tidyLPA: cannot estimate some models and number of parameters

[Paul Kong]

I had done a LPA using tidyLPA on my first set of variables and it went well. However, when I ran it on the second set of variables, I encountered the following problem, so that I would like to ask this:

tidyLPA analysis using mclust:

 Model Classes AIC     BIC     Entropy prob_min prob_max n_min n_max
 1     3       2386.24 2485.04 0.89    0.84     0.98     0.07  0.82
 2     3                                                            
 3     3       2419.15 2587.11 0.94    0.80     0.99     0.02  0.89
 6     3                                                            
 BLRT_p
 0.01  
       
 0.01  
       
Warning messages:
1: Mclust could not estimate model 2 with 3 classes.
2: Mclust could not estimate model 6 with 3 classes.
3:
One or more analyses resulted in warnings! Examine these analyses carefully: model_2_class_3, model_6_class_3 

Moreover, I also want to use the function calc_lrt() but I am not sure about the numbers of parameters in the models. I try to deduce them as follows. Are they correct?

- Number of parameters in Model 1 = Number of variables in the model

- Number of parameters in Model 2 = Number of variables x Number of variables

- Number of parameters in Model 3 = Number of variables x Number of profiles

- Number of parameters in Model 6 = Number of variables x Number of variables x Number of profiles

[Rosenberg, Joshua]

Hi Paul, 

In the case of not being able to estimate some models, this indicates that the estimation did not converge for those models. Generally, this indicates that simpler models are preferable/to be interpreted. You might try running these I'm MPlus to see whether you get a similar message.

For the number of parameters, I'm not sure, but I think it may be something other that what you described based on what is estimated for each of the models (see the vignette for more). There may be a value in the output or a way to determine from the output how many parameters are estimated - in addition to using a simple equation like those you proposed to determine the number.

Josh

--

Joshua M. Rosenberg, Ph.D.
Assistant Professor, STEM Education

The University of Tennessee, Knoxville
Department of Theory and Practice in Teacher Education

College of Education, Health, and Human Sciences

Homepage: https://joshuamrosenberg.com

He/Him/His

---

From: tid...@googlegroups.com <tid...@googlegroups.com> on behalf of Paul Kong <pau...@gmail.com>
Sent: Sunday, July 26, 2020 9:18 AM
To: tid...@googlegroups.com <tid...@googlegroups.com>
Subject: tidyLPA: cannot estimate some models and number of parameters

 

[External Email]

--
You received this message because you are subscribed to the Google Groups "tidyLPA" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tidylpa+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tidylpa/CAOKzw-tvcm-dn0n%2Bqud4Ve%2B2ye7RsuJQh8pWfFc1-DK4%2B7KaUg%40mail.gmail.com.

[Caspar van Lissa]

Josh is correct. The number of parameters can be found somewhere in the output, but in general I'd advise against using the lrt as it's known to be biased. That is why we output BLRT by default.

[Paul Kong]

Dear Josh,

Thank you so much for your reply. And again thank you for having developed the package which helps with the analysis.

Regards,
Paul

[Nadia Al-Dajani]

Hi all, 

I am currently in the process of writing a revise and resubmit, where I used tidyLPA to run a latent profile analysis (thank you for developing this package!) The reviewer asked for Vuong-Lo-Mendell-Rubin as an additional metric for determining the number of profiles. While I understand that this is not the best metric due to biased estimates, I would like to obtain this information anyways as part of my response to the reviewer. 

I am using calc_lrt and having some issues finding the number of parameters - it was mentioned that this can be found in the output, could you please advise? Are the number of parameters simply the number of rows in the output from "get_estimates"?

Thank you!

Nadia

[Marine Granjon]

Hi all,

I have the same issue as Nadia. Is it the model$df?

Thanks,

Marine

[Jet Buenconsejo]

Hi all,

I also can't find the number of parameters in the LPA output. 

The number of parameters is needed to calculate the Lo, Mendell, & Rubin likelihood ratio test.

Anyone who figured this out?