We do not find a significant difference in the behavior of treated versus control vendors, either in terms of how many other treated vendors they are acquainted with or in terms of the density of treated blocks in their proximity. Since this analysis does not lead to any new insight, we report the results in appendix A4. Finally, we estimate some alternative specifications as additional robustness tests. We estimate the effect of the training by using the assignment of vendors to a treatment group as an instrument for receiving the training. We also estimate a specification identical to equation (1), but controlling for the baseline levels of the outcome variables, off the data collected from the first monitoring period onward. In both cases, the results are qualitatively identical to the diff-in-diff specifications and available on request.36

C. Why Are Behavioral Improvements Difficult to Undertake?

As discussed thus far, the training had only a modest effect on vendors' actual behavior, despite the marked improvement in awareness and claimed actions taken. In this and the following section, we explore some of the reasons the vendors' improved awareness may not have translated into actions.

First, concerning the supply side of the market, it may be the case that relaxing the information constraint did not necessarily make it easier for the vendors to potentially pursue the courses that the workshops advocated and offer a broader array of more hygienic choices to consumers. In order to directly test for this, we included in the questionnaire a set of questions aimed at capturing the vendors' own assessment of the resources needed to adopt these behavioral practices. During the baseline and the endline surveys, vendors were asked how “costly” they thought it was for them to provide the following services at their stalls: (i) provide clean drinking water, (ii) maintain a trash can that was emptied regularly, (iii) use a clean cloth to wipe their hands, (iv) keep the cooking area clean and free of food debris, and (v) use disposable plates and cups. The vendors could rate the difficulty level of making these improvements on an ordinal scale of very easy (1) to very difficult (5). Note that since we are only comparing the answers of the same vendors from the baseline and the endline, it is not a problem for our purposes if all vendors did not have the same interpretation of the difficulty levels, as long as the interpretation was consistent over time.

Table A10 shows that the training has the expected negative sign most of the time, but the effect is almost never significant. That is, the workshops did not significantly reduce how difficult the vendors think it is to follow these practices, despite improving their awareness about them.37 As discussed earlier, a possible explanation is that many of the constraints that vendors identify as the major problems in their daily business operations have to do with the lack of basic infrastructure, which our intervention did not directly help mitigate since it did not involve any monetary transfers or other financial benefits.38 We conclude, therefore, that providing information is not a sufficient condition to substantially decrease the perceived cost of tackling food safety hazards.39

Second, concerning the demand side of the market, it may be the case that consumers are unable to distinguish between hygienically prepared food and contaminated fare, which would imply that vendors may not have enough incentives to alter their practices in order to retain their customer base. In the following section, we elaborate on our consumer survey and shed light on why vendors may not have been prompted to improve their behavior despite the presence of demand for safer street food.

VI. Evidence on Consumers

In order to provide direct evidence regarding the demand side of the market, we conducted a follow-up survey on 1,480 street-food consumers across Kolkata. The purpose of the survey was twofold. First, we gathered extensive information on their street-food habits, thereby giving us both a holistic picture of their stated preferences and beliefs, as well as a means of analyzing such preferences as a function of the street vendor's or stall's characteristics. Second, we administered a comprehensive series of choice scenarios, in the spirit of a discrete choice experiment. This allows us to obtain quantitative information on the relative importance of different attributes that influence their choice of street food, as well as the trade-offs between price and vendor characteristics and the probability of take-up of pricier but “better” options. The survey was conducted between June and August 2019 across our original blocks.40 In this section, we first discuss the patterns that emerge from the consumer survey about self-reported habits and preferences and then discuss the discrete choice experiment in detail.

A. Consumer Survey

Table A15 in appendix A6 reports summary statistics from the consumer survey. The sample is divided into four groups, depending on the popular street-food items used in the choice scenarios.41 The table reports means for customers in each group, as well as overall averages. Column 5 of panel B gives an idea of the overall demographics of the street-food customers we interviewed. Slightly more than half of the consumers were male, the average age was around 36 years, and around 25% of them reported earning more than 20,000 rupees per month.42

In panel C, we document the usual street-food habits of these consumers. Street-food consumption is common, with 65% of them consuming it more than once a week, as is having a regular vendor from whom one buys. Getting ill from street food is pervasive even among these extremely frequent customers, and consumers make an effort to identify the cause or source of such food-borne illnesses, illustrating some scope for accountability among the food vendors. Forty-four percent of the consumers report having fallen sick at some point from consuming unsafe street food. Among these consumers, 47% claimed to have definitely been able to identify the source of the contaminated food. Changing the regular vendor is also quite common as a result: 38% of the consumers have a regular stall they visit that is different from a previous regular vendor.43

In panel D, we document the factors that determine consumers' street food choices. Consumers care about several factors such as the taste of food, price, location, whether the food is healthy, and whether the stall is hygienic. In order to get a more precise idea about the relative importance of these attributes, we specifically asked consumers what they considered the single most important factor in their decision. Though almost 75% seemed to be influenced to some extent by the consideration of hygiene, a modest 37% reported that it was in fact the most crucial factor, highlighting the role of other attributes in their decision-making process.

Finally, we examine whether customers believed they were capable of distinguishing between safe and unsafe food.44 We find that only 31% of the sample reported that it was easy or possible for them to detect whether food was contaminated. Consumers' decision of whether and where to consume street food may therefore be guided more by ex post considerations as a result of contracting some food-borne illness rather than ex ante informed decisions about whether a particular stall appears safe. Since these numbers are self-reported, the actual number of consumers who can realistically tell safe food from unsafe food is likely to be even lower. Coupled with the fact that less than half of them believe that they can trace the source of food-borne illnesses, this suggests that there may not be sufficient demand-side pressures to cause vendors to undertake a change in their behavior.

B. Discrete Choice Experiment

We now provide a quantitative assessment from our discrete choice experiment setup45 and show that consumers' stated preferences indicate a positive willingness to pay for food that they perceive as safer.

