Samenvatting
Measurement error causes a bias towards zero when estimating a panel data linear regression model. The panel data context offers various opportunities to derive instrumental variables allowing for consistent estimation. We consider three sources of moment conditions: (i) restrictions on the covariance matrix of the errors in the equations, (ii) nonzero third moments of the regressors, and (iii) heteroskedasticity and nonlinearity in the relation between the error-ridden regressor and another, error-free, regressor. In simulations, these approaches appear to work well. (C) 2017 Elsevier B.V. All rights reserved.
Originele taal-2 | English |
---|---|
Pagina's (van-tot) | 169-180 |
Aantal pagina's | 12 |
Tijdschrift | Journal of Econometrics |
Volume | 200 |
Nummer van het tijdschrift | 2 |
DOI's | |
Status | Published - okt.-2017 |