The GiveWell Blog

New research on cash transfers

Summary

  • There has been a good deal of discussion recently about new research on the effects of cash transfers, beginning with a post by economist Berk Özler on the World Bank’s Development Impact blog. We have not yet fully reviewed the new research, but wanted to provide a preliminary update for our followers about our plans for reviewing this research and how it might affect our views of cash transfers, a program implemented by one of our top charities, GiveDirectly.
  • In brief, the new research suggests that cash transfers may be less effective than we previously believed in two ways. First, cash transfers may have substantial negative effects on non-recipients who live near recipients (“negative spillovers”). Second, the benefits of cash transfers may fade quickly.
  • We plan to reassess the cash transfer evidence base and provide our updated conclusions in the next several months (by November 2018 at the latest). One reason that we do not plan to provide a comprehensive update sooner is that we expect upcoming midline results from GiveDirectly’s “general equilibrium” study, a large and high-quality study explicitly designed to estimate spillover effects, will play a major role in our conclusions. Results from this study are expected to be released in the next few months.
  • Our best guess is that we will reduce our estimate of the cost-effectiveness of cash transfers to some extent, but will likely continue to recommend GiveDirectly. However, major updates to our current views, either in the negative or positive direction, seem possible.

More detail below.

Background

GiveDirectly, one of our top charities, provides unconditional cash transfers to very poor households in Kenya, Uganda, and Rwanda.

Several new studies have recently been released that assess the impact of unconditional cash transfers, including a three-year follow-up study (Haushofer and Shapiro 2018, henceforth referred to as “HS 2018”) on the impact of transfers that were provided by GiveDirectly. Berk Özler, a senior economist at the World Bank, summarized some of this research in two posts on the World Bank Development Impact blog (here and here), noting that the results imply that cash transfers may be less effective than proponents previously believed. In particular, Özler raises the concerns that cash may:

  1. Have negative “spillovers”: i.e., negative effects on households that did not receive transfers but that live near recipient households.
  2. Have quickly-fading benefits: i.e., the standard of living for recipient households may converge to be similar to non-recipient households within a few years of receiving transfers.

Below, we discuss the topics of spillover effects and the duration of benefits of cash transfers in more detail, as well as some other considerations relevant to the effectiveness of cash transfers. In brief:

  • If substantial spillover effects exist, they have the potential to significantly affect our cost-effectiveness estimates for cash transfers. We are uncertain what we will conclude about spillover effects of cash transfers after deeply reviewing all relevant new literature, but we expect that upcoming midline results from GiveDirectly’s “general equilibrium” study will play a major role in our conclusions. Our best guess is that the general equilibrium study and other literature will not imply that GiveDirectly’s program has large negative spillovers, but we remain open to the possibility that we should substantially negatively update our views after reviewing the relevant literature.
  • Several new studies seem to find that cash may have little effect on recipients’ standard of living beyond the first year after receiving a transfer. Our best guess is that after reviewing the relevant research in more detail we will decrease our estimate of the cost-effectiveness of cash transfers to some extent. In the worst (unlikely) case, this factor could lead us to believe that cash is about 1.5-2x less cost-effective than we currently do.

Spillovers

Negative spillovers of cash transfers have the potential to lead us to majorly revise our estimates of the effects of cash; we currently assume that cash does not have major negative or positive spillover effects. At this point, we are uncertain what we will conclude about the likely spillover effects of cash after reviewing all relevant new literature, including GiveDirectly’s forthcoming “general equilibrium” study. Our best guess is that GiveDirectly’s current program does not have large spillover effects, but it seems plausible that we could ultimately conclude that cash either has meaningful negative spillovers or positive spillovers.

We will not rehash the methodological details and estimated effect sizes of HS 2018 in this post. For a basic understanding of the findings and methodological issues, we recommend reading Özler’s posts, the Center for Global Development’s Justin Sandefur’s post, GiveDirectly’s latest post, or Haushofer and Shapiro’s response to Özler’s posts. The basic conclusions that we draw from this research are:

  • Under one interpretation of its findings, HS 2018 measures negative spillover effects that could outweigh the positive effects of cash transfers.1From Sandefur’s post: “Households who had been randomly selected to receive cash were much better off than their neighbors who didn’t. They had $400 more assets—roughly the size of the original transfer, with all figures from here on out in PPP terms—and about $47 higher consumption each month. It looked like an amazing success.
     
