| Message/Author |
|
|
| Lydia Zhang posted on Thursday, February 01, 2018 - 3:08 pm
|
|
|
I have a quick question about the new 3-step approach for predicting distal outcomes from latent classes. The examples are all about linear regression between latent classes and distal outcomes for the third step.
I am wondering if I can do growth curve modeling for the third step, that is, predicting trajectories over time from latent classes and covariates in Mplus? |
|
|
|
|
| Yes, you should be able to do that. |
|
|
|
|
| We have a 5-class model and 20 multiply imputed data sets. Question: Can we do a manual 3-step approach with multiple imputation? If so, can you point us to the best resource for carrying out this analysis? Many thanks! |
|
|
|
|
Manual 3-step can't be done very well with imputation because the model changes across the data sets. You can switch to BCH, see web note 21, http://statmodel.com/examples/webnotes/webnote21.pdf. That would be easier to do with imputation, similar to User's Guide example 13.13. If you want to do Manual 3-step you can run the 20 sets with the 20 different models separately then combine them manually like in the bottom of page 3 https://www.statmodel.com/download/MI7.pdf
If your measurement data is not imputed (is the same across the 20 sets) then you can use the same model in the third step and apply UG 13.13 method with the manual 3-step (but only in this special case when the measurement part is not being imputed). |
|
|
|
|
Hi,
I want to run a LTA with covariates and distal outcomes using multiple imputation. Can I use the manual three step approach? If not, what would you suggest?
Thank you very much!! |
|
|
|
|
Take a look at Section 4 in
http://statmodel.com/download/webnotes/webnote15.pdf
and Appendices F-N
http://statmodel.com/download/AppendicesOct28.pdf |
|
|
|
|
| Hi, I already know both documents. My question is whether I can use this approach with multiple imputation. Thank you. |
|
|
|
|
Or more concretely with type:imputation
Thank you! |
|
|
|
|
If the imputed variables are in the auxiliary model only then it should be pretty straight forward and you would just use type=imputation for the last step.
If you have imputed variables among the class indicators you would have to repeat the full muti-step analysis for each data set separately and combine the results manually. |
|
|
| Sara Suzuki posted on Thursday, March 05, 2020 - 9:25 am
|
|
|
I have a three time-point LTA and want to estimate distal outcomes using manual 3-step. However the missingness in the class indicators is causing problems for me.
As I am doing manual 3-step, FIML does not work.
Currently, is there no way to do multiple imputation for the class indicators with manual 3-step in Mplus?
How does one go about doing this manually...
Or, do you have a recommendation for an alternative method for estimating a distal outcome that invokes FIML or allows for automatic multiple imputation? |
|
|
|
|
Q1: No.
I would use the 1-step method and therefore benefit from FIML. |
|
|
|
|
Hello,
I am running an LTA model and to test which predicting factor relates to the transition probabilities. I used an ANOVA version of a manual Vermunt’s 3-step approach. Is there a way to obtain effect sizes for the conducted analysis?
I have looked through the manual, perhaps I have missed it. Thanks! |
|
|
|
|
I don't know about effect size for transitions. Why not use odds ratios as in Web Note 13:
Muthén, B. & Asparouhov, T. (2011). LTA in Mplus: Transition probabilities influenced by covariates. Mplus Web Notes: No. 13. July 27, 2011. |
|
|
| Back to top |
