Department of Statistics Seminar
North Carolina State University

presents

Dr. Eva Petkova

Columbia University

"Correcting for Omitted Variables and Measurement Error Biases in the Estimation of the Effect of Lead Exposure on the IQ of Children"

ABSTRACT

The talk will begin with an overview of the article by ML Marais and WE Wecker, "Correcting for Omitted-Variables Bias and Measurement-Error Bias in Regression with an Application to the Effect of Lead on IQ", JASA, 1998, pp. 495-505. In summary, this article shows how biases in OLS estimates due to omitted variables and/or measurement error can be corrected by using auxiliary information. The method is illustrated with an application to four studies of the effect of childhood exposure to lead on the IQ. These studied did not correct for measurement error in the predictors (lead exposure and mother's IQ) and they did not use as predictors father's IQ and any assessment of home environment that can potentially have an effect on child?s IQ. Based on auxiliary information from other studies the authors impute the unavailable correlations and variances necessary to compute the OLS estimates of the effect of lead exposure on IQ. After this imputation the estimates of the effect of lead exposure on child's IQ in all four studies change and become not statistically significantly different from zero. In the second part of this talk I will discuss the appropriateness of the imputations performed in the above article. In particular, questions about the selection of studies for collecting unavailable data and the risk associated with imputing population specific correlations will be addressed. Contemporary methods for data imputation will be presented and contrasted with these used in the article. The talk will conclude with remarks on the misuse and abuse of valid statistical methodologies.

Friday, April, 20, 2001

3:35 - 4:35 pm

206 Cox Hall

Refreshments will be served on the second floor of Dabney Hall (left of Room 222) at 3:00 pm.