Common Errors in Statistics (and How to Avoid Them), 4th Edition

PANEL DATA

When multiple observations are collected for each principal sampling unit, we refer to the collected information as panel data, correlated data, or repeated measures. For example, we may collect information on the likelihood that banks offer certain types of loans. If we collect that information from the same set of banks in multiple instances over time, we should expect that observations from the same bank might be correlated.

The dependency of observations violates one of the tenets of regression analysis: that observations are supposed to be independent and identically distributed or IID. Several concerns arise when observations are not independent. First, the effective number of observations (that is, the effective amount of information) is less than the physical number of observations since, by definition, groups of observations represent the same information. Second, any model that fails to specifically address correlation is incorrect, which means that statistics and tests based on likelihood are based on a faulty specification. Third, although the correct specification of the correlation will yield the most efficient estimator, that specification is not the only one to yield a consistent estimator.