Add more estimators and change the optioning for estimators, test, and SEs

[juliuspfadt]

So essentially what happens through this PR:

• all four options (estimator, test, information, standard error) have a default setting which can change in the background depending on the scale of the variables involved or the mimic option or the estimator option

• add estimators from lavaan: see here https://rdrr.io/cran/lavaan/man/lavOptions.html

• add a tooltip explaining the estimators

• add more model test options https://rdrr.io/cran/lavaan/man/lavTest.html

• add a few more footnotes under the model fit table showing the background options used when the setting is initially default

• add a footnote under the model fit table if there are warnings encountered

• add a checkbox to show those warnings as html text in the output window

• there is not really a lot of conditioning going on what test, se, information, estimator goes together or not (although I left the contains in place that were already there). When options do not match lavaan gives an error which means JASP will give an error and ideally that error is informative (I hope). Sometimes lavaan changes options in the background but gives a warning, we display that warning now in the html.

• adds thresholds table and intercepts for some special cases, that is, missing=fiml, or some but not all ordinal variables

• restructure the additional fits tables

• adds standardized CIs, all standardization now goes through standardizedSolution. The qml now offers to choose the standardization, and the tables will present show only one estimate, either unstd, std.all, std.lv, or std.nox

• there was an error in the bootstrapping, where the SEs in the parameter tables were still the SEs based on the initial model fit, that is, usually based on a single ML estimation. Those are now replaced with the bootstrapped estimated SEs

• the standardization for the bootstrapped option is a bit tricky, this doesnt just work through standardizedSolution. For now we have

  • in every bootstrap run the standardised solution is computed and the coefficient estimates are saved, then parameterEstimates is called on the resulting object with the CI and the boot.ci.type work just like for the unstandardized solution
  • this also means that the bootstrapped CIs do not always yield completely admissible solutions, that is, loadings larger than 1, or negative variances

• fixes [Bug]: WLSMV corrected statistics are not reported in the output jasp-issues#2644

• fixes JASP results DWLS - Discrepancies with Lavaan jasp-issues#1259

• fixes robust ML and Satorra Bentler Chi square jasp-issues#959

• fixes [Feature Request]: MLR Estimator for CFA/SEM jasp-issues#1484

• fixes [Bug]: computation of standardized estimates (mediation analysis) jasp-issues#2740

• fixes CIs for std(all) in SEM jasp-issues#1102

• fixes https://github.com/jasp-stats/INTERNAL-jasp/issues/2182

• fixes [Bug]: SEM Analysis Terminated Unexpectedly  jasp-issues#2879

• fixes [Bug]: SEM model jasp-issues#2905

• fixes [Bug]: SEM: Error is Node Stack Overflow jasp-issues#2908

What is a bit difficult is that some of the estimators change not only the estimation method, but also the model test and the standard error calculation. What is interesting is, if model test and standard errors are actually specified, beyond the default, their specification overwrites the estimator specification.

@Kucharssim maybe you have some feedback :)

toDo:

• More testing
• extend the options to the other SEM analyses (lgcm, mediation etc.)
• handle ordinal data
• help file
• I wonder about Upgrades.qml... if we need changes there. So far it seems fine.

[FBartos]

holly molly, I don't wanna be reviewing this!

[FBartos]

Mostly snake_case to camelCase changes needed, still gonna check the issues in JASP

[FBartos]

coefWithCallbackStd ?

[FBartos]

errId ?

[FBartos]

covMat ?

[FBartos]

lavaanArgs ?

[FBartos]

• for the bootstrap, shouldn't be there an option to specify the number of bootstrapped samples? I tried putting it on and it takes ages (also including a progress bar?)
• the information messages could be more informative by using plain English instead of the lavaan syntax names? (e.g., robust.sem)
  

[juliuspfadt]

• for the bootstrap, shouldn't be there an option to specify the number of bootstrapped samples? I tried putting it on and it takes ages (also including a progress bar?)

Taking ages can very well be. Especially if you have a model that does not fit very well. It does include a progress bar does it not? How do you not see an option for the bootstrap samples:

• the information messages could be more informative by using plain English instead of the lavaan syntax names? (e.g., robust.sem)

Agreed

[juliuspfadt]

I am patiently waiting for your review

[juliuspfadt]

Ready :)

[FBartos]

So, I dinally managed to get this running on 19.0 release --- you were right about that.

[FBartos]

• there is one too many closing brackets in the end
• the boostrap progress bar and settings is missing for the "Model test" (feel free to make an issue for future for this)
  

otherwise seems to be ok

[juliuspfadt]

Approve?