`.sanitize_Q_ids()` determination of whether each unit has one observation in the model fit

[jwasserman2]

I don't think it should compare nrow(uoa_cls_df) == nrow(x$model) here:

propertee/R/teeMod.R

Line 769 in 223247d

if (nrow(uoa_cls_df) == nrow(x$model)) {

My thinking is that x$model removes NA's, and since uoa_cls_df will have a row for every unit regardless of whether its individuals are included in the model fit, we want to compare the number of rows in it to the number of rows in the model-fitting dataframe that includes rows that are excluded from the fit. Therefore, I think we should replace nrow(x$model) with nrow(expand.model.frame(x, uoa_cols)), where uoa_cols are obtained by var_names(x@StudySpecification, "u").

[benthestatistician]

If there is an na.action associated with x or x$model, perhaps we could still avoid the situation you're concerned with by comparing to nrow(expand.model.frame(x, uoa_cols)) only in cases where the na action was non-null. In cases where it was null, we'd save some computing by continuing to compare against nrow(x$model). The code would be a bit more complex but the chain of conditions would reveal what you're concerned about.

[benthestatistician]

...better yet: just compare to nrow(x$model) + length(x$model$na.action). (Or nobs(x$model) + length(na.action(model$x) or whatever it should be.)