Skip to contents

Robust score (Rao) tests with finite-sample correction

robust_score_test.Rd

The default behavior is to do no finite sample correction (for the covariance of the score) for correlated data, but to do it for uncorrelated data. This choice performed best under a small simulation study. See Guo et al for details; the proposed modification is in equation 20.

Usage

robust_score_test(glm_object, call_to_model, param = 1, id = NA)

Arguments

glm_object

The fitted glm under the alternative.

call_to_model

The call used to fit the model. Used internally.

param

which parameter do you want to test? Used internally.

id

observations with the same id are in the same cluster

References

Guo, X., Pan, W., Connett, J. E., Hannan, P. J., & French, S. A. (2005). Small-sample performance of the robust score test and its modifications in generalized estimating equations. Statistics in Medicine, 24(22), 3479–3495. Wiley Online Library. doi:10.1002/sim.2161