Affiliation:
1. London School of Economics, London, U.K.,
Abstract
Stata’s two-stage least-squares (2SLS) computation procedures are sensitive to near collinearity among regressors, allowing situations in which reported results depend upon factors as irrelevant as the order of the data and variables. This article illustrates this claim with the public-use data of Oreopoulos (2006, American Economic Review 96: 152–175), where the instrumented coefficient estimate can be made to vary between 0.012 and 30.0 in one specification by permuting the order of the variables. Different methods for improving the accuracy of 2SLS estimates are reviewed, and a Stata command for collinearity-robust 2SLS estimation is provided.
Reference12 articles.
1. Baum C. F., Schaffer M. E., Stillman S. 2002. ivreg2: Stata module for extended instrumental variables/2SLS and GMM estimation. Statistical Software Components S425401, Department of Economics, Boston College. https://ideas.repec.org/c/boc/bocode/s425401.html.
2. Forced to be Rich? Returns to Compulsory Schooling in Britain
3. The Existence of Moments of k-Class Estimators
4. Estimating Average and Local Average Treatment Effects of Education when Compulsory Schooling Laws Really Matter