Discrete choice experiments are an established approach to elicit consumers preferences (WHO, 2012). Specifically, they are a commonly advocated quantitative method to assess different factors determining consumers' choices, as it provides information on the relative importance of various products' characteristics that influence consumers' choices. This method has been used in several contexts, including in the choice of food and health items in developing countries (Lancsar & Louviere, 2008; Otieno, 2011; and Mühlbacher & Johnson, 2016, among others). The key advantage of choice experiments is that they measure willingness to pay from stated preferences when reliable measures of revealed preferences are not available (de Bekker-Grob, Ryan, & Gerard, 2012). This makes them ideal in our setting, where we want to evaluate the relevance of each attribute (e.g., hygiene, price) from a series of hypothetical choices. The underlying idea is that stated preferences represent the expected utility from each combination of attributes, and thus we can evaluate the utility of each attribute level, as well as the overall utility of each alternative.

We follow a standard design, where each customer was asked to make a series of choices under a set of hypothetical scenarios, between a baseline good and an alternative good. Table 6 summarizes the main features of the design. The options were characterized by four attributes: (i) known, meaning whether the vendor is known to the consumer;46 (ii) hygiene, meaning whether the vendor and the stall appear hygienic;47 (iii) poster, meaning whether the vendor displays some poster or certification of training in food handling methods; and (iv) price. As mentioned before, the choices offered to each customer pertained to one out of four items,48 to allow us to check that the results were not simply true for one narrow food category or price range. We included both variation in terms of kind of food for the same price range (items 3 and 4), as well as variation in price ranges for the same kind of food (items 2 and 4).49

Table 6.

Design of Our Discrete Choice Experiment

(1)(2)(3)(4)
KnownHygienePosterPrices
Baseline:     
Option K  × × 
Alternatives:     
Option K,H   × A1, A2 
Option K,P  ×  A1, A2 
Option K, H, P    A1, A2 
Option U, H ×  × B, A1, A2 
Option U, P × ×  B, A1, A2 
Option U, H, P ×   B, A1, A2 
(1)(2)(3)(4)
KnownHygienePosterPrices
Baseline:     
Option K  × × 
Alternatives:     
Option K,H   × A1, A2 
Option K,P  ×  A1, A2 
Option K, H, P    A1, A2 
Option U, H ×  × B, A1, A2 
Option U, P × ×  B, A1, A2 
Option U, H, P ×   B, A1, A2 

The table provides a scheme of our design for the discrete choice experiment. There are four characteristics: (i) known, meaning whether the vendor is known to the consumer; (ii) hygiene, meaning whether the vendor and the stall appear hygienic; (iii) poster, meaning whether the vendor displays some poster or certification of training in food-handling methods; and finally (iv) price. K and U refer to known and unknown vendor. H refers to hygienic vendor. P refers to vendor with a poster or certificate. B, A1, and A2 refer to the baseline and two alternative prices (B<A1<A2). The default vendor is a known vendor who has bad hygiene, no poster, and is selling at a baseline price.

The set of choices are as follows. Consumers were asked to choose between two hypothetical vendors selling the respective food (e.g., a vegetarian meal for item 1): a default vendor and an alternative option. The default vendor remained the same across all choices: it was a known vendor, selling the food at the baseline price (as reported in panel A of table A15 in the appendix), who did not seem to have very good hygiene and did not display a poster or certification of training. The alternative options varied each of the attributes, individually as well as together, as well as the price. There were six types of alternative vendors, enumerated in table 6. B, A1, and A2 indicate the baseline price and the two alternative prices (as listed in panel A of table A15 in the appendix with B being the lowest price and A2 being the highest price). As borne out by the survey, consumers tend to favor regular vendors and value a relationship with the vendors: the first three alternatives therefore strictly dominate the default option and were therefore characterized only by each of the two alternative prices. The last three alternatives involve a trade-off between better facilities and knowing the vendor and are therefore characterized by the baseline price as well as the alternative prices, in order to allow for the possibility of a relationship with the vendor offsetting the improved stall conditions.50

The idea of good hygiene was communicated to the consumers in terms of our behavioral outcomes. Moreover, since the hygienic condition of the stall was explicitly described to all consumers instead of having them infer it, our approach allows us to gauge the intrinsic preferences of all consumers irrespective of their actual ability to detect stall conditions. One caveat is that survey participants may have been able to gauge that these are “desirable” practices from the questions, so their answers may reflect experimenter-demand effects to some extent. However, as we shall show, the valuation of perceived food safety is indeed higher for consumers who are more engaged with hygiene (either considering it the key determinant of their street-food-purchase decisions or being able to distinguish between safe and contaminated fare), thus suggesting that this should not be a major concern.

Figure 1 documents the proportion of consumers who opted for the alternative over the default vendor, for different combinations of characteristics, across the four items. A few patterns emerge from this picture. First, as one would expect, for each set of alternative options, the share of customers who choose the alternative declines with an increase in price. The decline is, however, fairly modest, and the share of customers who claim they would prefer the alternative is consistently high (between 65% and 90%), indicative of a positive willingness to pay for improved hygiene and a relatively flat demand curve. Second, the share of customers who state a preference for the alternative bundles are much smaller in the hypothetical scenarios where the vendor displays a poster or certification of training unaccompanied by an actual improvement in hygiene, compared to the choices involving better hygiene. This suggests that vendors should not experience an improvement in their business outcomes simply by displaying the poster without in fact improving behavior, and we show evidence for this in the next section.
Figure 1.
Demand for Alternative Options

This figure shows the proportions of consumers who report that they would choose the alternative option over the default vendor, for different combinations of characteristics. B, A1, and A2 refer to the baseline and 2 alternative prices (B<A1<A2). The default vendor is a known vendor who has bad hygiene and no poster. The alternative options correspond to those listed in table 6.

