IVBOUNDS: Stata module providing instrumental variable method to bound treatment-effects estimates with potentially misreported and endogenous program participation

ivbounds provides an estimate of the bounds of the average causal effect for compliers (Imbens and Angrist, 1994) when both noncompliance and misreporting of treatment status are present. The approach follows the estimation procedure developed in Tommasi and Zhang (2020). When iv(varname_iv) is binary, the command estimates the bounds of the local average treatment effect (LATE). When iv(varname_iv) is binary and covariates are included, the command estimates the bounds of the unconditional LATE (Frölich, 2007). In both cases, the estimated bounds coincide with those in Ura (2018). When iv(varname_iv) is discrete, the command estimates the bounds of the weighted average of LATEs (WLATE). When iv(varname_iv) is discrete and covariates are included, the command estimates the bounds of the conditional WLATE. In the latter case, the user must specify the number of strata to discretise the propensity score. The inference procedure follows Dehejia and Wahba (1999) and Battistin and Sianesi (2011). ivbounds assumes a binary treat(varname_t), a binary or discrete iv(varname_iv), and a continuous or discrete depvar. The bounds can be improved if external information regarding the treatment misclassification probability in treat(varname_t) is available.