CHANGES IN BayesFactor VERSION 0.9.11-1
CHANGES
* Fixed memory bug causing importance sampling to fail.
CHANGES IN BayesFactor VERSION 0.9.11
CHANGES
* Added support for prior/posterior odds and probabilities. See the new vignette for details.
* Added approximation for t test in case of large t
* Made some error messages clearer
* Use callbacks at least once in all cases
* Fix bug preventing continuous interactions from showing in regression Gibbs sampler
* Removed unexported function oneWayAOV.Gibbs(), and related C functions, due to redundancy
* gMap from model.matrix is now 0-indexed vector (for compatibility with C functions)
* substantial changes to backend, to Rcpp and RcppEigen for speed
* removed redundant struc argument from nWayAOV (use gMap instead)
CHANGES IN BayesFactor VERSION 0.9.10-2
CHANGES
* Removed "see also" to package BAS, due to its being archived
CHANGES IN BayesFactor VERSION 0.9.10-1
CHANGES
* Fixed issue causing Solaris build to fail
CHANGES IN BayesFactor VERSION 0.9.10
CHANGES
* Fixed bug in model enumeration code in generalTestBF (affected "withmain" analyses with neverExclude argument)
* Various bug fixes
* Analyses are more error tolerant (problem analyses will yield NA in BayesFactor object)
* Fixed some typos in citation information
* Improved numerical stability of proportional error estimates
CHANGES IN BayesFactor VERSION 0.9.9
CHANGES
* Added "simple" argument to ttest.tstat, oneWayAOV.Fstat, and linearReg.R2stat; when TRUE, return only the Bayes factor (not the log BF and error)
* When sampling Bayes factors, recompute() now increases the precision of BayesFactor objects, rather than simply recomputing them. Precision from new samples is added
* Added test for single proportion; see proportionBF()
* Added support for contingency tables; see contingencyBF()
* Added Hraba and Grant (1970) data set; see ?raceDolls
* Added model.matrix method for BayesFactor objects; allows for extracting the design matrix used for an analysis
* recompute() now has multicore and callback support, as intended
* Refactored t test and meta-t test code
* Moved many backend functions to Rcpp from R C API
* t test samplers now sample from interval null hypotheses and point null hypotheses where appropriate
* fixed bug in in meta t test sampler which wouldn't allow sampling small numbers of MCMC samples
CHANGES IN BayesFactor VERSION 0.9.8
CHANGES
* Fixed bugs in model enumeration code
* Fixed bug leading to wrong computation of number of covariate when interactions between continuous variables were included
* Corrected typos/old information in the documentation
* Fixed a memory allocation bug that affected computing Bayes factors with lots of data
* Added meta-analytic Bayes factor for t tests (see meta.ttestBF)
* Fixed bug in ttestBF that yielded Bayes factor of NaN for very extreme posterior interval probabilities
* Fixed several bugs causing infinite integrals; generally improved integration
* Added check to ensure no missing data before analyses
* Added callbacks for access by third-party interfaces
CHANGES IN BayesFactor VERSION 0.9.7
CHANGES
* Fixed a bug causing posterior sampling of t test to fail when data was defined as integer type
* Fixed a bug causing a two-way interaction to sometimes not be included in 3-way ANOVA analyses
* Reworked model enumeration code to be more efficient and faster; 4-way analyses no longer take a long time to build
* Moved 'methods' to Depends so that Rscript will work without the user having to explicitly load the package methods
CHANGES IN BayesFactor VERSION 0.9.6
CHANGES
* Fixed a bug causing Bayes factors to evaluate to NA when matrix inversion fails (very rare)
* Fixed a bug causing computation to fail if importance sampling fails in nWayAOV. Uses method="simple" as a fallback.
