The Price Effects of Cash Versus In-Kind Transfers

The Price Effects of Cash Versus In-Kind Transfers Abstract This article examines the effect of cash versus in-kind transfers on local prices. Both types of transfers increase the demand for normal goods; in-kind transfers also increase supply in recipient communities, which could lead to lower prices than under cash transfers. We test and confirm this prediction using a programme in Mexico that randomly assigned villages to receive boxes of food (trucked into the village), equivalently-valued cash transfers, or no transfers. We find that prices are significantly lower under in-kind transfers compared to cash transfers; relative to the control group, in-kind transfers cause a 4% fall in prices while cash transfers cause a positive but negligible increase in prices. In the more economically developed villages in the sample, households’ purchasing power is only modestly affected by these price effects. In the less developed villages, the price effects are much larger in magnitude, which we show is due to these villages being less tied to the outside economy and having less competition among local suppliers. 1. Introduction A central question in anti-poverty policy is whether transfers should be made in kind or as cash. The rationales for in-kind transfers include encouraging consumption of particular goods or inducing the less needy to self-select out of the programme (Nichols and Zeckhauser, 1982; Besley, 1988; Blackorby and Donaldson, 1988; Besley and Coate, 1991; Bearse et al., 2000). These potential benefits of in-kind transfers are weighed against the fact that cash transfers typically have lower administrative costs and give recipients greater freedom over their consumption. Another potentially important but less discussed aspect of this policy trade-off is the effect that in-kind and cash transfers have on local prices. Cash transfers increase the demand for normal goods, which will lead to price increases. This prediction holds either with perfect competition and marginal costs that are increasing in quantity, or with imperfect competition even if marginal costs are constant or decreasing (under certain assumptions about demand, as we discuss in detail later). In-kind transfers similarly increase demand through an income effect, but, in addition, if they increase local supply (e.g. the government trucks food aid into a village), then local prices should be lower under in-kind transfers, relative to cash transfers.1 From local suppliers’ viewpoint, an in-kind transfer consists of a negative shock to the residual demand they face because the transfer has met some of consumers’ demand, plus a positive demand shock due to consumers having higher income. The pecuniary effects could potentially be a useful policy lever, as noted by the previous literature.2 For example, the price declines caused by in-kind transfers could serve as a second-best way to tax producers and redistribute to consumers (Coate et al., 1994). Similarly, Coate (1989) discusses how price effects could make an in-kind food aid programme more effective than a cash programme, depending on the market structure. And even if the main rationale for in-kind transfers is paternalism or self-targeting and the pecuniary effects are an unintended consequence, they might significantly enhance or diminish the programme goal of assisting the poor.3 Note that under perfect competition, the price effects shift wealth between buyers and sellers, while with imperfect competition and prices above the first-best level, lower prices induced by in-kind transfers could represent an increase in efficiency, relative to cash transfers. This article tests for price effects of in-kind transfers versus cash transfers in rural Mexico and compares both to the status quo of no transfers. We study a large food assistance programme for poor households, the Programa de Apoyo Alimentario (PAL). When rolling out the programme, the government selected around 200 villages for a village-level randomized experiment. The poor in some of the villages received monthly in-kind transfers of packaged food (rice, vegetable oil, canned fish, etc.) that were trucked in by the government. The market value of the food transfer was about 200 pesos (20 US dollars) per household per month; most of the in-kind transfer was inframarginal to households’ consumption.4 In other villages, the poor households received monthly cash transfers of similar value to the in-kind transfer. A third set of villages served as a control group. The vast majority of households in the villages, 89% on average, were eligible for the programme. A comparison of the cash-transfer villages to the control villages provides an estimate of the price effect of cash transfers, which should be positive for normal goods since the income effect shifts the demand curve outward. The PAL in-kind transfer has a higher nominal value than the cash transfer (due to the idiosyncratic way that PAL administrators calculated the cost of the in-kind bundle). The in-kind bundle’s true value to recipients is, coincidentally, very similar to the cash transfer on average (Cunha, 2014). Therefore, the income effect in the in-kind villages should be similar to that in the cash villages, and a comparison of in-kind and cash villages isolates the supply effect of an in-kind transfer—the change in prices caused by the influx of goods into the local economy. This supply effect should cause a decline in prices. We use pre- and post-programme price data collected from households and food stores to test these predictions. We find no detectable increase in prices under cash transfers (though the point estimate suggests a small increase), while in-kind transfers cause prices of the transferred goods to fall by 3.7%. Across several specifications, we consistently find that providing transfers in kind rather than as cash causes prices to be lower by 3–4%. These effects are not limited to the short run; over the full range of programme duration in the data, from 8 to 22 months, the effects persist. Thus, the price effects do not appear to be undone by exit or entry of grocery shops in the village or other changes in market structure induced by the intervention, or alternatively, such adjustments take several years to materialize. Goods that are not part of the transfer programme are also subject to pecuniary effects. The supply influx from the in-kind transfer should lower demand and prices for food items that are substitutes of the in-kind items. Empirically, the price effects for these other goods are small. Therefore, all told, the price effects have only modest implications for many households’ purchasing power. This is noteworthy because programme eligibility is very high and the transfer is large relative to food expenditures, both of which result in a large aggregate shock to the local economy. This finding of, on average, small price effects suggests that for typical transfer programs, price effects may not be economically significant in many communities. The exception is the less developed villages in our sample, as proxied by low average income, small population, and physical remoteness. In fact, the average effects we find are driven almost entirely by the subsample of villages with below-median development.5 These villages could have larger price effects for at least two reasons. First, their goods markets could be less integrated with the regional or world economy, so local supply and demand determine prices. Second, there could be less competition among local suppliers (e.g. among grocery shops or distributors supplying those shops). We find evidence that both mechanisms help explain the result. Furthermore, surveys of store owners in a subsample of villages point to imperfect competition as a key feature of the market structure and an important factor in understanding the pronounced price effects in less developed villages. For the less developed villages, in-kind transfers cause prices of the transferred goods to fall by 5% relative to cash transfers. In addition, cash transfers lead to a 1.5% increase in overall food prices; this implies an elasticity of prices with respect to income of 0.15, as the cash transfers in less developed villages constitute a 10% increase in aggregate income, on average. Choosing in-kind rather than cash transfers generates extra indirect transfers to consumers in the form of lower prices worth about 14% of the direct transfer itself in less developed villages; these effects have the opposite implication for food-producing households in the recipient villages. We should note that our estimates of the programme’s total effects have wide confidence intervals, but they are nonetheless suggestive of quantitatively important price effects in poor communities. Mexico’s very poor villages have a similar level of development—income level and physical remoteness—as many villages in Africa, Asia, and Central America (World Bank, 1994). Our results suggest that transfer programs in ultra-poor communities in developing countries may have important pecuniary effects. Meanwhile, if the recipient community is well-connected with larger markets or has a competitive supply side, or in general is more developed, then pecuniary effects are likely to be small relative to the direct benefits of transfer programs. This article contributes to the literature on in-kind transfers, which has mostly focused on the consumption effects of in-kind transfers and on the political economy of transfer programs. (See Currie and Gahvari (2008) for a nice review of this literature.) Several studies have examined the consumption effects of the PAL programme in Mexico (Gonzalez-Cossio et al., 2006; Skoufias et al., 2008; Leroy et al., 2010; Cunha, 2014). They broadly find that cash and in-kind transfers lead to similar increases in total expenditures, although of different types of foods and non-foods. There is also extensive work on the consumption effects of other transfer programs, such as the U.S. Food Stamp programme (Moffitt, 1989; Hoynes and Schanzenbach, 2009). Other work examines whether in-kind transfers are effective at self-targeting (Reeder, 1985; Currie and Gruber, 1996; Jacoby, 1997). Another branch of the literature examines the political economy of in-kind programs, including their degree of voter support and how they affect producer rents (De Janvry et al., 1991; Jones, 1996). Fewer studies provide evidence on the question this article addresses, namely the price effects of in-kind transfers, and those that do often focus on voucher programs in which the government does not act as a supplier (Murray, 1999; Finkelstein, 2007; Hastings and Washington, 2010).6 Another related literature is on the international food aid and local prices, but few of the papers in this literature aim to establish causality; for example, Levinsohn and McMillan (2007) use estimates of the supply and demand elasticity of food from the literature to gauge the potential price effect of food aid, and Garg et al. (2013) examine food aid and prices, but emphasize that their estimates are correlations and not necessarily causal effects. Our article is also one of the first to measure the price effects of social programs. There is a vast literature that studies the direct effects of social programs, but fewer studies examine the indirect effects of programs and in particular their market-level price effects (Angelucci and De Giorgi, 2009; Lise et al., 2004; Kaboski and Townsend, 2011; Attanasio et al., 2012; Imbert and Papp, 2015; Muralidharan et al., 2017). Our finding that the pecuniary effects of social programs can be quite large in underdeveloped communities is relevant when thinking about the impacts of many other programs in developing countries.7 Finally, our findings also contribute to an active area of policy debate. One of the largest and most prominent in-kind programs worldwide, the World Food Programme, is increasingly shifting towards cash transfers, and in many developed and developing countries there is policy debate about providing a universal basic income (UBI) (World Food Programme, 2011). Meanwhile, other major programs are moving away from cash towards in-kind transfers. For example, in the U.S. much of the welfare support under the Temporary Assistance for Needy Families programme is now in the form of child care, job training, and other in-kind services (Pear, 2003). For policy makers choosing between cash and in-kind transfers, our work highlights that their choice could have non-trivial implications for local prices in markets with imperfect competition. Moreover, when local suppliers have market power, changes in local prices are not just pecuniary externalities, but have efficiency implications too. These lessons are relevant in developing countries where most of the poor live in rural villages. They may also be applicable in developed countries: High-poverty neighborhoods in the U.S. have high participation in transfer programs such as SNAP, and would experience large increases in average income through a UBI programme; meanwhile, they are often characterized as having few grocery stores and high food prices (Bell and Burlin, 1993; Talukdar, 2008). The remainder of the article is organized as follows. Section 2 lays out the theoretical predictions. Section 3 describes Mexico’s PAL programme, other aspects of the context, and the experimental design. Section 4 describes our empirical strategy and data. Section 5 presents the results, and Section 6 offers concluding remarks. 