MiPowerFit: not detecting some group constraints

[anna.deca...@upf.edu]

Hi!

I am using the function MiPowerFit from semTools_0.4-15.926 (R version: 3.5.0) to detect local misspecifications on a multiple group model in lavaan. 

I attach an example (R script), in which I first input the data as covariance matrices, then run the model in lavaan and then I compute the MiPowerFit test. 

Following the model specified in the example, one can observe that there are some parameters constraint to be equal across groups. 

* In the case where the parameter is already fixed to a certain value in group 1 (i.e. r(i), m(i), varF(i)), there is no problem. I get the all parameters in the different groups in the MiPowerFit output. 

* However, in the case where the parameter is free in group 1 or 2 but is later constraint to be equal in the following groups, these do not appear in the MiPowerFit output. These are: the loadings from the F(i) to the TS(i) factors, the covariances between the F(i) factors, the error variances of the y(i) and the variances of the M(i) factors.

Is there a way I can get these parameters in the MiPowerFit output?

Thank you very much in advance.

Best, 

Anna

[Terrence Jorgensen]

Is there a way I can get these parameters in the MiPowerFit output?

No, miPowerFit() is only designed to work with output from the modificationIndices() function, which are only provided for adding a parameter that is fixed to a certain value (see the add= argument in the ?lavTestScore() help page).  It does not provide statistics for parameters that are already estimated (albeit with constraints).  You can obtain them using the more general score test (see the release= argument in the ?lavTestScore() help page), but note that a single equality constraint only gives you a single test statistic because it only buys you a single df.  So you will not get a "modification index" or (S)EPC for each parameter that is constrained to equal another.  You will get one test per constraint (df), but no (S)EPC for that constraint because the constraint itself is not a parameter, although you can request epc=TRUE to see how all estimated parameters would be affected by releasing the constraint(s) specified in the release= argument.  

It is not generally the case that a constraint is something as simple as an equality constraint, so I'm not sure I can update miPowerFit() to optionally work with lavTestScore() output, but I will think about this some more and post here if I update it.

Terrence D. Jorgensen

Postdoctoral Researcher, Methods and Statistics

Research Institute for Child Development and Education, the University of Amsterdam

UvA web page: http://www.uva.nl/profile/t.d.jorgensen

[ANNA DE CASTELLARNAU ILLARRAMENDI]

Dear Terrence,

Thank you for your response. I've used the lavTestScore() as you suggested and it indeed provided the EPC for the constraint parameters. However, to judge on the parameters to introduce I would need the MI and the power of MI. Furthermore, it would be necessary to replicate the results one can get with JRule (Saris, Satorra and van der Veld 2009). I attach the JRule output that I get with the same model as in the previous example. 

Do you think that would be possible?

Best, 

Anna

--
You received this message because you are subscribed to the Google Groups "lavaan" group.
To unsubscribe from this group and stop receiving emails from it, send an email to lavaan+un...@googlegroups.com.
To post to this group, send email to lav...@googlegroups.com.
Visit this group at https://groups.google.com/group/lavaan.
For more options, visit https://groups.google.com/d/optout.

--

Anna DeCastellarnau

Researcher at Research and Expertise Centre for Survey Methodology (RECSM)

Department of Political and Social Sciences

Universitat Pompeu Fabra

Ramon Trias Fargas, 25-27 | 08003 Barcelona | Edifici Mercè Rodoreda | Office: 24.406

Tel: +34 93 542 1163 | www.upf.edu/survey/ | sqp.upf.edu |

[Terrence Jorgensen]

Do you think that would be possible?

Not with the way the software is designed.