Functions, data sets, examples, demos, and vignettes for the book Christian Kleiber and Achim Zeileis (2008), Applied Econometrics with R, Springer-Verlag, New York. ISBN 978-0-387-77316-2. (See the vignette "AER" for a package overview.)
Changes in Version 1.2-6
o Added update() method for "ivreg" objects that correctly handles the two-part right-hand side of the formula (suggested by Matthieu Stigler).
o Use pdata.frame() instead of plm.data() as preparation for modeling with "plm" and certain "systemfit" functions (requires systemfit 1.1-20).
o 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).
o 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.
Changes in Version 1.2-5
o Support for aliased coefficients in ivreg() (suggested and tested by Liviu Andronic).
o 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).
o Small improvements in DESCRIPTION/Imports and NAMESPACE for R CMD check.
Changes in Version 1.2-4
o Reference output updated for recent versions of R.
Changes in Version 1.2-3
o Package "splines" is loaded explicitly if needed (rather than assuming that it is loaded along with "survival").
o Some URLs in the manual pages had been outdated and are updated (or omitted) now.
Changes in Version 1.2-2
o 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.
o If there is more than one endogenous variable, summary(ivreg_object, diagnostics = TRUE) now reports separate tests of weak instruments for each endogenous variable.
Changes in Version 1.2-1
o 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.
o 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).
Changes in Version 1.2-0
o 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).
o Added new data set EquationCitations provided by Fawcett & Higginson (2012, PNAS).
o 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).
Changes in Version 1.1-9
o Recompressed data sets in package to reduce file storage requirements.
o ivreg() failed when used without instruments. Now fixed.
o The summary() for ivreg() displayed the degrees of freedom of the overall Wald test incorrectly (although the p-value was computed correctly).
o Some technical changes for new R 2.14.0, e.g., adding [email protected] in DESCRIPTION, recompressing data, etc.
Changes in Version 1.1-8
o 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.
o Added nobs() method for "survreg" objects (and thus "tobit" objects). Modified "ivreg" objects so that default nobs() methods works.
o Labeling in coeftest() method for "multinom" objects with binary responses has been fixed.
o Example 21.4 in ?Greene2003 now employs the scaled regressor fincome/10000.
Changes in Version 1.1-7
o Adapted some example in ?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.
Changes in Version 1.1-6
o Adapted code/examples/tests to new car version which has deprecated linear.hypothesis() in favor of linearHypothesis().
Changes in Version 1.1-5
o 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").
o 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").
o 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.
o 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")
o 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 http://statmath.wu.ac.at/~zeileis/grunfeld/.
o Added lrtest() method for "fitdistr" objects with intelligible model name (instead of the usual formula for formula-based models).
Changes in Version 1.1-4
o 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.
o Numerous spelling improvements in the documentation (thanks to the new aspell() functionality in base R).
Changes in Version 1.1-3
o Added PSID7682 data set which contains the full Cornwell & Rupert (1988) panel data for the years 1976-1982. This should be used for estimation of the Hausman-Taylor model in Exercise 6 from Chapter 3 (instead of PSID1982 which does not provide panel data but only the cross-section for 1982). See the errata and the manual page for more details.
o Fixed overall Wald test in summary() for "tobit" models with intercept only.
Changes in Version 1.1-2
o 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).
o Enhancements of the CigarettesSW examples from Stock & Watson.
o Fixed overall Wald test in summary() for "tobit" models without intercept.
o Improved "rgl" code in the SIC33 example.
o The variable "gender" in the Parade2005 data set was wrong for observation 70. It is now "male" (not "female").
Changes in Version 1.1-1
o A new improved version of the "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).
Changes in Version 1.1-0
o 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.
o 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 http://epub.wu-wien.ac.at/.
o Multinomial logit examples using Yves Croissant's "mlogit" package have been added for the TravelMode and BankWages data sets.
o Vignette/errata updated.
Changes in Version 1.0-1
o Small changes for R 2.8.0.
Changes in Version 1.0-0
o official version accompanying the release of the book (contains all code from the book in demos and tests)
o See the new vignette("AER", package = "AER") for an overview of the package and a list of errata.
Changes in Version 0.9-0
o release of the version used for compiling the final version of the book for Springer
Changes in Version 0.2-0
o first CRAN release of the AER package