I can’t go more than a few minutes through my inbox, or my Twitter, Facebook, and RSS feeds without running into yet another piece about the promise of cash transfers. It doesn’t look like the traditional media outlets are alone in propagating this: see this and this from This American Life and NYT Magazine on GiveDirectly’s work (large lump-sum unconditional cash transfers, or UCTs, to the poor in Kenya); Chris Blattman here, here, and here mainly on his study with Fiala and Martinez (large grants to groups of unemployed youth for skills training and microenterprise formation in Uganda); and IPA blog and Atlantic.com here and here on the new study by Benhassine et al. (conditional vs. labeled cash transfers to families with school-aged children to improve primary school completion in rural Morocco).
In many, but not all, of these pieces, these interventions are discussed together in a rather clumsy attempt to usher in a new era in development aid, in which things are as simple as dropping money into poor communities. But, the three interventions that are lumped together in support of UCTs are, in fact, quite different from each other. Over the next few paragraphs, I will summarize each intervention and the evidence associated with it – to draw out any parallels and contrast the differences.
Let’s start with GiveDirectly. I admire what they’re doing, which is something no others have dared to do: after some identification of the beneficiaries (through geographic targeting and proxy-means testing), they text a large sum of money, like US$500, via mobile phones using M-PESA. That’s the intervention. In the simplified world of textbook economics, individuals’ choice sets are larger when they’re given cash without any strings attached, hence their welfare cannot be lower than if they were given CCTs (or in-kind transfers, which is really the same thing). People sometimes ask me what I think about this or that in-kind transfer to households. I always reply by asking them to justify why they would give that good instead of cash. Of course, there are answers that justify either choice (more on these below, after I discuss the Morocco study), but GiveDirectly has chosen the simple, uncluttered, and transparent route: to focus on income effects to increase household welfare now (if not also in the future) and to leave it to others to address market failures and externalities.
However, before we get too excited, it’s too early to say what kind of an effect they’re having on poverty reduction. They are rigorously evaluating the impact of the program and I hope that we see sustained long-term impacts that we can compare to alternative uses of such sums. But, we don’t have this evidence yet and can’t quite declare victory simply because people are not spending the money on sin goods.
The study by Blattman, Fiala, and Martinez is quite different. It examines a program by Northern Uganda Social Action Funds (NUSAF) that targeted unemployed young people, which encouraged them to form groups to put together proposal for skills development that might lead to employment, self-employment, or formation of microenterprises. Groups that complied with the requirements were then given large sums of lump-sum cash grants – say, $10,000 to a group of 20 individuals. After four years, they find large effects of this program on earnings among the treatment group: it looks like many of these groups did spend a non-negligible portion of their grants on skills acquisition (in the form of training) and started small businesses in skilled trades or got employment in those sectors. I won’t go too much into the evidence as Chris has written plenty on this – see the links above.
A few things to note here: First, notice how different this intervention is than the approach of GiveDirectly. It is absolutely nothing like waking up one morning and finding $500 in your M-PESA account. Would such an intervention at the individual level for this target group of unemployed youths have produced such impacts as Blattman, Fiala, and Martinez are finding? We don’t know, but it would not be hard to imagine that it would not. Second, we don’t know whether poverty traps exist – a theoretical possibility for multiple equilibria, where the lowest level equilibrium is a poverty trap and a minimum level of wealth is needed to escape poverty (see Ravallion 2012). If they do exist, however, it is likely that there are other causes for being stuck in a low-level equilibrium than just the lack of money or credit constraints: poor households also have capital constraints (both human and physical); they suffer from lack of access to information, infrastructure, and markets, and so on. Perhaps, intervening in any one of these constraints is insufficient to escape poverty for good: the trick may be in finding the right combination of interventions. This study, which is currently a working paper, provides us with a small piece of that puzzle: examining grants to groups who had to come up with proposals for start-ups in skilled sectors. Compared to no grants, it found impressive gains in employment and earnings among the beneficiaries. It is important in a proof-of-concept kind of way. But, again, it is too early to tell what kinds of effects on poverty reduction we can expect from such an approach and what the study can teach us more generally about sustainable paths out of poverty. In his latest post, Chris writes that preliminary evidence from two other interventions “that give plain cash” that he is studying do not give him high hopes.
