Imran Rasul On The Economics Of Poverty Alleviation & How To Foster Innovation in Organizations

Field Experiments & the Economics of Poverty Alleviation

Aiden Singh: Much of your research uses field experiments to address outstanding economic questions and to help explain empirical patterns observed in data.

One area in which you have focused considerable attention is identifying what works and what does not work in efforts to alleviate poverty.

Could you explain for our readers what field experiments are, how economists design and implement them, and how this approach first came to interest you?

Imran Rasul: Most of my research focuses on understanding how policies can be designed and implemented in ways that ultimately improve people's economic lives. Much of this work has been conducted through the lens of poverty alleviation policies. In that context, I have been evaluating different interventions to determine whether they generate lasting impacts on the outcomes we aim to improve, and whether they do so in a cost effective way.

To obtain answers that can be causally attributed to a specific intervention, an ideal setup is to use something like a randomized controlled trial. This approach is very similar to what is commonly done in medical research. A group of individuals, households, or communities is randomly assigned to receive a treatment, while a comparable group is held aside as a control. Researchers then compare outcomes between the two groups in order to identify the causal impact of the intervention.

From there, we also consider how the initial data collection and research design can help us understand the mechanisms underlying the results. In other words, if an intervention has desirable effects, we want to understand why. Likewise, if it fails to produce meaningful improvements, we want to understand the reasons for that outcome as well.

The way economists design interventions can vary significantly depending on the context. I can illustrate this with a few examples from my own experiences.

In some cases, economists become involved when an organization, such as an NGO or a government agency, has already developed a program. These partners often possess detailed knowledge of the local context and understand the constraints individuals face in their daily lives. They may have designed a program that they intend to implement or potentially scale up. In that case, the economist’s role is primarily to evaluate a pre-existing intervention, often using a method such as a randomized controlled trial.

At the other extreme, partners sometimes approach researchers without a clear sense of what the most effective intervention might be. In those cases, economists may become involved from the very beginning in designing the intervention itself. One approach I have used in some of my work is to begin with a baseline survey before committing to any particular program. The purpose of this survey is to document the main issues people face and identify potential solutions. That initial data collection can then inform the design of an intervention, which can subsequently be evaluated.

A third possibility arises when a partner approaches researchers with a broad policy objective but without clarity about how best to achieve it. For example, a partner might want to design a policy that helps firms grow but may be uncertain about which specific policies would be most effective in their context. In such cases, economists often draw on an existing body of evidence. By examining similar contexts and comparable populations, researchers can identify interventions that have previously shown promise and use that knowledge to guide the design of a new policy.

There are therefore many ways in which economists can get involved in the design of these interventions.

For me, what makes this work particularly rewarding is that it brings together several different elements. It involves working closely with partners in the field who possess deep contextual knowledge. At the same time, it requires stepping back and asking which fundamental economic principles suggest that certain interventions may be more promising than others. From the outset, we must also think carefully about how to measure outcomes. If we find that an intervention is effective, we need to understand why it works. That understanding is what allows the knowledge to be applicable in other contexts.

This means thinking early on about what data must be collected in order to trace the entire causal chain, from the intervention itself to the ultimate outcomes we care about, such as whether people escape poverty.

Another aspect of field experiments that I have always found particularly compelling is the opportunity to study outcomes over long periods of time. Many of the most interesting experiments track individuals, households, or firms for many years. This often reveals that short term impacts can differ significantly from long-term effects. Studying those dynamics helps us understand how interventions evolve over time and how unintended consequences may emerge.

Economists have made important contributions in anticipating these kinds of dynamics and incorporating them into research design from the outset. Taking a long-term perspective can ultimately help policy-makers design interventions that are more effective and sustainable over time.

- - - - - - -

The Limits of Community-Based Development

Aiden Singh: You have also examined the effectiveness of what is often described as the “local delivery model” of development intervention. Could you explain how this approach works, and whether your research has found it to be effective?

Imran Rasul: That body of literature begins with a basic observation. In many very low-income settings, the state often lacks the capacity to reach the most vulnerable populations, particularly those living in remote areas or far from administrative centers. This represents a fundamental constraint on what governments are able to do. Simply put, the state cannot effectively deliver every service everywhere.

