Note: * Significant at 10 percent level; ** significant at 5 percent level; *** significant at 1 percent level.

Note: * Significant at 10 percent level; ** significant at 5 percent level; *** significant at 1 percent level.

sample selectivity regressor, and the other is corrected for both. Joint hypothesis tests based on Wald's method are given in table 11.5. Asterisks on the corrected standard deviations indicate the significance of the coefficients when the corrected standard deviations are used.

Basically the results indicate that the principle determinants of referrals for an opted-in general practitioner are (not surprisingly) the patient load, the referral behavior of peers, and whether the physician has a certificate from the College of Family Physicians. The overall fit of the equation is indicated by the 2SLS R2 = 0.4255.

Surprisingly there is little evidence of sample selectivity. Given the corrected variance estimator clgg = 58913.0, it is seen that the estimated correlation between u, and &lgt is 5391/(58913.0)1/2 = 2.221, which exceeds unity by a substantial margin.11 Hence the point estimate for clg Gi + 1 is both implausible and insignificant.

Tables 11.6and 11.7containestimatesanalogousto tables 11.4and 11.5, but for opted-out physicians. The results indicate that the principle determinants of referrals for an opted-out physician are family characteristics, the referral behavior of his peers, and his political orientation. The strong negative effect of the number of dependents is consistent with the hypothesis that an opted-out physician with family dependents will be hesitant to make referrals and run the risk of losing patients. The nature of the employed spouse effect is somewhat unclear and does not correspond to the a priori expected sign. The negative political orientation effects indicate that opted-out physicians who consider themselves to be more liberal or more conservative than the Ontario Medical Association are also less likely to make referrals. For opted-out physicians R2 — 0.6922.

Once again there is little evidence of sample selectivity. Given the corrected variance estimator c2gg = 23274.5, the estimated correlation between ut and e2gt is-35.07/(23274.5)1/2 = -0.02299. Taking into account the estimated nature of the sample selectivity regressor has a negligible effect on the estimated standard deviations.

Formal testing of identical behavior across the regimes is of little consequence here since the regimes were specified to be different on a priori

11. The fact that the estimated correlation exceeds unity explains why the corrected standard deviations in table 11.4 are sometimes smaller than the uncorrected standard deviations.

Table 11.5

Test of joint effects in referral equation (opted-in general practitioners)

Test description

Degrees of freedom

Was this article helpful?

## Post a comment