Ordinal regression with bootstrap
Message/Author
 Vera posted on Wednesday, November 06, 2013 - 12:01 pm
Hi there,

I am running an ordinal logistic regression, observed variables only, and need to test the significance of 4 mediators. I am new to Mplus and would like to make sure I am inputting and interpreting things properly.

1. Does this part of the input look correct?
categorical = Y;
ANALYSIS: bootstrap=10000;
Model:
M1 M2 M3 M4 on X c1 c2 c3 ;
Y on X M1 M2 M3 M4 c1 c2 c3 ;
MODEL INDIRECT:
Y IND M1 X ;
Y IND M2 X ;
Y IND M3 X ;
Y IND M4 X;
output: CINTERVAL (BCBOOTSTRAP);

2. In order to see whether the total indirect effect and specific indirects were significant, I need to look at CONFIDENCE INTERVALS OF TOTAL, TOTAL INDIRECT, SPECIFIC INDIRECT, AND DIRECT EFFECTS, right?
 Linda K. Muthen posted on Wednesday, November 06, 2013 - 1:34 pm
1. The best way to check an input is to run the analysis and see if you get what you expect.

2. You can look at the p-value for the z test or at confidence intervals.
 Jetty posted on Thursday, November 07, 2013 - 12:39 pm
Thanks, Linda. I am puzzled by a slight discrepancy between the CIs for the indirect effects and actual p-value for the z test. I have one effect for which p=.137 but the 95% CI (I am looking at lower and upper 2.5%) do not include 0 (lower 2.5%=.001 and upper 2.5%=.015). Am I missing something?
Thank you
 Linda K. Muthen posted on Thursday, November 07, 2013 - 1:18 pm
These should be the same for symmetric confidence intervals but not for BOOTSTRAP and BCBOOTSTRAP confidence intervals.
 Jetty posted on Thursday, November 07, 2013 - 3:23 pm
Thanks, symmetry was the first thing that came to my mind as well. I will report the CIs as they result from bootrapping.
 Jetty posted on Friday, November 08, 2013 - 7:44 am
Dr. Muthen,
I ran the model from my first post without ANALYSIS, MODEL INDIRECT, and OUTPUT commands because I wanted to get the fit indices for it, as I saw recommended on this board. The estimator is still WLSMV, parametrization DELTA, WRMR is identical, but some of the SE for the parameters and their associated p value differ. Why is that?
Thank you
 Jetty posted on Friday, November 08, 2013 - 8:43 am
bootstrapped SEs..got it