|
WORKING PAPER
Sweeping fewer things under the rug: tis often
(usually?) better to model than
be robust
Beck, NathanielAbstract The use of ``robust'' standard errors is now commonplace in political science. This paper considers one such type of errors, those that are robust to clustering of the data. While these give accurate estimates of parameter variability, we often can do better by direct modeling of the clustering process; such modeling can give insight into important sources of cluster effects. Applications are to grouped data with group level variables, difference in difference designs and time-series--cross-section data. Analysts should always ask whether clustering can be no more than an estimation nuisance before simply resorting to cluster robust standard errors. Keywords Cluster Robust Standard Errors Difference in Difference Moulton Problem Random Effects Time Series Cross Section Data File
Uploaded 07-16-2012 Document ID Number 1353 |
|
|