Finally, I’ll discuss (in some detail) the intervention that experimented with the role of conditions in conditional cash transfer programs (CCTs) for schooling. In Morocco, Benhassine et al. worked with the Ministry of Education to design an experiment where small cash transfers (similar to those in other countries with similar CCT programs) were given – some conditional on regular school attendance, some not. The authors describe one arm as a CCT, where explicit conditions for school attendance were announced and monitored, and the other arm as a ‘labeled cash transfer’ (LCT). It is ‘labeled’ because the program was run by the Ministry of Education, where parents of young children had to enroll in the program at the local school and where the headmaster registered and enrolled the children in school while registering them in the program: it was made salient to households that the funds were coming from the Ministry of Education under a pro-education program that was trying to raise primary school completion rates in rural Morocco. Notice that this intervention is yet again a different animal than the two described above.
This study has a lot going for it. In particular, imposing conditionalities in cash transfer programs to deal with market failures or externalities may be a blunt instrument (authors call it a shove) that may sometimes be the second- or third-best solution (setting political economy issues aside). If the reason we’re not satisfied with just giving cash to households is the presence of market failures, then there may be better ways of directly addressing such problems. For example, if the problem is misinformation about the value of schooling, then the provision of information may be better than CCTs, given that CCTs may cause distortions and trade-offs. Making education salient by enrolling children in school automatically but not monitoring attendance may be better than running a CCT. This study attempts to test this idea and finds that CCTs and LCTs were equally effective in improving school participation.
I have some doubts about this interpretation for a number of reasons. First, it looks like the problem in Morocco is enrollment, not attendance. About 60% of children complete primary school in rural Morocco, but conditional on enrollment the attendance rates are extremely high (96% in the control group at follow-up, based on surprise school visits). So, the problem is with getting people enrolled in school and not getting them to come to school conditional on being enrolled. In that case, it is not surprising that LCTs and CCTs were equally effective in reducing dropouts, as both were de facto conditional on enrollment. Second, while the LCT was de facto conditional on school enrollment from the start, by the end of the two-year program the CCT program was de facto unconditional – as seen by the parents (Table 4). Consider me de facto confused, but there does not seem to be a lot of difference between the two programs in terms of priming the importance of schooling. Add to this the fact that teachers did not want to advertise the lack of conditionality in the LCT group; and the fact that students in both groups may have thought that their attendance is being monitored (by their teachers or through the surprise school visits for data collection), it is hard to make a clear distinction between the nudge and the shove.
Third, and finally, and the authors should correct me if I am wrong, I believe that this study was designed as a CCT vs. a UCT, where the object of experimental variation was more on the monitoring of the conditions – with the CCT arms divided into light, moderate, and full monitoring (the last option with biometric machines at school doors). As the authors discuss in a footnote, this part of the experimentation failed and all three groups were lumped into a CCT group. This is fine – one should not underestimate the difficulty of running complex interventions. However, I believe that the UCT arm becoming an LCT one is also after the fact: once it became clear that the UCT was not a pure UCT but instead one that heavily suggested the desirability of sending children to school, it became an LCT. This is also fine – it would be a terrible waste to discard the findings of this study because the program did not get implemented exactly as planned. These may sound like small details, but they matter in RCTs.
After reviewing these three interventions and the existing (and evolving) evidence on their impacts on various outcomes, it is important to step back, take a deep breath, and realize that there is both elementary theory and a large body of evidence on the subject of cash transfer programs and the role of conditions within them. I don’t want to rehash this large literature – Francisco Ferreira and I wrote a two-part blog post in late 2011, titled: “Conditions Work: But Are they a Good Thing?” (see Part I and Part II). If your interest in this topic is serious, I invite you to read these two posts, but the bottom line is that while the conditionality in CCT programs, as predicted by theory, improves the conditioned-on outcome through a substitution/price effect, it also poses some serious trade-offs with respect to gains in overall welfare. Over the past decade, in addition to the countless evaluations of CCT programs, we have had two books, two studies that utilized structural models combined with microsimulation, two quasi-experimental studies, three new experimental studies, and a systematic review on the topic of conditional vs. unconditional cash transfers. When it comes to schooling CCTs, the evidence suggests (with some variation) that conditions make a positive difference in schooling outcomes, which may be increasing in the intensity with which school participation is made salient by the program. As rigorous studies moved from Latin America to Africa, the size of the income effect increased – from being practically non-existent in studies of Bolsa Escola and PROGRESA to playing an important role in increasing school enrollment in Africa where poverty is higher.
