bread() method for tobit() objects in
case the latter already used a “robust” sandwich variance. In that case
naive.var rather than var has to be used for the bread (reported by
Daniel Klinenberg).stop() in non-check settings.Improvements in coeftest() methods for
multinom and polr objects to be compatible
with recent extensions in lmtest >= 0.9-38.
Fixed a bug in the tests for weak instruments in
ivdiag(): Interactions like x1:x2 and
x2:x1 were considered to be different variables although
they are in fact equivalent. (Reported by Arne Henningsen and Ghislain
Dossou.)
Use fully-qualified calls to survival::survreg() and
survival::Surv() within the tobit() function
so that AER::tobit() can be called without attaching the
package (reported by Bill Denney).
Started replacing examples with repeated with()
variable transformations with a single
transform().
Some examples from the AER book lead to small numeric differences
across different platforms. Hence, these have been excluded now from
tests/Ch-*.R with an IGNORE_RDIFF comment for
CRAN.
The examples on the manual pages for the books
CameronTrivedi1998, Greene2003,
WinkelmannBoes2009 are now excluded from testing to reduce
the computational demands on CRAN. The corresponding analyses are mostly
available on the manual pages for the underlying data sets,
though.
survival >= 3.1-6 the survival
provides (or corrected) some standard S3 methods for
survreg objects: fitted(),
nobs(), weights(), vcov().
Previously, these were registered by AER” in order to work for “tobit”
objects. For now, AER registers the methods for tobit
objects. In the future, the methods will be dropped by AER altogether
and inherited from survival (when AER depends on survival >=
3.1-6).Diagnostic tests ivdiag() for weighted
instrumental variables regression was incorrect as weights were
erroneously ignored (detected and tested by Jonathan
Siverskog).
vcov() method for survreg objects in
survival currently (version 2.44-1.1) provides no
row/column names. Hence, a new vcov() method for
tobit objects was added in AER mimicking the
naming conventions from survival:::summary.survreg.
Analogously, a bread() method was added for
tobit objects that calls the vcov() method
internally.
linearHypothesis() method for tobit()
objects no longer assumes that a scale parameter was estimated as part
of the model. This enables the somewhat unusual but possible case of
summary(tobit(..., dist = "exponential")).
In examples/demos/tests with calls to set.seed(...)
the RNG version is now fixed to
suppressWarnings(RNGversion("3.5.0")) to keep results
exactly reproducible after fixing RNG problems in R 3.6.0.
New errata item in vignette("AER", package = "AER"):
The formula interface for pgmm() has changed. As of
plm version 1.7-0, the function dynformula()
is deprecated. Examples, tests, and demos have been adapted
accordingly.
In tests/Ch-Intro.R some quantreg
computations are now wrapped into try() because the
summary() method yields non-finite standard errors on some
platforms.
In ?CameronTrivedi1998 one version of the
negbin-negbin hurdle model is now wrapped into try()
because evaluating the log-likelihood at the given start values becomes
too unstable on some platforms.
Added update() method for ivreg objects
that correctly handles the two-part right-hand side of the formula
(suggested by Matthieu Stigler).
Use pdata.frame() instead of plm.data()
as preparation for modeling with plm and certain
systemfit functions (requires systemfit
1.1-20).
The model.matrix() method for ivreg
objects erroneously dropped the dimension of single-column projected
regressor matrices. This propagated to the estfun() method
and hence lead to problems with sandwich covariances (reported by Justus
Winkelmann).
In a bug fix the survival package changed the
internal structure of survreg()$y starting from
survival 2.42-7. This leads to incorrect
summary() output for tobit objects which has
been worked around now. Thanks to Terry Therneau for pointing out the
problem and suggesting a fix.
Support for aliased coefficients in ivreg()
(suggested and tested by Liviu Andronic).
New data sets GoldSilver, MotorCycles2
and MSCISwitzerland, taken from Franses, van Dijk, Opschoor
(2014): “Time Series Models for Business and Economic Forecasting”, 2nd
ed. For replication of the corresponding examples, several packages were
added to the list of ‘suggested’ packages (including
fGarch, forecast, longmemo,
rugarch, vars).
Small improvements in
DESCRIPTION/Imports and NAMESPACE
for R CMD check.
Package splines is loaded explicitly if needed
(rather than assuming that it is loaded along with
survival).
Some URLs in the manual pages had been outdated and are updated (or omitted) now.
Another bug fix in the new
summary(ivreg_object, diagnostics = TRUE). If sandwich
standard errors (or other vcov) were used, the chi-squared
form rather than the F form of the diagnostic Wald tests was computed
and hence the p-values were incorrect.
If there is more than one endogenous variable,
summary(ivreg_object, diagnostics = TRUE) now reports
separate tests of weak instruments for each endogenous
variable.
Bug fix in the new
summary(ivreg_object, diagnostics = TRUE). If there is more
than one endogenous variable, the degrees of freedom (and hence the
associated p-values) for the Sargan test were too large.
