freeiv: Instrument-free estimation of a linear structural model with an endogenous regressor

freeiv estimates the coefficient on an endogenous regressor when no external instrument is available, for the linear triangular model in which the endogeneity is generated by a single latent confounder. The command is organised around a result that costs no distributional assumption: under a single restriction on the confounder's two loadings, the coefficient lies between half the ordinary least-squares slope and that slope itself, and that interval is exactly the region in which the implied variances are non-negative. Every point estimator then buys identification with one further assumption, so freeiv reports the routes side by side against the interval, together with the diagnostic that says whether each one's assumption is carried by the data. Nine routes belong to the maintained model — closed forms at second and third order, a quantile-based extrapolation, and two generalized method of moments estimators on the moments of orders two, three, and four, each with a Hansen J statistic — and seven more, among them the estimators of Lewbel (2012), of Lewbel, Schennach, and Zhang (2024), and of Oster (2019), are computed on the same data for comparison. Two endogenous regressors loading on one confounder switch the command to a two-indicator model in which the confounder's direct loading is free and estimated, which is what makes the maintained restriction itself testable. Companion commands report what the data can carry before any estimation, what a given coefficient implies about the unobservables, and a full diagnostic table.