    “But when Haushofer and Shapiro compared the whole sample in these villages—half of whom had gotten cash, half of whom hadn’t—they looked no different than a random sample of households in control villages. In fact, their consumption was about $6 per month less ($211 versus $217 a month).
     
    “There are basically two ways to resolve this paradox:
     
    “1) Good data, bad news. Cash left recipients only modestly better off after three years (lifting them from $217 to $235 in monthly consumption), and instead hurt their neighbors (dragging them down from $217 to $188 in monthly consumption). Taking the data at face value, this is the most straightforward interpretation of the results.
     
    “2) Bad data, good news. Alternatively, the $47 gap in consumption between recipients and their neighbors is driven by gains to the former not losses to the latter. The estimates of negative side-effects on neighbors are driven by comparisons with control villages where—if you get into the weeds of the paper—it appears sampling was done differently than in treatment villages. (In short, the $217 isn’t reliable.)”
  • We do not yet have a strong view on how likely it is that the negative interpretation of HS 2018’s findings is correct. This would require having a deeper understanding of what we should believe about a number of key methodological issues in HS 2018 (see following footnote for two examples).2One methodological issue is how to deal with attrition, as discussed in Haushofer and Shapiro 2018, Pg. 9: “However, there is a statistically significant difference in attrition levels for households in control villages relative to households in treatment villages from endline 1 to endline 2: 6 percentage points more pure control households were not found at endline 2 relative to either group of households in treatment villages. In the analysis of across-village treatment effects and spillover effects we use Lee bounds to deal with this differential attrition; details are given below.”
     
    Another potential issue as described by Özler’s post: “The short-term impacts in Haushofer and Shapiro (2016) were calculated using within-village comparisons, which was a big problem for an intervention with possibility of spillovers, on which the authors had to do a lot of work earlier (see section IV.B in that paper) and in the recent paper. They got around this problem by arguing that spillover effects were small and insignificant. Of course, then came the working paper on negative spillovers on psychological wellbeing mentioned above and now, the spillover effects look sustained and large and unfortunately negative on multiple domains three years post transfers.
     
    “The authors estimated program impacts by comparing T [treatment group] to S [spillover group], instead of the standard comparison of T to C [control group], in the 2016 paper because of a study design complication: researchers randomly selected control villages, but did not collect baseline data in these villages. The lack of baseline data in the control group is not just a harmless omission, as in ‘we lose some power, no big deal.’ Because there were eligibility criteria for receiving cash, but households were sampled a year later, no one can say for certain if the households sampled in the pure control villages at follow-up are representative of the would-be eligible households at baseline.
     
    “So, quite distressingly, we now have two choices to interpret the most recent findings:
     
    “1) We either believe the integrity of the counterfactual group in the pure control villages, in which case the negative spillover effects are real, implying that total causal effects comparing treated and control villages are zero at best. Furthermore, there are no ITT [intention to treat] effects on longer-term welfare of the beneficiaries themselves – other than an increase in the level of assets owned. In this scenario, it is harder to retain confidence in the earlier published impact findings that were based on within-village comparisons – although it is possible to believe that the negative spillovers are a longer-term phenomenon that truly did not exist at the nine-month follow-up.
     
