Visiting Professor
Emmanuel Skoufias served as a Lead Economist in the Poverty and Equity Global Practice of the World Bank Group, from 2004 until his retirement in 2022. His area of expertise includes the use of microdata to analyse the determinants of poverty and household welfare, the impacts of risk and risk management strategies, and the targeting of social protection programs. We asked him to join us for a deep dive into how microdata informs policy planning and decisions.
David Austin: Thank you Professor Skoufias, for joining us today. I really appreciate it.
Emmanuel Skoufias: Thank you very much for the opportunity to be here.
David Austin: We're going to be speaking about microdata and its potential in policy planning and decisions, but I was hoping that you could start with just an explanation of what is meant by microdata in this context?
Emmanuel Skoufias: Microdata is data collected from individuals, or households, or firms, which allows one to get all these individual responses to the questions posed. For example, when you go to a household, you interview and you collect information from every member of the household, provided of course they're at the right age and they can answer the questions, or from a firm, you sometimes you get information from the owner, but also the managers, but also sometimes the workers.
So, microdata is this data collected at the unit level, basically, and of course, census is also microdata because information is collected on every household member. Now all this microdata can be, what you call "cross-sectional data," meaning that this is data collected at one point in time from many different individuals or households. Or it can be repeated cross-sectional data, which means you do the same kind of survey at different points in time, different years, or different months. Or it can be what is called panel or longitudinal microdata. Which means you interview the same individuals or the same households, at different points in time. So, you get repeated observations on the same individual.
The last one, the panel longitudinal data offers many advantages in terms of statistical analysis and sort of credibility of the results that one gets from the analysis of these data, because basically you get the same individual responding to questions in different points in time. and that allows you to control for a lot of unobservables that don't vary over time.
David Austin: Can you tell us how you use this microdata to administer greater intelligence in policy planning and outcomes?
Emmanuel Skoufias: Sure. Well, designing policy programs is one thing. Finding out what reactions these programs bring in terms of behaviour of people is a very difficult thing to do, and microdata becomes very useful in helping us determine what would be the potential impact of the public policy on outcomes of interest.
A lot of policies have failed because they didn't do a good job at predicting how individuals are going to react to the policy. So, microdata makes it much easier for us to analyse before the policy is implemented, how individuals are likely to react to the policy. And allows us to sort of estimate or predict the potential impact of the policy on outcomes of interest.
Now, the past is probably the best predictor of the future, but again, it's not the only predictor of the future. So, some things can still go wrong, but certainly with microdata and using microdata properly, in proper analysis and proper methods, one is able to get a much better sort of estimate of the impact of the policy.
Broadly defined, microdata is very useful at helping us understand the economic consequences of policies and the behavioural consequences of policies.
So, I will just expand a little bit on the economic consequences of policies.
For example, during the Covid crisis there were a lot of cash transfer programs in many countries where basically cash was provided to families, and the important thing to analyse there is whether this cash that was provided to the families or to the individuals helped the family protect itself from the Covid crisis.
Predicting the economic consequences of a cash transfer program certainly relies on microdata, using collected data from the past, for example, one can analyse, from the historical data, how households reacted to increases in their income or to, to receiving cash transfers. And that can give us sort of an estimate, ex ante or before the Covid transfer takes place, of how the welfare of the household will be affected by the cash transfer.
So, and again, it was previous evidence that cash transfers are quite effective for helping individuals maintain their welfare that led to the decision from policymakers to say, "Well, okay, the Covid crisis is hitting, we need to give cash to people to maintain their welfare because we know, we have evidence from the past that this is an easy and effective way of maintaining welfare."
So that's with regard to the economic consequences of the programs. Now, I should mention that, in the past ten to 20 years, the economics profession has sort of witnessed an explosion of studies and collection of data with a very particular study design.
These are the so-called randomised control trials, which are basically, you collect microdata, but in order to understand what the causal impact of the policy is, you divide the households that you survey into two groups, what is called the treatment group and the control group. And the treatment group sort of benefits from the policy and the control group does not benefit from the benefits of the policy or it gets delayed, it receives the transfers, for example a couple years later.
This way allows you to study, what would be the causal impact of a cash transfer on economic outcomes by having a group that receives the transfer and a group that does not receive the transfer.
This randomised control trial design comes from the medical literature where usually drugs, or the potential effectiveness of drugs is tested, by this random assignment into treatment and control groups. And then the treatment group receives the medicine of interest and the control group does not or receives a placebo.
So that's the revolution, if you like, that the economics profession, public policy profession has sort of witnessed in the last few years with regard to the uses of microdata. Microdata has always been collected for many, many years, but not necessarily under this specific type of study design.
David Austin: So, when you talked about the kind of questions that would be asked, for a household regarding cash transfer, those are the kind of questions. I assume that sometimes you see published in a newspaper that a certain percentage of a population suffered food insecurity during a certain time. Am I getting this right?