The examples employing rgl for 3d visualizations
(e.g., for the SIC33 data) are not tested anymore in
R CMD check (as rgl currently has some
problems on CRAN’s checks for OS X).
The summary() method for ivreg now has
a diagnostics = FALSE argument. If set to
TRUE, three diagnostic tests are performed: an F test of
the first stage regression for weak instruments, a Wu-Hausman test for
endogeneity, and a Sargan test of overidentifying restrictions (only if
there are more instruments than regressors).
Added new data set EquationCitations provided by
Fawcett & Higginson (2012, PNAS).
Changes in Depends/Imports/Suggests due to new CRAN check
requirements. In particular, the Formula package is now
only imported (but not loaded into the search path).
Recompressed data sets in package to reduce file storage requirements.
ivreg() failed when used without instruments. Now
fixed.
The summary() for ivreg displayed the
degrees of freedom of the overall Wald test incorrectly (although the
p-value was computed correctly).
Some technical changes for new R 2.14.0, e.g., adding
Authors@R in DESCRIPTION, recompressing data,
etc.
The hat values for instrumental variables regressions are now
computed in the hatvalues() method and not within
ivreg.fit() to save computation time for large data
sets.
Added nobs() method for survreg objects
(and thus tobit objects). Modified ivreg
objects so that default nobs() methods works.
Labeling in coeftest() method for
multinom objects with binary responses has been
fixed.
Example 21.4 in ?Greene2003 now employs the scaled
regressor fincome/10000.
?Greene2003 in order to work
both with the old and new dynlm package.
dynlm() now provides convenient support for linear time
trends via dynlm(y ~ trend(y)) etc.car version which
has deprecated linear.hypothesis() in favor of
linearHypothesis().CPS1985 now has 534 observations (not 533 as in
prior releases), the original observation 1 had been omitted
inadvertently. See also the errata in
vignette("AER", package = "AER").
Data and examples for Winkelmann and Boes (2009), “Analysis of
Microdata” (2nd ed.) have been added. For details and extensive (but not
quite complete) replication code see
help("WinkelmannBoes2009") as well as
help("GSS7402") and
help("GSOEP9402").
As announced in the changes for version 1.1-0 of the
AER package, the variable earnings has now
been removed from the PSID1976 (aka Mroz) data. In 1.1-0 it
was renamed to wage to avoid confusion with other data
sets.
The coeftest() method for polr objects
used to return wrong standard errors (and hence wrong z tests) for the
last intercept. This was caused by an inconsistency between the
summary() and vcov() methods for
polr objects which has been improved in recent versions of
the MASS package. The correct results are computed by
coeftest() for polr objects computed with
MASS version >= 7.3-6. See also the errata in
vignette("AER", package = "AER").
The paper describing the various versions of the
Grunfeld data has been accepted for publication in the
German Economic Review. An updated version of the manuscript and
associated replication files – mostly based on
data("Grunfeld", package = "AER") – is available from https://www.zeileis.org/grunfeld/.
Added lrtest() method for fitdistr
objects with intelligible model name (instead of the usual
formula for formula-based models).
ivreg() now uses the Formula package
(>= 0.2-0) for processing of its model formulas. However, this only
affects the internal computations, the user interface has remained
unchanged.
Numerous spelling improvements in the documentation (thanks to
the new aspell() functionality in base R).
PSID7682 data set which contains the full
Cornwell & Rupert
PSID1982 which does not provide panel data but
only the cross-section for 1982). See the errata and the manual page for
more details.summary() for
tobit models with intercept only.New errata item in vignette("AER", package = "AER"):
The comment regarding the output from the Johansen test (p. 169) is in
error. The null hypothesis of no cointegration is not rejected at the
10% level (only at 15% level).
Enhancements of the CigarettesSW examples from Stock
& Watson.
Fixed overall Wald test in summary() for
tobit models without intercept.
Improved rgl code in the SIC33
example.
The variable gender in the Parade2005
data set was wrong for observation 70. It is now "male"
(not "female").
plm package is available
from CRAN (version 1.1-1). This fixes a bug in the
summary() of plm objects, see the
vignette/errata for details. Furthermore, there is now a
vcovHC() method for panelmodel objects: It
gives equivalent results to pvcovHC() but is now the
recommended user interface and hence used in the examples of some manual
pages (see e.g. ?Fatalities).Some variable names in the PSID1976 (aka Mroz) data
have been renamed: earnings is now called wage
(to avoid confusion with other data sets), the previous variable
wage is renamed as repwage (reported wage).
Currently, earnings is kept; it will be removed in future
releases.
Documentation of the Grunfeld data has been enhanced
and updated. Much more details are available in a recent technical
report: Kleiber and Zeileis (2008), “The Grunfeld Data at 50”, available
from https://doi.org/10.57938/c1925b32-0b27-490f-b593-42da650c196c.
Multinomial logit examples using Yves Croissant’s
mlogit package have been added for the
TravelMode and BankWages data sets.
Vignette/errata updated.
Official version accompanying the release of the book (contains all code from the book in demos and tests)
See the new vignette("AER", package = "AER") for an
overview of the package and a list of errata.
AER package.