To understand relationship between two constructs Confirmatory Factor Analysis was performed in LISREL, which reports all the R2 value between variables and constructs are all above 25% (considered good), Degrees of Freedom = 34, Minimum Fit Function Chi-Square = 63.69 (P = 0.0015), Root Mean Square Error of Approximation (RMSEA) = 0.050, Normed Fit Index (NFI) = 0.99, Non-Normed Fit Index (NNFI) = 0.99, Comparative Fit Index (CFI) = 0.99,Goodness of Fit Index (GFI) = 0.97, Adjusted Goodness of Fit Index (AGFI) = 0.92. This result was achieved after applying all the modification suggested by LISREL.
The two constructs regarding Social Media i.e., "Consumption of Information" and "Generation of Information" were tested for Convergent, Discriminant and Construct Validity using EFA and MS Excel, which were found satisfactory. However, CFA reports correlation between the two constructs with r=0.60 (at P=0.04).
Now I am unable to understand how to interpret the following result and whether the CFA is actually providing any meaningful conclusion. Please help.
All model fit parameters are perfect except for p-value because it's value should be grater than 0.1.
Deep,
I would suggest that "improving" model fit by simply changing the model based on modification indices (e.g., adding residual covariances) is not the best approach, for at least two reasons:
1. Modification indices focus on the change of a single parameter, but there is no way of telling what happens when you change multiple parameters, which I am guessing you did.
2. Model fit is partially dependent on the sample. By adapting your model to the sample data, you are doing what is called "capitalizing on chance". If you applied the modified model to a different sample, it is unlikely to fit well, just by chance.
A better way of handling misfit is to understand its sources, and then very carefully think about how to deal with each one. A good place to start is requesting the residual covariance matrix from LISREL to see where in the model the misfit occurs.
Just in case: one should never conduct EFA and CFA on the same data, because this is akin to a self-fulfilling prophecy. The model structure determined exploratorily through EFA will always fit well in CFA, because the latter is no longer truly confirmatory!
Dear Heiko,
Keeping your advice in mind. The modification indices were used keeping in mind that none of the parameters cross over to other latent variables. Some literature were further consulted to support the limited changes done . Thank you for your advice
- Badges
- Science topic
- Similar topics
- Social Science
- Media
- Social Media