Random assignment is widely seen as one of the strongest tools for identifying cause and effect in economics. A new study shows that this confidence may be misplaced when researchers use random placement to study the impact of local conditions on individual outcomes.
The discussion paper “Random Placement but Real Bias” by Marco Schmandt, Constantin Tielkes, and Felix Weinhardt examines whether random placement truly delivers unbiased estimates of local effects. The authors ask a simple but important question: even if people are randomly assigned to places, can estimates of local conditions – such as unemployment or social attitudes – still be misleading?
Random placement policies are commonly used in research on education, neighborhoods, and migration. A prominent example is the allocation of refugees to regions, where governments assign people to locations rather than allowing them to choose freely. Because the assignment is random, many studies treat it as an “ideal experiment” for measuring the effects of local conditions on later outcomes.
The paper challenges this assumption. The authors argue that random placement breaks the link between individual characteristics and places, but not between places and the many characteristics that define them. A region is not just one feature, such as a labor market; it combines many related factors that often move together.
To clarify this problem, the authors develop a simple theoretical framework. It shows that standard estimates based on random placement typically combine three elements: the true causal effect, bias from correlated local factors, and bias from later mobility. The first bias, called multiple-treatment bias, arises because local characteristics – like unemployment, income, and political attitudes – are often correlated. The second, mobility bias, occurs when people move after initial assignment, which changes the local conditions they actually experience.
Importantly, the paper does not remain at a theoretical level. The authors propose a practical checklist that researchers can use to assess whether these biases are likely to matter in a given study. The checklist relies on observable data, such as correlations between local variables and patterns of post-assignment mobility.
The empirical analysis applies this framework to Germany, which has one of the largest refugee populations in Europe. Refugees arriving between 2015 and 2018 were assigned to counties under a dispersal policy that is close to random. The study uses administrative panel data from the Central Register of Foreigners, covering more than 69,000 individuals and tracking their locations and legal outcomes over time.
This data is particularly informative because it captures all subsequent moves. About one third of refugees leave their initially assigned county, often early after arrival. This makes it possible to study mobility directly, rather than treating it as a minor complication.
The authors focus on a clear outcome measure: whether a refugee obtains a permanent settlement permit. This permit reflects several aspects of integration, including employment, language skills, and legal stability. It therefore serves as a broad indicator of successful integration.
What the researchers find
Even in this strong random-placement setting, both multiple-treatment bias and mobility bias are substantial. Estimates of common local factors – such as the unemployment rate, the size of co-ethnic networks, or the local vote share for a right-wing party – vary widely depending on which other local variables are included. In some cases, estimated effects shrink to zero or even reverse sign.
Adding more controls or using regional fixed effects does not solve the problem. Local characteristics remain correlated over time, and mobility continues to link initial conditions to later outcomes. As a result, different reasonable model specifications can lead to very different conclusions about which local factors matter for integration.
What can we learn about policy-variables?
Policymakers might not care about the causal effect of local factors. It could be enough to know that a specific factor is linked to better outcomes, independent of the precise mechanism. The study shows that local conditions relate to each other differently in different settings: For example, the right-wing vote share is positively related to the unemployment rate in Germany, not related in Denmark and Sweden, and negatively related in Switzerland. As a result, even interpretations that ignore causality cannot inform future policy without additional assumptions about multiple treatments and onward mobility.
The implications go beyond refugee research. Many studies in economics rely on random placement to estimate peer effects, neighborhood effects, or regional influences. The paper shows that random placement alone does not eliminate the need for strong assumptions about correlated local factors and post-assignment mobility.
For policymakers, the findings suggest caution. Policies based on estimates from random-placement studies – such as where to place refugees or which local investments matter most – may rest on fragile evidence if these biases are ignored. More generally, the study encourages researchers to look beyond simple randomization tests and to assess the full structure of local conditions and mobility.
Conclusion
Random placement is valuable, but it is not a shortcut to causal certainty. Understanding how local factors interact and how people move between places is essential for drawing reliable conclusions about policy-relevant effects.
About the Authors
Marco Schmandt
Doctoral Researcher at Technische Universität Berlin. His research focuses on applied econometrics, housing markets, and migration, with particular interest in housing affordability and internal migration.
Constantin Tielkes
Economist at Europa-Universität Viadrina. His research interests include labor economics, migration, and applied econometrics.
Felix Weinhardt
Professor of Economics (Public Economics) at the European University Viadrina and affiliated with the Berlin School of Economics, IZA, CESifo, and CEP at the LSE. His research covers education economics, migration, and urban and regional economics.
