Interpreting gologit2 model outputs

[Pip Roddis]

Hello,

I have a question about interpreting the outputs of a generalised ordered logit model estimated using the gologit2 programme.

I am investigating public support for different types of energy. My outcome variable has 3 categories: support, neutral, oppose. The model therefore estimates two sets of coefficients: one for support vs. neutral/oppose, and one for support/neutral vs. oppose. Am I correct in interpreting the second set of coefficients as indicators of people more likely to be in the neutral category?

Basically, I am struggling to understand why the outcome categories are cumulative, and whether this means I can or cannot report the results as 'people are more likely to be in the support category' and 'people are more likely to be in the neutral category' (when technically the two sets of coefficients refer to people more likely to be in one category vs. cumulative other categories).

I hope this makes sense, please let me know if any clarification is required.

Many thanks in advance,
Pip

[Richard Williams]

I generally recommend that people check out the following:

Williams, Richard. 2006. "Generalized Ordered Logit/ Partial Proportional Odds Models for Ordinal Dependent Variables." The Stata Journal 6(1):58-82. The published article is available for free at
http://www.stata-journal.com/article...article=st0097.

Williams, Richard. 2016. 2016. "Understanding and interpreting generalized ordered logit models." The Journal of Mathematical Sociology, 40:1, 70. http://www.tandfonline.com/doi/full/...X.2015.1112384. If you can't access it then email me directly.

Williams, Richard. 2018. "Adjusted predictions and marginal effects for multiple outcome commands and models." https://www3.nd.edu/~rwilliam/stats3/Margins05.pdf. If not familiar with margins you may first want to read the Margins01-Margin04 handouts on the same page. You might even read this first.

[Pip Roddis]

Hi Richard,

Thanks for your reply. I've read the recommended articles but I'm still not quite clear on my question.

Basically, given the cumulative nature of the comparisons made between outcomes (and the fact I only have 3 outcome categories), is it correct to interpret the first set of coefficients as people who fall into category 1 (support), and the second set as people who fall into category 2 (neutral)?

The way that it tends to be phrased in the articles is the 'likelihood of getting a higher score', but this isn't very meaningful when there are only two 'higher scores' that can be achieved (support or neutral), and what I'm interested in is the difference between these two categories.

Can I therefore conclude that the first set of coefficients refer to those more likely to fall into the support category, and those in the second set refer to those more likely to fall into the neutral category?

Thanks again!
Pip

[Richard Williams]

The different panels in gologit2 are like a series of binary logits. The first set of results is for categories 1 vs 2 and 3, the second is for 1 and 2 versus 3. Put another way, multiple categories are always being contrasted with at least one category.

I find my third link listed above to be a handy way of keeping everything straight.