A wish: Confirmatory Factor Analysis

[mozny]

• Enhancement: Confirmatory Factor Analysis
• Purpose: It is used to test whether measures of a construct are consistent with a researcher's understanding of the nature of that construct (or factor). As such, the objective of confirmatory factor analysis is to test whether the data fit a hypothesized measurement model. This hypothesized model is based on theory and/or previous analytic research. In confirmatory factor analysis, the researcher first develops a hypothesis about what factors he believes are underlying the measures he has used Details e.g. here: https://en.wikipedia.org/wiki/Confirmatory_factor_analysis

[abnormally-distributed]

I second this as a useful feature. It'd also be really cool if there was the ability to create factor scores, like in SPSS.

[CasperAtUU]

Seconded. We use CFA in a course on measurement, it's a very useful method. It's part of the Lavaan package in R, and I'd love to see it added to JASP since students have a pretty hard time using R to perform the analysis (they are familiar with SPSS).

[EJWagenmakers]

@vankesteren ?

[Kucharssim]

Note that it is possible to conduct CFA in JASP using the SEM module. However, the model has to be specified using the lavaan syntax (so it is only slightly easier than doing it in R) and currently it is not possible to obtain factor scores directly. Maybe it’s a good idea to create a CFA in the Factor analysis menu to enable some simple factor models by point and click...

[vankesteren]

We will implement this reasonably soon!

[WilhelmKuntner]

Hello! This is my first post here, so I wanted to thank for JASP!!

In the SEM tab there is the advanced option to define several models (up to nine). Intuitively I would expect to be able to select and re-edit a model previously entered. However, JASP seems to overwrite the content of the model-slot selected by the drop-down menu with the actual content of the editing window. Intuitively I would expect to have the option to either read a previously entered model or to overwrite it (or rather delete it). Moreover, if the editing window is empty, selecting a previously defined model will not delete it.

Many thanks again for this great project and kind regards!

[EJWagenmakers]

Thanks! @vankesteren, any comments on this request? :-)

[vankesteren]

Hey @WilhelmKuntner,

yes, this advanced option is some legacy code. In the future, we will edit this so that the user interface will be more intuitive in this regard. It works in a way, but indeed it is currently implemented such that the model is overwritten every time, so you'll have to start with the simplest model and build up from there (and never touch your original model again). Hence we relegated it to the "advanced" tab.

Something you could do is write down your models in a text editor and copy/paste them once.

Might I ask which model comparisons you're specifically thinking about? I am currently working on an intuitive CFA module that includes some model comparisons such as measurement invariance testing. I could implement your ideas if it's relevant to more users!

[WilhelmKuntner]

Hi @vankesteren,

at the moment I am using RStudio with R Markdown, where the work flow in fact comprises lots of copy-and-paste operations.

I am trying to ascertain the dimensionality of a psychometric scale. Accordingly, a one-factor-model is tested against a two-factor-model by means of a CFA. Once I will be sure about the dimensionality (see below), I'm planning to check for measurement invariance across the German and the Italian version of the items as well as across females and males (within a German speaking sample). While doing so, I've been trying to find the best way to deal with non-normally distributed data on a 5 point Likert scale, and sample sizes ranging between ca. 600 and 100 cases, depending on language and sex. Therefore, while consulting some relevant literature (Bandalos, 2014; DiStefano & Morgan, 2014; Li, 2016a, 2016b), I am also comparing the results obtained by using the diverse estimators provided by lavaan (ML vs. MLM vs. MLR vs. DWLS vs. WLSMV).

So, based on my special case, an ideal work-flow with JASP would permit to store both the model syntax and all other sem()/cfa() settings. The most tangible advantage compared to the work-flow within RStudio would be:

1. Avoiding the sequential structure of an R Markdown (which makes sense when making automated reports) file while being able to enter and select diverse configurations of a CFA (or SEM).
2. Obtaining automatically a comparison table of the diverse fit.measures.

References:

• Bandalos, D. L. (2014). Relative Performance of Categorical Diagonally Weighted Least Squares and Robust Maximum Likelihood Estimation. Structural Equation Modeling: A Multidisciplinary Journal, 21(1), 102–116. https://doi.org/10.1080/10705511.2014.859510
• DiStefano, C., & Morgan, G. B. (2014). A Comparison of Diagonal Weighted Least Squares Robust Estimation Techniques for Ordinal Data. Structural Equation Modeling: A Multidisciplinary Journal, 21(3), 425–438. https://doi.org/10.1080/10705511.2014.915373
• Li, C.-H. (2016a). Confirmatory factor analysis with ordinal data: Comparing robust maximum likelihood and diagonally weighted least squares. Behavior Research Methods, 48(3), 936–949. https://doi.org/10.3758/s13428-015-0619-7
• Li, C.-H. (2016b). The performance of ML, DWLS, and ULS estimation with robust corrections in structural equation models with ordinal variables. Psychological Methods, 21(3), 369–387. https://doi.org/10.1037/met0000093

[WilhelmKuntner]

(@vankesteren) ... but then, given that the JASP SEM-module is probably interesting especially to scholars who still are rather inexperienced in terms of SEM (including myself), my above suggestion could lead to a so-called "data-torturing". The counterpart is, of course, a theory-led procedure. Anderson and Gerbing (1988) proposed a two-step approach to CFA. An optional template for running CFA following this (or any other, more up-to-date) procedure may be useful in order to keep things consistent and systematic. I cannot say how that fits in with JASP's goals. This is indeed to be understood as a spontaneous idea not thought through in detail. However, this idea seems to resemble an imperative I read somewhere on the JASP website, which claims that JASP shouldn't leave alone inexperienced (or lazy) users with typical pitfalls of statistical analysis.

• Anderson, J. C., & Gerbing, D. W. (1988). Structural equation modeling in practice: A review and recommended two-step approach. Psychological Bulletin, 103(3), 411–423. https://doi.org/10.1037/0033-2909.103.3.411

[vankesteren]

@WilhelmKuntner Thank you for your detailed feedback!

• Robust estimation & various estimators will be part of the CFA module
• I'll look into whether and how we want to incorporate model comparisons such as the dimensionality one you describe
• A part of the CFA procedure will be lavaan model code output. I will also look into extracting the specific lavaan settings for running the model directly in R

[vankesteren]

As of JASP 0.10, we have CFA!