HI. I am trying to develop a model to see how some factors from three domain of resources- Social, Cognitive, Non-cognitive affects the peoples CP. I have many such latent factors from each of the resources and measured them through test items. I then did EFA of data to identify and validate the factors of each resources separately. Next I did CFA of those latent factors again separately for each resources. Finally i did the path analysis taking all latent factors together to explore how it predict the CP
- My questions/doubts are:
- 1) Is it necessary to run the EFA and CFA taking all resources together before doing the path analysis of all resources with CP? As CFA is compulsory step before doing path analysis.
- 2) Is there any statistical or logical error in my path analysis (please see the attachment sl no 5)?
- 3) Is there any other better way to develop my model? Please suggest me if I am missing any steps or not?
- 4) Any comments regarding the fitness score of my model???
Its very difficult to explain what i actually mean so i am attaching a part of my data analysis and questions/doubts with this question. Please go through it and help me out. I am very new to statistics so need my fellow researchers'/professors' help.
Thank you.
Sir,
It is not compulsory to conduct CFA before modelling but always advised to conduct CFA to know standardised factor loadings which should be more than .6. It will help you if you are calculating CV, DV, MSV and ASV in excel.
You only have to report the evidence of reliability of the instruments, your structural model goes beyond the path analysis (only total scores) because it includes latent variables, in your case it looks like a hierarchical factorial model, because you have put a total score in the IV (it can be simulated as a higher order variable), it would be more advisable to include their factors, most predictive models have more IV than DV unlike yours, this is complex because they include many indirect routes between general and specific factors, which probably is why it presents low and negative factorial loads despite the AFC.
Cristian Ramos-Vera Thanks for response sir,
Yes i have used hierarchical model (specifically second order path analysis) as my number of items for each first order latent variable was too high. And i have two levels of latent variables to explore. So I want to know
1) Is it logically wrong to make 1st order latent variable as the observed endogenous variable?
2)What are the problems of using second order path modeling?
3) is my method of drawing paths between the two observed endogenous variable predicting two different 2nd order unobserved variables are statistically permissible or not? (like between LTS----RIBS)
4) yes, i am getting negative factor loading for some paths. As they are the standardized regression score so is their any harm in having negative but significant score? So far i understand the (single headed arrow) factor loading are showing the strength of effect between the variables that is how well one variable can predict the other variable.
5) In my all CFA the P value of chi square is less than 0.05 so cant i claim the fitness of my model ??? as its representing that my model is significantly different from ideal saturated model.
- Badges
- Science topic
- Similar topics
- Education