Do I treat polychoric correlations like I would other correlations when reporting zero, low, or high correlations before running EFA on my scale items? Would there be an instance where if two variables were highly correlated or not correlated at all that I would give pause and not include the items in the EFA that I run?
I hope this updated question is a clearer. Thank you.
Cut-offs for what? As long as your correlations are within the admissible range, I don't see how you would need to consider any additional kinds of cut-off values for (or before) running an EFA. Of course, if many or most of the correlations are close to zero, that would indicate that the items may not measure one or more common factor(s).
Christian Geiser I realize my wording was not very clear. I updated my question and hope it makes more sense.
The only situations that would potentially raise a concern would be:
1) all (or almost all) correlations are near zero. In that case, factor analysis probably won't make much sense because the items don't have enough "in common" and/or contain too much measurement error (are unreliable).
2) one or more correlation is very close to 1.0 indicating close-to-perfect redundancy of items.
3) The sign of one or more correlation coefficients does not match your theoretical expectations (e.g., sign of correlation is negative even though the relationship is expected to be positive based on theory).
Otherwise, I would simply proceed with the EFA and not worry too much until I see the results :-)
- Badges
- Science topic
- Similar topics
- Linguistics
- Composition Studies
- Publication