However, Chico and I also wrote the following in the concluding paragraph of our blog post: “Even when CCTs are correcting a market failure, they may only be a second- or third-best solution. If we could really pinpoint the externality (or information or agency failure) being addressed by the conditionality in CCT programs, could we design alternative programs that would provide unconditional cash alongside another cheap intervention designed to address the market failure? For example, if information or attitudes are part of the problem for school dropouts, would a UCT plus a social marketing campaign aided by communication scientists, psychologists, and behavioral economists perform better than a CCT by avoiding the trade-offs presented above? We believe that alternative designs should be squarely on the table when the next generation of cash transfer programs is being designed.”
Giving poor people handouts with no strings attached is not a panacea. What plain cash transfers would accomplish is not always a good benchmark for every intervention, although it may be for some. Consider the following thought experiment: would you be able to obtain the same gains attained from the distribution of free deworming pills at schools by replacing that intervention with a $1/year transfer to the families of all students? Now replace deworming pills with insecticide-treated bed nets and the cash amount with the market price of an ITN: how about now? UCTs won’t eradicate malaria, but gentler versions of CCTs combined with supply-side public health interventions might have a better shot. Now, think about giving people cash vs. a nutritionally balanced basket of food of equal value. Are we getting closer? A recent RCT in Mexico suggested that while people ate different bundles in such an experiment, their nutritional intakes were equally good: neither group had an advantage over the other on micronutrient intake, stunting, anemia, etc. I am guessing that you did not prefer plain cash transfers under each scenario and that’s the point: the appropriate intervention depends on the problem at hand.
When people are credit constrained and the removal of that constraint helps with your outcome of interest, go ahead and give cash to people and forget about the rest. But, if there are other failures causing people to be poor, sick, unhappy, etc. then money alone won’t be enough. Money alone will not solve information failures, intra-household bargaining problems, access to contraception, agency problems, etc. It seems like a simple message, but one that seems to be getting lost in the current debate.
And, what if we don’t know what the constraints are, like one of the co-founders of GiveDirectly who said: “…the truth is, I don’t think I have a very good sense of what the poor need.” Then, by all means, experiment – as these interesting and important studies are doing.
In many, but not all, of these pieces, these interventions are discussed together in a rather clumsy attempt to usher in a new era in development aid, in which things are as simple as dropping money into poor communities. But, the three interventions that are lumped together in support of UCTs are, in fact, quite different from each other. Over the next few paragraphs, I will summarize each intervention and the evidence associated with it – to draw out any parallels and contrast the differences.
Let’s start with GiveDirectly. I admire what they’re doing, which is something no others have dared to do: after some identification of the beneficiaries (through geographic targeting and proxy-means testing), they text a large sum of money, like US$500, via mobile phones using M-PESA. That’s the intervention. In the simplified world of textbook economics, individuals’ choice sets are larger when they’re given cash without any strings attached, hence their welfare cannot be lower than if they were given CCTs (or in-kind transfers, which is really the same thing). People sometimes ask me what I think about this or that in-kind transfer to households. I always reply by asking them to justify why they would give that good instead of cash. Of course, there are answers that justify either choice (more on these below, after I discuss the Morocco study), but GiveDirectly has chosen the simple, uncluttered, and transparent route: to focus on income effects to increase household welfare now (if not also in the future) and to leave it to others to address market failures and externalities.
However, before we get too excited, it’s too early to say what kind of an effect they’re having on poverty reduction. They are rigorously evaluating the impact of the program and I hope that we see sustained long-term impacts that we can compare to alternative uses of such sums. But, we don’t have this evidence yet and can’t quite declare victory simply because people are not spending the money on sin goods.
