===============================
===== The 'games' package =====
===============================

Maintained by Brenton Kenkel (brenton.kenkel@gmail.com).

See also http://github.com/brentonk/games


games 1.1-0 (2013-01-30)
------------------------

* "ultimatum" has new argument "minOffer"

* Fixed bug where "predict.ultimatum" computed E[offer] incorrectly; now using
  simulation due to absence of closed-form expression


games 1.0-5 (2012-03-04)
------------------------
(thanks to Journal of Statistical Software reviewers for suggestions)

* All model classes gain new element "localID", a logical indicator for whether
  the fitted model is locally identified (i.e., negative definite Hessian).
  A warning is issued during fitting, and in "print" and "summary" methods, when
  a model is not locally identified.

* "profile.game" gains argument "use.se", allowing "dist" to be specified
  absolutely instead of as the number of standard errors (useful for models
  where local identification fails and estimated standard errors are
  ill-defined)


games 1.0-4 (2011-12-18)
------------------------

* Fixed bug in the computation of the Vuong test statistic in "vuong"


games 1.0-3 (2011-11-09)
------------------------
(thanks to Julian Wucherpfennig for the suggestion)

* "predProbs" has new argument "type" to select whether to generate
  probabilities for outcomes (the default) or actions

* The name of the corresponding argument in the "predict" methods has been
  changed from "probs" to "type" (for consistency with "predict.glm" and others)


games 1.0-2 (2011-08-03)
------------------------

* Fixed typo in "egame12" documentation


games 1.0-1 (2011-03-23)
------------------------

* Added arguments "caption" and "label" to "latexTable"


games 1.0-0 (2011-03-04)
------------------------

* Major additions:

  - New model "egame123", a three-player extensive form game (and associated
    simulated dataset "data_123")

  - New functions "clarke" and "vuong" for non-nested model testing

  - New interactive function "makeFormulas" for constructing model formulas

  - Added dataset "leblang2003", for replication of David Leblang's 2003 article
    "To Devalue or to Defend? The Political Economy of Exchange Rate Policy"

  - All fitting functions now include argument "method" for selection of
    optimization algorithm from any of the options available in the "maxLik"
    package

  - New dependencies: package "stringr"

* Minor additions and changes:

  - Added option "report" to "profile.game" for printing status bar while
    computing profile log-likelihoods

  - Fitted model objects' "convergence" component now contains information on
    the optimization routine used and the number of iterations to convergence

  - Fitted ultimatum objects include both offer and acceptance (if applicable)
    in "y" component as matrix, rather than separately

  - Log-likelihood component of fitted ultimatum objects where outcome == "both"
    includes attributes with individual log-likelihood contributions from offers
    and from acceptance

  - Fitted models include component "xlevels" to store which levels of each
    factor variable were used in fitting (similar to the attribute of the same
    name in "lm" and "glm" objects)

  - For consistency with previous software and published results, utilities
    fixed to 0 (e.g., u23 in egame12) are now assumed to have a stochastic
    component in the private-error case

  - For consistency, the variable name for a utility equation containing only a
    constant is now displayed in the form "u1(1):(Intercept)", rather than being
    truncated to "u1(1)"

* Bug fixes:

  - "predict.game" works even if "newdata" does not contain variables
    corresponding to the left-hand side of the original fit

  - "predProbs" uses the modal value of binary variables by default when
    constructing a regressor profile, rather than the first value that appears
    in the dataset

  - "profile.game" no longer refits starting at originally estimated
    coefficients (this would yield infinitesimal improvements in the
    log-likelihood from nearly identical coefficients, causing the
    non-convergence warning to be issued inappropriately)

  - "plot.profile.game" uses smarter defaults for y-axis limits

  - In fitting functions, "formulas" specified as a list will no longer fail if
    the list has names

  - "latexTable" output now places the intercept in the first row


games 0.6-0 (2010-12-09)
------------------------

* Introduced functions "profile.game" and "plot.profile.game" for diagnosing
  failure to converge to a global maximum

* All fitting functions (egame12, egame122, ultimatum) now include an argument
  "profile", for using profile output from a previous fit attempt to move closer
  to the global maximum

* New dataset "student_offers" for illustrating convergence failure in the
  ultimatum model

* Added indicator for whether gradient was used in fitting to the convergence
  element of game objects

* Fitted ultimatum objects now include the outcome of interest (offer or both),
  the acceptance vector (if any), and the offer tolerance

* Minor bug fixes:

  - Fixed error with starting value generation in ultimatum model when
    acceptance is not specified


games 0.5-1 (2010-11-10)
------------------------

* Fixed error with starting value generation in ultimatum model when offer
  variable is of class "integer"

* Fixed numerical issues in ultimatum model


games 0.5-0 (2010-10-13)
------------------------

* Initial release of package