Configural Frequency Analysis and Latent Class Analysis: Do Their Results Line Up?

This study set out to put two methods side by side. We ran Configural Frequency Analysis (CFA) against solutions produced by Latent Class Analysis (LCA), looking for where the two agree and where they part ways. The question driving the comparison was straightforward: when you analyze the same data with each approach, do you end up describing the same structure?

For the material, we used the Big Five dataset that Lautsch and Thöle write about elsewhere in this issue. LCA was applied in exploratory fashion, with no model imposed in advance about how many classes should emerge or what they ought to look like.

Once the LCA solutions were in hand, we compared them against the CFA output. The two dominant Configuration-Cluster-Analysis types turned out to match, almost case for case, the typical response patterns sitting inside certain latent classes. Those classes were the ones carrying high recruitment probabilities, meaning respondents had a strong tendency to fall into them. So on the central patterns, the methods converge.

That convergence was not the whole story, though. Each technique also reported something the other missed when it came to pinning down types. CFA surfaced detail that LCA did not, and LCA returned information CFA left out. Neither result was simply a subset of the other.

From that we draw a practical point: running CFA alongside LCA gives you more than running CFA by itself. The second method fills in gaps rather than merely confirming the first. The analyses throughout were carried out with the lem software.