Fall 2020 Featured Research Profile
Impactful differences and differential impacts: Georgetown Economist Louise Laage shows how to recover average effects in panels with unobserved and unexplained sources of heterogeneity.
A researcher wants to assess the effect of maternal inputs on a child’s cognitive development. The effects might differ across families due to, for instance, variation in maternal skills. However, these skills could be unobservable to the researcher. Moreover, measurable inputs like a mother’s time spent with the child may be correlated with unobserved factors that vary over time, for instance her stress from the workplace.
Heterogeneity in the impact of these kinds of inputs can be large. Yet, in such situations the typical linear regression toolbox, even when the data is a panel, will not generally yield consistent estimates of the average effect.
GCER Fellow and Economics Professor Louise Laage examines methods for addressing problems like this in her recent paper, “A Correlated Random Coefficient Panel Model with Time-Varying Endogeneity.” Professor Louise Laage shows how the effects of variables on outcomes can be recovered when there may be substantial unobserved heterogeneity in the population.
Her analysis focusses on scenarios where the effects of a variable could vary across individuals, and the researcher has no prior information on where this heterogeneity comes from. More precisely, the researcher has no information, a priori, on the joint distribution of the regressor and this unobserved heterogeneity. Measuring the average impact is a complicated task. Moreover, the regressor itself could be correlated with unobserved variables that affect the outcomes.
In her paper, Professor Laage proposes a compelling solution to these problems. She lays out an easy-to-compute, multi-step method that exploits variations over time in the instruments and the regressors. Laage also shows that her method can be used in situations where both sample selection issues and variation in the impact of the regressors arise.