I've been reading around on the creation of Semantic Differential scales and the manner in which their reliability and validity can be determined. It seems there are a fair few ways in which one can generate scales and test these features, but the two that are turning up most frequently are a mix of Cronbach's Alpha for reliability and Factor Analysis for Unidimensionality. I was thinking of utilising these methods.
Is this sound? My statistical background is fairly limited and there appears to be an intimidating variety of methods that can be used for this end.
Would appreciate expert opinion.
Cheers
Paul Wilson
.You could use coefficient alpha as a measure of internal consistency. Keep in mind the semantic differential as originally designed measures three properties Evaluation, Potency and Activity, so alpha should be computed separately for each one (if you included adjective pairs for all three). You can certainly conduct a factor analysis, and should find the three factor structure. You would do a separate factor analysis on each target that you get a semantic differential set of ratings on, assuming you have more than one target. These analyses address internal consistency and dimensionality. To address validity, you would need to measure additional variables to see if the semantic differential scores relate to them as you would expect.
It would also seem reasonable to test for consistency or stability as part of reliability and that would involve test-retest in a stable situation and potentially calculation of an intra class correlation (ICC). This approach would provide different information than internal consistency. Specifically if responses are stable or consistent over time.
- Badges
- Science topic