| Message/Author |
|
|
| Sara Suzuki posted on Thursday, September 24, 2020 - 11:49 am
|
|
|
I would like to run an LTA with covariates and distal outcomes, using the 3-step method. There are missing data in the indicators so I am incorporating the steps outlined in section 7.2 of webnote 21 (version 8, 9/15/2020).
There are also missing data in the covariates and distal outcomes, how can I invoke FIML for those missing data?
Thank you. |
|
|
|
|
Missing data for the distal outcome is handled automatically and it does not require any special treatment.
For missing data in the covariates see Section 7.1 |
|
|
| Sara Suzuki posted on Thursday, September 24, 2020 - 6:23 pm
|
|
|
Would you recommend doing a model with covariates (following section 7.1) and a different model with the distal outcomes (following section 7.2)?
I cannot figure out how it would be possible to combine those two methods. |
|
|
|
|
| You would run Figure 10 first, then all the runs in section 7.2, where in the last run (Figure 19), you will replace the data command with the data command of Figure 11. |
|
|
| Sara Suzuki posted on Friday, October 02, 2020 - 7:11 pm
|
|
|
Thank you Dr. Asparouhov. I think I managed to run this correctly. I want to check a couple of things:
1) When running figure 13-15 of section 7.2, would it be ok to fix the parameters to what I got in a full LTA without covariates and distal outcomes? Instead of estimating new parameters for each time point...
2) In the last step, where I am doing Figure 19 but with [data: type=inmpuation;], should I still restrict to [starts=0]? |
|
|
|
|
| Yes on both questions. |
|
|
| Sara Suzuki posted on Tuesday, October 06, 2020 - 2:52 pm
|
|
|
| My last step (Figure 19) has class counts and proportions that do not match those in my LTA without the predictors and outcomes. This is unexpected for me because the classification parameters are brought over from this original LTA without predictors and outcomes, through Figure 13-15 and Figure 16-18. Why do the class counts and proportions suddenly change at Figure 19? |
|
|
|
|
| Small changes are fine: within 10%. To understand why class sizes can change you should take a look at Section 5. |
|
|
| Sara Suzuki posted on Tuesday, October 06, 2020 - 5:44 pm
|
|
|
Upon closer inspection, it looks as if the class sizes are only changing slightly--within 10%--but only for my figure 19 model that has outcomes.
When I have predictors (of transitions) in figure 19, the class sizes are doing what I think are class shifts.
Is there a way I can prevent class shifting? |
|
|
|
|
| The only other reason (apart from Section 5) that I have seen would be very low entropy, less than 0.5. In that case the method sometimes falls apart. |
|
|
| Sara Suzuki posted on Wednesday, October 07, 2020 - 3:45 pm
|
|
|
| Thank you—that was helpful. |
|
|
| Back to top |
