Published on Development Impact

Using lab experiments to help understand why more people don’t migrate

This page in:

The number of international migrants in the world is estimated at around 272 million, which is up 51 million since 2010. One reason to study migration is not because this number is so high, but rather because it is so low! Given the vast differences in income across countries, the big question is why more people don’t move? Moreover, those who do migrate are positively selected on education from almost every sending country, despite returns to skill often being higher in many developing countries than in the main migrant destinations. These patterns of low levels of migration and positive sorting on skills present a challenge to classic theories of migration decisions that compare the income gained from migration to the cost of moving, such as the income maximization approach of Sjaastad or the Roy-model selection approach of Borjas.

Of course, your immediate answer might be “they don’t move because policy barriers prevent them”. Such barriers are certainly important, and can be incorporated into the standard framework as imposing large migration costs (such as the costs of visas, of getting skills certified, of recruitment agency fees, or of paying for smugglers and other costs of irregular movement etc.). But they are not the whole story. After all, we still see many people not migrating in areas of free movement with big income gaps, such as within the European Union, or between Micronesia, Puerto Rico and the United States. We can then think of many other reasons why people might not move, including the value attached to home, liquidity constraints, imperfect information, and risk. But teasing these different explanations apart and understanding their relative importance is empirically very difficult.

A migration lab experiment

In a new working paper, Catia Batista and I use lab experiments to investigate how potential migrants decide between working in different locations. We focused on undergrads in their last year of tertiary study, since the peak age for international migration is 22 to 24 years, and this was a population that would be making work and location choices in the near future. We worked in two locations – Lisbon, with 154 students from Catia’s home institution, the Nova School of Business and Economics; and Nairobi, working with the Busara Center to recruit 265 students from Strathmore University and the University of Nairobi.

Participants were given the following instructions to motivate the sets of decisions they were going to make:

We are interested in learning how people make decisions between different places to work. Imagine that you have just accepted a job offer from a multinational company which has branches around the world. Your employer has told you that you need to accept a company transfer to a different country for a year, but is offering you the choice between destinations. You should assume that everything else about the job and living conditions will be the same across destinations, apart from the information the employer tells you.

You will start the game with a certain endowment. You can use this money to pay the costs of moving, or to acquire information. This is reset for each decision. If we end up playing a game for real, any amount left from your endowment that you do not spend will be added to your game winnings. You should make each decision as seriously as you can, since at the end of the game we will randomly choose one of your choices to play for real money.

We then randomly assigned individuals an observed (to the econometrician) skill level, a second idiosyncratic unobserved skill level (as in the Borjas sorting model), and a random endowment of wealth. The wage offer they would get in each destination was then determined by these skill levels, and the wealth could be used to pay the costs of moving, and in some decisions, to also pay the costs of acquiring more information about a destination.

We then had participants make decisions between locations as we introduced different features to the choice. The most basic version involved just choosing between two destinations, with known wages at both and a cost of migrating to one of them. More complicated versions then added liquidity constraints, the risk of being unemployed and some social insurance, imperfect information, and multiple destinations. The order in which destinations were listed, and the order of the different blocks of games were randomized. Figure 1 below gives an example of one of the more complicated versions, where they could choose between three destinations, don’t have enough endowment to pay the moving costs to one of them, and face imperfect information and the choice of whether to pay to acquire some information. While the first screen looks complicated, they then get the decision summarized on a second screen that makes it easy to compare.

Figure 1: Example of multi-destination choice

Migration lab instructions 1

Migration lab instructions 2

What did we learn?

By varying the features of destinations and being able to randomize the skill levels and endowments people have, we are able to test several different theories of migration and learn about what drives the decision to migrate. Some of the key results are as follows:

1.       Adding additional real-world features like risk, liquidity constraints, and incomplete information dramatically lowers migration rates and makes selection less negative. Figure 2 provides an illustration. In our simplest game, in which participants just choose between getting a wage at home, or paying a cost and migrating to get a different wage abroad, we see very high levels of migration from the low-skilled, who face higher wages abroad – 97%  of those assigned the lowest skill level choose to migrate in Lisbon and 68% in Nairobi. But by game 8, where we have added these additional real-world features, migration rates are much lower, and less skill-selective.

Figure 2: Migration rates fall dramatically and become less skill-selective when we add more real-world features 

Migration rates in different games

Notes: Lisbon and Nairobi are the two different samples. See paper for details of games.

1.       Migration rates fall most when risk is introduced. We consider choices across our 27 different variants of the game and can test which features change migration choices the most. Adding either an exogenous or endogenous risk of unemployment in the migration destination lowers the migration rate - by more than 40 percentage points for the Lisbon sample, and by 19 (endogenous risk) to 26 (exogeneous risk) percentage points in the Nairobi sample. Liquidity constraints lower migration rates by 6 to 10 percentage points.

2.       We can directly test the independence of irrelevant alternatives assumption and find when it holds and when it doesn’t. The IIA assumption is a mainstay of both micro-level modeling of individual-level destination choices as well as providing the foundations for cross-country macro-level gravity equation models of migration stocks or flows. We can directly test this assumption by adding additional destinations and seeing whether it changes the relative preferences people have between the original destination set. We find that IIA holds for most people when migration decisions just involve wages, costs, and possible liquidity constraints, but starts to break down for some individuals when risk and incomplete information are added.

3.       We uncover an alternative to income maximization for some people, which is cost minimization. One of the puzzles in our Nairobi sample was individuals choosing not to migrate, even when there was a net gain to be had from doing so. In qualitative work, we learned that there was a particular aversion to paying costs for some students, and this drove their decision-making.

4.       Finally, we find evidence of home bias. We played a 28th game, which had individuals choose between two identical destinations, where the only thing that varied was calling one of them “home”. Simply labeling a destination as home without changing the payoffs reduces migration to other destinations. Moreover, we then follow these students six months after graduation, and find this home bias in the lab is a significant predictor of subsequent real-world migration behavior. Participants displaying “home bias” in the lab were less likely to look for jobs abroad by 29p.p. and 17p.p. in the Lisbon and Nairobi samples, respectively.

This was my first time doing a lab experiment, and I enjoyed learning how to use this method and think we gained some useful insights about the complexities involved in migration decisions. Hopefully others think this was useful/interesting as well – and if so, this set-up would lend itself to investigating other populations of potential migrants (e.g. those signing up with recruiters) and to testing additional aspects of migration behavior (e.g. by allowing for multi-year dynamics and for networks to form and evolve). Hopefully this is then just the first of such studies.



David McKenzie

Lead Economist, Development Research Group, World Bank

Join the Conversation

The content of this field is kept private and will not be shown publicly
Remaining characters: 1000