Non-error trials with zero looking time

palday commented 5 years ago

There are trials where trial_error == FALSE but looking_time == 0. It's not clear whether this reflects a preprocessing error or just a 'correct' trial (no equipment issue, no interference from parents, etc.) where the baby never looked/fixated at the target. Beyond the inferential issues, there's also the problem that log 0 isn't a real number.

Depending on the cause of this constellation and your inferential question, there are a few possibilities:

Replace the 0's with NA.
Replace the 0 with some small epsilon (e.g. 1ms).
Add some small epsilon to all looking times.
Model the "look vs. no look" as a separate part of the data generation process, e.g. as a separate binomial model or in a hybrid hurdle model.

I'm raising this as a separate issue to the list in #5 because I think it might warrant some discussion before implementing a solution.

christinabergmann commented 5 years ago

There should be a column called short_lt which is set to FALSE when looking times are above 2s, i.e. only those should be included. Cf. https://github.com/manybabies/mb1-analysis-public/blob/master/paper/exclusions.Rmd l 119-128

mcfrank commented 5 years ago

I can't reproduce this error in my own version of the pipeline - perhaps again due to the exclusions issue covered in #8?

manybabies / mb1-analysis-public

Non-error trials with zero looking time #7