Introduction

The professor introduces the topic of poverty traps and how to test for them. He also discusses the relationship between income, nutrition, and productivity.

• Poverty traps can be tested by examining the mapping between income or assets today and tomorrow.

• The effect of nutrition on productivity is there but not strong enough to outweigh the effect of income on nutrition.

• The idea of poverty traps is prevalent in different areas of economics, leading to programs designed to combat it.

Testing for Poverty Traps

The professor explains how to test for poverty traps and why it's important.

• One way to test for a poverty trap is by looking at the relationship between today's assets and assets next year.

• If there is a crossing of the 45-degree line from below, it ensures that there are two steady states: one low and one high.

• Understanding poverty traps is important because it informs policy-making decisions.

Graduation Approach

The professor introduces the graduation approach as a recent set of antipoverty programs designed to fight poverty.

• The graduation approach was designed by BRAC in Bangladesh as a way to reach out to some of the poorest people in villages who were not interested in getting microcredit loans.

• By giving people an infusion of capital as a gift, not as a loan, and helping them start small businesses, they can escape poverty in a durable way.

• The program has been expanded into other countries in the global South.

Ultra Poor Programs

The professor discusses ultra poor programs and their role in presenting evidence for or against reduced form views of poverty traps.

• Ultra poor programs are inspired by the idea that there might be a poverty trap.

• These programs have given an opportunity to directly present evidence for or against a reduced form view of poverty traps.

• The graduation approach is at the heart of ultra poor programs.

BRAC

The professor provides more information about BRAC, the organization that designed the graduation approach.

• BRAC is a microfinance organization and NGO in Bangladesh that runs various programs such as microcredit, schools, health centers, antidiarrheal programs, and more.

• They designed the graduation approach as a way to reach out to some of the poorest people in villages who were not interested in getting microcredit loans.

• Their founder passed away a couple of years ago and was a wonderful individual.

Conclusion

The professor concludes by stating that they will describe the program in more detail and try to answer whether or not there is a poverty trap.

• The program will be described in more detail later on.

• The question they will try to answer is whether or not there is a poverty trap.

The Graduation Approach

In this section, the speaker explains how the graduation approach works in identifying and supporting the poorest of the poor in a village.

Identifying the Poorest of the Poor

• The community comes together to identify who they consider to be the poorest of the poor in their village.

• A participatory resource mapping exercise is conducted where people are ranked by wealth using cards.

• The first half of people are cut in half by wealth, then again until four or five people are identified as ultra-poor.

Supporting the Ultra-Poor

• BRAC visits these families and provides them with an asset which they choose from a menu. This could be cows, goats, sewing machines or money to start a small business.

• They receive income support for a few weeks depending on their assets and technical support to take care of their assets.

• Group meetings are held where there is literacy training, small health components and encouragement for regular savings.

Poverty Trap Model

In this section, the speaker discusses what form a poverty trap takes and how it can be proven through programs like the graduation approach.

Key Predictions of Poverty Trap Model

• One key prediction is that if someone receives capital that increases their income in one period, it should not diminish over time if they are caught in a poverty trap.

• Another key prediction is that if it's truly a poverty trap, then positive shocks experienced by others should not affect those caught in it negatively over time.

Poverty Traps and Asset Accumulation

The professor discusses the potential for heterogeneity in asset accumulation post-program, based on a poverty trap model. The audience asks questions about the impact of the program on the poorest individuals and whether there is potential for capture or manipulation.

Potential Heterogeneity in Asset Accumulation Post-Program

• A poverty trap model based on low return to capital at very level, followed by a steep increase, followed by low again could result in heterogeneity of asset accumulation post-program.

• Long-term program effects could reveal divergence between households that are above or below the poverty trap.

• Short-term effects may not be obvious at first.

Impact on Poorest Individuals

• Program effect may be larger for households that are not so poor that their base level of assets plus the asset received bring them above the poverty trap.

• Verification ex-post is done to check that households receiving transfers are poor enough. Some households may refuse transfer even if they are poor enough to receive it. Capture and manipulation are concerns but probably not significant issues.

Poverty Trap and Impact Evaluation

In this section, the speaker discusses the poverty trap and impact evaluation of a program aimed at breaking it. The speaker also introduces the concept of potential outcomes and Stable Unit Treatment Value Assumption (SUTVA).

Large Effect of Programs on Poverty Trap

• The programs have large effects that are very persistent over time, suggesting that there might be a poverty trap at play.

• There is an S-shape relationship between capital today and capital tomorrow, which could underlie this persistent effect.

• Nutrition and credit markets are ruled out as possible causes for the poverty trap.

