Statistical inference for the linear model with clustered data Public Deposited
Downloadable ContentDownload PDF
- Last Modified
- March 22, 2019
Harden, Jeffrey J.
- Affiliation: College of Arts and Sciences, Department of Political Science
- Political scientists often confront clustered data, which can present problems for statistical inference. Through Monte Carlo simulation I examine the performance of standard error methods in clustered data for two linear estimators: Ordinary Least Squares (OLS) and Median Regression (MR). I consider changes to several parameters: sample size, number of clusters, intra-cluster correlation, and error term distribution (normal, which favors OLS as the most efficient estimator, and Student's t, which favors MR). Results indicate that conventional OLS and MR standard errors are often, but not always, biased downward in clustered data. Within OLS, the performance of the robust cluster standard errors (RCSE), which are designed for clustered data, is conditional on the level of covariate variation and the severity of cluster correlation. Regarding MR, two nonparametric methods perform well. I conclude that researchers should carefully examine the nature of the clustering in their data before choosing a standard error method.
- Date of publication
- May 2009
- Resource type
- Rights statement
- In Copyright
- Carsey, Thomas M.
- Degree granting institution
- University of North Carolina at Chapel Hill
- Open access
This work has no parents.
|Statistical inference for the linear model with clustered data||2019-04-10||Public||