[luchkina_waxman_2024] write import script from scratch

[adriansteffan]

Notes: has two labels, use the first one for label and target onset

[adriansteffan]

The dataset passes the validator, but there are a few open points

1. The mails state that they used 8 additional words in their cdi. How do we encode this, since the rawscore will be skewed? Also, I am still confused what "measure" and "instrument" to put here, since I am not 100% sure what Level 1 MCDI translates to. Maybe it would make sense to dedicate a slot in a meeting to drawing some sort of decision tree/diagram for me so that I don't ask this for the next 100 datasets.

2. monitor_size_x, monitor_size_y are missing (not sure how important these are with icatcher)

3. The lookingscore graph (sample average preference for target from -1 to 1 over time) looks like it's shifted, and there is never a preference for the distractor even before the onset of the target word. I haven't done anything to the time data other than normalize it and didn't see anything weird in the data itself, so another pair of eyes would be great here.

[adriansteffan]

seems to be wg short, I will poke at the raw data to get clean cdi scores for both prod and comp

[mzettersten]

Will reach out to Elena. Current working hypothesis:

• target items are more salient than distractor items (since these are two distinct sets)
• looking curve is less distinctive given age of infants
• there might also be some lagginess in the timing

[adriansteffan]

I fixed the CDI, but there was one suspicious thing:

The CDI data for Subject_55 is missing in their wide format table (all other cdi data is there). This is likely a result of Subject_55 missing in the underlying raw cdi data they provided us with. While Subject_55 is missing, Subject_5 appears twice with the exact same cdi pattern. Likely a typo that led to data loss, so the import assumes there is no CDI data for subject_55, but it might be worth asking the authors again