9308309 Flavin While robust regression estimation has been an active area of research in the statistics literature for the past twenty years, robust estimators have not been fully exploited in applied econometric work, despite the widespread recognition that economic data sets are very likely to contain highly influential observations or outright data errors. The first part of this project is to encourage the use of formal robust methods in applied econometric research by evaluating a robust estimator suitable for simultaneous equations settings, which is both extremely intuitive and very simple to implement computationally. In essence, the first part of the project is designed to convince applied econometricians that robust estimation is easier than they may have thought. The second part of the project is designed to convince the same audience that robust methods are more necessary than they may have thought. Instead of looking at the implications of using robust estimators in place of conventional ones in several isolated or unrelated applications, the proposed research would first replicate the published results of a set of three papers which addressed the same set of economic issues. Considered as a whole, the striking, and unsettling, feature of this group of papers is that even though each used the same data set, comparable estimation techniques, and addressed the same set of issues, essentially no consensus on the substantive economic questions emerged from these studies. After replicating the original, highly divergent results, the proposed research would reestimate the models, maintaining any differences in the specification, but using formal robust methods. If robust estimates are in much closer agreement across the different studies, then this would show that currently accept econometric practice is not an adequate substitute for the systematic and objective treatment of data errors and extreme observations provided by robust estimation. More specifically, this project analyzes the properties of the Krasker-Welsch robust simultaneous equations estimators and the instrumental variables(IV) version of the Huber estimator as a simpler alternative to Krasker-Welsch. By comparing the estimates, and the corresponding economic implications, which result when one applies a conventional (nonrobust) estimator, IV- Huber, and Krasker-Welsch, the research will demonstrate the extent to which the choice of conventional vs. robust methods can make an important difference to the economic conclusions which are drawn from the data. ***
