Adam Rogers posted on Wednesday, August 31, 2016 - 10:45 am
I am trying to estimate a growth model for a categorical/ordinal outcome variable. Specifically, I am predicting frequency of alcohol use (on ordinal scale 0-7) from adolescents' age. To do so, I used a logit link function which gives me parameter estimates that are expressed in terms of a latent “ability” on the trait, as well as a set of threshold parameters that denote the value of the trait at which the probability of scoring in or above a particular response category is 0.50. I have been able to successfully estimate the linear and quadratic growth models. However, when I try to predict the slope using another variable (e.g., parental monitoring), the sign of the estimate is flipped. This happens consistently for numerous variables that I use as predictors of the slope (bivariate correlations are all in the opposite direction). I would be appreciative of any insights or guidance as to what could be the problem. Thank you!
Define: age_c = age_bnd3-3; age_sq = age_c*age_c;
model: %within% slope|ause2 on age_c; quad|ause2 on age_sq;
%between% slope ause2 quad with slope ause2 quad; slope on c6parmon; ause2 on c6parmon; quad on c6parmon; ![ause2@0]; [ause2$1 ause2$2 ause2$3 ause2$4 ause2$5 ause2$6 ause2$7];
-age_bnd is the predictor -ause2 is the outcome -c6parmon is a predictor of the slope