Figure 1.
Demand for Alternative Options

This figure shows the proportions of consumers who report that they would choose the alternative option over the default vendor, for different combinations of characteristics. B, A1, and A2 refer to the baseline and 2 alternative prices (B<A1<A2). The default vendor is a known vendor who has bad hygiene and no poster. The alternative options correspond to those listed in table 6.

We also specifically examine differences in demand responses by a number of consumer attributes, by reproducing the picture for different subsamples (figure A4 in appendix A6). An important takeaway is that consumers who consider hygiene the most important factor and those who believe they can detect whether food is contaminated are in fact also more likely to opt for food that appears safer (panels C and D). Finally, these patterns also hold when we examine demand separately by item (figure A5 in appendix A6.1).

We further supplement these graphs with an analysis of the the take-up rates of the alternative option by regressing whether the consumer chooses the alternative option on each of the attributes characterizing it. The results are reported in table A17 in the appendix.51 Consumers do exhibit willingness to pay for each of the attributes and, unsurprisingly, demand declines in price. Comparing among the attributes, the probability of opting for a hygienic vendor is the highest, followed by a known vendor, but consumers do not seem to assign much intrinsic valuation to external certifications or endorsements of vendor quality. Again, this supports the idea that vendors should not encounter better business outcomes in the absence of an actual improvement in hygiene.52

VII. Conclusion and Policy Implications

In this paper, we implement a randomized control trial to test whether a cheap intervention aimed at increasing food safety awareness among street vendors can improve food safety behavior. We also conduct a survey on consumers of street-food in this market and elicit their stated preferences for food attributes using a discrete choice experiment. Although we find substantial improvements in vendors' knowledge about food safety hazards and actions that they claim to have undertaken, we observe limited effects on vendors' actual practices. Our design allows us to investigate some possible explanations regarding the small effect on food safety behavior. We provide evidence that this is likely due to the training not significantly reducing vendors' perceived difficulty of providing more hygienic services to customers. We also show that though consumers exhibit a positive marginal willingness to pay for higher perceived hygiene, they struggle to detect whether street food is contaminated.

Our results highlight that information alone to the vendors is not sufficient to alter hygienic practices in this sector. This is a useful, evidence-based policy conclusion for those organizations and institutions whose job is to provide guidance to governments on setting up long-term strategies to control food safety. For instance, the ongoing formalization processes in India are uniquely focusing on licensing and delimiting vending areas. Conversely, we point out that future research, as well as policymakers, should focus on mitigating especially the supply-side constraints, as well as improving food safety awareness among consumers, in order to achieve improvements in the street-food vending sector. Lack of basic facilities, such as clean water, waste disposal, and electricity, are all factors that hinder self-improvement in food safety, and should be at the core of investigation in future studies. Governments interested in formalizing this sector should seriously take into account the supply-side constraints when drafting street-food vending regulations and when designing vending areas.

Notes

1

InnoAid is active in addressing development challenges supported by academic collaborations. They started working on street-food vendors in Kolkata in 2008. Besides InnoAid, the project was implemented jointly with two local organizations active in promoting sustainable development for the poor and marginalized communities: Joygopalpur Gram Vikash Kendra (JGVK) and Gana Unnayan Parshad (GUP). For more details on InnoAid and its partners, see the online appendix and http://www.innoaid.org/partners/existing-partners/.

2

We explain in more detail in section IIIC what we mean by “block” of vendors.

3

Crucially, our results are not going to be driven by spillover effects because we collected precise information about each vendor's network on the street and hence we can control for it. We elaborate on this in section IV.

4

In this regard, we relate to the social-dilemma literature on the effectiveness of rewards versus punishments (e.g., Sigmund, Hauert, & Nowak, 2001; Van Lange, Rockenbach, & Yamagishi, 2014).

5

“Self-regulation” refers to the set of actions undertaken by the private sector to regulate itself. This might arise for several reasons such as: (i) self-interest, e.g., a firm deciding to sell a “green” product because consumers are willing to pay for it; (ii) to enhance ethical beliefs (Baron, 2010); (iii) or to deter public regulation (Maxwell, Lyon, & Hackett, 2000; Lyon & Maxwell, 2004) or boycotting (Baron, 2009; Baron & Diermeier, 2007).

6

Recent experiments document that such threats may not always work positively (Casaburi & Macchiavello, 2015). Nevertheless, these findings support our argument that sanctions would prevent us from identifying the (pure) effect of increased awareness on vendors' behavior such that it is not confounded by other effects. For our specific context, details on the street food sector in Kolkata are provided in the Background section.

7

Both results are the average effects across several categories. We report results separately for each outcome variable in section V.

8

A corollary of this finding is the following. If one were to interpret the market failure in the street-food market as a standard adverse selection outcome, where the poorer good prevails because of consumers' inability to distinguish among qualities, this problem cannot be mitigated in the context of this sector simply by providing more information to consumers.

9

Note that, ideally, this channel could be directly tested by randomizing a monetary compensation (cash or in-kind) to vendors. However, since we deal with a complex sector characterized by widespread extortion suffered by street vendors, monetary incentives are difficult to test in our context. This is confirmed by the information we collected at baseline and also by InnoAid's local partners, who were highly skeptical about providing monetary incentives to street vendors. More details in section III.

10

The Municipal Act in the city of Kolkata prevents any type of street vending, and the punishment for the same is rigorous imprisonment for up to three months or a fine. For more details about the Municipal Act, see “section 371 of Kolkata Municipal Act, 1997.” As for the progress in the implementation of the National Act on Street Vending, see https://nasvinet.org/newsite/progress-in-the-implementation-of-the-street-vendors-act-2014-2.

11

See Karlan and Valdivia (2011); Mano et al. (2012); de Mel, McKenzie, and Woodruff (2014); Argent, Augsburg, and Rasul (2014); Valdivia (2015); and Brooks, Donovan, and Johnson (2018) for more recent work.

