Much of econometrics is devoted to modeling and estimating the relationship between a response variable (e.g., wages) and measured variables that influence the response (e.g., education, experience). Econometric models also include an error term meant to capture the effect of unmeasured variables on the response. Often the error term is assumed to have a known probability distribution. This can be a restrictive assumption, usually made for computational convenience, and seldom supported by economic theory. If, as often happens, this assumption fails to hold, estimators of unknown parameters, such as those that quantify the effect of measured variables on the response, can be unreliable. Some of my research involves developing semiparametric estimators of parameters in these models. These estimators can be reliable even when the distribution of the error term is unknown. I am especially interested in developing computationally attractive semiparametric estimators, as well as developing the large sample theory required to do asymptotic inference with these estimators.
In nearly every scientific discipline, data collected for making inferences about a population of interest can suffer from various deficiencies, rendering conclusions based on standard inference procedures invalid. For example, survey data in the social sciences are often biased due to various types of misreporting, nonresponse, and measurement error. Some of my research involves developing ways to detect and correct biases that result from flawed data-generating mechanisms, as well as developing and analyzing estimation procedures that are valid when bias cannot be removed.
Last updated: March 20, 2009 14:59