Hello! I am running a Multigroup analysis with Amos 29 using the Theory of plan behaviour. My latent constructs are Attitude, subjective norms, perceived behavioural control, and intention (intention would be my dependent variable). My CFA and Structural models have a reasonable Fit. However, during the construct validity analysis, the AVE of some latent variables is below 0.5, and the construct reliability is between 0.6 and 0.88.
How can this impact the interpretation of the results? The structural model has a reasonable fit. CFI 0.93 IFI 0.93 TLI 0.89 RMSEA 0.06 NFI 0.92.
I highly appreciate some literature suggestions.
Hi,
how is the model's chi square test? Given that attitude and subj. norm are usually highly correlated I would rather focus on a clan fit and substantial loadings as they indicate the validity. I don't know whether the literature in which you want to publish gives AVE so much weight. If not, ignore it. It was invented by folks who searched for side criteria beyond the chi square test (as they tend to ignore the test). Further, I don't know what "construct reliability" should mean as reliability concerns *measurement* error :)
Best,
Holger
Thanks for your response, the CFA fit results are as follows, chi square is a bit hight.
Model 1 Model Fit Statistics (χ2 = 658.12, df = 70; CFI = 0.952, TLI = 0.928, NFI= 0.946, RMSEA = 0.03)
Model 2 Fit Statistics (χ2 = 1189.88, df = 125; CFI = 0.942, TLI = 0.921, RMSEA = 0.057).
Yes, both models have a sign. chi square. Perhaps you could try to find out where the problems are. In a paper with Tom Rosman and others, I illustrated such a "detective work" with R.
Article Will COVID-19-related economic worries superimpose health wo...
Only study 1 is relevant. If you search, there is a R markdown file that you an dowload.
Best,
Holger
Holger Steinmetz, thanks for sharing the paper. I highly appreciate it! I am working with Amos. Is there any procedure you will suggest using Amos?
Hi,
as far as I know AMOS still cannot print the "standardized residuals"--that is, the cell-wise differences between the empirical covariance matrix and the model implied matris (stanardized). These are essential for any diagnostic where (with regard to which covariances) the model fails. I would really spend one week (you don't need more) to learn R (especially "tidyverse" framework) and then use the lavaan package. It is worth it!
The lavaan.org website has many tutorials which make it easy to run a model.
Best,
Holger
- Badges
- Science topic
- Similar topics
- Organizational Psychology
- Training