12

Medeiros et al. (2011) take into account three case studies in which food safety is measured through bacteriological analysis. However, they concern very different settings, for example, hospital catering food handlers.

14

This literature documents the high levels of bacterial contamination in street food across urban areas in developing countries and identify Staphylococcus, Escherichia coli, and Salmonella as the main microorganisms related to food-borne diseases in this sector. Importantly, the most frequent food safety hazards among street-food vendors are related to unhygienic handling; raw material inadequate cleaning of the machines used to cut the food, contact surfaces, clothes; and airborne contamination.

15

For instance, in the Indian context, Choudhury et al. (2011) find that only 30% of vendors were aware of hygienic practices related to food handling and only 8% were aware of biological sources of food contamination. Similar studies, which report street vendors' low levels of food safety knowledge, have been run in several developing countries, such as Ghana (Mensah et al., 2002), South Africa (Lues et al., 2006), Haiti (Samapundo et al., 2015), Brazil (Cortese et al., 2016), and Vietnam (Samapundo et al., 2016). As shown in the next section, we find low levels of food safety awareness among vendors in our sample.

16

Over the years, several public institutions around the world have embraced a similar approach, requiring all food handlers to be trained in food handling. Examples are the Brazilian Health Surveillance Agency (de Freitas Saccol et al., 2016), the Dubai Municipality Food Control Department (Mohammed Sharif Al-Awadhi et al., 2011), the Romanian Health Department (Jianu & Goleţ, 2014), as well as the European Food Safety Authority (Commission, 2004), and many others.

17

We thank an anonymous referee for helpful suggestions about the interpretation of the mechanism.

18

A simple conceptual framework is included in appendix A2.

19

We elaborate on this in section VI.

20

One of the first documented research on this city is a study by the All Indian Institute of Hygiene and Public Health (AIIH and PH), with support from FAO (Chakravarty & Canet, 1996), which has shown that food safety hazards are prevalent in Kolkata due to limited food hygiene awareness, unclean environments, working practices, and food adulteration. A more recent case study reports that 47% of the samples of water used by street-food vendors were contaminated (Walvekar, 2017). Importantly, this study highlights that the water was safe at the source but became contaminated through poor handling practices.

21

A “block” is arbitrarily defined by us for convenience, on the basis of the presence of vendors and its natural location on the street. For example, a street (or portion of) between two large infrastructures that naturally divides vendors may constitute a block. The definition of a block also facilitated the choice of the optimal time and location for the workshops, and fostered cooperation among vendors in the same block by providing them with a platform to come together. Our choice is sound and motivated also by the fact that vendors do not move across blocks, as they work in permanent built-up stalls.

22

The selection of the number of blocks and vendors was decided on the basis of a conservative power analysis, assuming an effect size of 0.20 SD, whereas we assume an intracluster correlation coefficient of 0.10. Furthermore, the pilot group was evaluated between March 2015 and October 2015. This group was crucial to improve several aspects of the research design such as the length and accuracy of the survey, the reliability of the direct monitoring, and the quality of the workshops. We control for the proximity to the pilot blocks and use them in additional robustness checks.

23

This includes the presence of other stall representatives at the workshops if the owners themselves could not always attend. In our analysis, we can always observe whether the owner of the stall attended all the workshops and who else (such as a family member) attended the workshop on his or her behalf. All results that follow are robust to these controls.

24

This treatment group attended the workshops only after the conclusion of the workshops for the first treatment group. We deal with this point in section V.

25

These questions are included only in the baseline survey, at the beginning of the project, with the purpose of providing suggestive evidence on the lack of food-safety awareness within our sample. Conversely, in the endline survey, we collect a different and more comprehensive set of questions covering all aspects of food safety knowledge (see section IV).

26

In particular, the vendors who seemed more aware of contaminants were not necessarily more likely to exhibit better practices at the baseline, underscoring the need to train them about the importance of basic hygiene and safe food handling.

27

The locations (e.g., union offices, building premises) varied depending on the local availability of indoor areas close to the vendors' workplace.

28

Examples of light snacks are bread, eggs, samosas, and pakoras. Examples of heavy snacks are rolls, momos, dosas, fried rice, and noodles.

29

The list of answers was (a) food hygiene: covering food, separating cooked and raw food, storing food at the proper temperature, handling food with proper implements, washing ingredients before cooking, using clean water, and using clean utensils; (b) personal hygiene: having clean hands, using soap, wearing clean clothes, having clean and short nails, wearing aprons, wearing caps, wearing gloves, taking regular baths, and using clean water; and (c) possible contaminants: bacteria in food, kerosene oil, or detergent; food coloring; insects; dirt; and contaminated water; and (d) utensils: it is good to use utensils to serve food, to avoid germs and bacteria, and to avoid sickness.

30

In the robustness checks, we also calculate an alternative proxy for awareness: we count the number of relevant points the vendor mentioned and normalize this number. The qualitative results of our analysis do not change.

31

Since these data on behavior are obtained only by observation and not by interacting with the vendors, individual outcomes are missing in the data both when a question is not relevant for a particular vendor as well as when the data collector cannot determine what the vendor's usual behavior is in some particular aspect (e.g., if the data collector happened to visit the stall when there were no customers, he or she might not be able to ascertain whether the vendors were serving the food on disposable plates). Since our purpose was to have an overall idea about whether the vendors improved their food-handling methods, and we should ideally use all of our sample to have enough statistical power, we aggregate the outcome variables for each category so as to maximize the number of observations.

32

Notice that these indices are balanced between groups at the time of the baseline survey. Results are reported in panel E of table 1.

33

On the other hand, if we were interested in estimating whether the information provided changes the behavior of the vendors, a more appropriate parameter to recover would be LATE, using treatment status as an instrument for actual workshop attendance . We report these estimates in appendix tables A11, A12, and A13. The results are qualitatively identical.

