Correlation coefficient is one of the most commonly used indicators to assess the construct validity. My question is "is there any cutoff of score indicating weak, moderate, and strong/very strong convergent/discriminant validity based on correlation coefficient"?
Reference-based discussion would be quite helpful.
You can use Cohen's effect size measures but these vary according to clinical measures and sample size, so I recommend using the Bayes factor which refers to the magnitude of the conclusive evidence for your hypothesis given the effect according to your sample (Bayes factor greater than 10) of convergent validity correlation.
Article Replicación bayesiana: cuán probable es la hipótesis nula e ...
Preprint La alternativa metodológica del factor Bayes en la investiga...
Article Dos ejemplos de replicación bayesiana para estimar el grado ...
Article La inferencia bayesiana como replicación y cuantificación en...
Sadiq Hussain, I'm sorry that I didn't see your question before now. It's a good question, and one that I find difficult to deal with / answer. I have confronted the same issue, and can't find anything authoritative concerning it.
In my experience, different researchers use different levels of correlation coefficient to claim either convergent or discriminant validity.
To make matters worse, some researchers use much the same levels of correlation to claim convergent validity that they use to claim discriminant validity - or even higher correlations for claiming discriminant validity than they use for claiming convergent validity.
It's a mess - and researchers often seem to choose whichever correlations that suit their purposes.
For what it's worth, in some research that I'm currently preparing for publication, I have set correlations at |.40| to |.60| as indicating satisfactory convergent validity (greater than |.60| might indicate too high a degree of similarity in my view) and correlations < |.15| as indicating satisfactory discriminant validity (i.e., adequate conceptual separation) - though I wonder whether I'm being too conservative and should perhaps raise the criterion to < |.25|.
The following article might help:
Grobler, A., & Joubert, Y. T. (2018). Psychological capital: Convergent and discriminant validity of a reconfigured measure. South African Journal of Economic and Management Sciences, 21. https://doi.org/10.4102/sajems.v21i1.1715
However, these researchers say that any correlation greater than .50 indicates satisfactory convergent validity. I think that's too lenient a perspective.
If anyone could provide a better reference for criteria concerning convergent and discriminant validity, I'd join you in being appreciative.
All the best with your research.
Robert Trevethan Obliged for you detail explanation, it proved quite helpful for me.
- Badges
- Science topic
- Similar topics
- Correction