High-Performance Open-Source Archive
New methods for dffits() generic in R-devel (to be
4.6.0) for ivreg and influence.ivreg objects.
Internally, these just call influece(...)$dffits which is
the recommended usage for earlier versions of R.
The test scripts have been updated accordingly to avoid errors in the CRAN checks using R-devel.
A warning is issued now in ivreg.fit() if there are
any collinear instruments.
If a model is specified where the instruments perfectly predict all regressors, a warning is now issued that no endogenous variables could be detected and that all regressors appear to be exogenous (reported by Alex Hayes in #23).
Additionally, the documentation clarifies that fitting a model with only a single right-hand side (i.e., treating all regressors as exogenous) is possible for convenience, e.g., to facilitate model comparisons. This now also works correctly for M and MM estimation.
The numeric checks of which variables are exogenous/endogenous/instruments have been improved and are a little more rigorous now.
The summary() method now also works correctly if
there are aliased coefficients that are NA (reported by
Alex Hayes in #24).
An error is issued now if
vcovHC(..., component = "stage1") is requested because
currently only "stage2" (the default) is supported
(reported in #26 by Matthew Bhagat-Conway).
Now weights(..., type = "working") also works for
ivreg objects. These are simply the weights already
available previously as weights(..., type = "variance") as
these are now used in computations of lm.influence(), e.g.,
dffits().
ivreg() objects with a dedicated manual page
?summary.ivreg. Along with the summary()
method this also documents the methods for confint(),
anova(), Anova(), and
linearHypothesis(). All of these take an argument
vcov. so that alternative (e.g., so-called “robust”)
covariance matrices can be plugged in. The vcov. processing
is made somewhat more convenient and consistent (suggested by Diogo
Ferrari).hatvalues.ivreg() (reported by Vasilis Syrgkanis).Enhanced predict.ivreg() method, which optionally
provides standard errors, confidence intervals, and prediction intervals
for predicted values.
The tinytable rather than the
kableExtra package (recently not actively maintained) is
used now for the modelsummary table shown in the package
vignette (contributed by Vincent Arel-Bundock).
Further small improvements in the package vignettes.
Improve non-anchored links in manual pages (prompted by CRAN).
Achim Zeileis took over maintenance, both on CRAN and on GitHub. The GitHub source repository is now at https://github.com/zeileis/ivreg/ with the web page at https://zeileis.github.io/ivreg/.
Avoid partial argument matches by calling
model.matrix(..., contrasts.arg = ...) rather than just
contrasts (reported by Kevin Tappe).
Make names of arguments of influencePlot.ivreg() and
outlierTest.ivreg() consistent with the corresponding
generic functions from the car package.
method is now an explicit argument to
ivreg() and not just passed through ... to
ivreg.fit().
More efficient computation of regression diagnostics (thanks to improvements implemented by Nikolas Kuschnig).
In models without any exogenous variables (i.e., not even an
exogenous (Intercept)) the $instruments
element in the fitted model object was erroneously empty, leading to
some incorrect subsequent computations. Also the
$endogenous element was an unnamed (rather than named)
vector. Both problems have been fixed now. (Reported by Luke
Sonnet.)
In the summary() method the default is now
diagnostics = NULL (rather than always TRUE).
It is now only set to TRUE if there are both endogenous and
instrument variables, and FALSE otherwise. (Reported by
Brantly Callaway.)
Small fixes.
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 regressor
ex 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 freedom
df 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 ivreg
objects with the residual degrees of freedom from the stage-1
regression.
Add coef(..., component = "stage1"),
vcov(..., component = "stage1"), and
confint(..., 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(), and
confint() methods gained a complete = TRUE
argument 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 modelsummary
package.
Small edits to the Diagnostics vignette.
Initial version of the ivreg package: An
implementation of instrumental variables regression using two-stage
least-squares (2SLS) estimation, based on the ivreg()
function previously in the AER package. In
addition to standard regression functionality (parameter estimation,
inference, predictions, etc.) the package provides various regression
diagnostics, including hat values, deletion diagnostics such as
studentized residuals and Cook’s distances; graphical diagnostics such
as component-plus-residual plots and added-variable plots; and effect
plots with partial residuals.
An overview of the package, documentation, examples, and
vignettes are provided at
https://john-d-fox.github.io/ivreg/.
Need mirroring services?
Contact our team at info@vpspulse.com.
Mirror powered by VPSpulse
Infrastructure sponsored by VPSPulse & Secure Payments by ArionPay.