34

Since the random assignment implies that the pretraining trends in outcome should be identical for treatment and control groups, the diff-in-diff estimate essentially measures the difference in treatment and control after the training. The β coefficient in equation (1) would then be β1 in equation (3) weighted by the probability of being observed after the workshop. One caveat with the diff-in-diff specification is the following. If the effect of the treatment is not stable over time, then the coefficient of interest would be biased because “treated” units would serve as “control” units in subsequent periods. Hence, changing effects of treatment in combination with parallel trends in the pretreatment implies a failure of parallel trends in these subsequent periods (Goodman-Bacon, 2018). We thank an anonymous referee for pointing this out.

35

Notice that we observe some swinging in the time that the effect of the training takes shape, with the training sometimes only taking effect once the vendors have had more time to absorb the new information and implement it. In order to deal with this, in our comparison of the effects on the two treatment groups, we rerun specification (1) for all three outcomes controlling for the amount of time elapsed since the vendor had their first workshop. Since our aim here is to only get a comparison between the two treatment groups, we estimate these regressions only off the treated sample. The results are reported in table A9 of the appendix. We find that while the vendors who were provided with the opportunity to signal do in fact perform better on the awareness front, the first treatment group exhibits statistically similar, or even marginally better, outcomes in terms of both claimed and actual behavior.

36

Interestingly, we observe an increase in the observed behavioral outcomes from the baseline to the endline across both treated and control groups. For instance, the Total Index goes from 0.58 (in table 1) up to 0.63 (in table 2). A possible explanation would be that vendors changed their behavior because of the monitoring, which could explain the improvement also among vendors in the control group. This improvement should therefore depend on the frequency of the monitoring. To discard this hypothesis, we test whether vendors monitored more times show better performances. The results, available on request, do not highlight any evidence in favor of this hypothesis.

37

A series of questions in the baseline had sought to identify what the vendors considered the greatest problems in running their business (such as unavailability of running water or persecution by the local authorities), which hinted at different noneconomic costs of altering their behavior. As a further robustness test, we separately examine how the perceived cost changed for vendors who identified no threats to their business and those that identified multiple threats, and find no significant improvement for either group.

38

Though vendors expected that the workshops would improve their business outcomes ex ante, it may be the case that providing these hygienic services appeared more expensive than they had anticipated since they did not receive any financial help.

39

Notice that one could interpret these results as showing that the training just provided a clearer idea about the costs. However, the negative sign of most of these effects suggests that the training worked in the expected direction, but it was not sufficiently strong to lower the other “costs of production” for the vendors.

40

While surveying customers who are consumers of street food at the same locations where we studied the vendors ensures that they are representative of the kind of demand that those vendors cater to, the time lag makes it unlikely that the attitudes of the consumers are influenced by the behavior of the vendors immediately pre- or posttraining. Since our intervention was only on the supply side and did not include any workshops for the customers themselves, our consumer survey is intended to reflect their inherent opinions and preferences without any regard to our vendor training program and consequent vendor compliance. Indeed, as table A16 shows, the consumers across the treatment and control blocks look almost identical in all dimensions.

41

Consumers were asked about the different items sequentially through the survey, and the samples are not balanced across item. There is also substantial variation within the consumers in each item category, allowing us to explore heterogeneity in consumer choices.

42

While there is some variation in terms of these socioeconomic factors in aggregate among consumers who were offered different items for their choice experiment, there is also substantial variation among the consumers in each item category. This allows us to explore whether consumers with different attributes make different choices.

43

Besides having fallen ill from the food, other reasons include an increase in price, finding a better option, and not visiting that area any longer.

44

Despite the undeniable importance that consumers seem to attach to food hygiene, it is necessary for our purposes to understand how exactly it is that customers perceive and judge the level of hygiene. For instance, vendors should be more likely to improve their behavior if it is indeed the case that consumers believe that they are at all capable of discerning whether food is contaminated.

45

We thank Erwin Bulte for suggesting this approach.

46

We framed this question as someone from whose stall they have consumed food with some regularity in the past.

47

In the program sense.

48

Specifically: Veggie Meal (Item 1), Chicken Biriyani (Item 2), Rice and Fish (Item 3), and Chicken Chowmein (Item 4).

49

It is worth mentioning that we administered the choice experiments at the beginning of the survey, so as to prevent (as much as possible) priming the consumers about food hygiene considerations with the follow-up questions relating to their street food habits and safety awareness.

50

More precisely, given our design, there are 24 (2 × 2 × 2 × 3) possible combinations of characteristics. In our questionnaire, we included the fifteen combinations that were not strictly dominated by the baseline option. The optimal length and design of the survey were drawn up based on (i) an orthogonal design, as recommended by WHO (2012), as well as (ii) piloting. In order to check that consumers were in fact answering in a rational manner, we also included two “test scenarios,” where the default option strictly dominated the alternatives but the alternatives were characterized by higher prices. Following their guidelines, we do not include the two choices for consistency checks in our econometric analysis.

51

We report results and graphs based on our most conservative sample: consumers who made rational choices, and whose interview durations were between five minutes and twenty minutes. Using the entire sample yields similar results.

52

As a complementary approach, as is standard for such choice experiment setups, we also use the survey data to estimate consumers' willingness to pay. This is described in detail in section A6.2.

