I have a construct 'actual usage' in a model. when i checked and tested model for adopters firms the construct 'actual usage' have alpha values and composite reliability are above 0.7. But when i include non-adopter firms the alpha value and composite reliability are below 0.5 that are 0.455 and 0.422. But it has acceptable convergent and discriminant validity. Kindly tell me can i continue with this construct in my study.
The problem with low reliability measures is that they will nor correlate with other variables, due to the large amount of random error that they contain (the technical term for this is "attenuation"). In your case, this would mean that meaningful statistical relationships in one subsample could easily be obscured by what amount to random responses in the other subsample.
This situation can be analyzed in terms of an interaction effects, where there is a statistically significant difference between the effects in the two subsamples.
You should try to conduct reliability analysis for each item in the construct to see which item you should keep or delete. In SPSS the output table is called Item-Total Statistics (the “answer” is in the last column)
Hi Tripathi, Claudia Abreu Lopes and Prof. Morgan answers make good sense. I don't know exactly what do you mean by actual usage (of what) and by whom? However, you need to see if the construct actual usage tends to be reflective or formative. In formative constructs, the underlying structure and hence the items, may not stick together causing low alpha values. You may like to check inter-item correlations to examine problomatic questions (with low scores).
- Badges
- Science topic
- Similar topics
- Medicine
- Physiology
- Biosignals
- Biomedical Signal Processing
- Statistical Analysis