2. Conceptual Framework In this section, we lay out the predictions about how cash and in-kind transfers affect prices. We do not present a formal model but instead informally derive the predictions that we take to the data. In a small open economy, changes in the local demand or supply should have no effect on prices since supply is infinitely elastic with prices set at the world level. However, the rural villages that are our focus are more typically partially-closed economies in which prices depend on local conditions. In our empirical application, an economy is a Mexican village, and the main goods we examine are packaged foods. The local suppliers are shopkeepers in the village, and they procure their inventory from outside the village.8 We discuss, in turn, two possibilities: that the supply side has perfect or imperfect competition. In our empirical setting, imperfect competition appears to be the more relevant scenario. 2.1. Perfect competition If the local market is perfectly competitive, then if the supply curve is positively sloped—that is, with increasing marginal costs—shifts in the demand for a good will affect its price. For local suppliers in Mexican villages, high transportation costs to other markets is one potential reason for increasing marginal costs, at least in the short run; to meet higher demand, a shopkeeper in a remote village might need to travel to a neighbouring village to buy supply from a shop there. Figure 1A depicts the market for a normal good in a village. The demand curve represents the aggregate demand faced by local suppliers. The figure shows, first, the effect of a cash transfer: The demand curve shifts to the right via an income effect, and the equilibrium price, $$p$$, increases.9 Denoting the amount of money transferred in cash by $$X_{\rm Cash}$$, our first prediction is that a cash transfer will cause prices to rise: Figure 1 View largeDownload slide Effect of cash and in-kind transfers on prices in different competitive environments. (A) Perfect competition; (B) Imperfect competition. Figure 1 View largeDownload slide Effect of cash and in-kind transfers on prices in different competitive environments. (A) Perfect competition; (B) Imperfect competition.   \begin{equation} \frac{\partial p}{\partial X_{\rm Cash}}> 0. \label{result-cash}\end{equation} (1) In-kind transfers also generate an income effect, so demand will again shift to the right. We define the in-kind transfer amount $$X_{\rm InKind}$$ in terms of its equivalent cash value.10 Thus the demand shift caused by a transfer amount $$X$$ is by definition the same for either form of transfer. With an in-kind transfer, however, some of consumers’ demand is now provided to them for free by the government, so the residual demand facing local suppliers shifts to the left by the amount provided in kind. While the net price effect of an in-kind transfer relative to the original market equilibrium is, in general, theoretically ambiguous, one can sign the price effect of in-kind transfers relative to cash transfers.11 For transferred goods, the price should be lower under in-kind transfers:   \begin{equation} \frac{\partial p}{\partial X_{\rm InKind}} - \frac{\partial p}{\partial X_{\rm Cash}} < 0. \label{result-ik} \end{equation} (2) Empirically, we will be better positioned to test Prediction (2) than Prediction (1). To detect the effect of the supply influx, we can concentrate on the nine specific goods provided in kind in the Mexican transfer programme we study. In contrast, the increased demand due to income effects will be spread across several food and non-food items. The cash transfer programme we study placed no restriction on how recipients could use the money, and it led to a small amount of extra demand per good, spread across many goods (Cunha 2014).12$$^,$$13 2.2. Imperfect competition In the setting we study, the supply side consists of food shops in the village and the distributors who supply the shops, trucking in food from outside the village. There are neither many shops nor distributors serving the typical village, so the degree of competition may be limited. Predictions (1) and (2) can also hold in the case of imperfect competition. Importantly, in contrast to the case of competitive firms, under imperfect competition, transfer programs can have price effects even if marginal costs are constant. Figure 1B depicts, for simplicity, the case of constant marginal cost for a monopolist facing linear demand, but the same predictions of price effects hold more generally, as we discuss below. Consider a Cournot–Nash model with $$N$$ firms that have constant marginal cost $$c$$ and face linear demand $$p = d - Q,$$ where $$Q$$ indicates quantity and $$d$$ represents factors that shift demand. The equilibrium price is $$p = (d + Nc)/(N+1).$$ Suppose the transfer changes the amount demanded from the local firms by an amount $$\Delta d$$; $$\Delta d$$ is positive for a cash transfer and negative or less positive for an in-kind transfer. Then the change in price is given by $$\Delta p/p = \Delta d/(d + Nc),$$ which has the property that the higher $$N$$ is (more competition), the smaller the magnitude of the price effects. More generally, the price effects under imperfect competition depend on the shape of the demand curve. For example, if the programme causes a multiplicative shift in demand, then there would be no effect on prices in the standard Cournot model (Cowan, 2004). In other cases, an increase in demand can cause oligopolistic prices to fall; greater competition would still dampen the magnitude of the price effects. Appendix A presents a Cournot model with a generalized demand function and shows conditions under which an increase in demand leads to a higher price. A sufficient condition for Predictions 1 and 2 to hold is a downward-sloping demand curve where the transfers represent an additive shift in demand. The price effects then vary with the degree of competition as follows:   \begin{equation} \frac{\partial^2 p}{\partial N \partial X_{\rm Cash}} < 0, \label{N-cash} \end{equation} (3) and   \begin{equation} \frac{\partial}{\partial N}\left(\frac{\partial p}{\partial X_{\rm InKind}} - \frac{\partial p}{\partial X_{\rm Cash}} \right) > 0 \label{N-ik}. \end{equation} (4) The higher $$N$$ is (more competition), the smaller in magnitude the price effect of a demand shift. Note that price effects under perfect and imperfect competition have different efficiency implications. If lack of competition causes prices to be above their efficient level, then in-kind transfers can increase total surplus. Local suppliers’ strategic rationing of supply is partly undone by the government provision of goods. (Note, however, that these potential welfare gains could be undone by inefficiencies in how the government runs the transfer programme.) The discussion above takes the market structure as given. The programme could also affect how many stores stock a given product as well as entry and exit of stores and thus the degree of competition. For example, in response to a supply influx from the government, a shop might stop carrying a product or go out of business, reducing competition and causing prices to return to, or even exceed, the counterfactual price level without the programme. A positive demand shock (e.g. due to a cash transfer) could cause stores to open or more stores to stock a given good, increasing competition. The theoretical predictions are not clear-cut in many cases. For example, the in-kind programme also made villagers richer, so the net effect on store entry and exit or inventory decisions is ambiguous. In addition, the price effect of a store beginning to or ceasing to stock a product is not easy to predict because firms do not profit maximize separately for each product. Nonetheless, in general these responses on the supply side would cause price effects to be smaller. These changes would likely not occur immediately, but as they occur, the price effects would fade. Thus, we also examine whether the price effects dissipate over time. The above are the main testable implications we take to the data. We next describe the transfer programme we study and discuss some of the above assumptions in the context of this programme. 3. Description of the PAL Programme and Context 3.1. PAL programme and experiment We study the Programa de Apoyo Alimentario (PAL) in Mexico. Started in late 2003, PAL operates in about 5,000 very poor, rural villages throughout Mexico. Villages are eligible to receive PAL if they have fewer than 2,500 inhabitants, are highly marginalized as classified by the Census Bureau, and do not receive aid from either Liconsa, the Mexican subsidized milk programme, or Oportunidades, the conditional cash transfer programme. Therefore PAL villages are typically poorer and more rural than the widely-studied Progresa/Oportunidades villages.14 Households within programme villages are eligible to receive transfers if they are classified as poor by the national government. PAL provides a monthly in-kind allotment consisting of seven basic items (corn flour, rice, beans, pasta, biscuits (cookies), fortified powdered milk, and vegetable oil) and two to four supplementary items (including canned tuna fish, canned sardines, lentils, corn starch, chocolate powder, and packaged breakfast cereal). All of the items are common Mexican brands and are typically available in local food shops. The basic goods are dietary staples for poor households in Mexico. The supplementary goods are foods typically consumed by fewer households in a village or less frequently; one goal of the programme was to encourage households to add diversity to their diet and consume more of these supplementary goods.15 Most recipient households consumed a larger quantity of the in-kind items, particularly the basic goods, than was provided in the transfer. That is, absent the transfer, their monthly quantity consumed exceeded the PAL in-kind allotment. The fact that recipients made out-of-pocket purchases of these goods even when receiving the in-kind transfer means that they were affected by the price effects; otherwise, price effects would only be relevant for non-recipients. Figure 2 shows the net-of-transfer expenditures on PAL goods (calculated using post-intervention expenditure in the control group).16 The poorest quartile of households spends slightly more than the richest quartile on these items, and spends more as a proportion of total food expenditures. Most of the PAL items are staple goods, which explains why they comprise a larger share of food spending for the poor. Figure 2 View largeDownload slide Expenditure on PAL goods across households. Means by quartile of per capita expenditure (Q1 are the poorest, Q4 the richest). Figure 2 View largeDownload slide Expenditure on PAL goods across households. Means by quartile of per capita expenditure (Q1 are the poorest, Q4 the richest). PAL is administered by the public/private agency, Diconsa. The Diconsa agency also maintains subsidized grocery shops in some villages (38% of the villages in our sample), which are run by a resident of the village. The government provides suggested prices to Diconsa store operators; the Diconsa stores are not obliged to use the suggested prices, but they must maintain prices that are 3–7% lower than market prices. Thus, prices at Diconsa stores should be responsive to market conditions, but to a lesser degree than at fully private stores.17 The local supply side of the market is mostly composed of small private stores that stock food products, including the packaged foods that PAL provided, as well as sundry items. Small villages typically have one to six of these types of stores. Some households in the village also grow food which is substitutable with the PAL packaged foods. Concurrent with the national roll-out of the programme, 208 villages in southern Mexico were randomly selected for inclusion in an experiment.18 Each study village was then randomly assigned to an in-kind treatment arm, cash treatment arm, or the control group; the village-level randomization was not stratified on any characteristics. Eligible households in the in-kind villages received a monthly in-kind food transfer (50% of villages); those in the cash villages received a 150 peso per month cash transfer (25% of villages); and those in the control group villages received nothing (the remaining 25% of villages). About 89% of households in the in-kind and cash villages were eligible to receive transfers (and received them). Due to administrative capacity constraints, experimental villages were rolled into the programme over the course of 14 months, beginning in December of 2003. This gradual rollout creates variation in how long the programme had been running when endline data collection occurred in 2005. Of the 208 villages in the experiment, 14 are excluded from the analysis. Eight villages do not have follow-up price data; in two villages, the PAL programme