One proposed solution has been to rely on local individuals within communities to serve as agents of change. These individuals are trained to deliver interventions directly to their neighbors. The appeal of this approach lies partly in its scalability. If it works in one location and relies on local capacity, it can potentially be replicated elsewhere without requiring a large and permanent expansion of state infrastructure. At the same time, it can build knowledge and human capital within the communities themselves. For these reasons, the local delivery model has often been viewed as a promising strategy.

There is evidence that this approach has been effective in a number of areas of economic development. For example, colleagues at the Institute for Fiscal Studies recently studied an intervention in Ghana designed to improve early childhood human capital development. In that case, mothers were trained to deliver developmental support to young children within their communities, and the program proved to be highly effective.

My own work examined this approach in the context of Uganda. We focused on a common type of intervention: the delivery of agricultural extension services to farmers. These programs have been implemented many times across Sub-Saharan Africa and other parts of the developing world. Typically, a farmer within the community is designated as a “contact farmer” or “lead farmer,” receives training, and then provides advice or inputs to other farmers.

Despite the apparent simplicity of these programs, the evidence on their effectiveness has often not been great. That was precisely why we chose to study them. We wanted to understand whether the challenges these programs face might be related to their reliance on local delivery.

Our study in Uganda used a randomized design. When we first approached communities, we randomly assigned some communities to receive the intervention and others not to receive it. Within all communities, before anyone knew whether their community would ultimately receive the program, we asked residents whether they would be willing to serve as the contact farmer if the program were implemented.

This process produced a shortlist of two individuals in each community who were willing to take on the role and who met the criteria typically used by the NGO implementing the program. From that final pair, one individual was randomly selected to become the contact farmer. The other individual effectively served as a control within the treated community.

At the same time, we conducted a detailed analysis of social networks within each community. We gathered information about who farmers turned to for advice, who their friends were, and how they interacted in different contexts, including religious and political activities. This allowed us to analyze how social relationships influenced the way the program was implemented.

With this design and dataset, we were able to examine how the selected contact farmer chose which farmers to target. For example, we studied who received seeds or advice and how those decisions were shaped by the contact farmer’s social connections.

We found that the selected farmer tended to prioritize individuals they knew well. In some respects, this is not necessarily problematic as farmers may be more successful at persuading people they know to adopt new techniques. However, when we examined the patterns more closely, we also found that contact farmers were more likely to target relatively better off farmers within their social networks and less likely to target the friends of the individual who had been the runner-up in the selection process.

This pattern was even more pronounced in communities where there was greater conflict or tension between the selected farmer and the runner-up. In other words, the more fragmented the community, the stronger the bias in how the intervention was delivered.

These findings led us to step back and consider the broader implications. When policy-makers rely on local individuals as delivery agents, they often do so precisely because the state has limited presence in those areas. However, this also means there is very limited oversight of how those agents carry out their responsibilities. Policy-makers must therefore rely on them to act in ways that align with the program’s objectives.

On the one hand, these local agents possess valuable private information about their communities. They often know which farmers are struggling and which are not. Our results suggest that they do make use of this information to some extent. On the other hand, they also rely heavily on their social relationships when deciding whom to help.

This creates an important trade-off. If the chosen contact farmer has a large and well-connected social network, the program may reach a greater number of farmers overall. However, the benefits may not be concentrated among the poorest farmers. Some resources may be directed toward individuals who are relatively less in need.

The policy question, therefore, becomes how to balance these objectives. Should interventions prioritize reaching as many individuals as possible, or should they focus more narrowly on ensuring that the most vulnerable populations receive assistance? The local delivery model highlights that this trade-off is often unavoidable.

Toward the end of the paper, we return to the broader question of whether these dynamics might help explain why such programs have often struggled to achieve consistent success. 

Our findings suggest that the answer depends in large part on the degree of fragmentation within communities. In communities that are relatively cohesive, local delivery may function reasonably well. However, when communities are divided into distinct social groups, the incentives facing a lead farmer or contact farmer can become more complicated.

In those circumstances, the contact farmer may have strong incentives to direct resources and advice toward the segment of the community with which they are most closely connected. As a result, the intervention may end up serving a particular subset of the community rather than prioritizing those who are most in need.

