Hey everyone, I’m having trouble with my confirmatory factor analysis (CFA) in Stata. I’ve got survey data from 139 people and I’m trying to check the validity of six variables.
Three of them are fine, but the other three are giving me weird results. The Chi Square is 0, RMSEA is 0, and CFI and TLI are 1. This seems too good to be true for my small sample.
These problematic variables only have 3 items each. Could this be the issue? I thought CFA was okay with 3-item variables.
Here’s my Stata code for one of the variables (optimism):
sem(OPTIMISM -> op01_01 op01_02 op01_03)
estat gof, stats(all)
Has anyone run into this before? Any ideas on how to fix it? I’m pretty stumped here.
I can share some of my data if that helps. Just let me know if you need more info. Thanks for any help you can give!
hey emma, try combining those vars. with just 3 items, ur model might be just-identified so it always fits perfectly. might be worth building a bigger CFA to get meaningful stats. what exactly are you studying?
Hey Emma_Brave! That’s a tricky situation you’ve got there. I’ve actually run into something similar before with CFAs on small samples.
Have you considered that with only 3 items per variable, your model might be just-identified? That could explain the perfect fit statistics you’re seeing. It’s not necessarily wrong, but it doesn’t tell you much about the actual fit of your model.
What about trying a more complex model? Maybe you could combine all six of your variables into one CFA? That way, you’d have more degrees of freedom to work with. Something like:
Just a thought! Have you explored any alternative approaches? I’m really curious to hear more about your research. What’s the overall goal you’re trying to achieve with this CFA?
I’ve encountered similar issues with CFAs on small samples before. Your model is likely just-identified with only 3 items per factor, leading to those perfect fit statistics. This doesn’t necessarily indicate a problem, but it’s not very informative either.
To get more meaningful results, you could try combining all six variables into a single CFA model. This would increase your degrees of freedom and provide more robust fit statistics. Something like:
Additionally, consider exploring alternative model specifications or using bootstrapping techniques to assess model stability given your small sample size. It might also be worth examining the correlation matrix and descriptive statistics of your items to gain insights into their relationships.
What’s the broader context of your research? Understanding your overall objectives might help in suggesting more tailored approaches to validate your measures.