began before the baseline survey; two villages are geographically contiguous and cannot be regarded as separate villages; and two villages were deemed ineligible for the experiment because they were receiving the conditional cash programme, Oportunidades, contrary to PAL regulations.19 Observable characteristics of the excluded villages are balanced across treatment arms. (Results available from the authors.) Of the remaining 194 villages, three received the wrong treatment (one in-kind village did not receive the programme, one cash village received both in-kind and cash transfers, and one control village received in-kind transfers). We include these villages and interpret our estimates as intent-to-treat estimates. The aggregate impact of the PAL programme on a recipient village was large, both because the eligibility rate was high and because the transfer per household was sizeable. The in-kind transfer represented 18% of a recipient household’s baseline food expenditures on average and 11% of total expenditures. Including the ineligible households, the injection of food into the village through the programme was equivalent to 16% of baseline aggregate food expenditures and 10% of total expenditures for the village. Similarly, the cash transfer represented an 8% increase in recipients’ income and, in aggregate, a 7% increase in total village income. In the in-kind experimental villages, the transfer comprised the seven basic items and three supplementary goods: lentils, breakfast cereal, and either canned tuna fish or canned sardines. There is some ambiguity about whether the in-kind villages always received these three supplementary items, so, in some of our analyses, we separate the basic PAL goods from the supplementary ones. Another reason to examine the basic goods separately is that they isolate the simple income and supply effects of in-kind transfers; if the government succeeded in increasing households’ taste for the supplementary goods, then the supplementary goods would have an additional effect of changing preferences (which goes in the direction of increasing demand and prices). The market for basic goods is also thicker, so the price effects might be easier to detect for the basic goods. Both the in-kind and cash transfers were, in practice, delivered bimonthly, two monthly allotments at a time per household. A woman (the household head or spouse of the head) was designated the beneficiary within the household, if possible. The transfer size was the same for every eligible household regardless of family size. Resale of in-kind food transfers was not prohibited, nor were there purchase requirements attached to the cash transfers. The monthly box of food had a market value of about 206 pesos in the programme villages, and the cash transfer was 150 pesos per month, based on the government’s wholesale cost of procuring the in-kind items.20 The items included in the in-kind transfer are not produced locally.21 Thus, the main welfare effects on the local supply side of the market will be felt by shopkeepers. There will also be welfare effects for local agricultural producers in cases where there is a high degree of substitutability (or complementarity) between the in-kind goods and the local products. An inconvenient feature of the programme for our purposes is that the cash villages and a randomly selected half of the in-kind villages were assigned to receive health, hygiene, and nutrition classes, as well. This programme feature could create two potential problems for the interpretation of our results. First, the difference between the price effects of cash and in-kind transfers, which we interpret as due to the injection of supply, could be partly driven by differential exposure to the classes. Second, the impact of cash transfers on prices could be partly driven by the classes, rather than being a pure income effect. These concerns appear to be small in practice. Regarding the first concern (in-kind versus cash), as documented in the Appendix, when we restrict the sample to in-kind villages assigned to receive classes—that is, if we analyse in-kind and cash villages that do not differ in their assignment to classes—the cash-versus-in-kind price effect is very similar to our main results that use all of the in-kind villages. This finding is not surprising given that classes were actually offered in almost all of the in-kind villages assigned not to receive them (Cunha, 2014).22 Thus, in practice, the cash and in-kind treatment arms were essentially identical vis-$$\grave{a}$$-vis classes, and it seems valid to interpret the in-kind versus cash comparison as due to the supply effect. For the second concern (cash versus control), there is no experimental variation to exploit, but when we compare class attendees to non-attendees in the cash arm, there is no evidence that the classes shifted food consumption, either overall or towards the PAL foods (as shown in the Appendix). This evidence makes us doubtful that the classes affected prices in the cash treatment arm, though attendance is endogenous so this evidence is only suggestive. Therefore, the caveat that the classes may have played some role in the price effect of cash transfers should be kept in mind when interpreting our cash versus control effect as a pure income effect. We abstract from this component of the programme for the remainder of our analysis. 3.2. Assumption of identical income effects for cash and in-kind transfers In Section 2, we expressed the size of the in-kind transfer $$X_{\rm InKind}$$ in terms of its cash equivalent to recipients. If one compares a cash transfer programme and an in-kind transfer programme, and the cash equivalent of the in-kind transfer is exactly the same amount as the cash transfer, then the income effect for both transfer programs is the same. Coincidentally, this is quite close to being the case in our empirical setting. The market value of the in-kind transfer in the recipient villages averaged 206 pesos (based on pre-programme prices). The in-kind bundle would have had a cash-equivalent value of 206 pesos if the transfer was inframarginal to consumption or resale was costless, that is, if the in-kind nature of the transfers did not distort recipients’ consumption choices. However, the transfers did alter consumption patterns, so the cash equivalent was less than the nominal value of 206 pesos. We estimate that recipients valued it at 146 pesos on average, or 71 cents on the dollar, as detailed in the next paragraph. The Mexican government made the (peculiar) decision to set the cash transfer in its randomized experiment equal to its wholesale cost of procuring the in-kind goods, which was about 27% lower than the cost at consumer prices in the recipient villages. The government also did not adjust for the fact that its estimated distribution cost was 30 pesos per in-kind box but 20 pesos per recipient for the cash transfer. The cash transfer was set at 150 pesos per month. There are three conceptually distinct ways that recipients use goods provided to them in kind. First, they consume some amount of it that they would have consumed anyway; they value this inframarginal portion at market prices. By comparing the control group’s consumption to transfer recipients’ consumption, Cunha (2014) estimates that 116 pesos worth of the 206-peso bundle falls in this category. Second, recipients consume an additional amount of the transferred foods, more than they would have consumed absent the in-kind transfer. PAL recipients consumed an estimated 35 pesos more of food in the transferred categories as a result of the in-kind transfer. Third, recipients received an additional 55 pesos worth of goods that they did not consume and presumably resold instead.23 For the latter two categories—the “extramarginal” portion—there is deadweight loss, and recipients will value the goods at less than their market value. For the extra goods they consume, they would not have been willing to purchase them at market prices, and for the goods they resell, they likely incur transaction costs. We assume, first, that consumers value the extramarginal consumption at a two-thirds discount relative to its market value, and second, that for goods that are resold, transaction costs erode two thirds of their value. Thus, the 90 pesos of extramarginal transfers are valued at only 30 pesos. Under these assumptions, the PAL in-kind transfer is worth 146 pesos to recipients (116 for the inframarginal portion + 30 for the extramarginal portion). To recap, while it is impossible to pinpoint the precise value of the in-kind transfer to recipients—its nominal value minus the deadweight loss relative to an unconstrained transfer—the value of the PAL in-kind transfer was likely quite similar to the value of the cash transfer to which we compare it (146 pesos versus 150 pesos).24 Moreover, even if consumers place zero value on the extramarginal portion of the in-kind transfer, valuing only the 116 pesos of inframarginal consumption, this difference in the income effect is much too small to explain the magnitude of the cash-versus-in-kind price effects that we estimate in Section 5, as we show in that section. It is also worth noting that flypaper effects could be especially strong when transfers are made in-kind: By giving households particular goods, the government might signal the high quality of these goods (e.g. their nutritional value) and also make these items more salient to households. In other words, with an in-kind transfer relative to a cash transfer, not just the supply but also the demand for the transferred goods might increase. This extra effect of in-kind transfers would counteract the supply effect, and our estimated price effects would give a lower bound for the pure supply-shift effect of in-kind transfers.25 3.3. Market structure As the data collected by the Mexican government for the PAL experiment did not include information on market structure, we conducted surveys of store owners in a subsample of 52 villages to qualitatively understand the market structure, stores’ cost curves, and their price-setting behaviour. (See Appendix B for further details on the data collection.) Several facts are worth highlighting. First, there are few food stores per village. The median number of stores in 2015 was 4, and while respondents could not reliably recall the number of stores at the time the PAL experiment began in 2003, they reported that the number of stores was lower at that time. Second, there are fewer stores in less economically developed villages. Third, marginal cost curves appear to be upward-sloping over the short run (e.g. 1 month), but flat over a longer duration. Store owners report that they meet unexpectedly high demand by travelling to a neighbouring village or town to buy goods, which is costly, but for a permanent demand shock, they readjust the amount they procure from their distributors on a regular basis. Finally, store owners report that they adjust their prices quickly in response to increases or decreases in demand, usually within a week. We interpret these facts as pointing to stores having market power and facing a flat marginal cost curve over the one- to two-year time horizon for which we test for price effects. 4. Empirical Strategy and Data 4.1. Empirical strategy Our analysis treats each village as a local economy and examines food prices as the outcome, using variation across villages in whether a village was randomly assigned to in-kind transfers, cash transfers, or no transfers. We begin by focusing on the food items included in the in-kind programme. Our first prediction is that prices will be higher in cash villages relative to control villages since a positive income shock shifts the demand curve out (under the assumption that the items are normal goods). The second prediction is that relative to cash villages, prices will be lower in in-kind villages because of the supply influx. Our main data consists of prices collected in experimental villages both pre- and post-programme. We estimate the following regression where the outcome variable is $$p_{gsv}$$, the price for good $$g$$ at store $$s$$ in village $$v$$:   \begin{equation}p_{gsv} =\alpha +\beta_1 {\rm InKind}_{v} +\beta_2 {\rm Cash}_{v}+ \phi p_{gv,t-1} + \sigma I_{gv}+\epsilon_{gsv}. \label{eqn-1} \end{equation} (5) Our two predictions correspond to $$\beta_2>0$$ (cash transfers increase prices), and $$\beta_1<\beta_2$$ (prices are lower under in-kind transfers than cash transfers). In our main specification, we control for the baseline price, denoted $$p_{gv,t-1}$$, which does not vary within a village (see below). (The subscript $$t-1$$ is shorthand for the variable being constructed from the baseline data; the estimation sample is cross-sectional, not a panel over time.) We also include the dummy variable $$I$$ to indicate whether the pre-programme price is imputed (again, see below). We cluster standard errors at the village level, the level at which the treatment was randomized. Note that a difference between the two predictions is that the first one—a positive price effect of cash transfers—applies to all normal goods, whereas the second one—a negative price effect of in-kind relative to cash transfers—applies to the goods provided in kind. We therefore have a more focused (and possibly higher-powered) way to test the second prediction, namely by examining the prices of PAL goods rather than all goods. 