Expensive Intervention

• The intervention is super expensive compared to other alternatives such as giving cash to people or directly affecting their health or education.

• The theory of change underlying the intervention is that it's one big push that will pay off over the lifetime of the person and potentially their children as well.

Evaluating Impact

• The first question is whether the intervention works in breaking the poverty trap.

• To evaluate impact, we need to consider what it means to have an effect and what impact means.

• We want to know whether our particular treatment has an impact on a specific outcome Y.

• Imbens and Rubin hold up potential outcome if treated as something we can observe by multiplying it with a dummy variable indicating treatment status.

Stable Unit Treatment Value Assumption (SUTVA)

• SUTVA assumes that someone else's treatment doesn't affect my treatment, allowing us to write potential outcomes as just my treatment and not everyone else's.

• SUTVA is necessary to avoid having multiple potential outcomes for each person depending on the treatment status of others.

Introduction to Treatment Effects

In this section, the speaker introduces the concept of treatment effects and explains why it is important to understand them.

Understanding Treatment Effects

• The speaker explains that in order to understand treatment effects, we need to consider SUTVA (Stable Unit Treatment Value Assumption).

• It is impossible to observe both potential outcomes for an individual, so we can only estimate the treatment effect.

• We may be able to estimate treatment effects for individuals with similar observable characteristics if we have enough data.

• We can also be interested in average treatment effects for the population or for those who received the treatment.

• Personalized medicine and machine learning aim to bypass the limitations of estimating individual treatment effects by exploiting old information about individuals.

• Quantile treatment effect allows us to compare how different quantiles are affected by a certain treatment.

Estimating Average Treatment Effects

• To estimate average treatment effects, we start by looking at the difference between means of treated and control groups.

• We can add and subtract an object that is not observed but conceptually exists - the average of untreated outcome for treated population.

Selection Problem

The section discusses the selection problem in program evaluation and how it affects the estimation of treatment effects.

Potential Outcome Framework

• The potential outcome framework is used to estimate treatment effects.

• It assumes that each individual has two potential outcomes, one for receiving treatment and one for not receiving treatment.

• The difference between these two outcomes is the treatment effect on the treated.

Selection Problem

• The selection problem arises when individuals are not randomly assigned to receive treatment.

• This can lead to differences between the treated and untreated groups that are not due to the treatment itself.

• These differences can bias estimates of treatment effects.

Best Case Scenario

• In the best case scenario, the program was randomly assigned.

• This means there should be no systematic differences between those who received treatment and those who did not.

• In this case, an unbiased estimate of the effect of the treatment on the treated can be obtained.

Conditional Independence Assumption

• When random assignment is not possible, a conditional independence assumption can be made.

• This assumption states that once all relevant covariates are controlled for, there is no systematic difference between those who received treatment and those who did not.

• If this assumption holds, an unbiased estimate of the effect of the treatment on the treated can still be obtained.

Matching and Research Design Strategies

In this section, the speaker discusses different matching estimates and research design strategies used to solve selection bias in samples.

Matching Estimates

• Different types of matching estimates are more or less parametric and believable.

• The underlying assumption is always that of unconfoundedness.

Double Machine Learning Estimators

• Double machine learning estimators add sophistication to econometric analysis but are conditional on assumptions.

• Examples include double post-lasso and generic double machine learning method.

Research Design Strategy

• Random assignment is the easiest research design strategy.

• It solves selection bias in a sample because it's a known assignment rule.

• However, there are issues with randomized evaluation such as context dependence and noncompliance issues.

Concerns with Randomized Evaluation

• Results may not be predictive of what would happen in another setting due to context dependence.

• Noncompliance issues such as some people refusing treatment or sneaking into the program can affect results.

Randomized Controlled Trials: Issues and Concerns

In this section, the professor discusses some of the issues and concerns related to randomized controlled trials (RCTs). These include site selection bias, cherry-picking outcomes, uncertainty, compliance, spillovers, external validity, and power issues.

Site Selection Bias

• RCTs may be biased towards selecting sites where treatment effects are highest.

• This can lead to overestimating treatment effects in those specific sites.

• A paper by Hunt Alcott showed that the first experiment by Opower had large effects in the first few sites but much lower effects in later sites due to site selection bias.

Cherry-Picking Outcomes

• With many outcomes, there is a risk of cherry-picking for only a few outcomes.

• People have limited attention span and cannot focus on too many outcomes at once.

• This issue is particularly relevant for programs like the ultra poor program which measures improvements in many ways.

Uncertainty and Compliance

• There is always uncertainty associated with RCT results due to sample size limitations.