    “2) Or, we find the pure control sample suspect, in which case we have an individually randomized intervention and need to assume away spillover effects to believe the ITT estimates.”
    HS 2018 reports that the potential bias introduced by methodological issues may be able to explain much of the estimated spillover effects.3Haushofer and Shapiro 2018, Pgs. 24-25: “These results appear to differ from those found in the initial endline, where we found positive spillover effects on female empowerment, but no spillover effects on other dimensions. However, the present estimates are potentially affected by differential attrition from endline 1 to endline 2: as described above, the pure control group showed significantly greater attrition than both treatment and spillover households between these endlines. To assess the potential impact of attrition, we bound the spillover effects using Lee bounds (Table 8). This analysis suggests that differential attrition may account for several of these spillover effects. Specifically, for health, education, psychological well-being, and female empowerment, the Lee bounds confidence intervals include zero for all sample definitions. For asset holdings, revenue, and food security, they include zero in two of the three sample definitions. Only for expenditure do the Lee bounds confidence intervals exclude zero across all sample definitions. Thus, we find some evidence for spillover effects when using Lee bounds, although most of them are not significantly different from zero after bounding for differential attrition across treatment groups.”
  • The mechanism for what may have caused large negative spillovers (if they exist) in HS 2018 is uncertain, though the authors provide some speculation (see footnote).4Haushofer and Shapiro 2018, Pg. 3: “We do not have conclusive evidence of the mechanism behind spillovers, but speculate it could be due to the sale of productive assets by spillover households to treatment households, which in turn reduces consumption among the spillover group. Though not always statistically different from zero, we do see suggestive evidence of negative spillover effects on the value of productive assets such as livestock, bicycles, motorbikes and appliances. We note that GiveDirectly’s current operating model is to provide transfers to all eligible recipients in each village (within village randomization was conducted only for the purpose of research), which may mitigate any negative spillover effects.” We would increase our credence in the existence of negative spillover effects if there were strong evidence for a particular mechanism.

One further factor that complicates application of HS 2018’s estimate of spillover effects is that GiveDirectly’s current program is substantially different from the version of its program that was studied in HS 2018. GiveDirectly now provides $1,000 transfers to almost all households in its target villages in Uganda and Kenya; the intervention studied by HS 2018 predominantly involved providing ~$287 transfers to about half of eligible (i.e., very poor) households within treatment villages, and HS 2018 measured spillover effects on eligible households that did not receive transfers.5See this section of our cash transfers intervention report. GiveDirectly asked us to note that it now defaults to village-level (instead of within-village) randomization for the studies it participates in, barring exceptional circumstances. Since GiveDirectly’s current program provides transfers to almost all households in its target villages, spillovers of its program may largely operate across villages rather than within villages. These changes to the program and the spillover population of interest may lead to substantial differences in estimated spillover effects.

Fortunately, GiveDirectly is running a large (~650 villages) randomized controlled trial of an intervention similar to its current program that is explicitly designed to estimate the spillover (or “general equilibrium”) effects of GiveDirectly’s program.6From the registration for “General Equilibrium Effects of Cash Transfers in Kenya”: “The study will take place across 653 villages in Western Kenya. Villages are randomly allocated to treatment or control status. In treatment villages, GiveDirectly enrolls and distributes cash transfers to households that meet its eligibility criteria. In order to generate additional spatial variation in treatment density, groups of villages are assigned to high or low saturation. In high saturation zones, 2/3 of villages are targeted for treatment, while in low saturation zones, 1/3 of villages are targeted for treatment. The randomized assignment to treatment status and the spatial variation in treatment intensity will be used to identify direct and spillover effects of cash transfers.”
 
Note that this study will evaluate a variant of GiveDirectly’s program that is different from its current program in that it will not provide transfers to almost all households in target villages. The study will estimate the spillover effects of cash transfers on ineligible (i.e., slightly wealthier) households in treatment villages, among other populations. Since GiveDirectly’s standard program now provides transfers to almost all households in its target villages, estimates of effects on ineligible households may need to be extrapolated to other populations of interest (e.g., households in non-target villages) to be most relevant to GiveDirectly’s current program.
Midline results from this study are expected to be released in the next few months.

Since we expect GiveDirectly’s general equilibrium study to play a large role in our view of spillovers, we expect that we will not publish an overview of the cash spillovers literature until we’ve had a chance to review its results. However, we see the potential for negative spillover effects of cash as very concerning and it is a high-priority research question for us; we plan to publish a detailed update that incorporates HS 2018, previous evidence for negative spillovers (such as studies on inflation and happiness), the general equilibrium study, and any other relevant literature in time for our November 2018 top charity recommendations at the latest.