4.2. Data The data for our analysis come from surveys of stores and households conducted in the experimental villages by trained enumerators from the Mexican National Institute of Public Health both before and after the programme was introduced. Baseline data were collected in the final quarter of 2003 and the first quarter of 2004, before villagers knew they would be receiving the programme. Follow-up data were collected two years later in the final quarter of 2005, one to two years after PAL transfers began in these villages. The Mexican government’s purpose in running the experiment was to measure the programme’s impacts on food consumption, and what type of data they collected was determined accordingly. Our measure of post-programme prices comes from a survey of local food stores. From each store, enumerators collected prices for fixed quantities of sixty-six individual food items. They were instructed to first identify all the food stores in the village and then survey a maximum of three stores per village; unfortunately, no data were recorded from the step where they identified all of the stores. If more than three stores existed per village, they were instructed to randomly select three to survey, if possible one from each of three store types: general stores with posted prices, general stores without posted prices (e.g. small corner shops, butcher shop, or bakery), and the village market, taken as a unit. For 37% of villages in our sample, one store was surveyed; for 47% of villages, two stores were surveyed; and three stores were surveyed in the remaining 16% of villages.26 Some of the stores surveyed were part of the Diconsa agency (21%) while the majority were independent stores (79%). We also use measures of pre-programme food prices. Baseline data collection on store prices are missing for 40% of the sample because, first, data were collected for only forty of the 66 food items, and, second, even among the sampled goods, there are missing data for 19% of village-good observations (see Appendix B for details). Therefore, we also use the household survey to construct the pre-programme unit value (expenditure divided by quantity purchased) for each food item. In each village, a random sample of thirty-three households was interviewed about purchase quantities and expenditures on sixty food items. We use the median unit value among households in the village as a measure of the village’s pre-programme price.27 In cases where the pre-programme village median unit value is missing, we impute it using the median unit value in other villages within the same municipality (or within the same state in the few cases where there are no data for other villages in the municipality). Despite the missing data, we also use pre-programme store prices in some specifications to check the robustness of our results. The data do not allow us to match stores between waves; therefore, we use the median store price within a village and good as a measure of the pre-programme price. When the village median store price is missing, we impute the price using, first, the village median unit value, and then the geographic imputation of village median unit values (as above). To facilitate comparisons across goods with different price levels, we normalize the price for each good by the sample mean for the good within the control group, by survey wave. (If one good is ten times the price of another good, we would not expect the programme to have the same effect in levels for these two goods, but we would expect it to have the same proportional effect, all else equal.) The mean price for each good is thus roughly 1, and exactly 1 for the control group. The empirical results are nearly identical if we normalize by the mean value across all the villages, but using the control villages seems preferable so that the normalization factor is not affected by the treatments. We also show the results using the logarithm of the price as the outcome. We exclude some food items from the analysis due to missing data. Among the PAL goods, the store price survey mistakenly did not include biscuits; for the non-PAL items, chocolate powder, nixtamalized corn flour, salt, and non-fortified powdered milk were not included in the household survey and corn starch was not included in the store survey.28 Finally, two pairs of goods were asked about jointly in the household survey (beef/pork and canned fish) but separately in the store survey (beef, pork, canned tuna, canned sardines). To address this discrepancy, we use the aggregated categories and take the median across all observed store prices for either good as our post-programme price measure. Our final data set comprises six basic PAL goods (corn flour, rice, beans, pasta, oil, fortified milk), three supplementary PAL goods (canned fish, packaged breakfast cereal, and lentils), and fifty-one non-PAL goods. Appendix Table A2 lists all of the goods in our analysis. Table 1 presents descriptive statistics for the PAL goods. Column 2 shows the quantity per good of the monthly household transfer, and column 3 shows its monetary value measured using our pre-programme measure of prices. Column 4 presents each good’s share of the total calories in the transfer bundle. As can be seen, the supplementary items were transferred in smaller amounts with lower value and fewer calories than the basic goods. Table 1 Summary of PAL food box    Type  Amount per box (kg)  Value per box (pre-programme, in pesos)  Calories, as % of total box  Village change in supply ($$\Delta$$Supply)  Item  (1)  (2)  (3)  (4)  (5)  Corn flour  Basic  3  15.7  20  1.00  Rice  Basic  2  12.7  12  0.61  Beans  Basic  2  21.0  13  0.29  Fortified powdered milk  Basic  1.92  76.2  17  8.62  Packaged pasta soup  Basic  1.2  16.2  8  0.93  Vegetable oil  Basic  1 (lt)  10.4  16  0.25  Biscuits  Basic  1  18.7  8  0.81  Lentils  Supplementary  1  10.3  2  3.73  Canned tuna/sardines  Supplementary  0.6  14.8  2  1.55  Breakfast cereal  Supplementary  0.2  9.3  1  0.90     Type  Amount per box (kg)  Value per box (pre-programme, in pesos)  Calories, as % of total box  Village change in supply ($$\Delta$$Supply)  Item  (1)  (2)  (3)  (4)  (5)  Corn flour  Basic  3  15.7  20  1.00  Rice  Basic  2  12.7  12  0.61  Beans  Basic  2  21.0  13  0.29  Fortified powdered milk  Basic  1.92  76.2  17  8.62  Packaged pasta soup  Basic  1.2  16.2  8  0.93  Vegetable oil  Basic  1 (lt)  10.4  16  0.25  Biscuits  Basic  1  18.7  8  0.81  Lentils  Supplementary  1  10.3  2  3.73  Canned tuna/sardines  Supplementary  0.6  14.8  2  1.55  Breakfast cereal  Supplementary  0.2  9.3  1  0.90  Notes: (1) Value is calculated using the average of pretreatment village-level median unit values. 10 pesos $$\approx$$ 1 USD. (2) $$\Delta$$Supply measures the PAL supply influx into villages, relative to what would have been consumed absent the programme. It is constructed as the average across all in-kind villages of the total amount of the good transferred to the village divided by the average consumption of the good in control villages in the post-period. (3) We do not know whether a household received canned tuna fish (0.35 kg) or canned sardines (0.8 kg); the analysis assumes the mean weight and calories throughout. (4) Biscuits are excluded from our analysis as post-programme prices are missing. Table 1 Summary of PAL food box    Type  Amount per box (kg)  Value per box (pre-programme, in pesos)  Calories, as % of total box  Village change in supply ($$\Delta$$Supply)  Item  (1)  (2)  (3)  (4)  (5)  Corn flour  Basic  3  15.7  20  1.00  Rice  Basic  2  12.7  12  0.61  Beans  Basic  2  21.0  13  0.29  Fortified powdered milk  Basic  1.92  76.2  17  8.62  Packaged pasta soup  Basic  1.2  16.2  8  0.93  Vegetable oil  Basic  1 (lt)  10.4  16  0.25  Biscuits  Basic  1  18.7  8  0.81  Lentils  Supplementary  1  10.3  2  3.73  Canned tuna/sardines  Supplementary  0.6  14.8  2  1.55  Breakfast cereal  Supplementary  0.2  9.3  1  0.90     Type  Amount per box (kg)  Value per box (pre-programme, in pesos)  Calories, as % of total box  Village change in supply ($$\Delta$$Supply)  Item  (1)  (2)  (3)  (4)  (5)  Corn flour  Basic  3  15.7  20  1.00  Rice  Basic  2  12.7  12  0.61  Beans  Basic  2  21.0  13  0.29  Fortified powdered milk  Basic  1.92  76.2  17  8.62  Packaged pasta soup  Basic  1.2  16.2  8  0.93  Vegetable oil  Basic  1 (lt)  10.4  16  0.25  Biscuits  Basic  1  18.7  8  0.81  Lentils  Supplementary  1  10.3  2  3.73  Canned tuna/sardines  Supplementary  0.6  14.8  2  1.55  Breakfast cereal  Supplementary  0.2  9.3  1  0.90  Notes: (1) Value is calculated using the average of pretreatment village-level median unit values. 10 pesos $$\approx$$ 1 USD. (2) $$\Delta$$Supply measures the PAL supply influx into villages, relative to what would have been consumed absent the programme. It is constructed as the average across all in-kind villages of the total amount of the good transferred to the village divided by the average consumption of the good in control villages in the post-period. (3) We do not know whether a household received canned tuna fish (0.35 kg) or canned sardines (0.8 kg); the analysis assumes the mean weight and calories throughout. (4) Biscuits are excluded from our analysis as post-programme prices are missing. There is considerable variation across the PAL goods in the size of the aggregate village-level transfer. One measure of the size of this supply shift is listed in column 5. Here, the village change in supply, $$\Delta {\rm Supply}$$, is constructed as the average across in-kind villages of the total amount of a good transferred to the village (i.e. average number of eligible households per village times allotment per household) divided by the average consumption of the good in control villages in the post-programme period. For example, there was almost exactly as much corn flour delivered to the villages each month as would have been consumed absent the programme ($$\Delta {\rm Supply} = 1.00$$ for corn flour), while the allotment of beans was 29% of what would have been consumed absent the programme ($$\Delta {\rm Supply} =0.29$$ for beans). Our final data set contains 360 stores in 194 villages and 12,940 good-village-store observations. The number of goods varies by store since many stores sell only a subset of goods. Table 2 presents summary statistics by treatment group. The baseline characteristics are for the most part balanced across groups. For three variables, there are significant differences across groups at the five percent level: The presence of a Diconsa store differs between control and in-kind, the share of food-producing households differs between control and cash and between in-kind and cash, and farm costs differ between control and in-kind and between control and cash. For our primary comparison—between the cash and in-kind treatments—no variable is unbalanced at baseline at the 5% level and only one variable is unbalanced at the 10% level.29 Table 2 Baseline characteristics by treatment group    Control  In-kind  Cash  $$(1)=(2)$$$$p$$-value  $$(1)=(3)$$$$p$$-value  $$(2)=(3)$$$$p$$-value  $$(1)=(2)=(3)$$$$p$$-value     (1)  (2)  (3)  Prices, basic PAL goods     Median village unit-value, normalized  1.00  0.98  0.98  0.28  0.31  0.95  0.48     (0.014)  (0.012)  (0.015)                 Missing median village unit-value  0.13  0.14  0.13  0.94  0.80  0.72  0.94     (0.018)  (0.013)  (0.020)                 Observations (good level)  486  1,092  582              Prices, all PAL goods     Median village unit-value, normalized  1.00  1.02  1.00  0.39  0.88  0.46  0.64     (0.017)  (0.016)  (0.016)                 Missing village unit-value  0.18  0.17  0.16  0.72  0.48  0.64  0.77     (0.016)  (0.013)  (0.021)                 Observations (good level)  729  1,638  873              Prices, all goods     Median village unit-value, normalized  1.00  1.02  1.00  0.23  0.98  0.18  0.30     (0.015)  (0.010)  (0.013)                 Missing village unit-value  0.23  0.23  0.23  0.84  0.99  0.84  0.97     (0.017)  (0.012)  (0.016)                 Observations (good level)  4,860  10,920  5,820              Village level characteristics     Missing median store price  0.13  0.10  0.16  0.69  0.66  0.36  0.65     (0.048)  (0.034)  (0.046)                 Diconsa store in the village  0.26  0.45  0.39  0.03**  0.16  0.51  0.08*     (0.71)  (0.049)  (0.068)                 Travel time to nearest market (hours)  0.77  0.69  0.74  0.55  0.86  0.69  0.82     (0.108)  (0.076)  (0.104)                 Village population  682.83  580.39  543.90  0.29  0.21  0.70  0.42     (79.65)  (55.14)  (75.65)                 Number of stores  1.70  1.82  1.8  0.33  0.47  0.88  0.62     (0.102)  (0.072)  (0.098)                 Median months for which  –  13.21  12.96  –  –  0.52  –     transfers were received     (0.224)  (0.305)                 Observations (village level)  47  96  51              Household level characteristics     Monthly per capita expenditure (pesos)  570.54  535.10  529.54  0.31  0.26  0.85  0.50     (29.02)  (18.90)  (21.77)                 Food-producing