search

joaquinanguera / aceR

An R package for processing ACE data
MIT License
3 stars 14 forks source link

count vs length check (check both ACE and SEA) #9

Open monicathieu opened 5 years ago

monicathieu commented 5 years ago

We noticed that the 'count' and 'length' metrics are the same. in my notes I have that count is the number of trials attempted and length is the number of trials presented. Count should be reflecting number of trials that are not 'no response' correct? If so, I think that because we started labeling no responses explicitly, it is now counting all trials and the count and length metrics are effectively the same.

monicathieu commented 5 years ago

Mmm, good catch. Count is intended to discard no-responses, but you're right, the original ace_count() function is just doing this by counting non-NA values. If no-responses are now marked with strings, not with NAs, this doesn't work! This is quick to fix though.

monicathieu commented 5 years ago

OK, I also found a herpy derp in the code that I left there: no-responses are marked with character strings in the correct_button column (just fixed ace_count() to take care of this and am about to push it), but I DON'T scrape out the ensuing RTs with NA (this is the thing you told me about a while ago where no-response RTs were being imputed in the task code with mysterious values divisible by 10). I realized I never actually added anything to NA out those RTs. This means that count and length will return different things now for ACCURACY, but not for RT. Should I update the raw ACE loading code to NA out those mysterious divisible-by-10 RTs?

monicathieu commented 5 years ago

➤ Jessica Younger commented:

Ahah yes please NA out the mysterious divisible by 10 RTs as I've confirmed they are actually no responses. It's complicated to explain exactly what happened, but I know they don't reflect actual responses. This only happens in older versions of ACE Classroom.

I think that ACE Explorer isn't using decimal precision for RT though, so I need to put in a flag to update that on Oliver's end so that a response isn't accidentally categorized this way in ACE Explorer. Unless it's easy to implement one set of processes for ACE Classroom and one for ACE Explorer?

monicathieu commented 5 years ago

➤ Jessica Younger commented:

Update on this one: Current process for ACE Explorer doesn't report decimal for RT, which could potentially lead to us throwing out some real trials in ACE Explorer. I'm checking with our developer about getting it to increase precision. Otherwise I think we'd have to do something to detect build/version and only implement this extra cleaning step for older versions of ACE.

monicathieu commented 5 years ago

➤ Jessica Younger commented:

OK - ace explorer reports RT in integers (which probably makes more sense given refresh rates right?), but this means we'd want some way of determining whether output is ace classroom or ace explorer and only applying these flagging/scrubbing rules to ace classroom data and not ace explorer data.

monicathieu commented 5 years ago

Yes, currently the data are assumed to be Ace Explorer if there is a separate demographics file found in the data, but there can be a different flag (or manual argument in load_ace_bulk) if appropriate

monicathieu commented 5 years ago

➤ Jessica Younger commented:

Ok so let's scrub the mysteriously divisible by 10 RTs in the ace classroom branch (and not the primary ace explorer brance). That solves the hold up on this one correct?