Duration of benefits

Several new studies seem to find that cash may have little effect on recipients’ standard of living beyond the first year after receiving a transfer. Our best guess is that after reviewing the relevant research in more detail we will decrease our estimate of the cost-effectiveness of cash to some extent. In the worst (unlikely) case, this could lead us to believe that cash is about 1.5-2x less cost-effective than we currently do.

In our current cost-effectiveness analysis for cash transfers, we mainly consider two types of benefits that households experience due to receiving a transfer:

  1. Increases in short-term consumption (i.e., immediately after receiving the transfer, very poor households are able to spend money on goods such as food).
  2. Increases in medium-term consumption (i.e., recipients may invest some of their cash transfer in ways that lead them to have a higher standard of living in the 1-20 years after first receiving the transfer).

Potential spillover effects aside, our cost-effectiveness estimate for cash has a fairly stable lower bound because we place substantial value on increasing short-term consumption for very poor people, and providing cash allows for more short-term consumption almost by definition. In particular:

  • Our current estimates are consistent with assuming little medium-term benefit of cash transfers. We estimate that about 60% of a typical transfer is spent on short-term goods such as eating more food, and count this as about 40-60% of the benefits of the program.7For our estimate of the proportion of the benefits of cash transfers that come from short-term consumption increases, see row 30 of the “Cash” sheet in our 2018 cost-effectiveness model.
     
    For our estimate of the proportion of transfers that is spent on short-term consumption, we rely on results from GiveDirectly’s randomized controlled trial, which shows investments of $505.94 (USD PPP) (within villages, or $601.88 across villages) on a transfer of $1,525 USD PPP, or about one-third of the total. See Pg. 117 here and Pg. 1 here for total transfer size.
    If we were to instead assume that 100% of the transfer was spent on short-term consumption (i.e., none of it was invested), our estimate of the cost-effectiveness of cash would become about 10-30% worse.8See a version of our cost-effectiveness analysis in which we made this assumption here. The calculations in row 35 of the “Cash” tab show how assuming that 0% of the transfer is invested would affect staff members’ bottom line estimates. We think using the 100% short-term consumption estimate may be a reasonable and robust way to model the lower bound of effects of cash given various measurement challenges (discussed below).
  • Nevertheless, our previous estimates of the medium-term benefits of cash transfers may have been too optimistic. Based partially on a speculative model of the investment returns of iron roofs (a commonly-purchased asset for GiveDirectly recipients), most staff assumed that about 40% of a transfer will be invested, and that those investments will lead to roughly 10% greater consumption for 10-15 years.9See rows 5, 8, and 14, “Cash” sheet, 2018 Cost-Effectiveness Analysis – Version 1. Some new research discussed in Özler’s first post suggests that there may be little return on investment from cash transfers within 2-4 years after the transfer, though the new evidence is somewhat mixed (see footnote).10See this section of Özler’s post: “This new paper and Blattman’s (forthcoming) work mentioned above join a growing list of papers finding short-term impacts of unconditional cash transfers that fade away over time: Hicks et al. (2017), Brudevold et al. (2017), Baird et al. (2018, supplemental online materials). In fact, the final slide in Hicks et al. states: ‘Cash effects dissipate quickly, similar to Brudevold et al. (2017), but different to Blattman et al. (2014).’ If only they were presenting a couple of months later…”
     
    See also two other recent papers that find positive effects of cash transfers beyond the first year: Handa et al. 2018 and Parker and Vogl 2018. The latter finds intergenerational effects of a conditional cash transfer program in Mexico, so may be less relevant to GiveDirectly’s program.
    Additionally, under the negative interpretation of HS 2018’s results, it finds that cash transfers did not have positive consumption effects for recipients three years post-transfer, though it finds a ~40% increase in assets for treatment households (even in the negative interpretation).11Haushofer and Shapiro 2018, Abstract: “Comparing recipient households to non-recipients in distant villages, we find that transfer recipients have 40% more assets (USD 422 PPP) than control households three years after the transfer, equivalent to 60% of the initial transfer (USD 709 PPP).”
     