household  0.68  0.75  0.82  0.11  0.00***  0.05*  0.01***     (0.04)  (0.02)  (0.03)                 Farm costs (pesos)  413.76  664.92  784.65  0.03**  0.00***  0.32  0.01***     (82.46)  (76.91)  (93.22)                 Farm profits (pesos)  211.72  319.13  289.61  0.24  0.38  0.70  0.50     (72.52)  (56.80)  (52.08)                 Asset index  2.24  2.18  2.27  0.78  0.87  0.59  0.86     (0.16)  (0.10)  (0.13)                 Indigenous household  0.21  0.18  0.15  0.66  0.39  0.56  0.68     (0.06)  (0.03)  (0.04)                 Household has a dirt floor  0.32  0.31  0.32  0.77  0.95  0.70  0.92     (0.04)  (0.03)  (0.03)                 Household has piped water  0.65  0.57  0.50  0.23  0.06*  0.33  0.16     (0.05)  (0.04)  (0.06)                 Observations (household level)  1291  2810  1473                 Control  In-kind  Cash  $$(1)=(2)$$$$p$$-value  $$(1)=(3)$$$$p$$-value  $$(2)=(3)$$$$p$$-value  $$(1)=(2)=(3)$$$$p$$-value     (1)  (2)  (3)  Prices, basic PAL goods     Median village unit-value, normalized  1.00  0.98  0.98  0.28  0.31  0.95  0.48     (0.014)  (0.012)  (0.015)                 Missing median village unit-value  0.13  0.14  0.13  0.94  0.80  0.72  0.94     (0.018)  (0.013)  (0.020)                 Observations (good level)  486  1,092  582              Prices, all PAL goods     Median village unit-value, normalized  1.00  1.02  1.00  0.39  0.88  0.46  0.64     (0.017)  (0.016)  (0.016)                 Missing village unit-value  0.18  0.17  0.16  0.72  0.48  0.64  0.77     (0.016)  (0.013)  (0.021)                 Observations (good level)  729  1,638  873              Prices, all goods     Median village unit-value, normalized  1.00  1.02  1.00  0.23  0.98  0.18  0.30     (0.015)  (0.010)  (0.013)                 Missing village unit-value  0.23  0.23  0.23  0.84  0.99  0.84  0.97     (0.017)  (0.012)  (0.016)                 Observations (good level)  4,860  10,920  5,820              Village level characteristics     Missing median store price  0.13  0.10  0.16  0.69  0.66  0.36  0.65     (0.048)  (0.034)  (0.046)                 Diconsa store in the village  0.26  0.45  0.39  0.03**  0.16  0.51  0.08*     (0.71)  (0.049)  (0.068)                 Travel time to nearest market (hours)  0.77  0.69  0.74  0.55  0.86  0.69  0.82     (0.108)  (0.076)  (0.104)                 Village population  682.83  580.39  543.90  0.29  0.21  0.70  0.42     (79.65)  (55.14)  (75.65)                 Number of stores  1.70  1.82  1.8  0.33  0.47  0.88  0.62     (0.102)  (0.072)  (0.098)                 Median months for which  –  13.21  12.96  –  –  0.52  –     transfers were received     (0.224)  (0.305)                 Observations (village level)  47  96  51              Household level characteristics     Monthly per capita expenditure (pesos)  570.54  535.10  529.54  0.31  0.26  0.85  0.50     (29.02)  (18.90)  (21.77)                 Food-producing household  0.68  0.75  0.82  0.11  0.00***  0.05*  0.01***     (0.04)  (0.02)  (0.03)                 Farm costs (pesos)  413.76  664.92  784.65  0.03**  0.00***  0.32  0.01***     (82.46)  (76.91)  (93.22)                 Farm profits (pesos)  211.72  319.13  289.61  0.24  0.38  0.70  0.50     (72.52)  (56.80)  (52.08)                 Asset index  2.24  2.18  2.27  0.78  0.87  0.59  0.86     (0.16)  (0.10)  (0.13)                 Indigenous household  0.21  0.18  0.15  0.66  0.39  0.56  0.68     (0.06)  (0.03)  (0.04)                 Household has a dirt floor  0.32  0.31  0.32  0.77  0.95  0.70  0.92     (0.04)  (0.03)  (0.03)                 Household has piped water  0.65  0.57  0.50  0.23  0.06*  0.33  0.16     (0.05)  (0.04)  (0.06)                 Observations (household level)  1291  2810  1473              Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) Standard errors in parentheses. For normalized median village unit values and household level characteristics, standard errors are clustered at the village level. (2) Median village unit values are normalized with the good-specific control group mean and are imputed geographically if missing (see text). (3) Travel time to the nearest market is the time in hours needed to travel to a larger market that sells fruit, vegetables, and meat. It is constructed as the village median of household self-reports. (4) Expenditure is the value of non-durable items (food and non-food) consumed in the preceding month, measured in pesos; six households are missing expenditure data. (5) Food producing households are those that, at baseline, either auto-consume their production or report planting or reaping produce or grain or raising animals. (6) Farm costs and profits are for the preceding year. Samples are trimmed of outliers greater than 3 SDs above the median (about 1% of observations). (7) The asset index is the sum of binary indicators for whether the household owns the following goods: radio or TV, refrigerator, gas stove, washing machine, VCR, and car or motorcycle; two households are missing the asset index. (8) A household is defined as indigenous if one or more members speak an indigenous language. Table 2 Baseline characteristics by treatment group    Control  In-kind  Cash  $$(1)=(2)$$$$p$$-value  $$(1)=(3)$$$$p$$-value  $$(2)=(3)$$$$p$$-value  $$(1)=(2)=(3)$$$$p$$-value     (1)  (2)  (3)  Prices, basic PAL goods     Median village unit-value, normalized  1.00  0.98  0.98  0.28  0.31  0.95  0.48     (0.014)  (0.012)  (0.015)                 Missing median village unit-value  0.13  0.14  0.13  0.94  0.80  0.72  0.94     (0.018)  (0.013)  (0.020)                 Observations (good level)  486  1,092  582              Prices, all PAL goods     Median village unit-value, normalized  1.00  1.02  1.00  0.39  0.88  0.46  0.64     (0.017)  (0.016)  (0.016)                 Missing village unit-value  0.18  0.17  0.16  0.72  0.48  0.64  0.77     (0.016)  (0.013)  (0.021)                 Observations (good level)  729  1,638  873              Prices, all goods     Median village unit-value, normalized  1.00  1.02  1.00  0.23  0.98  0.18  0.30     (0.015)  (0.010)  (0.013)                 Missing village unit-value  0.23  0.23  0.23  0.84  0.99  0.84  0.97     (0.017)  (0.012)  (0.016)                 Observations (good level)  4,860  10,920  5,820              Village level characteristics     Missing median store price  0.13  0.10  0.16  0.69  0.66  0.36  0.65     (0.048)  (0.034)  (0.046)                 Diconsa store in the village  0.26  0.45  0.39  0.03**  0.16  0.51  0.08*     (0.71)  (0.049)  (0.068)                 Travel time to nearest market (hours)  0.77  0.69  0.74  0.55  0.86  0.69  0.82     (0.108)  (0.076)  (0.104)                 Village population  682.83  580.39  543.90  0.29  0.21  0.70  0.42     (79.65)  (55.14)  (75.65)                 Number of stores  1.70  1.82  1.8  0.33  0.47  0.88  0.62     (0.102)  (0.072)  (0.098)                 Median months for which  –  13.21  12.96  –  –  0.52  –     transfers were received     (0.224)  (0.305)                 Observations (village level)  47  96  51              Household level characteristics     Monthly per capita expenditure (pesos)  570.54  535.10  529.54  0.31  0.26  0.85  0.50     (29.02)  (18.90)  (21.77)                 Food-producing household  0.68  0.75  0.82  0.11  0.00***  0.05*  0.01***     (0.04)  (0.02)  (0.03)                 Farm costs (pesos)  413.76  664.92  784.65  0.03**  0.00***  0.32  0.01***     (82.46)  (76.91)  (93.22)                 Farm profits (pesos)  211.72  319.13  289.61  0.24  0.38  0.70  0.50     (72.52)  (56.80)  (52.08)                 Asset index  2.24  2.18  2.27  0.78  0.87  0.59  0.86     (0.16)  (0.10)  (0.13)                 Indigenous household  0.21  0.18  0.15  0.66  0.39  0.56  0.68     (0.06)  (0.03)  (0.04)                 Household has a dirt floor  0.32  0.31  0.32  0.77  0.95  0.70  0.92     (0.04)  (0.03)  (0.03)                 Household has piped water  0.65  0.57  0.50  0.23  0.06*  0.33  0.16     (0.05)  (0.04)  (0.06)                 Observations (household level)  1291  2810  1473                 Control  In-kind  Cash  $$(1)=(2)$$$$p$$-value  $$(1)=(3)$$$$p$$-value  $$(2)=(3)$$$$p$$-value  $$(1)=(2)=(3)$$$$p$$-value     (1)  (2)  (3)  Prices, basic PAL goods     Median village unit-value, normalized  1.00  0.98  0.98  0.28  0.31  0.95  0.48     (0.014)  (0.012)  (0.015)                 Missing median village unit-value  0.13  0.14  0.13  0.94  0.80  0.72  0.94     (0.018)  (0.013)  (0.020)                 Observations (good level)  486  1,092  582              Prices, all PAL goods     Median village unit-value, normalized  1.00  1.02  1.00  0.39  0.88  0.46  0.64     (0.017)  (0.016)  (0.016)                 Missing village unit-value  0.18  0.17  0.16  0.72  0.48  0.64  0.77     (0.016)  (0.013)  (0.021)                 Observations (good level)  729  1,638  873              Prices, all goods     Median village unit-value, normalized  1.00  1.02  1.00  0.23  0.98  0.18  0.30     (0.015)  (0.010)  (0.013)                 Missing village unit-value  0.23  0.23  0.23  0.84  0.99  0.84  0.97     (0.017)  (0.012)  (0.016)                 Observations (good level)  4,860  10,920  5,820              Village level characteristics     Missing median store price  0.13  0.10  0.16  0.69  0.66  0.36  0.65     (0.048)  (0.034)  (0.046)                 Diconsa store in the village  0.26  0.45  0.39  0.03**  0.16  0.51  0.08*     (0.71)  (0.049)  (0.068)                 Travel time to nearest market (hours)  0.77  0.69  0.74  0.55  0.86  0.69  0.82     (0.108)  (0.076)  (0.104)                 Village population  682.83  580.39  543.90  0.29  0.21  0.70  0.42     (79.65)  (55.14)  (75.65)                 Number of stores  1.70  1.82  1.8  0.33  0.47  0.88  0.62     (0.102)  (0.072)  (0.098)                 Median months for which  –  13.21  12.96  –  –  0.52  –     transfers were received     (0.224)  (0.305)                 Observations (village level)  47  96  51              Household level characteristics     Monthly per capita expenditure (pesos)  570.54  535.10  529.54  0.31  0.26  0.85  0.50     (29.02)  (18.90)  (21.77)                 Food-producing household  0.68  0.75  0.82  0.11  0.00***  0.05*  0.01***     (0.04)  (0.02)  (0.03)                 Farm costs (pesos)  413.76  664.92  784.65  0.03**  0.00***  0.32  0.01***     (82.46)  (76.91)  (93.22)                 Farm profits (pesos)  211.72  319.13  289.61  0.24  0.38  0.70  0.50     (72.52)  (56.80)  (52.08)                 Asset index  2.24  2.18  2.27  0.78  0.87  0.59  0.86     (0.16)  (0.10)  (0.13)                 Indigenous household  0.21  0.18  0.15  0.66  0.39  0.56  0.68     (0.06)  (0.03)  (0.04)                 Household has a dirt floor  0.32  0.31  0.32  0.77  0.95  0.70  0.92     (0.04)  (0.03)  (0.03)                 Household has piped water  0.65  0.57  0.50  0.23  0.06*  0.33  0.16     (0.05)  (0.04)  (0.06)                 Observations (household level)  1291  2810  1473              Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) Standard errors in parentheses. For normalized median village unit values and household level characteristics, standard errors are clustered at the village level. (2) Median village unit values are normalized with the good-specific control group mean and are imputed geographically if missing (see text). (3) Travel time to the nearest market is the time in hours needed to travel to a larger market that sells fruit, vegetables, and meat. It is constructed as the village median of household self-reports. (4) Expenditure is the value of non-durable items (food and non-food) consumed in the preceding month, measured in pesos; six households are missing expenditure data. (5) Food producing households are those that, at baseline, either auto-consume their production or report planting or reaping produce or grain or raising animals. (6) Farm costs and profits are for the preceding year. Samples are trimmed of outliers greater than 3 SDs above the median (about 1% of observations). (7) The asset index is the sum of binary indicators for whether the household owns the following goods: radio or TV, refrigerator, gas stove, washing machine, VCR, and car or motorcycle; two households are missing the asset index. (8) A household is defined as indigenous if one or more members speak an indigenous language. In some of our auxiliary analyses, we use household-level data to either construct village-level variables or to estimate household-level regressions. For example, we calculate the median household expenditures per capita in a village at baseline as a measure of the income level in the village. Also, when we test for heterogeneous welfare effects for households that produce agricultural goods, we use household-level outcomes such as farm profits and expenditures per capita. We present more detail on other relevant data as we introduce each analysis in the next section. Note that the data collection was designed to measure the PAL programme’s impact on food consumption, not its price effects. It is fortunate that the price data from stores were collected, enabling our analysis of the programme’s price effects. However, other data that ideally we would have are unavailable, for example, a census of grocery shops in each village. Thus, we do not have data on market structure to include in the empirical analysis. (Our survey of store owners in a subset of the villages, described in Section 3.3, provides a qualitative understanding of the typical market structure in the study villages.) 