Closed christinabergmann closed 4 years ago
Can you specify the datasets that have this issue?
I think I found it, it's in StatSeg in a Hay & Saffran entry, one of them has an impossible correlation over 1, so I am removing the correlation value. This fixes the issue.
Follow-up: We need to implement checks for correlations provided (eg by authors).
@christinabergmann could this be another check we do when validating the data, and reporting if there is an invalid correlation value? I will add it to the checks if so.
These values are calculated but indeed, there can be a check whether the calculation makes sense. But first, I need to further track whether (1) this still happens and (2) it's not an error in our math. I'll update with affected datasets
check 'r' column for [-1, 1], add warning, same with corr
fixed in metalabr package
Some calculated variances seem negative, wanted to flag this for further investigation.