I designed a scale to measure a pattern of negative leadership leadership behavior in the arab context, and tried to examine its convergent validity with similar constructs such as despotic leadership and immoral management behaviors. I conducted a correlation analysis and the results are .9, and .89 respectively.
One of the reviewers told me these coefficients are so high that they may reveal that all the constructs mean the same thing and wondered about the value of my model.
in order to handle this issue, I conducted CFA which involves all the constructs and my scale altogether (five first order factor model). the model fit values are good, however, my original model (three first order factor model) provides better fit values. I looked at the BIC and AIC. they indicate that my original model with only three factors (and without the other similar constructs) better explain the data.
Can I claim, based on the previous findings, that although other constructs measure close phenomena, the original model provides a meaning above what other similar constructs deliver?
Please help me and thank you in advance.
Yes. you can claim for your first Model as it is evidenced by CFA and the different indices
Dear Hawass
Ros-Galvez is right!
You must conduct a discriminant validity test (Fornell-Larcker criterion or the new HTMT). See the SmartPLS website (www.smartpls.com) for the literature that is available for these tests.
Often reanalysis of their scale may reveal that respondents have a strong tendency to agree (or disagree) on items. In this case we have to review the scale and its application.
Perhaps, you can create a constriction by aggregating the two highly correlated constructs and searching the literature or juxtaposed specialists for it.
hope this help
Best Regards
Dirceu
Dear Hawass,
After checking based on Fornell Larcker, one possible approach is to see if you can introduce a second-order (or higher order) factor for the highly correlated constructs. However, there should be theoretical support in backing for second-order model. If you have theoretical base for that, it may be a better model.
- Badges
- Science topic