interpretation of zero/one-inflated beta regression

[joonsdad]

Hi - apologies for the bad username - Mike Mulcahy here. [Tried to update my account to comply with guidelines, but I don't have access to Stata on this computer, didn't work...]

• I'm using Prof. Buis' user-written zoib program (2010) to analyze success rates of union representation elections, aggregated to the city level, in a panel dataset of about 575 cities over about 9 years.
  • I emailed Prof. Buis with my question a few weeks ago - he's probably just too busy to too annoyed with these dumb questions to answer...
• The dependent variable is the proportion of elections in the city-year that were successful for the union;
• the distribution includes about 25% 0's, 25% 1's and the rest proportions such that 0 < y < 1.
• The focus is on the direct and interaction effects of a municipal living wage campaigns,
• The municipal policy variable is categorical, with 3 values: 0 = no campaign; 1 = successful municipal policy campaign; 2 = failed campaign.
• The policy indicator is interacted with measures of the local presence and prior activity of unions and social movement organizations in the city.

My questions:

• First: is it possible to summarize the net marginal effect effect of our policy variable, for example when it is significant in two of the zoib submodels? For example, if the policy variable is statistically significant and positive in both the 0<y<1 model and the Pr(y=0)?

• Second: I have tried to create interaction graphs using margins compare and marginsplot, comparing outcomes for successful and failed campaigns to the base category of no campaign, but the results do not seem consistent with the model results - for example, the margins of the successful and failed campaign lines overlap the 0 line (the reference category of no campaign); I interpret that to mean not significant, but in the results, it is significant. Could this be a function of choosing the wrong IV for the X-axis? Any help would be appreciated.

Thank you,
Mike Mulcahy

Buis, Maarten. 2010. "ZOIB: Stata module to fit a zero-one inflated beta distribution by maximum likelihood" https://ideas.repec.org/c/boc/bocode/s457156.html

[Maarten Buis]

I am not annoyed, but busy.