Hurdle model?

[Serena Smith]

Hello.
I am trying to model healthcare seeking behaviour. I am mainly considering three key variables, whether sickness is reported, then treatment seeking, then healthcare spending. These decisions are conditional on the other.

What I am thinking is
1. Probit model for first stage, whether or not sickness is reported
2. Conditional on being sick, probit model for whether treatment is sought
3. Conditional on 1 and 2, what's the health care spending. Here I am thinking of using Tobit model given the censoring of data at 0.

Now I am confused about what such modelling can be called, is this triple hurdle model? I have seen double hurdle model which is used instead of Tobit in case Sequential decisions and also for allowing different factors to influence decisions at each stage. So, my question is what can I refer to this type modeling given that I am using Tobit at one stage, will it be appropriate to refer to it as triple hurdle model?

Thanks in advance!

[Joao Santos Silva]

Dear Serena Smith,

I cannot advise you on the name for the approach, but I suggest that in the final stage you use Poisson regression instead of a Tobit because you actually do not have censoring.

Best wishes,

Joao

[Serena Smith]

Thank you for replying Joao Santos Silva

I do not understand why do you suggest to use Poisson regression because health care spending is not a count data. But it's different types of costs of medical care in aggregate form in dollars. It would be great if you could kindly explain me the reasoning.

[Rich Goldstein]

see https://blog.stata.com/2011/08/22/us...tell-a-friend/

[Joao Santos Silva]

Dear Serena Smith,

Poisson regression can be used to model any data with a positive mean, which is your case, and has been used in similar contexts (see, e.g., John Mullahy's paper).

Best wishes,

Joao
PS: Of course, you need to use a suitable estimator of the standard errors.