The study by Blattman, Fiala, and Martinez is quite different. It examines a program by Northern Uganda Social Action Funds (NUSAF) that targeted unemployed young people, which encouraged them to form groups to put together proposal for skills development that might lead to employment, self-employment, or formation of microenterprises. Groups that complied with the requirements were then given large sums of lump-sum cash grants – say, $10,000 to a group of 20 individuals. After four years, they find large effects of this program on earnings among the treatment group: it looks like many of these groups did spend a non-negligible portion of their grants on skills acquisition (in the form of training) and started small businesses in skilled trades or got employment in those sectors. I won’t go too much into the evidence as Chris has written plenty on this – see the links above.
A few things to note here: First, notice how different this intervention is than the approach of GiveDirectly. It is absolutely nothing like waking up one morning and finding $500 in your M-PESA account. Would such an intervention at the individual level for this target group of unemployed youths have produced such impacts as Blattman, Fiala, and Martinez are finding? We don’t know, but it would not be hard to imagine that it would not. Second, we don’t know whether poverty traps exist – a theoretical possibility for multiple equilibria, where the lowest level equilibrium is a poverty trap and a minimum level of wealth is needed to escape poverty (see Ravallion 2012). If they do exist, however, it is likely that there are other causes for being stuck in a low-level equilibrium than just the lack of money or credit constraints: poor households also have capital constraints (both human and physical); they suffer from lack of access to information, infrastructure, and markets, and so on. Perhaps, intervening in any one of these constraints is insufficient to escape poverty for good: the trick may be in finding the right combination of interventions. This study, which is currently a working paper, provides us with a small piece of that puzzle: examining grants to groups who had to come up with proposals for start-ups in skilled sectors. Compared to no grants, it found impressive gains in employment and earnings among the beneficiaries. It is important in a proof-of-concept kind of way. But, again, it is too early to tell what kinds of effects on poverty reduction we can expect from such an approach and what the study can teach us more generally about sustainable paths out of poverty. In his latest post, Chris writes that preliminary evidence from two other interventions “that give plain cash” that he is studying do not give him high hopes.
Finally, I’ll discuss (in some detail) the intervention that experimented with the role of conditions in conditional cash transfer programs (CCTs) for schooling. In Morocco, Benhassine et al. worked with the Ministry of Education to design an experiment where small cash transfers (similar to those in other countries with similar CCT programs) were given – some conditional on regular school attendance, some not. The authors describe one arm as a CCT, where explicit conditions for school attendance were announced and monitored, and the other arm as a ‘labeled cash transfer’ (LCT). It is ‘labeled’ because the program was run by the Ministry of Education, where parents of young children had to enroll in the program at the local school and where the headmaster registered and enrolled the children in school while registering them in the program: it was made salient to households that the funds were coming from the Ministry of Education under a pro-education program that was trying to raise primary school completion rates in rural Morocco. Notice that this intervention is yet again a different animal than the two described above.
This study has a lot going for it. In particular, imposing conditionalities in cash transfer programs to deal with market failures or externalities may be a blunt instrument (authors call it a shove) that may sometimes be the second- or third-best solution (setting political economy issues aside). If the reason we’re not satisfied with just giving cash to households is the presence of market failures, then there may be better ways of directly addressing such problems. For example, if the problem is misinformation about the value of schooling, then the provision of information may be better than CCTs, given that CCTs may cause distortions and trade-offs. Making education salient by enrolling children in school automatically but not monitoring attendance may be better than running a CCT. This study attempts to test this idea and finds that CCTs and LCTs were equally effective in improving school participation.
I have some doubts about this interpretation for a number of reasons. First, it looks like the problem in Morocco is enrollment, not attendance. About 60% of children complete primary school in rural Morocco, but conditional on enrollment the attendance rates are extremely high (96% in the control group at follow-up, based on surprise school visits). So, the problem is with getting people enrolled in school and not getting them to come to school conditional on being enrolled. In that case, it is not surprising that LCTs and CCTs were equally effective in reducing dropouts, as both were de facto conditional on enrollment. Second, while the LCT was de facto conditional on school enrollment from the start, by the end of the two-year program the CCT program was de facto unconditional – as seen by the parents (Table 4). Consider me de facto confused, but there does not seem to be a lot of difference between the two programs in terms of priming the importance of schooling. Add to this the fact that teachers did not want to advertise the lack of conditionality in the LCT group; and the fact that students in both groups may have thought that their attendance is being monitored (by their teachers or through the surprise school visits for data collection), it is hard to make a clear distinction between the nudge and the shove.