    Haushofer and Shapiro 2018, Pg. 28: “Since we have outcome data measured in the short run (~9 months after the beginning of the transfers) and in the long-run (˜3 years after the beginning of transfers), we test equality between short and long-run effects…Results are reported in Table 9. Focusing on the within-village treatment effects, we find no evidence for differential effects at endline 2 compared to endline 1, with the exception of assets, which show a significantly larger treatment effect at endline 2 than endline 1. However, this effect is largely driven by spillovers; for across-village treatment effects, we cannot reject equality of the endline 1 and endline 2 outcomes. This is true for all variables in the across-village treatment effects except for food security and psychological well-being, which show a smaller treatment effect at endline 2 compared to endline 1. Thus, we find some evidence for decreasing treatment effects over time, but for most outcome variables, the endline 1 and 2 outcomes are similar.”
    Note that any benefits from owning iron roofs were not factored in to the consumption estimates in HS 2018.12Haushofer and Shapiro 2018, pgs. 32-33: “Total consumption…Omitted: Durables expenditure, house expenditure (omission not pre-specified for endline 1 analysis)” If we imagine the potential worst case scenario implied by these results and assume that the ~40% of a cash transfer that is invested has zero benefits, our cost-effectiveness estimate would get about 2x worse.

Our best guess is that we’ll decrease our estimate for the medium-term effects of cash to some extent, though we’re unsure by how much. Challenging questions we’ll need to consider in order to arrive at a final estimate include:

  • If we continue to assume that about 40% of transfers are invested, and that those investments do not lead to any future gains in consumption, then we are effectively assuming that money spent on investments is wasted. Is this an accurate reflection of reality, i.e. are recipients failing to invest transfers in a beneficial manner?
  • Is our cost-effectiveness model using a reasonable framework for estimating recipients’ standard of living over time? Currently, we only estimate cash’s effects on consumption. However, assets such as iron roofs may provide an increase in standard of living for multiple years even if they do not raise consumption. How, if at all, should we factor this into our estimates?
  • GiveDirectly’s cash transfer program differs in many ways from other programs that have been the subject of impact evaluations. For example, GiveDirectly provides large, one-time transfers whereas many government cash transfers provide smaller ongoing support to poor families. How should we apply new literature on other kinds of cash programs to our estimates of the effects of GiveDirectly?

Next steps

We plan to assess all literature relevant to the impact of cash transfers and provide an update on our view on the nature of spillover effects, duration of benefits, and other relevant issues for our understanding of cash transfers and their cost-effectiveness in time for our November 2018 top charity recommendations at the latest.

Notes   [ + ]

1. From Sandefur’s post: “Households who had been randomly selected to receive cash were much better off than their neighbors who didn’t. They had $400 more assets—roughly the size of the original transfer, with all figures from here on out in PPP terms—and about $47 higher consumption each month. It looked like an amazing success.
 
“But when Haushofer and Shapiro compared the whole sample in these villages—half of whom had gotten cash, half of whom hadn’t—they looked no different than a random sample of households in control villages. In fact, their consumption was about $6 per month less ($211 versus $217 a month).
 
“There are basically two ways to resolve this paradox:
 
“1) Good data, bad news. Cash left recipients only modestly better off after three years (lifting them from $217 to $235 in monthly consumption), and instead hurt their neighbors (dragging them down from $217 to $188 in monthly consumption). Taking the data at face value, this is the most straightforward interpretation of the results.
 
