This function combines estimates, standard errors and p-values across imputed
datasets, for a single model (i.g., one vertex).
The function was largely taken from mice::pool() and
mice::summary.mipo() code. It averages the estimates of the
complete data model, and computes relevant statistics, following Rubin's
rules (Rubin, 1987, p. 76).
P-values are estimated using the *t-as-z* approach at the moment. This is known to the anti-conservative for small sample sizes. However we preferred a relatively lenient (and computationally inexpensive) solution at this stage. We will be addressing Type I error mores strictly at the cluster forming stage.
The residuals of the model are currently simply averaged across imputed datasets, for lack of a better idea of how to combine them.
Arguments
- out_stats
: Output of
single_lmm, i.e.: a list with two elements:"stats": a dataframe with estimates, SEs and p-values for each fixed effect term)"resid": a vector of residuals for the given model.
- m
: Number of imputed dataset (to avoid recomputing it)
Note
Used inside hemi_vw_lmm.