linearmodels.iv.results.IVResults.anderson_rubin¶

property IVResults.anderson_rubin : InvalidTestStatistic | WaldTestStatistic¶

Anderson-Rubin test of overidentifying restrictions

Returns:¶

Object containing test statistic, p-value, distribution and null

Return type:¶

linearmodels.shared.hypotheses.WaldTestStatistic

Notes

The Anderson-Rubin test examines whether the value of \(\kappa\) computed for the LIML estimator is sufficiently close to one to indicate the model is not overidentified. The test statistic is

\[n \ln(\hat{\kappa}) \sim \chi^2_{q}\]

where \(q = n_{instr} - n_{endog}\).