“2) Bad data, good news. Alternatively, the $47 gap in consumption between recipients and their neighbors is driven by gains to the former not losses to the latter. The estimates of negative side-effects on neighbors are driven by comparisons with control villages where—if you get into the weeds of the paper—it appears sampling was done differently than in treatment villages. (In short, the $217 isn’t reliable.)”
2. One methodological issue is how to deal with attrition, as discussed in Haushofer and Shapiro 2018, Pg. 9: “However, there is a statistically significant difference in attrition levels for households in control villages relative to households in treatment villages from endline 1 to endline 2: 6 percentage points more pure control households were not found at endline 2 relative to either group of households in treatment villages. In the analysis of across-village treatment effects and spillover effects we use Lee bounds to deal with this differential attrition; details are given below.”
 
Another potential issue as described by Özler’s post: “The short-term impacts in Haushofer and Shapiro (2016) were calculated using within-village comparisons, which was a big problem for an intervention with possibility of spillovers, on which the authors had to do a lot of work earlier (see section IV.B in that paper) and in the recent paper. They got around this problem by arguing that spillover effects were small and insignificant. Of course, then came the working paper on negative spillovers on psychological wellbeing mentioned above and now, the spillover effects look sustained and large and unfortunately negative on multiple domains three years post transfers.
 
“The authors estimated program impacts by comparing T [treatment group] to S [spillover group], instead of the standard comparison of T to C [control group], in the 2016 paper because of a study design complication: researchers randomly selected control villages, but did not collect baseline data in these villages. The lack of baseline data in the control group is not just a harmless omission, as in ‘we lose some power, no big deal.’ Because there were eligibility criteria for receiving cash, but households were sampled a year later, no one can say for certain if the households sampled in the pure control villages at follow-up are representative of the would-be eligible households at baseline.
 
“So, quite distressingly, we now have two choices to interpret the most recent findings:
 
“1) We either believe the integrity of the counterfactual group in the pure control villages, in which case the negative spillover effects are real, implying that total causal effects comparing treated and control villages are zero at best. Furthermore, there are no ITT [intention to treat] effects on longer-term welfare of the beneficiaries themselves – other than an increase in the level of assets owned. In this scenario, it is harder to retain confidence in the earlier published impact findings that were based on within-village comparisons – although it is possible to believe that the negative spillovers are a longer-term phenomenon that truly did not exist at the nine-month follow-up.
 
“2) Or, we find the pure control sample suspect, in which case we have an individually randomized intervention and need to assume away spillover effects to believe the ITT estimates.”
3. Haushofer and Shapiro 2018, Pgs. 24-25: “These results appear to differ from those found in the initial endline, where we found positive spillover effects on female empowerment, but no spillover effects on other dimensions. However, the present estimates are potentially affected by differential attrition from endline 1 to endline 2: as described above, the pure control group showed significantly greater attrition than both treatment and spillover households between these endlines. To assess the potential impact of attrition, we bound the spillover effects using Lee bounds (Table 8). This analysis suggests that differential attrition may account for several of these spillover effects. Specifically, for health, education, psychological well-being, and female empowerment, the Lee bounds confidence intervals include zero for all sample definitions. For asset holdings, revenue, and food security, they include zero in two of the three sample definitions. Only for expenditure do the Lee bounds confidence intervals exclude zero across all sample definitions. Thus, we find some evidence for spillover effects when using Lee bounds, although most of them are not significantly different from zero after bounding for differential attrition across treatment groups.”
4. Haushofer and Shapiro 2018, Pg. 3: “We do not have conclusive evidence of the mechanism behind spillovers, but speculate it could be due to the sale of productive assets by spillover households to treatment households, which in turn reduces consumption among the spillover group. Though not always statistically different from zero, we do see suggestive evidence of negative spillover effects on the value of productive assets such as livestock, bicycles, motorbikes and appliances. We note that GiveDirectly’s current operating model is to provide transfers to all eligible recipients in each village (within village randomization was conducted only for the purpose of research), which may mitigate any negative spillover effects.”
5. See this section of our cash transfers intervention report.
6. From the registration for “General Equilibrium Effects of Cash Transfers in Kenya”: “The study will take place across 653 villages in Western Kenya. Villages are randomly allocated to treatment or control status. In treatment villages, GiveDirectly enrolls and distributes cash transfers to households that meet its eligibility criteria. In order to generate additional spatial variation in treatment density, groups of villages are assigned to high or low saturation. In high saturation zones, 2/3 of villages are targeted for treatment, while in low saturation zones, 1/3 of villages are targeted for treatment. The randomized assignment to treatment status and the spatial variation in treatment intensity will be used to identify direct and spillover effects of cash transfers.”
 