• Compliance issues can also arise if participants do not follow through with the treatment as intended.

Spillovers and External Validity

• Spillover effects can occur when participants who did not receive treatment still benefit from it indirectly.

• External validity refers to how well RCT results generalize beyond the specific context of the study.

• Both spillovers and external validity are important considerations when interpreting RCT results.

Power Issues

• A criticism against experiments is that they are generally too small with low power.

• This means that significant results can only be obtained if treatment effects are very large.

Randomization and Spillovers

In this section, the professor explains how randomization was done in the Bangladesh studies. He also discusses how randomizing at the site level and then individually within site allows for checking spillovers.

Randomization Process

• Individuals were randomized at the individual level within village.

• A couple of villages were randomized at both levels.

• Randomization was done first at the site level, and then within site, at the individual level.

Checking for Spillovers

• The reason for randomizing first at the site level is to check for spillovers.

• By comparing untreated people in treated villages with their counterparts in control villages, it is possible to directly check for spillovers.

• There is no evidence of spillover in this case.

Results from Bangladesh Studies

This section presents some of the outcomes used in the Bangladesh studies.

Outcomes Used

• Outcomes used include below poverty line, consumption expenditure per capita or per adult equivalent, value of household assets, savings, whether or not they get loans, whether or not they give loans.

• Each row is a treatment effect with standard error and stars indicating significance.

Treatment Effects

• Treated households had on average 8.4% lower people below poverty line four years after program impact compared to control group (62%).

• This represents a 13.5 percentage reduction.

• Impact seems to increase over time.

• R-square is included along with number of women, observations in households and clusters.

Varying Fraction of People Treated

In this section, the professor discusses how varying the fraction of people treated can be used to study individual and market effects.

Varying Fraction of People Treated

• It is possible to vary the fraction of people treated in order to study equilibrium-type effects on prices and other factors.

• Sometimes researchers want to rule out externalities because they are interested in individual effects.

• Other times, they are interested in market effects and want to make sure that enough saturation is achieved.

• In the case of ultra poor studies, very few people were treated in each village, which may explain why there were no externalities observed.

The Context Dependent Project

In this section, the speaker discusses a project led by Dean Karlan and supported by six different organizations in six different countries. The goal of the project was to estimate the same program across all locations while accounting for contextual differences.

Same Program Across Different Locations

• The goal was to estimate the same program across all locations.

• Sameness was enforced through annual meetings where people from all over the program met in Paris to discuss their processes.

• Despite contextual differences such as staff pay and asset value, they arrived at the same program given each context.

Writing One Paper

• Instead of each team writing their own paper, a decision was made to write one paper with all data together.

• This addressed cherry-picking of results since everyone had to follow the same template.

Index Used

• An index was used that indexed components related to food security.

• This is how literature has evolved and is considered standard practice.

Results

• Effects were found across pretty much all outcomes with an effect for consumption at about 0.12 standard deviation.

• Q-value adjusted p-value was used as a method to account for false discovery rate.

External Validity and Country Heterogeneity

In this section, the speaker discusses how the results of an experiment can vary from country to country. They also talk about the impact of cash transfers versus asset transfers.

Variability in Results Across Countries

• The results of an experiment can vary from country to country.

• Honduras had a poor outcome due to all the chickens dying because of bird flu.

• In India, they avoided giving out chickens because they were deemed dangerous.

Cash Transfers vs Asset Transfers

• There is a question about why assets are used instead of cash transfers.

• The speaker explains that once it's established that there is an impact, policymakers may want to consider taking out less effective components or switching to cash transfers.

• It's important to understand the mechanism behind why certain things work together and what can be learned from these experiments.

Country Heterogeneity

• The speaker talks about how each site in an experiment is different for two reasons: treatment effect and noise in the world.

• Different treatment effects across countries could be due to fundamental differences or sample differences between treatment and control groups.

• Treatment effects across countries are difficult to compare since every estimate comes with its own standard error.

Bayesian Hierarchical Analysis

In this section, the speaker discusses the limitations of attributing differences in treatment effect to heterogeneity alone. They introduce Bayesian hierarchical analysis as a method for adding structure to the problem and understanding the concept of distribution of possible treatment effects.

Understanding Bayesian Hierarchical Analysis

• Bayesian hierarchical analysis is a method for adding structure to problems where there is underlying noise in the world.

• The basic idea behind Bayesian hierarchical analysis is to assume that there is a distribution of possible treatment effects, which are themselves drawn from a normal distribution.

• By finding out the variance in treatment effect across sites in the world, it's possible to estimate one pool treatment effect and use information from all other countries to increase precision.

