I thought it had to do with varying numbers of factors but based on the literature I have read it seems to do more with fixing/freeing of parameters. Any info would be greatly appreciated!

up vote 5 down vote accepted

Model A is nested in model B if some of the coefficients in B, or their combinations, can be restricted to obtain model A.

In case of CFA models, the number of factors is a moderately complicated example of nesting. Consider a two-factor model (taken from UCLA ATS website) CFA2

A simpler one-factor model will be nested in it with a linear constraint $${\rm Corr}[{\rm visual}, {\rm verbal}]=1.$$ Alternatively, we can say that it is nested with a nonlinear constraint that $${\rm Cov}({\rm visual},{\rm verbal}) = \mathbf{V}[{\rm visual}]\cdot\mathbf{V}[{\rm verbal}].$$ Depending on how exactly the model is parameterized, there may be small differences in the results (every 1 matters!). Conversely, you can move from 1-factor to 2-factor model by freeing the correlation coefficient, although you still had to present the model as a 2-factor model to begin with.

Note that testing one vs. two factors may give rise to complications: the natural range of the correlation parameter is $[-1,1]$, so it is tempting to restrict it to this range (as EQS does) to avoid Heywood cases. Vika Savalei and I argued that this is generally a poor testing strategy, because it produces weird likelihood ratio distributions that are not the traditional beloved $\chi^2$.

Moreover, in some factor presence testing situations (e.g., MTMM), the likelihood ratio has a wrong null distribution even if you do not restrict the range of the parameter: if you have zero variance of say the methods factor, then its loadings are not identified, so you get an effective mix of $\chi^2$s with different degrees of freedom asymptotically. This is a poorly understood topic, overall, not only in SEM, but in other areas of application of statistics.

  • Wow I like the graph in Latex! +1. – SRKX May 2 '12 at 20:00
  • @SRKX, I don't know if it is LaTeX, frankly. As I said, I took it from UCLA website, and they may have produced it in AMOS. I usually do my path diagrams in Dia software (and I have a little tutorial on how to use Graphviz, the engine behind R sem package, on my webpage), but I was sure I could just google a decent 2-factor model in three seconds. – StasK May 2 '12 at 21:28

Your Answer

By clicking "Post Your Answer", you acknowledge that you have read our updated terms of service, privacy policy and cookie policy, and that your continued use of the website is subject to these policies.

Not the answer you're looking for? Browse other questions tagged or ask your own question.