* Fixed a bug where quadrature failed for large F values in unbalanced one-way ANOVA designs
* Data frame for linear model analyses now get modified: character columns are converted to factors, and factors are re-factor()ed to get rid of levels with no observations
* Increased prior scales in one-sample t test by a factor of sqrt(2), in order to be consistent with the two-sample t test for the same effect size and effective sample size
* Added vignette with prior checks to show consistency across methods
* Added vignette with comparisons of BayesFactor results to arm/lme4
CHANGES IN BayesFactor VERSION 0.9.5
CHANGES
* Minor bug fixes
* Changed back to new R 3.0 vignette building (and moved files to vignettes directory)
* Restricted to R >= 3.0.0
CHANGES IN BayesFactor VERSION 0.9.4
CHANGES
* Fixed bug in sampling of posteriors in linear models
* Fixed bug computing the Bayes factor of unbalanced one-way ANOVA with two levels - caused function to fail
* Fixed occasional problem where optimization for the importance sampler would fail, giving a numerically singular matrix
* Fixed problem where extremely rare large or small g values from the sampler would cause the Bayes factor to be NA (again, creating a singular numerically singular matrix). These very rare samples are now disregarded.
* Added global option to turn progress bars on or off; use options(BFprogress = TRUE) or options(BFprogress = FALSE)
* Added MCMC chain thinning to nWayAOV - to use, pass "thin" argument to posterior() or lmBF()
* Added MCMC chain column filtering, useful for low memory systems. To use, pass "columnFilter" argument to posterior() or lmBF(). See the help for posterior() for more details.
* Added function generalTestBF(), which allows testing of restrictions on a full model in a manner similar to regressionBF() and anovaBF().
* Added "noSample" argument to many functions. These will disable time-consuming sampling, and return an object of the same structure as if sampling had been done. This allows for the culling of BayesFactor objects and preplanning chain analyses before sampling
* Added is.na() method for BFBayesFactor objects. When combined with recompute() and noSample (see above), this allows one to create lists of models with missing Bayes factors, remove uninteresting models, then recompute only the missing ones
CHANGES IN BayesFactor VERSION 0.9.3
CHANGES
* Restricted to R 3.0.0 (due to vignette compilation).
CHANGES IN BayesFactor VERSION 0.9.2
CHANGES
* Full support for linear models: continuous and categorical covariates can now be included in the same model using lmBF()
* Minor changes to the BayesFactor output to make it clearer
* Fixed display of very large and very small Bayes factors; no longer will display read something like "Inf (2%)" or the like
* Clearer labels on MCMC output
* When error is missing from the BayesFactor object, plot prints "?" next to the bar to indicate no error estimate is available
* Default prior scale setting changed for continuous covariates; scale now defaults to sqrt(2)/4, which corresponds to the ANOVA "medium" setting (and will give the same Bayes factor in special cases where they should)
* Default prior scale setting in one-sample t changed to 1/2 (it was erroneously changed to sqrt(2)/2). Two-sample t test default setting remains the same, at sqrt(2)/2
* Added new prior scale settings for random effects; default to "nuisance", which is the same as the old default (r=1)
* New prior scale setting: "ultrawide"
* Fixed bug with BFManual() which caused it not to start if dynamic help had not been started yet
* When doing an Bayes factor analysis that requires sampling, the new default setting (method="auto") will automatically try to select the best sampler for you so that you get the most efficient samples.
CHANGES IN BayesFactor VERSION 0.9.1
CHANGES
* Vignette compilation changed for compatibility with R 3.0.0
CHANGES IN BayesFactor VERSION 0.9.0
CHANGES
* New S4 classes representing Bayes factors, models, and MCMC chains. The output of all functions will now be objects of these classes
* Error estimates are now given for all Bayes factor outputs
* To accomodate the new system for creating and manipulating Bayes factors, the main function names have all changed. ANOVA is done via anovaBF(), multiple regression is done via regressionBF() and both can be done through lmBF()
* Posterior sampling is supported by calling the new posterior() method on Bayes factor objects. The result is an BayesFactor MCMC object, which inherits methods for for mcmc objects from the coda class
* New recompute() method will allow the reestimation of Bayes factors (for Bayes factor objects) and restimation of posteriors (for BayesFactor MCMC objects)
* New cleaned-up code base