• If there is huge variance in treatment effect itself, then there isn't heterogeneity and predictability of next site decreases.

Results of Bayesian Hierarchical Analysis

• The results show that assuming one treatment effect for the whole world decreases confidence in results. Predictive effects are quite wide with large variability in treatment effect from site to site.

• Despite this heterogeneity, most effects are positive. This result differs from what was found for microcredit studies where pooling was more common.

• Making assumptions about uniformity changes estimates because one observation becomes many observations.

Exploring Heterogeneity in Treatment Effect

In this section, the speaker discusses a paper that explores whether covariates can explain heterogeneity in treatment effect. The paper aims to predict heterogeneity in treatment effect across sites based on the fact that the covariates of the sample were different.

Covariates and Heterogeneity

• The paper tries to see if they can use heterogeneity in treatment effect by covariates to predict heterogeneity in treatment effect across sites.

• Using covariate data to predict different treatment effects is not very effective, even within countries.

Cost-Benefit Analysis

• The program being discussed is super expensive and only worth it if benefits persist into the future.

• A table is presented that calculates how long benefits need to last for the program to be cost-effective.

• An empirical question arises as to whether benefits persist over time.

Persistence of Effects

• Data from a study shows persistence of effects between year 7 and year 10, even with economic growth.

• By year 10, even if the effect were going to disappear fully, the program would have paid off for itself.

Outcome Analysis

• Treatment effects persist over time except for financial inclusion where there was never much impact.

The Mechanism of Persistence in Anti-Poverty Programs

In this section, the speaker discusses the mechanism behind the persistence of anti-poverty programs and how diversification from agriculture to non-agricultural businesses can lead to more stable wage work.

Persistence of Anti-Poverty Programs

• The persistence of anti-poverty programs is achieved through diversification from agriculture to non-agricultural businesses and then from small business to more stable wage work.

• This persistence is achieved via the following generation, which moves for a longer period, further away, and earns more money as they do that.

• These programs have highly heterogeneous effects by country and people. Treatment effect tends to be towards the highest quantile with less effect at the lowest quantile of moving the distribution.

• A key question for future program perspective is how to identify people with very high return.

S-Curve Analysis

• An attempt was made by researchers to draw an S-curve or see whether there is an S-curve at the individual level. Some people have been quite skeptical that those existed.

• In observational data, it's difficult to capture an S-curve in real-world situations unless it's actually individual-specific because people would have moved out of it by now.

Ultra Poor Program

• The ultra-poor program helps us nicely because it identifies people who are most likely to benefit from anti-poverty programs.

Ideal Experiment for Testing S-Curve

The ideal experiment for testing the S-curve would have been to give different amounts of money or goods to different people who start from the same assets.

• Different amount of money to people who start from the same assets would have been the ideal thing to do.

• In order to do this exercise, you have to believe there is one threshold out of which that people can jump over or not.

• There are two stable steady states in the real world and a lot of people at each of them, and then not too many people in the meantime.

• It's going to be very difficult to catch them because they don't see very many people there.

Steady States and Threshold

The model has two steady states - one low and one high - including an unstable one. People face the same thing but are shocked, which means that at a given point in time, they end up with either a high or low level of assets.

• In the model, there is two steady states--one low one, one high one, one unstable steady state, or two or three.

• Every person faces the same thing but then they are shocked which means that at a given point in time they end up with either a high or low level of assets.

• There are only two stable steady states in the real world: every person faces the same thing but then they are shocked which means that at a given point in time they end up with either a high or low level of assets.

Regression Exercise

The difference in assets is defined as delta i, and k star is the threshold level of capital. People grow faster when their level plus treatment is above the threshold in the treatment villages compared to the control villages.

• Define delta i as the difference in assets. And then k star is the threshold level of capital, which they estimate from that regression.

• People grow faster when their level plus treatment is above the threshold in the treatment villages compared to the control villages.

Poverty Trap and Psychology

In this section, the speaker discusses the poverty trap and how it may be related to psychology.

Poverty Trap

• The treatment for poverty is not food or credit.

• Frank will come for two lectures to talk about behavior as a potential explanation for the poverty trap.

• The poverty trap might have impacts that are functionally equivalent to the ideas that were in the nutrition-based poverty trap that comes from people's reasoning, as opposed to something else.

• For the ultra poor, there might be something like that which is really at play.

Unpacking

• There should be more energy and motivation to find what it is that can be at play, hence the unpacking.

• Several points will give us a chance to unpack the ultra-poor program.

Conclusion

The speaker concludes by saying that next week Frank will continue talking about behavior and he'll see everyone again in two weeks.