Emmanuel Skoufias: Yes. Correct. A related question, which would be very important for policy, is suppose a household suffers from food insecurity. Should I give the household food or transfer in kind, or should I give the household cash? These are two alternatives and the costs associated with these alternatives are quite different because, you know, transferring food from one place to the area that needs the food is quite costly. And there's a lot of losses involved, whereas transferring cash is much easier. Yeah? You know, just electronically transfer cash to people's accounts and you make it available.
So, these are the kinds of questions that microdata allows you to answer very easily. Especially if you have the right design like this, as I said, treatment control, randomised control, trial design and, once you learn from that studies, it helps you design policy for the future in a more effective manner.
David Austin: Okay.
Emmanuel Skoufias: I just wanted to say that another important use of microdata is for understanding the behavioural consequences.
For example, the behavioural consequences are very important for a policymaker to know so that in the future they would know what could be the likely constraint for the policy success or the likely facilitating factor for the success of the policy?
When people receive cash, for example, do they work less or they take more leisure or they just continue working the same amount? Again, prior beliefs, you know, conservatives or neo-conservatives or whatever you want to call it politically, always argue against the use of cash transfer programs because they say, "Oh, you know, the people are gonna become lazy and they're not gonna work."
And you had that, again, this argument a lot during the Covid crisis when a lot of cash transfers in many countries took place. People were arguing, "Oh, people have incentive to stay unemployed because they're gonna get unemployment benefits." So, use of microdata allows you to study these questions in much more detail. And with appropriate study design you can get answers to these questions.
Cash transfer programs have been very popular in the last 20 years, in developing countries, especially as a way of redistributing income.
And again, the argument has always been that “Oh, you know, if poor families get cash, they're gonna stop working and that's not gonna help them escape poverty. So, it's useless, don't give cash to the poor households. You don't give them fish…”
David Austin: You teach them how to fish, right?
Emmanuel Skoufias: That kind of thing.
A lot of studies have shown actually that in terms of behaviour, no, the poor actually don't change their labour supply, their work habits as a result of the cash transfer received. And so that's an important sort of behaviour or consequence. Then, you know, there's many others one can look at, whether the preferences for equality in the population sort of benefiting from the program, or in the society overall, are better, or whether there's more concern for efficiency or whether perhaps it leads to less selfishness or more selfishness. So, a lot of questions regarding behaviour can be answered by using microdata.
David Austin: Can you elaborate on an example of how microdata was able to give policymakers a good idea of how it will be received before they actually implement it?
Emmanuel Skoufias: Personally, I worked on the evaluation of what is called a conditional cash transfer program in Mexico. That was a government program that was targeting the poor areas of Mexico and then providing cash to poor families on the condition that they would send their children to school and take their children to health centres for regular check-ups.
The microdata collected and the study showed that actually households were very willing to fulfil the conditions of the program in exchange for getting the money, so that type of policy was very successful in Mexico.
Now, to what extent does context matter, right? So, whenever you get historically, you do a study in a certain population and you find, oh, there's a certain impact in that population. It doesn't necessarily follow that you're going to have the same impact or measurable equal impact, if you like, in a different context.
You know, if you go to a country in Africa or if you go to a country in Asia and you do the same program, will you get the same impact? One is to just say, okay, I don't know if this is gonna work, but we're gonna try it. That's one, that's one way of doing it.
And another one is to implement the same program in three or four different contexts. As general a context as possible, find out whether the program works in these different contexts, and then conclude with a fair amount of confidence that this indeed is gonna work no matter where you're going to implement.
And that's what is being done with these kinds of programs. It worked in Mexico, then they tried it in three or four other different regions of the world. It worked there. And then from then on it just expanded in many, many different countries, with great success in terms of design. This kind of design where you give money on the condition of sending your children to school, and the health centre check-ups is a design of policy that has been used in many, many different countries around the world.
So, then we can say that yes you can predict what the impact will be based on the historical data.
David Austin: Are there any risks that come from using microdata when you're planning or designing a policy?
Emmanuel Skoufias: Basically, there's nothing intrinsically wrong or bad with collecting microdata. I can only see benefits in that. But the problem is what you do with this microdata that you collect and for what purpose you use it. And I think the main example here is the digital data, collected from social platforms like Google or Facebook and other social media platforms where basically every click gets recorded and the location and the particular characteristics of the website that this person goes to with every click gets recorded and this information gets collected.
This data, called Big Data, is essentially microdata which, using the appropriate statistical techniques, has been used in the past very successfully for predicting or extracting the preferences of a household, trying to identify what this household likes and doesn't like.
