5.9.2.1 Model diagnostics

It is possible to perform model testing to check whether fixed or random effects estimation should be used in connection with panel regressions. This is done by using the hausman command.

Syntax and input follow the same logic as the associated regression command (regress-panel): The dependent variable is used as the first input, and then the independent ones are listed.

Example:

 
regress-panel wage age highedu wealth oslo
hausman wage age highedu wealth oslo
 

The result of the hausman run is a standard regression result for resp. fixed and random effect estimation. In addition, the difference between the coefficients in the alternative estimates are also shown, as well as an aggregate measure that indicates which variant is best to use for the current data set: P-value based on chi-square diagnostics.

P-values ​​less than 0.05 indicate that there are systematic differences in the coefficient estimates and that fixed effect modeling fits the data best. P-values ​​above this limit indicate the opposite (that random effect modeling should be used).

For more details, it is recommended to use the help command: help hausman

The regress-panel-predict command can also be used as a tool for model diagnostics, cf. section 5.9.2.2.