5. Results 5.1. Price effects of in-kind transfers and cash transfers Table 3, column 1, presents the main specification (equation (5)) using all nine PAL goods. The regression pools the effects for the different PAL food items. (See Appendix Table A4 for the results separately for each PAL good.) For cash villages, the point estimate suggests that the transfer programme caused prices to increase by 0.2% ($$\widehat{\beta_2}$$), though the coefficient is not statistically significant. In in-kind villages, prices fell by 3.9% relative to the cash villages ($$\widehat{\beta_1}-\widehat{\beta_2}$$), with a $$p$$-value of 0.02; the bottom of the table reports the difference between the in-kind and cash coefficients and the statistical significance of this difference. As mentioned above, theory is ambiguous about whether the supply or demand effect is bigger in magnitude, but unless a good has a particularly high income elasticity of demand, we would expect the supply effect to dominate. Empirically we indeed find that the net effect of the in-kind transfer on prices is negative (3.7% decline, significant at the 10% level). Table 3 Price effects of in-kind and cash transfers    All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price  $$\Delta$$price  $$\Delta$$price    (1)  (2)  (3)  (4)  (5)  (6)  In-kind  –0.037*  –0.033  –0.036*  –0.033  –0.062**  –0.025     (0.020)  (0.020)  (0.020)  (0.020)  (0.029)  (0.024)  Cash  0.002  0.014  0.003  0.012  0.000  0.039     (0.023)  (0.027)  (0.023)  (0.026)  (0.031)  (0.029)  Lagged normalized unit value  0.027  0.127***                 (0.021)  (0.042)              Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.039**  –0.047**  –0.038**  –0.045**  –0.063**  –0.064**  H0: In-kind = Cash (p-value)  0.02  0.04  0.03  0.04  0.02  0.02     All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price  $$\Delta$$price  $$\Delta$$price    (1)  (2)  (3)  (4)  (5)  (6)  In-kind  –0.037*  –0.033  –0.036*  –0.033  –0.062**  –0.025     (0.020)  (0.020)  (0.020)  (0.020)  (0.029)  (0.024)  Cash  0.002  0.014  0.003  0.012  0.000  0.039     (0.023)  (0.027)  (0.023)  (0.026)  (0.031)  (0.029)  Lagged normalized unit value  0.027  0.127***                 (0.021)  (0.042)              Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.039**  –0.047**  –0.038**  –0.045**  –0.063**  –0.064**  H0: In-kind = Cash (p-value)  0.02  0.04  0.03  0.04  0.02  0.02  Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable in columns 1–4 is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. (2) Lagged normalized unit value in columns 1–2 is the village median unit-value, imputed geographically if missing (see text), normalized using the good-specific control group mean; it varies at the village-good level. (3) Columns 3-4 do not control for the lagged normalized unit value. (4) The outcome variable in columns 5–6 is the difference between the normalized post-programme price (the outcome in columns 1–4) and the lagged normalized unit value (the baseline price measure in columns 1–2). (5) Regressions in columns 1–2 and 5–6 include an indicator for imputed pre-programme prices (see text). (6) Standard errors (in parentheses) are clustered at the village level. Table 3 Price effects of in-kind and cash transfers    All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price  $$\Delta$$price  $$\Delta$$price    (1)  (2)  (3)  (4)  (5)  (6)  In-kind  –0.037*  –0.033  –0.036*  –0.033  –0.062**  –0.025     (0.020)  (0.020)  (0.020)  (0.020)  (0.029)  (0.024)  Cash  0.002  0.014  0.003  0.012  0.000  0.039     (0.023)  (0.027)  (0.023)  (0.026)  (0.031)  (0.029)  Lagged normalized unit value  0.027  0.127***                 (0.021)  (0.042)              Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.039**  –0.047**  –0.038**  –0.045**  –0.063**  –0.064**  H0: In-kind = Cash (p-value)  0.02  0.04  0.03  0.04  0.02  0.02     All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price  $$\Delta$$price  $$\Delta$$price    (1)  (2)  (3)  (4)  (5)  (6)  In-kind  –0.037*  –0.033  –0.036*  –0.033  –0.062**  –0.025     (0.020)  (0.020)  (0.020)  (0.020)  (0.029)  (0.024)  Cash  0.002  0.014  0.003  0.012  0.000  0.039     (0.023)  (0.027)  (0.023)  (0.026)  (0.031)  (0.029)  Lagged normalized unit value  0.027  0.127***                 (0.021)  (0.042)              Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.039**  –0.047**  –0.038**  –0.045**  –0.063**  –0.064**  H0: In-kind = Cash (p-value)  0.02  0.04  0.03  0.04  0.02  0.02  Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable in columns 1–4 is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. (2) Lagged normalized unit value in columns 1–2 is the village median unit-value, imputed geographically if missing (see text), normalized using the good-specific control group mean; it varies at the village-good level. (3) Columns 3-4 do not control for the lagged normalized unit value. (4) The outcome variable in columns 5–6 is the difference between the normalized post-programme price (the outcome in columns 1–4) and the lagged normalized unit value (the baseline price measure in columns 1–2). (5) Regressions in columns 1–2 and 5–6 include an indicator for imputed pre-programme prices (see text). (6) Standard errors (in parentheses) are clustered at the village level. The in-kind-versus-cash difference is much too large to be due to just the income effect differing between the two types of transfer programs. As discussed in Section 2, recipients valued the in-kind bundle at roughly 146 pesos which is similar to the cash transfer amount of 150 pesos. The coefficient on $$Cash$$ of 0.002 is the effect of a 150 peso income transfer, suggesting that the 4 peso difference would generate an in-kind-versus-cash difference in the income effect on the order of $$-$$0.00005. Even if recipients only valued the in-kind goods that were purely inframarginal to their consumption, which account for 116 pesos of the bundle, and they placed zero value on the rest of the food transfer, the resulting 34 peso difference in the value of the in-kind and cash transfer would only lead to a coefficient difference of $$-$$0.00045, which is smaller by a factor of 80 than the actual difference of $$-$$0.039. Thus, the fact that prices are lower under in-kind transfers compared to cash transfers appears to be driven by the supply influx into the village, not by differing income effects. In column 2, we estimate the model excluding the supplementary PAL goods. The fact that canned fish, cereal, and lentils may not have been the supplementary goods in some experimental villages should not affect the cash or control villages but might attenuate our estimates of the in-kind-versus-cash effect. In addition, there is low consumption at baseline for the supplementary goods, and for very thin markets, prices are noisier. We find an in-kind-versus-cash coefficient difference that is somewhat larger in magnitude when we exclude the supplementary goods (magnitude of $$-$$0.047 with a $$p$$-value of 0.04). The remaining columns of Table 3 test the same predictions while varying the specification. In cases such as ours where the outcome variable is autocorrelated but noisy, controlling for the baseline outcome is more efficient than either using only post-programme data or using a difference-in-differences estimator, but we also show the results using these two alternatives (McKenzie, 2012). Columns 3 and 4 do not control for baseline prices, and columns 5 and 6 present the difference-in-differences estimates. 5.2. Robustness checks The results are also robust to using several other specifications, as shown in Appendix Table A5. First, we show that the results are nearly identical when we include good fixed effects. Second, rather than controlling for baseline unit values, we control for baseline store prices, imputing them for the 40% of cases where they are missing.30 The results are again very similar to the main specification. Third, we show the results using the log of (unnormalized) prices rather than the normalized price level. While the predictions are in terms of price levels rather than the log of prices, this robustness check is helpful to ensure that the results are not driven by outliers. The in-kind versus cash effect is slightly larger in magnitude in this specification and, again, significant at the 5% level. Fourth, we show that regressions that weight each observation by the expenditure share for the good (as observed in the control group post-programme) produce almost identical results. Fifth, we show that the results are similar when we drop half of the in-kind villages and focus on the cash and in-kind villages assigned to receive health and nutrition classes. Finally, we show that the results are robust to restricting the sample to privately-owned stores.31 In addition, the results are remarkably similar if we aggregate the data to the village-good or village level, estimating the model with one observation per village-good or per village (results available from the authors). We also investigate the potential concern that the effects we estimate reflect changes in quality within a product category—stores might have started stocking higher quality vegetable oil, for example—rather than changes in prices. Note, however, that if households upgrade quality when their income increases, this effect should apply to recipients of both cash and in-kind transfers. Nonetheless, in Table 4, we explore this concern by using proxies for the amount of quality variation there is for a good. First, we subjectively categorize the goods as having a high or low degree of product variation (each of the three authors independently categorized the goods, and we use the median of our answers). We categorized cereal, beans, corn flour, lentils, and pasta soup as having high quality variation, and vegetable oil, rice, canned fish, and powdered milk as having low variation. We run an interacted model, testing whether the price effects are driven by goods with more scope for quality upgrading (or downgrading). If quality were the explanation, the effects would be driven by the high-quality-variation goods. As seen in columns 1 and 2, the effects do not seem to vary with the likelihood of quality changes. The coefficient on the interaction of cash villages and quality variation is wrong-signed and insignificant, and the difference in the interaction terms for in-kind and cash villages is close to zero. Meanwhile, even among goods with little quality variation (the main effects), we find significantly lower prices in in-kind villages than in cash villages. Table 4 Robustness check testing for changes in product quality Measure of quality variation =  Subjective categorization  Good-specific coefficient of variation of baseline price  Village-good-specific coeff. of variation of baseline price    All PAL  Basic PAL  All PAL  Basic PAL  All PAL  Basic PAL     goods  goods only  goods  goods only  goods  goods only  Outcome =  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  High-quality variation $$\times$$ In-kind  –0.026  –0.034  –0.001  0.032  0.007  0.021     (0.025)  (0.027)  (0.029)  (0.033)  (0.024)  (0.037)  High-quality variation $$\times$$ Cash  –0.018  –0.029  –0.006  0.039  –0.004  0.027     (0.033)  (0.041)  (0.040)  (0.046)  (0.036)  (0.047)  In-kind  –0.022  –0.014  –0.036*  –0.044**  –0.040**  –0.038**     (0.021)  (0.029)  (0.021)  (0.019)  (0.018)  (0.018)  Cash  0.012  0.030  0.006  0.001  0.004  0.007     (0.025)  (0.034)  (0.028)  (0.031)  (0.027)  (0.027)  High-quality variation  –0.007  –0.002  –0.012  –0.031  –0.006  –0.002     (0.021)  (0.023)  (0.026)  (0.029)  (0.019)  (0.031)  Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.034*  –0.044*  –0.041*  –0.044  –0.044*  –0.045*  H0: In-kind = Cash (p-value)  0.08  0.09  0.08  0.13  0.06  0.06  Effect size: High-quality var.$$\times$$  –0.008  –0.005  0.005  –0.007  0.011  –0.006  In-kind - High-quality var.$$\times$$Cash                    H0: High-quality var.$$\times$$In-kind =  0.78  0.9  0.88  0.86  0.73  0.89  High-quality var.$$\times$$Cash (p-value)                    Measure of quality variation =  Subjective categorization  Good-specific coefficient of variation of baseline price  Village-good-specific coeff. of variation of baseline price    All PAL  Basic PAL  All PAL  Basic PAL  All PAL  Basic PAL     goods  goods only  goods  goods only  goods  goods only  Outcome =  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  High-quality variation $$\times$$ In-kind  –0.026  –0.034  –0.001  0.032  0.007  0.021     (0.025)  (0.027)  (0.029)  (0.033)  (0.024)  (0.037)  High-quality variation $$\times$$ Cash  –0.018  –0.029  –0.006  0.039  –0.004  0.027     (0.033)  (0.041)  (0.040)  (0.046)  (0.036)  (0.047)  In-kind  –0.022  –0.014  –0.036*  –0.044**  –0.040**  –0.038**     (0.021)  (0.029)  (0.021)  (0.019)  (0.018)  (0.018)  Cash  0.012  0.030  0.006  0.001  0.004  0.007     (0.025)  (0.034)  (0.028)  (0.031)  (0.027)  (0.027)  High-quality variation  –0.007  –0.002  –0.012  –0.031  –0.006  –0.002     (0.021)  (0.023)  (0.026)  (0.029)  (0.019)  (0.031)  Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.034*  –0.044*  –0.041*  –0.044  –0.044*  –0.045*  H0: In-kind = Cash (p-value)  0.08  0.09  0.08  0.13  0.06  0.06  Effect size: High-quality var.