And that has been used mainly for marketing purposes and sales, within limits. But unfortunately, there have been other actors or other players in this whole area, the prime example here is the Cambridge Analytica, which is a firm that borrowed the big data from Google and other social platforms, and it used this big, detailed data or microdata to construct psychological profiles of consumers or individuals.
But then that wasn't necessarily used for marketing and sales. It was used for influencing election outcomes.
So, it was basically an unethical use of the microdata by a certain company for a purpose that was used to influence political outcomes. And the examples here are the outcome of elections in Kenya, outcome of elections in the US. Apparently, the Trump organisation used the services of Cambridge Analytica. And in the UK, the whole Brexit sort of election outcome is believed to be influenced dramatically by the profiles Cambridge Analytica constructed. And therefore, that profile was bombarded with information and propaganda, if you like, against the EU to specific population groups that influence the outcome of the elections. Yeah. So yes, these are very terrible risks because they make you question the extent to which we have free elections and democracy in countries.
David Austin: Right. Well, let's talk about some more positive objectives. And I was hoping you could share how microdata fits into policy design when it comes towards action towards climate change.
Emmanuel Skoufias: Yeah. The collection of microdata can provide very useful information about the willingness and the ability of households to adapt to climatic shocks. So, by having detailed microdata, again looking historically, you can see how households react to, let's say, certain bad weather events.
What are some of the constraints that households face in protecting themselves from these climatic events and how properly designed policy can facilitate the maintenance of welfare in the event of climatic shocks or sometimes even the improvement of welfare in spite of climatic shocks.
Yeah. So, in that respect, microdata is particularly useful for understanding and designing adaptation policies that are effective.
David Austin: Okay. Do you have any examples you could share?
Emmanuel Skoufias: I'm currently working on a project we are trying to estimate, again with historic data, what is the impact of droughts in Africa, on poverty? Let's say a lack of rainfall by so many millimetres, how much does the poverty rate increase in a certain area? That's historic. And there's a bunch of methods and statistical techniques that one has to use to try to make this more forward looking in the sense that, we know that under climate, extreme weather events and droughts are going to be even more frequent than in the past.
So, what we do is we first estimate what the historic impact of droughts was, with microdata in the past. And then try to come up with simulations of what if the weather in the extremes, either top rainfall or lack of rainfall, becomes more frequent as it would be under climate change. And then try to predict what would be the fall in the poverty rate if these extreme events were to take place.
That gives you an estimate of how many resources would be needed to respond to this future disaster. So that's an example of the use of microdata combined with other methods and techniques of how it can be used to inform policy making in the context of climate change, in the present or in the future.
David Austin: That sounds like really important research,
Emmanuel Skoufias: Yes, we can do the research, but how do you convince the governments to take action now versus later? Before the disaster as opposed to after the disaster? There's a fundamental sort of constraint. The governments have a very short horizon and it has nothing to do with microdata. Governments would rather be seen shaking the hands of people affected by a disaster and giving them aid after the disaster, as opposed to giving them money before a disaster hits. Unfortunately, nobody has devised a way, you know, you can use microdata to change the lack of incentives in governments. To act now as opposed to after the disaster.
David Austin: There’s a lot of talk now about PeaceTech. What are your thoughts about integrating microdata into PeaceTech?
Emmanuel Skoufias: The integration of microdata into this whole peace building process, I think is very promising, the World Bank also, for example, has already managed to collect resources, and it's allocating these resources to collect household survey data from internally displaced populations from migrants, from people who moved because of climate change or because of conflict and the micro information collected from these specific population group is, is very important for us to, to understand what are the constraints and the incentives that people, these groups, it's not only data collected from these groups, but also from the hosting populations. You know, because usually whenever people move into an area, there's always resentment from the population in the area that was there before about all these migrants. So again, collecting the microdata provides the opportunity to understand what the windows of opportunity are for integrating the two groups.
And in taking advantage of the various sorts of assets and skills that the migrants bring to make the situation in the hosting area more productive and more welcoming. And that's to the betterment of, of everybody, not just the migrants, but also the, the hosting communities. So, microdata collected from these kinds of populations and groups is particularly informative for the design policy and the strengthening of peace efforts.
David Austin: Very good. Yes. You describe it as "Exposing the windows of opportunity." I think that seems to kind of describe a lot of the research that you've discussed, is that it's really letting policy makers know what's possible and where they can help the most.
Thank you so much for sharing all your insights with us, is there anything else that you wanted to say?
Emmanuel Skoufias: Just one more thing, which is, microdata in itself sometimes is, is not enough. The study design is also very important, to make it relevant for answering with confidence the impacts of a policy, But also microdata becomes even more powerful and more useful when it can be combined with other sources of data, like administrative data or satellite data, for the purposes of addressing the questions related to climate change that make it very useful and powerful.
David Austin: Very good. Well, thank you so much.
Emmanuel Skoufias: Thank you very much for the opportunity.