- - - - - - -

Vocational Training, Apprenticeships, & Long-Term Labor Market Outcomes

Aiden Singh: Closely related to economic development is the question of employment. You and your colleagues designed a field experiment to examine employment outcomes under two different types of labor market interventions: vocational training and job matching. What did your research find about the effectiveness of these two approaches?

Imran Rasul: Before discussing the matching intervention, it is helpful to begin with the comparison between vocational training and firm-sponsored training, which essentially takes the form of apprenticeships.

We focused on these two forms of training because, throughout Sub-Saharan Africa, they are the primary ways in which individuals acquire human capital after leaving the formal education system and entering the labor force. One option is to attend a vocational training institute, where individuals receive sector-specific instruction. The other, which is often more common, is to gain skills by being hired by a firm and trained as an apprentice.

Our goal was to compare the long-run impacts of these two types of training. One pathway involves attending a vocational training institute, while the other involves receiving training directly within a firm.

There are several reasons why these forms of training might produce different outcomes. When a firm trains a worker, the skills that worker acquires tend to be more specific to that firm. Those skills may be highly valuable within that particular workplace, but they may not translate as easily to other firms in the same sector. In contrast, vocational training programs tend to provide more general skills that can be applied across many different firms.

A second important difference relates to certification. Individuals who complete vocational training typically receive a certificate verifying that they have completed the program. This credential allows them to demonstrate their skills to potential employers. By contrast, workers who are trained within a firm as apprentices rarely receive a formal certificate confirming their training. The ability to credibly demonstrate one's skills can be important in the labor market, because it increases the likelihood that other employers will make job offers if they can clearly identify the worker’s qualifications.

Given these differences, and the prevalence of both types of training, we wanted to understand the returns associated with each approach and the mechanisms that might explain those differences.

We conducted a randomized controlled trial in Uganda. Young labor market entrants, both men and women, were offered the opportunity to receive vocational training. Half of the participants were randomly assigned to attend vocational training institutes, where they completed six months of intensive sector specific instruction. The training covered common occupations in manufacturing and services, including motor mechanics, catering, hairdressing, and plumbing. Participants attended these programs daily and received instruction from established vocational training institutes located throughout Uganda. At the end of the program, they received certificates verifying their training.

For the other group, we introduced an additional component. We connected them with firms that typically hire vocational trainees and offered those firms a wage subsidy if they agreed to take on the trainee and provide in-house training. We then tracked these workers over time.

In the short run, workers who entered the wage-subsidized apprenticeship pathway tended to perform better on basic measures such as employment and earnings. Firms were willing to hire them quickly because the training was subsidized, which meant that these workers found jobs almost immediately.

However, the long-run pattern looked quite different. Vocational trainees generally took longer to find their first job, but once employed, they experienced greater mobility in the labor market. Because their skills were certified and transferable across firms, they were more likely to move from one employer to another. This job-to-job mobility often occurred because other firms offered higher wages. Over time, this mobility generated a widening earnings gap between the vocational trainees and the firm-trained apprentices.

Another important mechanism involved unemployment spells. When workers lost their jobs, vocational trainees were able to return to employment more quickly because they could demonstrate their certified skills to potential employers. As a result, their periods of unemployment tended to be shorter.

By contrast, firm-trained apprentices faced a different dynamic. If they lost their jobs, whether because the firm closed, the subsidy ended, or they were dismissed, their subsequent labor market outcomes looked very similar to those of individuals in the control group who had received no training at all. Without a formal way to demonstrate the skills they had acquired, it was difficult for them to convince other employers of their qualifications.

Depending on the assumptions used in the analysis, we estimate that the internal rate of return to vocational training falls between 20-30%. These are extremely high returns compared with most other investments. The returns to firm-sponsored training are roughly half of that.

We also examined the effects of a job-matching intervention layered on top of vocational training. Among the vocational trainees, some individuals were randomly selected to receive additional assistance in connecting with employers. At the end of their training, they were asked whether they would like their details to be shared with firms that were recruiting in their sector.

We also included a small group of individuals from the control population who received matching assistance without vocational training.

The most striking result was that vocational trainees who received additional matching assistance actually performed slightly worse than those who simply entered the labor market and searched for jobs on their own.

The explanation appears to lie in expectations. After completing six months of intensive training, vocational trainees were well aware that their skills had value and expected their earnings to rise. However, they were also highly optimistic about how quickly they would find employment. Many believed they would receive a good job offer within a few months of completing the program.