REFERENCES

Argent
,
J.
,
B.
Augsburg
, and
I.
Rasul
, “
Livestock Asset Transfers with and without Training: Evidence from Rwanda,
Journal of Economic Behavior and Organization
108
(
2014
),
19
39
.
Ashraf
,
N.
,
E.
Field
, and
J.
Lee
, “
Household Bargaining and Excess Fertility: An Experimental Study in Zambia,
American Economic Review
104
(
2014
),
2210
2237
.
Banerjee
,
A.
,
S.
Barnhardt
, and
E.
Duflo
, “
Movies, Margins, and Marketing: Encouraging the Adoption of Iron-Fortified Salt
,” in
David A.
Wise
, ed.,
Insights into the Economics of Aging
(
Chicago
:
University of Chicago Press
,
2015
).
Baron
,
D. P.
, “
A Positive Theory of Moral Management, Social Pressure, and Corporate Social Performance,
Journal of Economics and Management Strategy
18
(
2009
),
7
43
.
Baron
,
D. P.
Morally Motivated Self-Regulation,
American Economic Review
100
(
2010
),
1299
1329
.
Baron
,
D. P.
, and
D.
Diermeier
, “
Strategic Activism and Nonmarket Strategy,
Journal of Economics and Management Strategy
16
(
2007
),
599
634
.
Bhaskar
,
J.
,
M.
Usman
,
S.
Smitha
, and
G.
Bhat
, “
Bacteriological Profile of Street Foods in Mangalore,
Indian Journal of Medical Microbiology
22
(
2004
), 197.
Bhowmik
,
S.
,
Street Vendors in the Global Urban Economy
(
London
:
Taylor & Francis
,
2012
).
Bromley
,
R.
, “
Street Vending and Public Policy: A Global Review,
International Journal of Sociology and Social Policy
20
(
2000
),
1
28
.
Brooks
,
W.
,
K.
Donovan
, and
T. R.
Johnson
, “
Mentors or Teachers? Microenterprise Training in Kenya
,”
American Economic Journal: Applied Economics
10
(
2018
),
196
221
.
Casaburi
,
L.
, and
R.
Macchiavello
, “
Loyalty, Exit, and Enforcement: Evidence from a Kenya Dairy Cooperative,
American Economic Review
105
(
2015
),
286
290
.
Chakravarty
,
I.
, and
C.
Canet
, “
Street Foods in Calcutta,
Food, Nutrition and Agriculture
29
(
1996
),
30
37
.
Chan
,
S.
, and
Z.
Chan
, “
A Review of Foodborne Disease Outbreaks from 1996 to 2005 in Hong Kong and Its Implications on Food Safety Promotion,
Journal of Food Safety
28
(
2008
),
276
299
.
Choudhury
,
M.
,
L.
Mahanta
,
J.
Goswami
,
M.
Mazumder
, and
B.
Pegoo
, “
Socioeconomic Profile and Food Safety Knowledge and Practice of Street Food Vendors in the City of Guwahati, Assam, India,
Food Control
22
(
2011
),
196
203
.
Chumber
,
S. K.
,
K.
Kaushik
, and
S.
Savy
, “
Bacteriological Analysis of Street Foods I., Pune,
Editorial Board
51
(
2007
), 51.
Clayton
,
D. A.
,
C. J.
Griffith
,
P.
Price
, and
A. C.
Peters
, “
Food Handlers' Beliefs and Self-Reported Practices,
International Journal of Environmental Health Research
12
(
2002
),
25
39
.
Cortese
,
R. D. M.
,
M. B.
Veiros
,
C.
Feldman
, and
S. B.
Cavalli
, “
Food Safety and Hygiene Practices of Vendors during the Chain of Street Food Production in Florianopolis, Brazil: A Cross-Sectional Study,
Food Control
62
(
2016
),
178
186
.
da Cunha
,
D. T.
,
A. R. C.
Braga
,
E. de Camargo
Passos
,
E.
Stedefeldt
, and
V. V.
de Rosso
, “
The Existence of Optimistic Bias about Foodborne Disease by Food Handlers and Its Association with Training Participation and Food Safety Performance,
Food Research International
75
(
2015
),
27
33
.
da Cunha
,
D. T.
,
E.
Stedefeldt
, and
V. V.
de Rosso
, “
The Role of Theoretical Food Safety Training on Brazilian Food Handlers' Knowledge, Attitude and Practice,
Food Control
43
(
2014
),
167
174
.
de Bekker-Grob
,
E. W.
,
M.
Ryan
, and
K.
Gerard
, “
Discrete Choice Experiments in Health Economics: A Review of the Literature,
Health Economics
21
(
2012
),
145
172
.
de Freitas Saccol
,
A. L.
,
A. L.
Serafim
,
L. H.
Hecktheuer
,
L. B.
Medeiros
, and
E. A. D.
Silva Jr.
, “
Food Safety in Feeding Services: A Requirement in Brazil,
Critical Reviews in Food Science and Nutrition
56
(
2016
),
1363
1369
.
de Mel
,
S.
,
D.
McKenzie
, and
C.
Woodruff
, “
Business Training and Female Enterprise Start-Up, Growth, and Dynamics: Experimental Evidence from Sri Lanka,
Journal of Development Economics
106
(
2014
),
199
210
.
Desenclos
,
J.
,
K. C.
Klontz
,
M. H.
Wilder
,
O. V.
Nainan
,
H. S.
Margolis
, and
R. A.
Gunn
, “
A Multistate Outbreak of Hepatitis A Caused by the Consumption of Raw Oysters,
American Journal of Public Health
81
(
1991
),
1268
1272
.
Duflo
,
E.
,
P.
Dupas
, and
M.
Kremer
, “
Education, HIV, and Early Fertility: Experimental Evidence from Kenya,
American Economic Review
105
(
2015
),
2757
2797
.
Duflo
,
E.
,
S.
Galiani
, and
M.
Mobarak
, “
Improving Access to Urban Services for the Poor: Open Issues and a Framework for a Future Research Agenda
,” J-PAL Urban Services review paper (
2012
), http://www/povertyactionlab.org/publication/improving-access-urban-services-poor.
Dupas
,
P.
, “
Do Teenagers Respond to HIV Risk Information? Evidence from a Field Experiment in Kenya,
American Economic Journal: Applied Economics
3
(
2011
),
1
34
.
Dupas
,
P.
, and
E.
Miguel
, “Impacts and Determinants of Health Levels in Low-Income Countries” (pp.
3
93
), in
Esther
Duflo
and
Abhijit
Banerjee
, eds.,
Handbook of Economic Field Experiments
2
(
Amsterdam
:
North-Holland
,
2017
).
European Commission, European Food Safety Authority
, “
Regulation (EC) No. 852/2004 of the European Parliament and of the Council of 29 April 2004 on the Hygiene of Foodstuffs
,”
OJEU
L13
9
(
2004
),
1
54
.
FAO
,
Promises and Challenges of the Informal Food Sector in Developing Countries
(
New York
:
Food and Agriculture Organization
,
2007
).
FAO
FAO/WHO Guidance to Governments on the Application of HACCP in Small and/or Less-Developed Food Businesses
(
New York
:
Food and Agriculture Organization
,
2009a
).
FAO
Good Hygienic Practices in the Preparation and Sale of Street Food in Africa
” (
New York
:
Food and Agriculture Organization
,
2009b
).
FAO
Food for the Cities Initiative
” (
New York
:
Food and Agriculture Organization
,
2013
).
FAO/WHO
,
Codex Alimentarius: Food Hygiene, Basic Texts
(
New York
:
Food and Agriculture Organization
,
2003
).
Feglo
,
P.
, and
K.
Sakyi
, “
Bacterial Contamination of Street Vending Food in Kumasi, Ghana,
Journal of Medical and Biomedical Sciences
1
(
2012
),
1
8
.
Flint
,
J. A.
,
Y. T.
Van Duynhoven
,
F. J.
Angulo
,
S. M.
Delong
,
P.
Braun
,
M.
Kirk
,
Peter
Braam
, “
Estimating the Burden of Acute Gastroenteritis, Foodborne Disease, and Pathogens Commonly Transmitted by Food: An International Review,
Clinical Infectious Diseases
