Internal consitency for multilevel items

[Stephanie Peschel]

Dear Stata-users,

I am looking for a way to calculate an index reflecting internal consistency of a scale consisting of multilevel items in Stata. More specifically, my scale consists of 3 items (measuring one construct), which participants filled out multiple times a day for multiple consecutive days. From my understanding, Cronbach's alpha is not the best option here and I would highly appreciate any ideas for obtaining a better index in Stata. I am using Stata 17.

Thank you all very much in advance!

[Weiwen Ng]

This is an interesting question. Internal consistency, in psychometrics, is how consistent the responses to the questions are. Alpha is the most frequently used measure of that concept, and it’s based on the inter-item correlations. So, if one question is not correlated with the others, that will reduce the scale’s alpha. If you have measures at three time points, in other contexts we would assume that the responses at different time points are correlated with each other. I think that would inflate the scale’s alpha.

a simple solution seems to be just calculate alpha at each time point, and report the most conservative value.

I am not aware of a version of alpha that accounts for multiple measures.

[Stephanie Peschel]

Hello Weiwen Ng, thank you very much for your reply! Sorry for responding late, haven't been on here for a couple of days. Your considerations sound very helpful. Thanks!