$$\times$$  –0.008  –0.005  0.005  –0.007  0.011  –0.006  In-kind - High-quality var.$$\times$$Cash                    H0: High-quality var.$$\times$$In-kind =  0.78  0.9  0.88  0.86  0.73  0.89  High-quality var.$$\times$$Cash (p-value)                    Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. Standard errors (in parentheses) are clustered at the village level. (2) Regressions control for the pre-period normalized unit value and an indicator for imputed pre-programme prices (see text). (3) High-quality variation is defined in three ways. First, we subjectively identified goods that had high-quality variation; these goods are beans, cereal, corn flour, lentils, and pasta soup (columns 1–2). Second, we use the coefficient of variation (C.V.) of pre-period unit values; a high C.V. is one that is above the median. We construct the within-village-good C.V. We average across villages to create a good-specific measure of quality variability (columns 3–4) and also use the village-good-specific measure (columns 5–6). When the village-good C.V. is missing, it is imputed with the good-specific C.V. Table 4 Robustness check testing for changes in product quality Measure of quality variation =  Subjective categorization  Good-specific coefficient of variation of baseline price  Village-good-specific coeff. of variation of baseline price    All PAL  Basic PAL  All PAL  Basic PAL  All PAL  Basic PAL     goods  goods only  goods  goods only  goods  goods only  Outcome =  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  High-quality variation $$\times$$ In-kind  –0.026  –0.034  –0.001  0.032  0.007  0.021     (0.025)  (0.027)  (0.029)  (0.033)  (0.024)  (0.037)  High-quality variation $$\times$$ Cash  –0.018  –0.029  –0.006  0.039  –0.004  0.027     (0.033)  (0.041)  (0.040)  (0.046)  (0.036)  (0.047)  In-kind  –0.022  –0.014  –0.036*  –0.044**  –0.040**  –0.038**     (0.021)  (0.029)  (0.021)  (0.019)  (0.018)  (0.018)  Cash  0.012  0.030  0.006  0.001  0.004  0.007     (0.025)  (0.034)  (0.028)  (0.031)  (0.027)  (0.027)  High-quality variation  –0.007  –0.002  –0.012  –0.031  –0.006  –0.002     (0.021)  (0.023)  (0.026)  (0.029)  (0.019)  (0.031)  Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.034*  –0.044*  –0.041*  –0.044  –0.044*  –0.045*  H0: In-kind = Cash (p-value)  0.08  0.09  0.08  0.13  0.06  0.06  Effect size: High-quality var.$$\times$$  –0.008  –0.005  0.005  –0.007  0.011  –0.006  In-kind - High-quality var.$$\times$$Cash                    H0: High-quality var.$$\times$$In-kind =  0.78  0.9  0.88  0.86  0.73  0.89  High-quality var.$$\times$$Cash (p-value)                    Measure of quality variation =  Subjective categorization  Good-specific coefficient of variation of baseline price  Village-good-specific coeff. of variation of baseline price    All PAL  Basic PAL  All PAL  Basic PAL  All PAL  Basic PAL     goods  goods only  goods  goods only  goods  goods only  Outcome =  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  High-quality variation $$\times$$ In-kind  –0.026  –0.034  –0.001  0.032  0.007  0.021     (0.025)  (0.027)  (0.029)  (0.033)  (0.024)  (0.037)  High-quality variation $$\times$$ Cash  –0.018  –0.029  –0.006  0.039  –0.004  0.027     (0.033)  (0.041)  (0.040)  (0.046)  (0.036)  (0.047)  In-kind  –0.022  –0.014  –0.036*  –0.044**  –0.040**  –0.038**     (0.021)  (0.029)  (0.021)  (0.019)  (0.018)  (0.018)  Cash  0.012  0.030  0.006  0.001  0.004  0.007     (0.025)  (0.034)  (0.028)  (0.031)  (0.027)  (0.027)  High-quality variation  –0.007  –0.002  –0.012  –0.031  –0.006  –0.002     (0.021)  (0.023)  (0.026)  (0.029)  (0.019)  (0.031)  Observations  2,335  1,617  2,335  1,617  2,335  1,617  Effect size: In-kind - Cash  –0.034*  –0.044*  –0.041*  –0.044  –0.044*  –0.045*  H0: In-kind = Cash (p-value)  0.08  0.09  0.08  0.13  0.06  0.06  Effect size: High-quality var.$$\times$$  –0.008  –0.005  0.005  –0.007  0.011  –0.006  In-kind - High-quality var.$$\times$$Cash                    H0: High-quality var.$$\times$$In-kind =  0.78  0.9  0.88  0.86  0.73  0.89  High-quality var.$$\times$$Cash (p-value)                    Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. Standard errors (in parentheses) are clustered at the village level. (2) Regressions control for the pre-period normalized unit value and an indicator for imputed pre-programme prices (see text). (3) High-quality variation is defined in three ways. First, we subjectively identified goods that had high-quality variation; these goods are beans, cereal, corn flour, lentils, and pasta soup (columns 1–2). Second, we use the coefficient of variation (C.V.) of pre-period unit values; a high C.V. is one that is above the median. We construct the within-village-good C.V. We average across villages to create a good-specific measure of quality variability (columns 3–4) and also use the village-good-specific measure (columns 5–6). When the village-good C.V. is missing, it is imputed with the good-specific C.V. As a second proxy for quality variation, we use data from the household survey on the unit value that different households report paying for the same good and construct the coefficient of variation of unit values for each village-good. The variation in unit values is likely due mostly to measurement error, not quality variation, so this is an imperfect measure, but it has the advantage of being more objective than our subjective categorization. We average the coefficient of variation across villages to create a good-specific measure of quality variation (columns 3 and 4) and also use the village-good-specific measure (columns 5 and 6). We again find that, first, the results are not driven by the goods with more quality variation, and, second, even for the goods with low quality variation, prices are lower in in-kind villages than in cash villages. In short, the price effects we estimate do not appear to be a result of quality upgrading. To summarize, we find that the influx of supply from in-kind transfers causes prices to fall relative to prices under cash transfers. The result is robust to several alternative specifications and does not appear to be driven by changes in product quality. The point estimates suggest that this price gap between transfer modalities results from in-kind transfers having a net negative effect on prices and cash transfers having a very small positive effect on prices, though these two individual effects relative to the control group are less precisely estimated than the cash-versus-in-kind gap. 5.3. Persistence of price effects In Table 5 we present evidence on whether the price effects dissipate over time, using the variation across villages in when the programme was launched. We calculate the duration of the treatment, which is the difference between the date of the follow-up survey and the start date of benefit receipts. This duration ranges from 8 to 22 months. Note that programme duration is undefined for the control group, so this analysis compares in-kind to cash villages only. Table 5 Price effects based on duration of intervention    All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price    (1)  (2)  (3)  (4)  In-kind  –0.031  –0.029  –0.038  –0.035     (0.022)  (0.022)  (0.031)  (0.030)  In-kind $$\times$$ Above median length of treatment  –0.021  –0.021  –0.022  –0.029     (0.034)  (0.034)  (0.040)  (0.038)  Above median length of treatment  0.004  0.000  0.018  0.015     (0.028)  (0.027)  (0.033)  (0.029)  In-kind $$\times$$ Development index     0.010     –0.005        (0.020)     (0.026)  Development index     –0.010     –0.007        (0.016)     (0.023)  Observations  1,818  1,798  1,258  1,245     All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price    (1)  (2)  (3)  (4)  In-kind  –0.031  –0.029  –0.038  –0.035     (0.022)  (0.022)  (0.031)  (0.030)  In-kind $$\times$$ Above median length of treatment  –0.021  –0.021  –0.022  –0.029     (0.034)  (0.034)  (0.040)  (0.038)  Above median length of treatment  0.004  0.000  0.018  0.015     (0.028)  (0.027)  (0.033)  (0.029)  In-kind $$\times$$ Development index     0.010     –0.005        (0.020)     (0.026)  Development index     –0.010     –0.007        (0.016)     (0.023)  Observations  1,818  1,798  1,258  1,245  Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. Standard errors (in parentheses) are clustered at the village level. (2) Regressions control for the pre-period normalized unit value and an indicator for imputed pre-programme prices (see text). (3) Length of treatment is defined as the village median number of months for which transfers were received prior to the follow-up survey. (4) The development index is the first principal component from a factor analysis of the following four variables: (1) the time required to travel to a larger market that sells fruit, vegetables, and meat; (2) the distance to the head of the municipality; (3) pre-programme village median monthly expenditure on non-durables, and (4) the village population. Table 5 Price effects based on duration of intervention    All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price    (1)  (2)  (3)  (4)  In-kind  –0.031  –0.029  –0.038  –0.035     (0.022)  (0.022)  (0.031)  (0.030)  In-kind $$\times$$ Above median length of treatment  –0.021  –0.021  –0.022  –0.029     (0.034)  (0.034)  (0.040)  (0.038)  Above median length of treatment  0.004  0.000  0.018  0.015     (0.028)  (0.027)  (0.033)  (0.029)  In-kind $$\times$$ Development index     0.010     –0.005        (0.020)     (0.026)  Development index     –0.010     –0.007        (0.016)     (0.023)  Observations  1,818  1,798  1,258  1,245     All PAL goods  Basic PAL goods only  Outcome =  price  price  price  price    (1)  (2)  (3)  (4)  In-kind  –0.031  –0.029  –0.038  –0.035     (0.022)  (0.022)  (0.031)  (0.030)  In-kind $$\times$$ Above median length of treatment  –0.021  –0.021  –0.022  –0.029     (0.034)  (0.034)  (0.040)  (0.038)  Above median length of treatment  0.004  0.000  0.018  0.015     (0.028)  (0.027)  (0.033)  (0.029)  In-kind $$\times$$ Development index     0.010     –0.005        (0.020)     (0.026)  Development index     –0.010     –0.007        (0.016)     (0.023)  Observations  1,818  1,798  1,258  1,245  Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. Standard errors (in parentheses) are clustered at the village level. (2) Regressions control for the pre-period normalized unit value and an indicator for imputed pre-programme prices (see text). (3) Length of treatment is defined as the village median number of months for which transfers were received prior to the follow-up survey. (4) The development index is the first principal component from a factor analysis of the following four variables: (1) the time required to travel to a larger market that sells fruit, vegetables, and meat; (2) the distance to the head of the municipality; (3) pre-programme village median monthly expenditure on non-durables, and (4) the village population. We interact programme duration with the in-kind treatment dummy in Table 5. For ease of interpretation, we use a dummy for above median duration (the average duration is 16 months in above-median villages and 12 in below-median villages), but the conclusion is similar if we use the duration in months: The coefficient on the interaction is insignificant and in fact negative, suggesting that the effects become if anything larger over time. In any case, we find no evidence that the effects fade away. The programme start date is not randomly assigned, so one concern is the endogeneity of the programme duration at follow-up. The one observable characteristic that we find is significantly correlated with programme duration is the level of development of the village (we define our measure of development in the next section). Thus, we reproduce the test above controlling for the level of development and its interaction with the in-kind indicator; as shown in columns 2 and 4, the results are similar. Many supply-side adjustments such as store owners altering their procurement would likely be complete by the one to two year mark. Thus, these results appear to be inconsistent with the village markets being perfectly competitive, as we would expect the marginal cost curve to be flat over this time span, and with a flat marginal cost curve and perfect competition, there would be no price effects of shifts in demand. Even with imperfect competition, one might expect the effects to fade over time as firms respond by entering or exiting the market, or local agricultural producers change their production levels. These adjustments would likely be underway after two years, so this finding of persistence suggests that such adjustments might not fully undo the price effects of transfer programs, at least in the medium run. Thus, while we cannot look at effects further out than two years, the price effects appear to persist beyond the short run. 5.4. Heterogeneity by the village’s level of development