41
(
2005
),
698
704
.
Ghosh
,
M.
,
S.
Wahi
,
M.
Kumar
, and
A.
Ganguli
, “
Prevalence of Enterotoxigenic Staphylococcus aureus and Shigella spp. in Some Raw Street Vended Indian Foods,
International Journal of Environmental Health Research
17
(
2007
),
151
156
.
Goodman-Bacon
,
A.
, “
Difference-in-Differences with Variation in Treatment Timing
,”
NBER working paper
25018
(
2018
).
Greig
,
J. D.
,
E. C.
Todd
,
C. A.
Bartleson
, and
B. S.
Michaels
, “
Outbreaks Where Food Workers Have Been Implicated in the Spread of Foodborne Disease. Part 1. Description of the Problem, Methods, and Agents Involved,
Journal of Food Protection
70
(
2007
),
1752
1761
.
Jianu
,
C.
, and
I.
Goleţ
, “
Knowledge of Food Safety and Hygiene and Personal Hygiene Practices among Meat Handlers Operating in Western Romania,
Food Control
42
(
2014
),
214
219
.
Karlan
,
D.
, and
M.
Valdivia
, “
Teaching Entrepreneurship: Impact of Business Training on Microfinance Clients and Institutions
,” this review 93 (
2011
),
510
527
.
Lancsar
,
E.
, and
J.
Louviere
, “
Conducting Discrete Choice Experiments to Inform Healthcare Decision Making,
Pharmacoeconomics
26
(
2008
),
661
677
.
Lues
,
J. F.
,
M. R.
Rasephei
,
P.
Venter
, and
M. M.
Theron
, “
Assessing Food Safety and Associated Food Handling Practices in Street Food Vending,
International Journal of Environmental Health Research
16
(
2006
),
319
328
.
Lyon
,
T. P.
, and
J. W.
Maxwell
,
Corporate Environmentalism and Public Policy
(
Cambridge
:
Cambridge University Press
,
2004
).
Ma
,
X.
,
S.
Sylvia
,
M.
Boswell
, and
S.
Rozelle
, “
Ordeal Mechanisms and Information in the Promotion of Health Goods in Developing Countries: Evidence from Rural China
,” Rural Education Action Project working paper (
2013
).
Mano
,
Y.
,
A.
Iddrisu
,
Y.
Yoshino
, and
T.
Sonobe
, “
How Can Micro and Small Enterprises in Sub-Saharan Africa Become More Productive? The Impacts of Experimental Basic Managerial Training,
World Development
40
(
2012
),
458
468
.
Maxwell
,
J. W.
,
T. P.
Lyon
, and
S. C.
Hackett
, “
Self-Regulation and Social Welfare: The Political Economy of Corporate Environmentalism,
Journal of Law and Economics
43
(
2000
),
583
618
.
Medeiros
,
C. O.
,
S. B.
Cavalli
,
E.
Salay
, and
R. P.
Proenca
, “
Assessment of the Methodological Strategies Adopted by Food Safety Training Programmes for Food Service Workers: A Systematic Review,
Food Control
22
(
2011
),
1136
1144
.
Mensah
,
P.
,
D.
Yeboah-Manu
,
K.
Owusu-Darko
, and
A.
Ablordey
, “
Street Foods in Accra, Ghana: How Safe Are They?
Bulletin of the World Health Organization
80
(
2002
),
546
554
.
Meredith
,
J.
,
J.
Robinson
,
S.
Walker
, and
B.
Wydick
, “
Keeping the Doctor Away: Experimental Evidence on Investment in Preventative Health Products,
Journal of Development Economics
105
(
2013
),
196
210
.
Mohammed Sharif Al-Awadhi
,
K.
,
A. AbdulRahman
Al Ali
,
O. P.
Snyder
,
A.
AlSheikh
,
B.
Krishna
, and
J.
Taylor
, “
Food Safety Challenges and Initiatives in the Dubai Hospitality Industry,
Worldwide Hospitality and Tourism Themes
3
(
2011
),
443
449
.
Mühlbacher
,
A.
, and
F. R.
Johnson
, “
Choice Experiments to Quantify Preferences for Health and Healthcare: State of the Practice,
Applied Health Economics and Health Policy
14
(
2016
),
253
266
.
NCEUS
, “
Report on Conditions of Work and Promotion of Livelihoods in the Unorganised Sector
” (
New Delhi
:
National Commission for Enterprises in the Unorganised Sector, Government of India
,
2007
).
NCEUS
The Challenge of Employment in India: An Informal Economy Perspective
,” Report of the National Commission for Enterprises in the Unorganised Sector, Government of India (New Delhi: National Commission for Enterprises in the Unorganised Sector, Government of India, 2009).
NPUSV
, “
National Policy on Urban Street Vendors
” (
New Delhi
:
Government of India Ministry of Housing and Urban Poverty Alleviation
2009
).
Otieno
,
D. J.
, “
Choice Experiments in Developing Countries: Implementation, Challenges and Policy Implications,
Journal of Agricultural Economics
62
(
2011
),
245
247
.
Rane
,
S.
, “
Street Vended Food in Developing World: Hazard Analyses,
Indian Journal of Microbiology
51
(
2011
),
100
106
.
Samapundo
,
S.
,
R.
Climat
,
R.
Xhaferi
, and
F.
Devlieghere
, “
Food Safety Knowledge, Attitudes and Practices of Street Food Vendors and Consumers in Port-au-Prince, Haiti,
Food control
50
(
2015
),
457
466
.
Samapundo
,
S.
,
T. C.
Thanh
,
R.
Xhaferi
, and
F.
Devlieghere
, “
Food Safety Knowledge, Attitudes and Practices of Street Food Vendors and Consumers in Ho Chi Minh City, Vietnam,
Food Control
70
(
2016
),
79
89
.
Sigmund
,
K.
,
C.
Hauert
, and
M. A.
Nowak
, “
Reward and Punishment,
Proceedings of the National Academy of Sciences
98
(
2001
),
10757
10762
.
Soon
,
J. M.
,
R.
Baines
, and
P.
Seaman
, “
Meta-Analysis of Food Safety Training on Hand Hygiene Knowledge and Attitudes among food Handlers
,”
Journal of Food Protection
75
(
2012
),
793
804
.
Sousa
,
C. P. D.
, “
The Impact of Food Manufacturing Practices on Food Borne Diseases,
Brazilian Archives of Biology and Technology
51
(
2008
),
615
623
.
Valdivia
,
M.
, “
Business Training Plus for Female Entrepreneurship? Short and Medium-Term Experimental Evidence from Peru,
Journal of Development Economics
113
(
2015
),
33
51
.
Van Lange
,
P. A.
,
B.
Rockenbach
, and
T.
Yamagishi
,
Reward and Punishment in Social Dilemmas
(
New York
:
Oxford University Press
,
2014
).
Walvekar
,
V.
, “
A Research and Design Initiative for the Informal Sector of Street Food Vending in India
,” in
International Conference on Research into Design
(
New York
:
Springer
,
2017
),
309
316
.
World Health Organization
, “
Essential Safety Requirement for Street-Vended Foods
” (
Geneva
:
WHO
,
1996
),
World Health Organization
A Guide to Healthy Food Markets
” (
Geneva
:
WHO
,
2006
).
World Health Organization
Foodborne Disease Outbreaks: Guidelines for Investigation and Control
(
Geneva
:
WHO
2008
).
World Health Organization
How to Conduct a Discrete Choice Experiment for Health Workforce Recruitment and Retention in Remote and Rural Areas: A User Guide with Case Studies
” (
Geneva
:
WHO
,
2012
).
Wongtada
,
N.
, “
Street Vending Phenomena: A Literature Review and Research Agenda,
Thunderbird International Business Review
56
(
2014
),
55
75
.
World Bank
,
World Bank 2012
(
Washington, DC
:
World Bank
,
2013
).

