using a hurdle model?

[Ontiq Dey]

Hello everyone,

I am a senior undergrad in economics from Dhaka, Bangladesh.

We are trying to analyze 'access to health' using a national household survey. Here we have 2 dichotomous variables: 1. was an individual sick with acute disease in the last 30 days? and 2. did that individual seek medical assistance? Our hunch is that the situation is a 'hurdle/double hurdle" one. these two decisions are likely to have different, yet not mutually exclusive sets of determinants. How do we handle this situation using stata?

Regards
Ontiq

[Ariel Linden]

This does not make sense from a hurdle perspective. A hurdle model generally has the first part which is binary, and the second part which is a continuous or count variable. As an example, if we model health care utilization, we would assume that people that have no utilization (zero) are fundamentally different than those who use any services (e.g. > zero). We'd thus model the zero using a model for binary variables (logit/probit), and we'd estimate the >zero utilization with the appropriate count or continuous variable model.

Here, you have two binary variables, so there is no "hurdle" that gets crossed to greater utilization.

[daniel klein]

I doubt that whether an individual was sick within the last 30 days (your first variable) has a lot to do with what we normally call a decision. So depending on the exact nature of your research question and hypotheses, this variable might just be a (strong) predictor in modeling the decision to seek medical assistance. You might even only be interested in those people suffering from illness, and thus only use this specific sub sample to fit your model.

Anyway, I guess your question does not so much concern software issues in the first place, but is more about econometric modelling. This is fine on Statlist, but to get better advice you probably need to tell us more about the substantive question(s) your are trying to answer.

Best
Daniel