Two-part model with two binary outcomes (Hurdle model?)

[Bryony Simmons]

Hello

I am planning to run a two-part model exploring healthcare utilisation but am a little lost on the methodology to use.

I have a dataset of individuals with a specific health condition and want to model two different decision processes:

1. The decision of the individual to seek formal healthcare (ie, go to a health facility)
2. For individuals seeking care, the decision of the healthcare provide to use a specific diagnostic test

In Applied Health Economics (2007; p291), Jones et al. describes this as a principal-agent type model, where the physician (agent) determines utilisation on behalf of the patient (principal) once initial contact is made. They suggest a Hurdle model might be suitable - but this seems to only be appropriate when the second equation is count.

In my situation I have 2 binary variables (ie, seek care (yes/no), receive diagnostic (yes/no)) & am not sure a Hurdle model can be run? I also want to be able to control for different sets of variables which may differentially impact the 2 decisions.

One option we have considered using is a bivariate probit model.

Does anyone have any advice on which model might be appropriate? Or could direct me to some literature/examples?

Thank you

[Mustafa Coban]

Hi Bryony,

maybe I haven't understood your estimation problem well enough, so let me outline two alternatives.

If both dependent variables are always observed and you assume that y2 has an impact on y1, you can estimate an rbiprobit model. More details are here

If you observe y1 only when y2 = 1 (sample selection model) and you don't assume that y2 has an impact on y1 than I would recommend to take a look at heckprobit .

Best regards