The instruments are highly significant in the first stage regression (J statistics between 13.2 and 22.2 corresponding to p-values of zero for the instruments’ joint significance in the first stage regressions); over-identification tests fail to reject the validity of our instruments and Wald test of exogeneity fail to reject the exogeneity of the popularity of smokers and non-smokers’ suggesting that a single equation probit model may be more appropriate (we can reject exogeneity at the 10% level in one regression, having ever smoked regularly by 2009; yet, results still hold).
1
—
—