Dear all,

I want to analyse spillover health effects among spouses following the decision to provide care (for relatives for instance) of one of the spouse. More particularly, I look at mental health outcomes of both spouses whose one might be a caregiver. Since health outcomes are highly correlated among spouses, I use a 3SLS approach to take into account simultaneity such that:

Depression_ i = Depression_ j + X_ i,j + Y_ i + Z_ i
Depression_ j = Depression_ i + X_ i,j + W_ j + Z_ i

Where X_ i,j are household variables (and so common for both spouses), W_ j and Y_ i individuals regressors and Z_ i the decision of the respondent to care. Another endogeneity issue arises: due to reverse causality between the decision to care (Z_ i) and depression, I have to use an added instrument called V (quite used in this literature but I would like to have more robust arguments).

The command is the following:

reg3 (Depression_ i Depression_ j X_ i,j Y_ i Z_ i) (Depression_ j Depression_ i X_ i,j W_ j Z_ i), exog (V) endog(Z_i)

Here are my two questions:

- is this command correctly specified ?
- how can I test the robustness of the instrument V ? There are two equations, that is to say, two dependent variables, and the instrument V might be valid in one case and not in the other (less likely to be valid in the first case I think..)

Any help is appreciated,
sincerely,

Marie Blaise