Proportional odds test
Message/Author
 Peter Taylor posted on Friday, October 06, 2017 - 8:11 am
I am running a model with three predictors: two latent variables (with categorical indicators) and one observed binary predictor. The outcome is binary. When i run this model I notice Mplus runs a "brant wald test for proportional odds". I assume this is the test of the proportional odds assumption that is used in ordinal regression.
1. i'm a little unsure why it is running this test for each indicator of my latent variables.
2. if the test is significant, and so the assumption violated what can i do about this?

The syntax is below:

VARIABLE: NAMES ARE sex age SI SP SC SA NI NA d1-d16 e1-e16 rs1-rs5 exp1 exp2 s1-s25 an1-an14 aa1-aa3 bu1-bu7 be1-be5 g bi nos min;

USEVARIABLES ARE an2 an4 an12 an14
be1-be5 d1-d16 min na an6r an8r an10r;
MISSING = ALL (999);
CATEGORICAL an2 an4 an6r an8r an10r an12 an14 d1-d16 na;
DEFINE: an6r=5-an6; an8r=5-an8; an10r=5-an10;

analysis:
estimator = MLR;
MODEl:
dep by an2 an4 an6r an8r an10r an12 an14;
BE by be1-be5;
def by d1-d16;
def be on min;
na on min def be dep;
 Bengt O. Muthen posted on Friday, October 06, 2017 - 2:55 pm
1. Your model has a covariate "min" and therefore ordinal variables influenced by that covariate are assumed to follow the proportional odds model wrt to this covariate.

2. Take a look at the literature on this test, either in our RMA book or in sources such as

https://us.sagepub.com/en-us/nam/logistic-regression-models-for-ordinal-response-variables/book226230