Note that this study will evaluate a variant of GiveDirectly’s program that is different from its current program in that it will not provide transfers to almost all households in target villages. The study will estimate the spillover effects of cash transfers on ineligible (i.e., slightly wealthier) households in treatment villages, among other populations. Since GiveDirectly’s standard program now provides transfers to almost all households in its target villages, estimates of effects on ineligible households may need to be extrapolated to other populations of interest (e.g., households in non-target villages) to be most relevant to GiveDirectly’s current program.
7. For our estimate of the proportion of the benefits of cash transfers that come from short-term consumption increases, see row 30 of the “Cash” sheet in our 2018 cost-effectiveness model.
 
For our estimate of the proportion of transfers that is spent on short-term consumption, we rely on results from GiveDirectly’s randomized controlled trial, which shows investments of $505.94 (USD PPP) (within villages, or $601.88 across villages) on a transfer of $1,525 USD PPP, or about one-third of the total. See Pg. 117 here and Pg. 1 here for total transfer size.
8. See a version of our cost-effectiveness analysis in which we made this assumption here. The calculations in row 35 of the “Cash” tab show how assuming that 0% of the transfer is invested would affect staff members’ bottom line estimates.
9. See rows 5, 8, and 14, “Cash” sheet, 2018 Cost-Effectiveness Analysis – Version 1.
10. See this section of Özler’s post: “This new paper and Blattman’s (forthcoming) work mentioned above join a growing list of papers finding short-term impacts of unconditional cash transfers that fade away over time: Hicks et al. (2017), Brudevold et al. (2017), Baird et al. (2018, supplemental online materials). In fact, the final slide in Hicks et al. states: ‘Cash effects dissipate quickly, similar to Brudevold et al. (2017), but different to Blattman et al. (2014).’ If only they were presenting a couple of months later…”
 
See also two other recent papers that find positive effects of cash transfers beyond the first year: Handa et al. 2018 and Parker and Vogl 2018. The latter finds intergenerational effects of a conditional cash transfer program in Mexico, so may be less relevant to GiveDirectly’s program.
11. Haushofer and Shapiro 2018, Abstract: “Comparing recipient households to non-recipients in distant villages, we find that transfer recipients have 40% more assets (USD 422 PPP) than control households three years after the transfer, equivalent to 60% of the initial transfer (USD 709 PPP).”
 
Haushofer and Shapiro 2018, Pg. 28: “Since we have outcome data measured in the short run (~9 months after the beginning of the transfers) and in the long-run (˜3 years after the beginning of transfers), we test equality between short and long-run effects…Results are reported in Table 9. Focusing on the within-village treatment effects, we find no evidence for differential effects at endline 2 compared to endline 1, with the exception of assets, which show a significantly larger treatment effect at endline 2 than endline 1. However, this effect is largely driven by spillovers; for across-village treatment effects, we cannot reject equality of the endline 1 and endline 2 outcomes. This is true for all variables in the across-village treatment effects except for food security and psychological well-being, which show a smaller treatment effect at endline 2 compared to endline 1. Thus, we find some evidence for decreasing treatment effects over time, but for most outcome variables, the endline 1 and 2 outcomes are similar.”
12. Haushofer and Shapiro 2018, pgs. 32-33: “Total consumption…Omitted: Durables expenditure, house expenditure (omission not pre-specified for endline 1 analysis)”

Comments

  • Keith on May 4, 2018 at 6:26 pm said:

    If an iron roof is one of the first/only durable goods that households buy when they get their transfer, there has to be some value to it. It might not be measurable in consumption changes and be very difficult to attach a numerical value to not getting rained on, but I don’t think it should be discounted to 0

  • Andy J on May 5, 2018 at 2:58 am said:

    If cash transfers turn out to be substantially worse than expected *despite* having the best evidence base, are you likely to make adjustments to your estimates of *other* interventions under the assumption everyone’s been over-optimistic about them too?

