You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
as per Li and Redden (2014), using the standard sandwich variance-covariance matrix to perform a Wald test of significance - as we do in waldtestGEE() - leads to inflated Type I error (false positive) rates when the number of subjects is small and / or the number of timepoints per-subject is large
this was seen in our simulation study, where we observed high numbers of false positives when using $n_s = 6$ subjects
scRNA-seq datasets quite often have low numbers of subjects compared to observations per-subject, making this a problem for the GEE mode
several small-sample bias corrections exist, the easiest of which to implement is the DF correction, where $p$ is the number of covariates:
thus, it makes sense to add an option to bias-correct the estimated sandwich variance-covariance matrix when the number of subjects is lower than some heuristic threshold e.g., $n_s = 30$
The text was updated successfully, but these errors were encountered:
waldtestGEE()
- leads to inflated Type I error (false positive) rates when the number of subjects is small and / or the number of timepoints per-subject is largeThe text was updated successfully, but these errors were encountered: