Bootstrapping_NOT running completely PreviousNext
Mplus Discussion > Categorical Data Modeling >
Message/Author
 Sanjoy posted on Thursday, May 26, 2005 - 12:03 am
Dear Professor/s ... I have tried bootstrapping with 5 different no. of replication (30,50,100,250, 500)

under all scenarios it stops roughly after half way
"
Number of bootstrap replications
Requested 30
Completed 16
"

and I'm getting "*****" under SE colum primarily

Q1. Why is it happening this way

Pagee 114, (Testing SEM edited by Bollen) Prof. Bollen raised a nice point "Bootstrap resampling from the observations does not* resemble sampling from a population in which the null hypothesis holds" ...then he proposed a refinement

* I think what he meant is "may not"

Q2. how does MPlus handle this situation? his refinement was in reference to ML, I'm using WLSMV ... what is your advice?

thanks and regards

command I used

ANALYSIS:
ESTIMATOR=WLSMV;
PARAMETERIZATION=THETA;
BOOTSTRAP=30;

MODEL:

R BY D31 D32 D33;
B by D43 D53 D63;
Bid1A on R B S B1SK F3 F4
F5 F6 F6B F7
F10;
R on B S F22 F5 F6 F7;
B on R B1SK F4 F5 F6 F6 F10;


OUTPUT:
CINTERVAL;
 Linda K. Muthen posted on Thursday, May 26, 2005 - 5:33 am
The series of asterisks means that the number is too large to print in the space provided. If you ask for TECH9, you will see why some of your data sets have problems.
 Sanjoy posted on Thursday, May 26, 2005 - 12:23 pm
Thanks madam ... but TECH9 works only when we run Monte carlo and "bootstrap" is not available in conjunction with Monte Carlo

"
*** ERROR in Output command
TECH9 option is only available with MONTECARLO.
Request for TECH9 is ignored.
1 ERROR(S) FOUND IN THE INPUT INSTRUCTIONS
"

the problem of "*****" remains in SE and therefore in Est/SE

Thanks and regards
 bmuthen posted on Friday, May 27, 2005 - 9:31 am
A side effect of bootstrapping is that is exposes any weaknesses in the model. An example maybe relevant your situation is that with categorical variables, with small samples or skewed data, some draws will give low-frequency bivariate tables and therefore run into problems in doing a good job in estimating correlations and even more so the weight matrix. This can lead to convergence failure. I would expect the WLSMV SEs to often be as good as the bootstrap SEs (and not as prone to failure).
 Sanjoy posted on Friday, May 27, 2005 - 11:00 am
Thank you Sir ... a great relief indeed, I have tried bootstrapping almost infinte time, but nothng was happening ... now I can mention this in my paper, and 'will stick to WLSMV

thanks and regards.
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: