Since the endogenous regressors represent a nonlinear function of the same variable, the potential of instrument weakness is a concern. To my knowledge, no procedure is available to test for instrument weakness in the nonlinear generalized method of moments (GMM) setting, whereas tests and implications in the linear model are better understood ( Stock et al., 2002 ). For this reason, I implement a two-stage least squares (2SLS) estimation of Equation (5) , although results are similar when using nonlinear GMM in- stead. Because Low relation and High relation are indicator variables, I follow the recommendation in Wooldridge (2002) to first estimate a probability model for the relationship dummies with the instruments described above and then use the predicted probabilities as the actual instruments in the first stage of the 2SLS estimation. 12 To ac- count for the dependence between the indicator variables, the estimation uses an ordered probit model.