Since the probit model does not provide consistent estimates
in the presence of heteroscedasticity (Greene 2000), we first
check for heteroscedasticity by conducting the Lagrange
multiplier (LM) test (Davidson and MacKinnon 1993, Donald
1995). Our test result suggests that we are not able to reject
the null hypothesis of homogeneous variance. Accordingly,
we present the estimates from the probit model for the
selection equation (see Table 5). All of the coefficients in the
probit model are statistically significant at the 5 percent or 1
percent levels with the expected signs. The predicted power
of our sourcing choice model is 80.2 percent, which implies
that a substantial portion of the firms in our sample are
correctly classified into the appropriate sourcing regime. We
conclude that our probit model adequately controls for
selection bias.