Measurement error and the effects of illegal drug use on birth weight: An econometric investigation.
Item
-
Title
-
Measurement error and the effects of illegal drug use on birth weight: An econometric investigation.
-
Identifier
-
AAI9924856
-
identifier
-
9924856
-
Creator
-
Ye, Yongjia.
-
Contributor
-
Adviser: Michael Grossman
-
Date
-
1999
-
Language
-
English
-
Publisher
-
City University of New York.
-
Subject
-
Health Sciences, Public Health | Statistics | Economics, General | Health Sciences, Obstetrics and Gynecology
-
Abstract
-
This paper mainly develops a theoretical criterion for assessing bias of OLS estimators when measurement errors in variables are ignored in the linear regression model. Several bias or variance correction indexes due to measurement errors are introduced in this paper. It also spells out some important empirical estimation issues on measurement errors. The author shows that the regression coefficients for continuous and binary measurement errors are not comparable. In estimating infant health production function, split sampling, bootstrap reshaping and simulating techniques, simultaneous equations model with the endogenous binary or latent variables, and two step estimation procedure combined with the corrections for standard errors in the second step, heteroscedasticity, multicollinearity, selectivity and specification errors are used in the analysis. The findings in this paper suggest that the OLS estimator is biased downwards due to binary measurement errors, and it is biased upwards except for Blacks due to continuous measurement errors. The direction and magnitude of change in the variance of OLS estimator for the variable given error can be determinate if one has information about measurement error. The restricted OLS estimator always has smaller variance than the unrestricted one if the omitted measurement error is non-stochastic in the finite sample. The heavy user are more likely to report their use. The treatment effect on heavy users will be underestimated if one uses the coefficient on the actual to evaluate drug program aimed at heavy users. The study indicates that one can get consistent estimator in a less expensive way by using out-sample information about measurement error. Additionally, this paper convinces or extends some interesting findings in the literature about human behaviors of illicit drug use among pregnant women.
-
Type
-
dissertation
-
Source
-
PQT Legacy CUNY.xlsx
-
degree
-
Ph.D.
