I am conducting a scale adaptation study on Amos. However, I am unable to validate the main model, which I attribute to the fact that several items are meaningless in our culture. The scale has a high number of items.
When I researched, many different options were given with item removal. But which one is more prioritized?
For example, should I first remove items according to Standardized Regression Weight or R-squared, and then, if the model still does not fit, how should I proceed?
Why not remove the items that are obviously theoretically/substantively meaningless first?
If you want a systematic method for removing items, you can look at your statistical analysis program's treatment of coefficient alpha so to see if it offers an option for "alpha if item deleted." This will allow you to delete items one-at-a-time until you have reached a maximum level of reliability.
I ran into a similar issue with a scale comprising numerous subscales, each measured by many items. However, a CFA revealed that neither the factorial structure nor unidimensionality could be confirmed for any of the scales. Therefore, we developed a short version of only 5 subscales that were theoretically meaningful with a smaller number of items.
I think it would be better to develop a short version tailored to your culture than to just drop some items because they do not fit. "Alpha if item deleted" is not a good idea, because then the model is modified to fit the data optimally, which is a theory-free procedure.
HTH, Karin
The advantage of using a statistical procedure such "alpha if item removed" is that does not involve subjective judgements.
Hopefully, the results will correspond to the authors' intuitive sense of what is not culturally appropriate.
In addition to Christian's recommendation of removing by theoretical reasoning first, you could also use algorithmic approaches such as the ones described by Schroeders et al. (2016, here: 10.1371/journal.pone.0167110) or me and my colleague for specifically cross-cultural settings (10.1027/1614-2241/a000155), which I also implemented in an R-Package called stuart.
- Badges
- Science method
- Similar topics
- Methodology
- Research Methodology
- Qualitative Research