  • Michael Plant on May 5, 2018 at 8:38 am said:

    Thanks for this write-up. My worry is that you didn’t discuss the really important finding, which is that cash-transfers didn’t have a mid-term (i.e. 3 year) effect on the psychological well-being index (across-village comparison, p22). The typical reason we think alleviating poverty is important is because we think people will be happier(/suffer less) if they aren’t so poor, and income is a proxy for happiness. If we find, as HS18 shows, cash transfers don’t increase mid-term psychological well-being, that is a reason to consider them ineffective.

    Of course, cash transfers might have a short-term psychological effect, and that is not to be ignored if we can find it. However, the concerns about spillovers are worrying there. Overall, it looks like there is little or no short- or mid-term psychological effect of cash transfers. That is damning. We might value poverty reduction for reasons besides increasing happiness, but it’s one of the main one.

    By contrast, there are long, very large sustained effects to treating mental health. These last up to 5 years without an obvious tail off. See: Uher, Rudolf and Barbara Pavlova. 2016. “Long-Term Effects of Depression Treatment.” The Lancet Psychiatry 3(2):95–96. Retrieved October 30, 2017 (http://linkinghub.elsevier.com/retrieve/pii/S2215036615005787).

  • What do recipients say if you ask them about temporary and permanent changes to their lives? That doesn’t answer any questions, but this research sounds like it’s still working on what questions to ask. If people buy roofs, why is that?

    If you summarize the results to the subjects, do they tell you what you’re missing?

  • Josh (GiveWell) on May 15, 2018 at 4:42 pm said:

    Hi all,

    Thanks for these comments!

    Keith: We agree that it probably doesn’t make sense to assign zero value to iron roofs even if they don’t end up having investment returns, but we have not yet decided on a framework for estimating the value of durable goods like these. This is one of our key remaining questions, and we hope to provide an update on how we’re thinking about it later this year.

    Andy: If we ended up concluding that cash is substantially less effective than we currently believe, then we would assess how this should affect our worldviews and whether it has any implications for how we’re valuing other programs, but I doubt we’d apply a blanket adjustment to all programs that we recommend solely for this reason.

    There may be fairly direct ways that our other analyses could be affected by this research on cash: e.g., if spillover effects for cash are larger than expected, then we should probably think they’re also larger than we expected for programs like No Lean Season. However, there may be other cases where our models seem independent enough that they shouldn’t be affected by a negative update on cash: for example, there is strong evidence from several RCTs that seasonal malaria chemoprevention is an effective program for reducing child deaths, and views on this kind of program seem fully distinct from one’s view of cash.

    Nevertheless, we take the point that an extremely surprising negative update might suggest that one should have more general skepticism and adjust more for “regression to the mean” across programs, so we might revisit making those kinds of adjustments in an extreme enough scenario.

    Michael: Thanks for mentioning the effects on reported psychological well-being; we also plan to review those more carefully when we revisit these papers. We expect that answering the research questions discussed in the above post will substantially affect our view on the likely welfare effects of transfers in addition to the directly reported measures, since they should give a stronger sense of the mechanisms for any well-being effects. Providing a detailed response on how to think about the importance and reliability of well-being measures compared to other kinds of outcomes is beyond the scope of this comment, but I know we’ve been in touch with you about this issue and will continue to consider how we might best do this kind of analysis.

    I also wanted to note that we hope to consider mental health interventions in the future as part of our intervention prioritization process.

    eub: We agree that asking recipients about their reasons for purchasing assets like iron roofs could provide valuable insight. Off the top of my head, I can’t recall any systematic analyses of people’s reported reasons for buying iron roofs.

  • A give directly post on the value of an iron roof.
    https://www.givedirectly.org/blog-post.html?id=2845341784910255488
    They claim that the iron roof prevents an annual ~$100 expense in repairing a thatched roof.

Comments are closed.