Third, and finally, and the authors should correct me if I am wrong, I believe that this study was designed as a CCT vs. a UCT, where the object of experimental variation was more on the monitoring of the conditions – with the CCT arms divided into light, moderate, and full monitoring (the last option with biometric machines at school doors). As the authors discuss in a footnote, this part of the experimentation failed and all three groups were lumped into a CCT group. This is fine – one should not underestimate the difficulty of running complex interventions. However, I believe that the UCT arm becoming an LCT one is also after the fact: once it became clear that the UCT was not a pure UCT but instead one that heavily suggested the desirability of sending children to school, it became an LCT. This is also fine – it would be a terrible waste to discard the findings of this study because the program did not get implemented exactly as planned. These may sound like small details, but they matter in RCTs.
After reviewing these three interventions and the existing (and evolving) evidence on their impacts on various outcomes, it is important to step back, take a deep breath, and realize that there is both elementary theory and a large body of evidence on the subject of cash transfer programs and the role of conditions within them. I don’t want to rehash this large literature – Francisco Ferreira and I wrote a two-part blog post in late 2011, titled: “Conditions Work: But Are they a Good Thing?” (see Part I and Part II). If your interest in this topic is serious, I invite you to read these two posts, but the bottom line is that while the conditionality in CCT programs, as predicted by theory, improves the conditioned-on outcome through a substitution/price effect, it also poses some serious trade-offs with respect to gains in overall welfare. Over the past decade, in addition to the countless evaluations of CCT programs, we have had two books, two studies that utilized structural models combined with microsimulation, two quasi-experimental studies, three new experimental studies, and a systematic review on the topic of conditional vs. unconditional cash transfers. When it comes to schooling CCTs, the evidence suggests (with some variation) that conditions make a positive difference in schooling outcomes, which may be increasing in the intensity with which school participation is made salient by the program. As rigorous studies moved from Latin America to Africa, the size of the income effect increased – from being practically non-existent in studies of Bolsa Escola and PROGRESA to playing an important role in increasing school enrollment in Africa where poverty is higher.
However, Chico and I also wrote the following in the concluding paragraph of our blog post: “Even when CCTs are correcting a market failure, they may only be a second- or third-best solution. If we could really pinpoint the externality (or information or agency failure) being addressed by the conditionality in CCT programs, could we design alternative programs that would provide unconditional cash alongside another cheap intervention designed to address the market failure? For example, if information or attitudes are part of the problem for school dropouts, would a UCT plus a social marketing campaign aided by communication scientists, psychologists, and behavioral economists perform better than a CCT by avoiding the trade-offs presented above? We believe that alternative designs should be squarely on the table when the next generation of cash transfer programs is being designed.”
Giving poor people handouts with no strings attached is not a panacea. What plain cash transfers would accomplish is not always a good benchmark for every intervention, although it may be for some. Consider the following thought experiment: would you be able to obtain the same gains attained from the distribution of free deworming pills at schools by replacing that intervention with a $1/year transfer to the families of all students? Now replace deworming pills with insecticide-treated bed nets and the cash amount with the market price of an ITN: how about now? UCTs won’t eradicate malaria, but gentler versions of CCTs combined with supply-side public health interventions might have a better shot. Now, think about giving people cash vs. a nutritionally balanced basket of food of equal value. Are we getting closer? A recent RCT in Mexico suggested that while people ate different bundles in such an experiment, their nutritional intakes were equally good: neither group had an advantage over the other on micronutrient intake, stunting, anemia, etc. I am guessing that you did not prefer plain cash transfers under each scenario and that’s the point: the appropriate intervention depends on the problem at hand.
When people are credit constrained and the removal of that constraint helps with your outcome of interest, go ahead and give cash to people and forget about the rest. But, if there are other failures causing people to be poor, sick, unhappy, etc. then money alone won’t be enough. Money alone will not solve information failures, intra-household bargaining problems, access to contraception, agency problems, etc. It seems like a simple message, but one that seems to be getting lost in the current debate.
And, what if we don’t know what the constraints are, like one of the co-founders of GiveDirectly who said: “…the truth is, I don’t think I have a very good sense of what the poor need.” Then, by all means, experiment – as these interesting and important studies are doing.
Join the Conversation