How many latent factors should you keep to prove a model fitness? Is there any limit on it?
The number of latent factors in a model is not a criterion for determining the plausibility of the model in a CFA. Instead, the number of latent factors is predetermined by previous studies on the instrument you are validating and it constitutes a primary hypothesis you are testing in your CFA. As such, previous studies determine the limit of the number of latent factors which can be modified depending on the results of your CFA.
Best wishes.
Latent variables should be based on theoretical framework and of course theory is usually evidence based. There should be a fixed frame of latent variables to make the study more authentic and realistic.
Yes. There are definitely limits on the number of latent variables. This is related to identification of the statistical model (there are a lot of statistical literature on this topic). The more latent variable you include, the more parameters need to be estimated. At the same time, you only have limited amount of information from your observed data. Think about it. If you only ask people one question, can you identify the levels of their intelligence, mental heath, SES, and whatever else you care about?
A model makes theoretical sense doesn't mean you can do it empirically with a statistical model.
Dear all thank you for wonderful responses. Based on literature review, I have a conceptual framework consisting of 26 different latent variables, under 5 domains or themes, which are related to quality of some learning experience. For those 26 latent variables, I have 100 items. After collecting data, what would be right choice or route, EFA leading to CFA and finally SEM, or I should directly move to CFA and SEM? Considering 26 latent variables and 100 items, what problems can I face while running it on SPSS Amos? Such a number of latent variables and items would affect the model fitness?
You mentioned that you have a theoretical framework from which you can proceed to the CFA-SEM. But considering the amount of variables it would be advisable to reduce the variables and items, by means of EFA to maintain a better adjustment of the proposed model, you can also recommend the method of plotting due to a great number of variables.
Hi Talat, if you have 26 latent variables and 100 items this means that the latent variables are going to have very few indicators. It's probably best to think about having at least three items per latent variable, but I'd opt for more as some may be lost due to poor performance. Going forward I think you need to reduce the number of latent variables or increase the number of items. Maybe you could look at the 26 latent variables and think about where these could be collapsed into broader conceptual groups and hence reduce the number of latent variables.
Mark Shevlin, I agree with you about three items per latent variable being "risky". I'm sure I've read about that being the case in some authoritative text, but offhand I can't recall where I'd read it, and I'd like to cite something authoritative to that effect in an article I'm currently working on.
Might you know of at least one good citation (maybe two), please?
Hi Robert
With only three items a CFA model has zero degrees of freedom and is therefore not testable, and 2 indicators makes the model not identified. However, if such a structure is embedded in a lager model then degrees of freedom are 'borrowed' from other parts of the model and factors with 2 or 3 indicators will be OK. I say OK, but this is not optimal. More indicators are better, but too many is not good either - see below.
Mark
Article Is More Ever Too Much? The Number of Indicators per Factor i...
Article Indicators per Factor in Confirmatory Factor Analysis: More ...
- Badges
- Science topic
- Similar topics
- Psychology