I´ve got a problem concerning a short scale measuring a perception of a social atmosphere.
The overall sample alpha of that scale is not that good (about a=.5) but at least ok for a two item scale. The problem now is, that even if the cfa model fit for the multigroup measurement model is ok, I get a really bad alpha for the high educated subsample.
So should I trust the latent factor solution, or the alpha value?
Any ideas?
Hi Michael,
I just recently joined researchgate. Good to see you here.
Regarding your question:
1.Low reliability should work against finding strong and consistent effects. If you find meaningful effects nonetheless, you are probably under-estimating their magnitude.
2. Also, I would compare spread statistics across subsamples. Perhaps the low reliability in the high education group is due to lower range / variance.
3. Run separate analyses using just each of the single items in place of the 2-item scale. If structures look similar, there's additional reason to trust your model.
Cheers
Gerd
- Badges
- Science topic
- Similar topics
- Mathematical Sciences
- Statistics
- Statistical Software
- SPSS