In reality, even with vocational training, it typically takes longer to secure the first job. Those who received matching assistance often misinterpreted the intervention. Because their details had been passed along to employers, they believed they were very likely to receive an offer. In practice, only about one in three of them was ever contacted for an interview.

Because of these misaligned expectations, many participants became discouraged when the expected job offers did not materialize. Some began to question whether the training had been worthwhile at all. By contrast, those who searched independently maintained their initial optimism and continued looking for opportunities, which ultimately produced slightly better outcomes.

The paper therefore illustrates a couple of broader points. One is that individuals can hold biased beliefs, and that these biases are not always harmful. In some cases there are benefits to what might be described as rational exuberance: a degree of optimism, or even overconfidence, can encourage people to persist in their job search.

Attempts to intervene in the presence of certain behavioral biases can therefore have unintended consequences. If an intervention undermines that optimism, or creates misunderstandings about what is happening in the labor market or why employers may not respond, it can actually discourage individuals and lead to worse outcomes.

This project is also an example of a study that we have been following for a very long time. We have continued to track the same vocational trainees over the course of the pandemic. The original study took place about six years before the pandemic, and some of our most recent work examines what happened to those individuals during that period compared with people who had been randomized out of the vocational training program at the time of the original study.

Interestingly, more skilled workers were sometimes the first to be laid off during the pandemic. However, they were also able to find new employment relatively quickly. This further highlights the long-term value of transferable and certifiable skills in helping workers navigate shocks in the labor market.

That is part of the most recent work we are doing within the same long-running field experiment. We have now been tracking these individuals for close to a decade.

- - - - - - -

Organizational Incentives & Innovation in the Workplace

Aiden Singh: Let me end with a question that is especially relevant to my own day-to-day work as someone trying to foster innovation within an organization.

You have studied the outcomes organizations achieve with their employees after they are hired, and in particular how organizational incentives and workplace culture can either encourage or discourage innovation. Based on your research, what can organizations do to foster innovation?

Imran Rasul: This study took place in a very specific setting. It was conducted in Ghana and focused on central government ministries based in Accra. In total, we worked with roughly 40 ministries, covering nearly every major area of government. The project was carried out in close collaboration with the head of the Ghanaian Civil Service, who was interested in understanding how to make the civil service more innovative.

From the outset, he had a clear intuition. Some ministries consistently performed much better than others, even though they operated under very similar conditions. They were located in the same place, recruited individuals with similar backgrounds and skill sets, and followed the same rules and incentive structures. Yet their performance differed substantially. His question to us was simple: why was this happening, and could the lower performing ministries replicate what the most effective ones were doing?

That question became the starting point for the project. At the beginning, we did not know what the appropriate intervention would be. Instead, we began by collecting baseline data. We conducted interviews with approximately 3,000 senior civil servants across these ministries. We asked about their backgrounds, how they generated ideas, what the organizational culture was like, and how new ideas were received and shared within their workplaces. We combined quantitative evidence with qualitative evidence to develop a fuller picture of how these organizations operated.

Much of the analysis focused on divisions within ministries. A typical division consists of roughly 6-10 individuals who work closely together. For example, a ministry might have a human resources division or a statistics and data division. Within these divisions we found strong evidence of hierarchical norms governing how ideas were raised. Junior staff members were often reluctant to propose new ideas because they believed those ideas would not be taken seriously by senior colleagues. In some cases they feared their suggestions would simply be ignored, and in others they worried that appearing overly innovative might even attract negative attention in what were otherwise very hierarchical organizations.

After presenting these findings to the head of the Civil Service, we proposed an intervention. The goal was not to change the hierarchical culture itself. Instead, we wanted to work within the existing norms and constraints to see whether innovation could still be encouraged.

We designed a randomized control trial in which individual bureaucrats received training on how to develop new ideas and solve complex problems by breaking them down into simpler components. The training also included role playing exercises that focused on persuasion. Participants were taught how to present new ideas in ways that would make them more acceptable to senior colleagues. For example, directly telling a supervisor that an organization has been doing things incorrectly for years is unlikely to be effective. Instead, the training emphasized how to frame suggestions so that managers might adopt them more readily, and ideally even feel a sense of ownership over the idea.

