Wes Bonifay

Yours & Preacher's (2006) article introducing the idea of "fitting propensity" are, in my opinion, MUST READS for anyone who does any kind of latent variable modelling.

Thanks! Flattered to be recommended alongside Kris.

Seconded. Dale Kim introduced me to the idea of fitting propensity and it bugs me it's not more widely known!

There are a lot of things that I feel are important but a little "out there". Aside from fit propensity I'd put in Raykov's proof of infinite equivalent models, growth mixture models that cannot distinguish non-normality from multiple latent classes, etc. I guess the main prob1/2

with this kind of things (and hence their lack of popularity) is that they're a constant reminder that there is no such thing as 'standard statistical analysis' that can be rote-performed HOWEVER if there are problems and nobody talks about them... do they even exist? 2/2

Another along those lines is Waller (2008) on fungible regression weights - coefficients can be drastically different (even with reversed signs!) and still yield near identical R^2s

Extended to SEM by Pek & Wu (2018), Lee, MacCallum, & Brown (2018) and others. Interesting work.

I've only read Waller (2008). I found it interesting from a theory point of view however... correct me if I'm wrong but, doesn't the whole thing about fungible regression coefficients allows for solutions that aren't optimal? As in they are not *the* least-squares solution?

Waiting on my library to access this pub, so apologies if you answer this question already: is there still utility in using model-based reliability indices (e.g., OmegasH, OmegaHS, % reliable variance) from bifactor models to inform use of composite scores?

Those indices aren’t discussed in this paper, but yes - bifactor (and imo all latent variable) models must be evaluated by criteria that supplement overall model-data fit. For those who don’t know, useful criteria are presented by Rodriguez & Reise (2015a/b in JPA/Psych Methods)