Suppose we fit a proportional odds model to some data with $Y$ the response variable being ordinal. If we get 0 estimated for all the coefficients (except the intercepts), what does that mean? How would we interpret the model?
3
-
$\begingroup$ Ross T, were you the same person who posted a nearly identical question: stats.stackexchange.com/questions/22700/…? Duplicate posts (and accounts) are discouraged. $\endgroup$– MacroCommented Feb 14, 2012 at 13:42
-
$\begingroup$ As I said in the other thread, if the regression coefficients are exactly 0, this means that the distribution of your predictors is exactly same for each level of $Y$. $\endgroup$– MacroCommented Feb 14, 2012 at 13:43
-
$\begingroup$ @Macro: If the distribution of the predictors is exactly the same for each level of Y, does this mean that the proportional odds assumption does not hold? $\endgroup$– Ross TCommented Feb 15, 2012 at 2:18
Add a comment
|
1 Answer
$\begingroup$
$\endgroup$
One possibility is that the scale is wrong - e.g. if one of your independent variables was (say) height of adult humans and you had measured it in millimeters, then the effect of each additional mm of height might be so close to 0 that it gets rounded.
Another is that you've just got a bad model.
Another is that there is no variation in the variables.
Are they exactly 0 or just really small?
answered Feb 14, 2012 at 12:05