Cross-lagged path analysis
Message/Author
 Gia Grisagoni posted on Thursday, June 11, 2015 - 3:39 am
Dear Drs. Muthen,
I have run the following path analysis with 5 binary variables (0=normal 1=abnormal scores), assessed at two waves, using WLSMV:

Grouping IS sex (0=boys 1=girls);
SAVEDATA: DIFFTEST IS model_sex.dat;

MODEL:
X1 ON SES;
X2-X5 ON t1_age SES;
Y1 ON SES;
Y2-Y5 ON t2_age SES;

X1 WITH X2-X5;
X2 WITH X3-X5;
X3 WITH X4-X5;
X4 WITH X5;
Y1 WITH Y2-Y5@0;

Y2 ON X2
X1 (EP1);
Y3 ON X3
X1 (PP1);
Y4 ON X4
X1 (HP1);
Y5 ON X5
X1 (CP1);
Y1 ON X1
X2 (EP2)
X3 (PP2)
X4 (HP2) ! significant in girls only
X5 (CP2);

Now I am not sure how to report the results (preferably STDY?), considering that I am primarily interested in the "abnormal to abnormal scores" associations (1-1). In this analyses, would it be appropriate, for example, to state that "in girls, abnormal scores on hyperactivity predicted overweight status over a 1-year period, after adjusting for baseline weight status". Moreover, I am wondering if I can derive odds ratios for these associations as well.

 Bengt O. Muthen posted on Thursday, June 11, 2015 - 12:53 pm
In your regressions of Y on X, StdYX=StdY.

WLSMV uses only probit so no odds ratios are available as with logit.
 Gia Grisagoni posted on Monday, June 15, 2015 - 5:55 am
Dear Bengt,

thanks for your quick response. Might I follow up on this issue, please. You are saying that no odds ratios can be derived when using WLSMV. But what about their calculation? For example, can I apply the following formula [P(u = 1/x) = F (a + b*x)] from the Mplus User's Guide 7 (Chapter 14, Page 492) to obtain odds ratios in my cross-lagged design? I'm just asking because my supervisor insisted on reporting odds ratios.

Thanks for your help and kind regards.
 Linda K. Muthen posted on Monday, June 15, 2015 - 6:24 am
You can use ML where you will get logistic regression and then can obtain odds ratios.