When we presented this design, the head of the civil service suggested an additional possibility. He pointed out that the civil service already offered many training programs, but that these programs often produced limited change. In his view, the problem was that when one individual returned from training with new ideas, their colleagues had not participated in the same experience and were often unreceptive to change. He proposed that it might be more effective to train entire divisions together.

As a result, we tested two approaches. In one group, individuals received training on their own. In the other, entire divisions were trained together as teams using a very similar curriculum and set of role playing exercises.

The results were somewhat surprising. Individual-level training proved far more effective than division-level training. Civil servants who received training individually were more willing to generate new ideas, share those ideas with colleagues, and ultimately implement changes that improved public service delivery. In contrast, divisions that received group training showed little improvement, and in some cases performance even declined slightly.

To understand what was driving these results, we looked more closely at the types of ideas participants proposed during and after the training.

From earlier qualitative interviews, we knew that bureaucrats tended to think about innovation in two broad ways.

The first consisted of simple, practical improvements. These are small changes that could increase productivity without fundamentally altering the organization. For example, in one ministry, reserving a conference room required several layers of approval. An employee might have to ask their supervisor, who would then consult another department before the request was approved. By the time permission was granted, the meeting might no longer even be necessary. In a neighboring ministry, by contrast, there was simply a sheet of paper posted on the door where employees could write their name to reserve the room. These kinds of small changes may seem minor, but they can have meaningful effects on day-to-day productivity.

The second type of ideas were much more ambitious. In these cases, bureaucrats often argued that meaningful improvements required broader institutional changes or additional resources from elsewhere in government. For example, they might say that the Ministry of Finance does not allocate enough budget, or that staff are not given sufficient training. In other words, the problem was framed as something that lay outside the division itself and would require more systemic change.

When we looked at the interventions, a clear pattern emerged. Individuals who received training on their own tended to propose the first type of idea: simple, practical innovations that they could implement themselves, or perhaps with one or two colleagues. By contrast, when an entire division was trained together, participants were more likely to emphasize the second type of idea, focusing on broader systemic problems that required major institutional change.

We think this reflects the hierarchical culture within these organizations. When people are trained individually, they may feel more comfortable proposing small improvements that do not challenge existing structures. But when the entire division is present, individuals may hesitate to suggest even modest changes in front of their colleagues. Instead, discussions tend to fall back on arguments everyone already agrees with - namely, that the real constraints lie elsewhere, such as limited resources or insufficient support from central ministries.

Interestingly, this pattern appeared regardless of whether the head of the division was physically present during the training. Even when senior managers were absent, participants behaved as though their views would eventually become known within the organization. As a result, group training often reinforced existing norms rather than challenging them. In fact, we find some evidence that division-level training may actually stifle innovation to some extent.

Returning to your original question about what organizations can do to foster innovation, one way to think about this is to start from the specific context we were studying, which involved very bureaucratic organizations. Of course, that is not representative of all organizations.

A useful way to classify organizations is along two dimensions.

The first concerns how easy it is to design incentives that encourage good performance.

In some organizations, outputs are clear and measurable, so it is relatively straightforward to link incentives to results.

In others, such as bureaucracies, this is much harder. Much of the work is done in teams, outcomes are difficult to measure, and the relationship between effort and output is not always clear. In that sense, it is somewhat similar to academia: there are many things people are asked to do, not all of which produce easily measurable outputs, even though they are still valuable. So organizations vary in how easy it is to design contracts or incentive systems that reward performance.

The second dimension concerns hierarchy: how hierarchical the organization is and whether people feel able to speak up or challenge more senior colleagues.

Taken together, these two dimensions produce a simple four-part typology of organizations. Our study looks at one particular cell within that framework.

There is still a great deal of work to be done to understand the other three. For example, if we looked at some well-known startups that claim to have well-designed incentives and relatively flat hierarchies, we might expect very different results. But it would be extremely valuable to have evidence on that and to understand what is happening across all four types of organizations.

So our work focuses on one specific setting rather than trying to change organizational culture more broadly.

Another line of research might instead ask why hierarchical cultures emerge in the first place, or whether there are interventions that could change those norms.

There are many directions this work could take in the future.

——————

Editing by Harpreet Chohan.