Author notes

We are extremely grateful to InnoAid for the support to conduct this research project, in particular the president Marie Louise Pollman-Larsen, and the project manager, Matias Pollmann-Larsen. This project was funded by CISU (Civil Society in Development), which is financed by the Danish Ministry of Foreign Affairs (DANIDA), and by Monash University, Faculty Research Grant. The experiment has been registered with AEA RCT registry with ID-0000750. The project was shortlisted for the WAF Award 2015. IRB approval was obtained from Georgetown University Qatar and from Monash University. We thank the program teams of Joygopalpur Gram Vikash Kendra and Gana Unnayan Parshad for their collaboration during the implementation of the project and data collection—in particular, the project leaders, Sudipta Barman and Sudipta Moitra; project coordinators Anindya Basu and Ashis Saha; our research assistant, Parswati Dasgupta; and the entire field team. We also thank Debayan Pakrashi and his team. We are indebted to Agnese Carrera, Luca Coraggio, Giorgia Moretta, Valentina Pretolani, and Marie Rueskov Grif for outstanding research assistance and countless contributions during several phases of the project. We thank Gani Aldashev, Samson Alva, Tommaso Aquilante, Anik Ashraf, Jean Marie Baland, Sascha Becker, Stefan Bergheimer, Erwin Bulte, Janjala Chirakijja, Ilaria D'Angelis, Georg Kirchsteiger, Claudio Labanca, Gordon Leslie, Luca Livio, Pushkar Maitra, Giovanna Marcolongo, Andreas Menzel, Pauline Rossi, Paulo Santos, Priyanka Sarda, Glenn Sheriff, Lorenzo Trimarchi, Marcella Veronesi, the editor Rohini Pande, and four anonymous reviewers, as well as seminar participants at Boston College, Monash University, Paris School of Economics, Université libre de Bruxelles, University of Copenhagen, University of Nottingham, International University of Japan, University of Rochester, and Binghamton University. G.D. gratefully acknowledges financial support from MIUR under the PRIN grant 2015TPM9. Denni Tommasi gratefully acknowledges financial support from FNRS, DANIDA, and Monash University. We have no relevant or material financial interests that relate to the research described in this paper. All errors are our own.

A supplemental appendix is available online at https://doi.org/10.1162/rest_a_00913.

Supplementary data