Augment shouldn't drop NA rows about broom

[stefanfritsch]

As I see it, if the parameter data=dataset is present, augment augments that dataset. So all rows of the input dataset should be present with new columns added.

That said, why not just add a column "na.action", that way the user has full control? Although in most cases that would just be an is.na(fitted.values), wouldn't it? :)

[gavinsimpson]

It might be better to have augment() respect the na.action requested by the user when the statistical model was fitted. Hence to achieve the behaviour you desire, the model would need to be fitted using the standard convention of na.action = na.exclude.

Where the model fitting function doesn't have an na.action argument, perhaps the augment() method could add an argument which would allow this NA preserving feature?

[alexpghayes]

Dave and I talk about this earlier and we've decided to strictly enforce:

• glance returns a single row summarizing model level information
• augment returns a single row for each observation present in data or newdata, and if neither of those are present, for each observation present in the original training data. When there's missing data for a particular row, the NA will get propagated to .fitted and any other relevant columns.

[alexpghayes]

This has now been codified in the strict version of the check_augment_function() test. Very few of the strict tests are passing at the moment, but we're about to start making changes so that they do. Sorry about the delays on this, and thanks for your patience.

[github-actions]

This issue has been automatically locked. If you believe you have found a related problem, please file a new issue (with a reprex: https://reprex.tidyverse.org) and link to this issue.