- Three-part right-hand side
formulas are supported now to facilitate specification of models with many exogenous regressors. For example, if there is one exogenous regressorex and one endogenous regressor en with instrument in, a formula with three parts on the right-hand side can now also be used: y ~ ex | en | in. This is equivalent to specifying: y ~ en + ex | in + ex.
- Robust-regression estimators are provided as an alternative to ordinary least squares (OLS) both in stage 1 and 2 by means of
rlm() from package MASS. Specifically, in addition to 2-stage least squares (2SLS, method = "OLS", default) ivreg() now supports 2-stage M-estimation (2SM,method = "M") and 2-stage MM-estimation (2SMM,method = "MM").
- Dedicated
confint() method allowing specification of the variance-covariance matrix vcov. and degrees of freedomdf to be used (as in the summary()method).
- Include information about which
"regressors" are endogenous variables and which "instruments" are instruments for the endogenous variables in the fitted model objects from ivreg() and ivreg.fit(). Both provide elements $endogenous and $instruments which are named integer vectors provided that endogenous/instrument variables exist, and integers of length zero if not.
- Include
df.residual1 element in ivregobjects with the residual degrees of freedom from the stage-1 regression.
- Add
coef(..., component = "stage1"),vcov(..., component = "stage1"), andconfint(..., component = "stage1") for the estimated coefficients and corresponding variance-covariance matrix and confidence intervals from the stage-1 regression (only for the endogenous regressors). (Prompted by a request from Grant McDermott.)
- Add
residuals(..., type = "stage1") with the residuals from the stage-1 regression (only for the endogenous regressors).
- The
coef(), vcov(), andconfint() methods gained a complete = TRUEargument assuring that the elements pertaining to aliased coefficients are included. By setting complete = FALSE these elements are dropped.
- Include demonstration how to use
ivreg() results in model summary tables and plots using the modelsummarypackage.
- Small edits to the Diagnostics vignette.