Suppose I estimate an ordinal logistic regression model: $Y$ ~ $\beta_1X + \beta_2Z$ where $Y$ is the ordinal-scale dependent variable with $y = 1, 2...k$ responses. $X$ and $Z$ are independent variables of any scale.

Under the proportional odds assumption (often known as the parallel lines assumption), I assume $\beta_1$ and $\beta_2$ are constant across all values $k$ of $Y$. Thus, I have one coefficient for each of $X$ and $Y$.

However, I want to relax the proportional odds assumption for the relationship between $X$ (similar to multinomial logistic regression) and $Y$ while maintaining the assumption for the relationship between $Z$ and $Y$. This is known as assuming partial proportional odds. I will now have $k-1$ coefficients for $X$ and one coefficient for $Y$.

How do I interpret each of the coefficients on $X$ in this new model?

For convenience, suppose $k=3$ and thus the coefficients on $X$ are $\beta_2X$, $\beta_3X$ (with $Y = 1$ being the baseline).


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.