and market structure We next test for heterogeneity in the price effects based on the village’s level of development. We hypothesize that less developed villages experience larger price effects because they are less integrated with the outside economy and have less competition among local suppliers. Moreover, understanding how the price effects vary with how impoverished the village is of policy interest per se. We combine several village characteristics to construct a measure of its “development”. Specifically, we use the average expenditures per capita, population, average self-reported travel time to a larger market that sells fruit, vegetables, and meat, and distance to the nearest municipality head (calculated using GIS software). We construct the first principal component of these variables. (See Appendix B for details on the construction of this variable.)32 Essentially, an underdeveloped village is poorer, smaller, and more physically remote. For convenience, we will refer to villages with a development index below the sample median as less developed or underdeveloped. Table 6 reports the results on how the price effects vary with development. Column 1 reports the results for less developed villages. In-kind transfers cause a 3.6% price decline, and cash transfers cause a 1.5% increase. The difference is statistically significant at the 5% level. Meanwhile, in more developed villages (i.e. above-median development index), in-kind transfers cause a 3.3% decline in prices, while cash transfers cause a 0.7% price decline, with the difference of $$-$$0.027 in the predicted direction but insignificant (column 2).33,34 These findings reveal that the average effects for the cash-versus-in-kind effect (Table 3) are mostly driven by less developed villages.35 Column 3 reports the interacted model which shows that the interaction is statistically insignificant. Table 6 Heterogeneous price effects by level of village development, market integration, and supply-side competition    Below-median development  Above-median development  All villages  Villages with market power  Villages without market power  Below-median price correlation  Above-median price correlation  All villages  All villages  Outcome =  price  price  price  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  In-kind  –0.031  –0.041  –0.031  –0.048*  –0.005  –0.060**  –0.019  0.009  0.018     (0.027)  (0.030)  (0.027)  (0.025)  (0.021)  (0.028)  (0.028)  (0.025)  (0.036)  Cash  –0.006  0.012  –0.006  0.007  –0.005  0.002  –0.014  –0.036  –0.038     (0.037)  (0.031)  (0.037)  (0.029)  (0.026)  (0.032)  (0.032)  (0.034)  (0.041)  Development index below-median $$\times$$ In-kind        –0.010                 –0.013           (0.041)                 (0.040)  Development index below-median $$\times$$ Cash        0.018                 0.006           (0.048)                 (0.048)  Market power village $$\times$$ In-kind                       –0.039  –0.037                          (0.036)  (0.037)  Market power village $$\times$$ Cash                       0.029  0.027                          (0.041)  (0.044)  Price correlation below-median $$\times$$ In-kind                       –0.039  –0.045                          (0.041)  (0.042)  Price correlation below-median $$\times$$ Cash                       0.022  0.022                          (0.047)  (0.048)  Observations  1,194  1,110  2,304  1,733  602  1,115  1,220  2,335  2,304  Effect size: In-kind - Cash  –0.053**  –0.025  –0.025  –0.055***  0.000  –0.063***  –0.006        H0: In-kind = Cash (p-value)  0.01  0.38  0.38  0.01  1.00  0.01  0.81        Effect size: Development index below-median$$\times$$        –0.028                 –0.019  In-kind - Development index below-median$$\times$$Cash                             H0: Development index below-median$$\times$$In-kind =        0.43                 0.60  Development index below-median$$\times$$Cash (p-value)                             Effect size: Price correlation below-median$$\times$$                       –0.061*  –0.067*  In-kind - Price correlation below-median$$\times$$Cash                             H0: Price correlation below-median$$\times$$In-kind =                       0.07  0.05  Price correlation below-median$$\times$$Cash (p-value)                             Effect size: Market power village$$\times$$In-kind -                       –0.067*  –0.064*  Market power village$$\times$$Cash                             H0: Market power village$$\times$$In-kind = Market                       0.05  0.09  power village$$\times$$Cash (p-value)                                Below-median development  Above-median development  All villages  Villages with market power  Villages without market power  Below-median price correlation  Above-median price correlation  All villages  All villages  Outcome =  price  price  price  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  In-kind  –0.031  –0.041  –0.031  –0.048*  –0.005  –0.060**  –0.019  0.009  0.018     (0.027)  (0.030)  (0.027)  (0.025)  (0.021)  (0.028)  (0.028)  (0.025)  (0.036)  Cash  –0.006  0.012  –0.006  0.007  –0.005  0.002  –0.014  –0.036  –0.038     (0.037)  (0.031)  (0.037)  (0.029)  (0.026)  (0.032)  (0.032)  (0.034)  (0.041)  Development index below-median $$\times$$ In-kind        –0.010                 –0.013           (0.041)                 (0.040)  Development index below-median $$\times$$ Cash        0.018                 0.006           (0.048)                 (0.048)  Market power village $$\times$$ In-kind                       –0.039  –0.037                          (0.036)  (0.037)  Market power village $$\times$$ Cash                       0.029  0.027                          (0.041)  (0.044)  Price correlation below-median $$\times$$ In-kind                       –0.039  –0.045                          (0.041)  (0.042)  Price correlation below-median $$\times$$ Cash                       0.022  0.022                          (0.047)  (0.048)  Observations  1,194  1,110  2,304  1,733  602  1,115  1,220  2,335  2,304  Effect size: In-kind - Cash  –0.053**  –0.025  –0.025  –0.055***  0.000  –0.063***  –0.006        H0: In-kind = Cash (p-value)  0.01  0.38  0.38  0.01  1.00  0.01  0.81        Effect size: Development index below-median$$\times$$        –0.028                 –0.019  In-kind - Development index below-median$$\times$$Cash                             H0: Development index below-median$$\times$$In-kind =        0.43                 0.60  Development index below-median$$\times$$Cash (p-value)                             Effect size: Price correlation below-median$$\times$$                       –0.061*  –0.067*  In-kind - Price correlation below-median$$\times$$Cash                             H0: Price correlation below-median$$\times$$In-kind =                       0.07  0.05  Price correlation below-median$$\times$$Cash (p-value)                             Effect size: Market power village$$\times$$In-kind -                       –0.067*  –0.064*  Market power village$$\times$$Cash                             H0: Market power village$$\times$$In-kind = Market                       0.05  0.09  power village$$\times$$Cash (p-value)                             Notes: ***$$p<0.01$$, **$$p<0.05$$, *$$p<0.1$$. (1) The outcome variable is the post-programme price; it varies at the village-store-good level. It is normalized by good; the price is divided by the average price of the good across all observations in the control group. Standard errors (in parentheses) are clustered at the village level. (2) Regressions include all PAL goods and control for the main effects of the interaction terms reported, and for the pre-period normalized unit value and an indicator for imputed pre-programme prices (see text). (3) Price correlation is the correlation coefficient of the pre- to post-programme change in village prices with the pre- to post-programme change in prices in Mexico City for all PAL goods, it varies at the village level. (4) The number of stores is the number of stores included in the baseline price survey; a maximum of three stores were surveyed per village. (5) The development index is the first principal component of a factor analysis of the following four variables: the time required to travel to a larger market that sells fruit, vegetables, and meat; the distance to the head of the municipality; pre-programme village median monthly expenditure on non-durables; and the village population. Table 6 Heterogeneous price effects by level of village development, market integration, and supply-side competition    Below-median development  Above-median development  All villages  Villages with market power  Villages without market power  Below-median price correlation  Above-median price correlation  All villages  All villages  Outcome =  price  price  price  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  In-kind  –0.031  –0.041  –0.031  –0.048*  –0.005  –0.060**  –0.019  0.009  0.018     (0.027)  (0.030)  (0.027)  (0.025)  (0.021)  (0.028)  (0.028)  (0.025)  (0.036)  Cash  –0.006  0.012  –0.006  0.007  –0.005  0.002  –0.014  –0.036  –0.038     (0.037)  (0.031)  (0.037)  (0.029)  (0.026)  (0.032)  (0.032)  (0.034)  (0.041)  Development index below-median $$\times$$ In-kind        –0.010                 –0.013           (0.041)                 (0.040)  Development index below-median $$\times$$ Cash        0.018                 0.006           (0.048)                 (0.048)  Market power village $$\times$$ In-kind                       –0.039  –0.037                          (0.036)  (0.037)  Market power village $$\times$$ Cash                       0.029  0.027                          (0.041)  (0.044)  Price correlation below-median $$\times$$ In-kind                       –0.039  –0.045                          (0.041)  (0.042)  Price correlation below-median $$\times$$ Cash                       0.022  0.022                          (0.047)  (0.048)  Observations  1,194  1,110  2,304  1,733  602  1,115  1,220  2,335  2,304  Effect size: In-kind - Cash  –0.053**  –0.025  –0.025  –0.055***  0.000  –0.063***  –0.006        H0: In-kind = Cash (p-value)  0.01  0.38  0.38  0.01  1.00  0.01  0.81        Effect size: Development index below-median$$\times$$        –0.028                 –0.019  In-kind - Development index below-median$$\times$$Cash                             H0: Development index below-median$$\times$$In-kind =        0.43                 0.60  Development index below-median$$\times$$Cash (p-value)                             Effect size: Price correlation below-median$$\times$$                       –0.061*  –0.067*  In-kind - Price correlation below-median$$\times$$Cash                             H0: Price correlation below-median$$\times$$In-kind =                       0.07  0.05  Price correlation below-median$$\times$$Cash (p-value)                             Effect size: Market power village$$\times$$In-kind -                       –0.067*  –0.064*  Market power village$$\times$$Cash                             H0: Market power village$$\times$$In-kind = Market                       0.05  0.09  power village$$\times$$Cash (p-value)                                Below-median development  Above-median development  All villages  Villages with market power  Villages without market power  Below-median price correlation  Above-median price correlation  All villages  All villages  Outcome =  price  price  price  price  price  price  price  price  price    (1)  (2)  (3)  (4)  (5)  (6)  (7)  (8)  (9)  In-kind  –0.031  –0.041  –0.031  –0.048*  –0.005  –0.060**  –0.019  0.009  0.018     (0.027)  (0.030)  (0.027)  (0.025)  (0.021)  (0.028)  (0.028)  (0.025)  (0.036)  Cash  –0.006  0.012  –0.006  0.007  –0.005  0.002  –0.014  –0.036  –0.038     (0.037)  (0.031)  (0.037)  (0.029)  (0.026)  (0.032)  (0.032)  (0.034)  (0.041)  Development index below-median $$\times$$ In-kind        –0.010                 –0.013           (0.041)                 (0.040)  Development index below-median $$\times$$ Cash        0.018                 0.006           (0.048)                 (0.048)  Market power village $$\times$$ In-kind                       –0.039  –0.037                          (0.036)  (0.037)  Market power village $$\times$$ Cash                       0.029  0.027                          (0.041)  (0.044)  Price correlation below-median $$\times$$ In-kind                       –0.039  –0.045                          (0.041)  (0.042)  Price correlation below-median $$\times$$ Cash                       0.022  0.022                          (0.047)  (0.048)  Observations  1,194  1,110  2,304  1,733  602  1,115  1,220  2,335  2,304  Effect size: In-kind - Cash  –0.053**  –0.025  –0.025  –0.055***  0.000  –0.063***  –0.006        H0: In-kind = Cash (p-value)  0.01  0.38  0.38  0.01  1.00  0.01  0.81        Effect size: Development index below-median$$\times$$        –0.028                 –0.019  In-kind - Development index below-median$$\times$$Cash                             H0: Development index below-median$$\times$$In-kind =        0.43                 0.60  Development index below-median$$\times$$Cash (p-value)                             Effect size: Price correlation below-median$$\times$$                       –0.061*