Three essays in applied microeconometrics

This dissertation contains three essays on problems in applied microeconometrics that arise from common shortcomings of survey data. After an introduction that situates the three essays in the literature on data problems, the first essay, “Misclassification in Binary Choice Models” (with Bruce Meyer), discusses the problem of misclassification of the dependent variable in binary choice models. We derive the asymptotic bias from misclassification that may be arbitrarily correlated with the covariates and the error term. Monte Carlo studies and an application to food stamp take-up show that the bias formulas can be used to interpret biased coefficients and conduct sensitivity analyses. The second part of the paper uses validation data to evaluate estimators for binary choice models that are consistent under misclassification. Estimators based on the assumption that misclassification is independent of the covariates are sensitive to their functional form assumptions and tend to aggravate the bias if the independence assumption is invalid. On the other hand, estimators that allow correlation with the covariates perform well if a model of misreporting or validation data is available. This underlines that simplifying assumptions should be avoided unless there is strong evidence that they hold, but sufficient information about the misclassification can be a good substitute for clean data. We propose two tests for misclassification that can help to choose an estimator.;The second essay, “Imputations: Benefits, Risks and a Method for Missing Data”, examines methods to deal with missing variables and missing observations. While the conditions under which missing data does not lead to bias as well as the conditions for missing data methods to yield consistent estimates are well understood, they often do not hold in practice. In order to provide guidance on whether to omit the missing data or to apply a method for missing data, the paper first examines conditions under which these methods can improve estimates and then discusses biases from frequent violations of their assumptions. I then discuss advantages and problems of common missing data methods. Two important problems are that most methods work well for some models, but poorly for others and that researchers often do not have enough information to use imputed observations in common data sets appropriately. To address these problems, I develop a method based on the conditional density and apply it to common problems to show that it works well in practice.;The third essay, “A Method of Correcting for Misreporting Applied to the Food Stamp Program”, demonstrates the consequences of misreporting in survey data and proposes a method to address this problem based on validation data. I first use administrative data on food stamp receipt and benefit amounts linked to American Community Survey data from New York State to show that survey data misrepresents the program in important ways. The measurement error is non-classical, so standard corrections cannot remove the bias. An extension of the conditional density method developed in the previous essay recovers the correct estimates using public use data only, which solves the problem that access to linked administrative data is usually restricted due to confidentiality. I provide evidence that the conditional density method is a promising approach to solve the problem that validation data is often based on a convenience sample, because it performs well at extrapolation across time and geography. For example, extrapolation to the entire U.S. reduces deviations from official aggregates by a factor of 4 to 9 compared to survey aggregates.