The Impact of Health Information Sharing on Duplicate Testing

Though our quasi-experiment approach (described in the “Results” section), using a treatment-to-control matching algorithm, is a well-accepted methodology to purge out confounding effects, one may still argue that the Treatmenth variables might still be subject to potential endogeneity. For example, providers with higher duplication rates may be more likely to implement health information sharing technologies. To address the concern of endogeneity, we adopted a control function estimation approach as discussed in Wooldridge (2011) and Rivers and Vuong (1988). The control function approach disintegrates the correlation between endogenous explanatory variables and unobservables affecting the outcome using additional regressors that do not appear in the structural equation (Wooldridge 2010). Suppose y1 refers to the outcome variable, y2 refers to endogenous explanatory variable, and z is the 1 × L vector of exogenous variables, where z1 is a 1 × L1 strict subvector of z. Then model becomes