Addressing convergence issues in multigroup CFA for factor invariance across time

I’m stuck with a multigroup CFA problem. I’m trying to check if two latent factors stay the same across three time points. My model has 7 items total, split between the two factors (4 and 3 items each). I’ve got 92 observations in all.

I tried the marker variable method first, fixing one loading at 1 and its intercept at 0 for all groups. But the model won’t converge. Any ideas why?

So I switched to the ‘in every group’ method, setting latent variable means to 0 and variances to 1 across groups. This one worked. Is this a good backup plan? Or should I maybe use IRT instead?

I used Stata for both tries. The first method didn’t work, but the second did. Any thoughts on what’s going on or what I should do next? Thanks for any help!

I’ve encountered similar issues with multigroup CFA models. Your sample size of 92 across three time points is likely the primary culprit for convergence problems. The marker variable method often requires larger samples to stabilize estimates.

The ‘in every group’ approach can work as a workaround, but be cautious about interpreting results. It constrains more parameters, potentially masking real differences across time points.

Before considering IRT, I’d suggest exploring a few alternatives:

  • Reduce model complexity by combining factors or dropping problematic items.
  • Use bootstrapping to get more stable estimates.
  • Try Bayesian estimation, which can perform better with small samples.

Additionally, examine modification indices and residuals to identify local areas of misfit. This may provide insights into which constraints are causing issues.

Remember, sometimes the data simply can’t support the desired model. Be prepared to revise your research questions if needed. Good luck with your analysis!

hey charlotte, that’s a tricky one! ive done similar stuff before. with only 92 observations, ur model might be too complex. have u tried simplifying it? maybe combine some items or use parcels? also, check for outliers - they can mess things up. the ‘in every group’ method could work but watch out for overfitting. keep at it, youll figure it out!

Hey Charlotte91! Convergence issues can be such a headache, right? :sweat_smile: I’m really intrigued by your problem - it’s got me thinking!

Have you considered the sample size issue? With only 92 observations split across three time points, you might be pushing the limits for a complex model like this. That could explain why the marker variable method didn’t converge.

The ‘in every group’ method working is interesting! It’s definitely a valid approach, but I wonder if it might be masking some underlying issues. Have you looked at your model fit indices? How do they compare between the two methods?

IRT could be an option, but before jumping ship, maybe try a few tweaks? Like, have you considered:

  1. Simplifying your model a bit?
  2. Checking for outliers or influential cases?
  3. Playing around with different estimators?

I’m super curious to hear more about your data and what you’re trying to achieve. What’s the broader research question you’re tackling? Sometimes taking a step back can help us see new solutions!

Keep us posted on how it goes, okay? This stuff can be tricky, but that’s what makes it fun too!