Hello,

I'm currently running a PCA on a polytomous data set (5 point Likert). I want to create a discrimination index to narrow down my items before running the PCA. After looking around, I'm undecided about how to go about this. Some research suggests using the Corrected Item-Total Correlation value as an item selection tool (Article Item-Score Reliability as a Selection Tool in Test Construction

) with a cut off of .3, or .2 for exploratory studies as a guide. Other articles I have come across seem to suggest a discrimination index isn't used in the same way as it is for dichotomous data, however I am unsure if using this corrected item-total correlation as an initial selection method is viable as it seems it's use lies in reliability testing instead.

Is anyone able to shed some light on the best way to perform an initial item discrimination on my data (with academic reference if possible) please?

Thank you

More Michael Ward's questions See All
Similar questions and discussions