When researchers are exploring convergent validity, they often compare a new questionnaire with an existing one. When both share the same theoretical underpinnings, are designed for similar populations (e.g., geriatrics) and are in the same language, what kind of figure might one expect for r2? A lot of researchers report highly significant correlations, but, when testing the psychometric properties of new questionnaires, they often have large sample sizes. So I'm wondering what kind of R2 might be a good indicator of convergent validity?
I appreciate it's not going to be 100%, as that would mean a new questionnaire did not improve upon an existing on, but a lot of researchers report 'highly significant' relationships, but with r in the order of .30 to .50.
Andrew, that depends on the similarity of the questionnaires. Correlations of .30 or .50 seem to be a bit low for very similar questionnaires, but in most cases the questionnaires are supposed to be more different with regsrd to content, number of items, reliability, and validity. It would be more important to demonstrate thst the new measure does not correlate higher with dissimilar constructs. Thst means that not only convergent but also discriminant validity should be examined.
Regards, Karin
Andrew,
The key is to find the brackets for convergent validity (maybe .7 is good) as it should not be too high as well because then someone may question the need for establishing a new questionnaire if already existing one is functioning pretty well on its own.
Hope it helps a bit and good luck
- Badges
- Science topic
- Similar topics
- Mathematics
- Statistics
- Sampling Design
- Sample Size