jasp-stats / jaspfactor Goto Github PK

So there is a bit of a cluttered output in the tables when running a multi group cfa:
For some tables the structure is:
Group1

• table1
• table2

Group2

• table1
• table2
  

For instance the residual Cov table is differently structured which aligns with the SEM module's multi group output:
Table1:

• group1
• group2

Table2:

• group1
• group2
  

I would like to make this consistent. But I rather start with it, once the other PRs are merged. Any thoughts @Kucharssim?

• add residual matrix output for EFA and PCA
• add fit measures SRMR and CFI for EFA

Unit tests:

• polychoric/tetrachoric correlation matrix
• parallel analysis table
• mardia test

Small things:

• clean up help files
• consistent camelCase naming for objects

jasp-stats/jasp-desktop#4320

To reproduce this one has to use the latest version for jaspFactor, which might not be in the nightlies yet.
However, in R one can:

library(lavaan)
dt <- read.csv("https://raw.githubusercontent.com/jasp-stats/jasp-desktop/development/Resources/Data%20Sets/Data%20Library/14.%20SEM/Grand%20Point%20Average.csv", header = T)

mod <- "
f1 =~ gpa1+ gpa2+ gpa3+ gpa4+ gpa5+ gpa6
f2 =~ job3+ job4+ job5+ job6
f1 ~ c(0, NA)*1
f2 ~ c(0, NA)*1
"
fit <- cfa(mod, dt, std.lv = TRUE, group = "admitted", meanstructure = TRUE)
fitmeasures(fit)
summary(fit)

this part is also what is produced through JASP:

f1 ~ c(0, NA) * 1
f2 ~ c(0, NA) * 1

stems from here: https://github.com/juliuspf/jaspFactor/blob/invarianceFix/R/confirmatoryfactoranalysis.R#L323-L333
I am confused and wonder why this piece of model syntax is added. It seems sufficient to just set meanstructure to TRUE.

I obtain standard error issues (can't be calculated) if the model syntax is specified like above, but no issues when only meanstructure=TRUE and the piece of code is omitted from the model syntax. All other estimates stay equal.

In terms of model syntax this should mean that the mean of the first factor for group 1 is set to 0 and the mean for the second group is forced to be a free parameter.... but why

As I understand this, with meanstructure you can either estimate the latent means or the manifest means. By default it is always the manifest means that are estimated, and only if one forces the manifest means to be zero in the model.syntax will the model estimate the latent means. But we do not have that option in JASP, so in any case the latent means will always be zero.

I suggest to not label the intercepts for the factors when we have multiple groups since meanstructure=TRUE is specified either way when the meanstructure option is chosen. The other parameter estimates won't change, it would only lead to fewer convergence issues since there are two degrees of freedom more.