Specify starting values for all param... PreviousNext
Mplus Discussion > Latent Variable Mixture Modeling >
Message/Author
 Hong Ling posted on Wednesday, September 13, 2017 - 1:44 pm
Dear Dr. Muthen,

Thank you for the nice technical manual (http://www.statmodel.com/download/Starts.pdf) that clarifies how the permutation of starting values is provided by the software.

I wonder whether Mplus supports the specification of the starting values for all estimated parameters? I plan to examine the effects of different sets of starting values on the estimation accuracy, which requires to manipulate the starting values of all parameters to specific values, e.g., a set of extreme values for the parameters to be estimated, or a random sampled from a specific parameter space.

If so, where can I find reference on how to achieve that? The closest syntax identified was the "STARTING =" under the MONTECARLO command. Yet it is insufficient information to achieve the goal.

If not, the follow-up question is that does the permutation of the starting values supplied by Mplus use the same seed for each estimation? In other word, if I replicate one estimation multiple times without specifying any seed for starting value, will the estimation process and results be exactly the same?


Best regards,
 Bengt O. Muthen posted on Wednesday, September 13, 2017 - 4:40 pm
Yes, you can specify starting values for each parameter to be estimated. Note that each example in the User's Guide has a corresponding Monte Carlo version on our website - this shows how you give starting values.
 Hong Ling posted on Wednesday, September 13, 2017 - 8:17 pm
Thank you very much for your speedy response! The information is very helpful!

PS: For others' reference, below is the link I found on how to specify starting values in LCA (e.g., Example 7.5): https://www.statmodel.com/HTML_UG/chapter7V8.htm
 Hong Ling posted on Monday, October 09, 2017 - 1:05 pm
Dear Dr. Muthen,

May I ask a follow-up question about how to verify whether the starting values have been implemented as specified in Mplus? I have tried to output TECH 8, but it does not include such details.

I get identical results regardless of very different starting values, over dozens of replication. So I want to double check the key step.

Thanks!
 Bengt O. Muthen posted on Monday, October 09, 2017 - 4:18 pm
TECH1 shows the starting values used.
 Hong Ling posted on Monday, October 09, 2017 - 7:47 pm
Thank you very much! The detailed information from TECH1 is very helpful to identify potential problems.
Back to top
Add Your Message Here
Post:
Username: Posting Information:
This is a private posting area. Only registered users and moderators may post messages here.
Password:
Options: Enable HTML code in message
Automatically activate URLs in message
Action: