==============================================================================
Changes in mvmeta 0.4.3 (08 Apr 2014)
=====================================
Major changes
-------------
o New argument 'bscov' in mvmeta() to structure the between-study
(co)variance matrix Psi. New functions initpar() and par2psi() to set the
initial values and derive Psi depending on the structure and type of
parameterization. New function checkPD to check positive definiteness.
Minor changes to functions mvmeta.ml(), mvmeta.reml() and other likelihood
functions. New help page mvmetaCovStruct with info.
o Added the option to input within-study correlations, when not provided,
thorugh the argument 'Scor' in the control list. Added function
inputcov() to input (co)variance matrices directly. Modifications to
mvmeta.fit(), mvmeta.control(), mkS().
o Starting values are now set through the new function initPar(). The values
can be submitted through the argument 'initPsi' in the control list. If
left NULL (the default), iter.igls() is called.
o Changes to mvmeta.control(): arguments 'initPsi', 'Psifix', 'Psicor' and
'Scor' to set initial values, to specify fixed Psi, correlations in Psi
and correlations in S.
o Bug fixed in mvmeta(): call to define argument 'S' now set with correct
environment. Previously possible conflicts with suggested packages.
o New data sets added: fibrinogen, hsls, hyp, p53, and smoking. Many
examples included in help pages.
Minor changes
-------------
o New code section in mvmeta() to set the fitting method and the structure
of Psi.
o Small changes in the code of the function mvmetaSim().
o Function mvmeta.igls() renamed iter.igls(). Functions mvmeta.ml* and
mvmeta.reml* renames mlprof* and remlprof*.
o Various internal function renamed without the initial dot '.'.
o Modification to print() method of summary.mvmeta objects.
o Special quotes "`" removed from names of the functions (this prevented
the debugging tools of RStudio to work).
o Documentation changed accordingly.
==============================================================================
Changes in mvmeta 0.3.5 (08 Dec 2013)
=====================================
Major changes
-------------
o A new estimation method based on variance components has been added.
o New arguments 'sd' and 'cor' added to mvmetaSim().
Minor changes
-------------
o Modified print method for summary.mvmeta objects.
o summary.mvmeta() and its print method improved.
o Argument set.negeigen added to mvmeta.control().
o Documentation changed accordingly.
==============================================================================
Changes in mvmeta 0.3.4 (23 Jan 2013)
=====================================
Major changes
-------------
o An estimation method based on method of moments has been added, using
method="mm" in mvmeta(). The algorithm is implemented in the function
mvmeta.mm().
o Function metaSim() and method simulate.mvmeta() added. These functions
simulate data for multivariate and univariate meta-analysis from
used-defined data or from a fitted model, respectively.
o New control parameters added in mvmeta.control().
Minor changes
-------------
o Bug fixed in mvmeta() with subset.
o Documentation changed accordingly.
==============================================================================
Changes in mvmeta 0.3.2 (25 Aug 2012)
=====================================
Minor changes
-------------
o Bug fixed in coef.mvmeta() when k=1.
==============================================================================
Changes in mvmeta 0.3.1 (24 Aug 2012)
=====================================
Minor changes
-------------
o Bug fixed in .mkS() when k=1 and S is a list.
==============================================================================
Changes in mvmeta 0.3.0 (01 Aug 2012)
=====================================
Major changes
-------------
o The internal structure of the package has been substantially revised, with
the aim to improve efficiency, stability and reliability. The function
mvmeta() now resembles the standard regression functions such as lm() and
glm(). These changes will also hopefully make easier further extensions
and improvements. Specific changes are described below.
o Estimation procedure divided in steps: mvmeta() calls mvmeta.fit(), a
wrapper for specific fitting functions mvmeta.fixed(), mvmeta.ml(), and
mvmeta.reml(). The last two in turns call optim(), based on new likelihood
functions with suffix .fn and .gr to compute the likelihood and the
partial derivatives. The procedure is 15-25% faster.
o The structure of mvmeta() now is that of a proper regression function,
properly based on model frames and terms objects. This allows icluding
offset, extracting residuals, fitted values and the model matrix, among
other benfits. Several default methods for regression functions are now
available for mvmeta objects.
o Handling of missing values exploits the new structure based on model
frame with additional class "data.frame.mvmeta". Method function na.omit()
and na.exclude() have been added to properly define missing observations.
o The method summary() now returns an object of class "summary.mvmeta", with
a similar structure and printing methods of those of lm() and glm().
o New function mvmeta.control() to provide different options in several
steps of the model fitting. Likely to be extended in future versions.
o Several new internal functions have been added but not exported in the
namespace. All the method functions are now exported and documented.
Minor changes
-------------
o Changed dependencies: not based anymore on functions in package Matrix.
o AIC() and BIC() are now based on the default methods.
o Coefficients are now reported in matrix form.
o Labels are now internally defined and handled.
o The help page mvmetaObjects have been added.
==============================================================================
Changes in mvmeta 0.2.4 (06 Jan 2011)
=====================================
Minor changes
-------------
o Computation of quantities for Q test moved from qtest to mvmeta.
o Q test output now includes also tests for single outcomes.
o Warning added to mvmeta when convergence not reached.
o Bug fixed in mvmeta when formula is not specified.
o beta changed to coef for coherence with other regression functions.
o Simplified output of predict() and blup(): vector for meta-analysis or
single prediction. Also label pred changed to fit, coherently with other
regression functions.
==============================================================================
Changes in mvmeta 0.2.3
========================
Minor changes
-------------
o Depends changed to R (>= 2.13.0).
==============================================================================
Changes in mvmeta 0.2.2
========================
Major changes
-------------
o The argument 'formula' now accept also matrix-type objects for simple
meta-analysis.
o Included 'fnscale=-1' for optim() and changed the sign of the
estimation algorithms: now the function maximizes.
o Also 'S' and 'mlab' now can be stored in 'data'.
o Included labels in the dataframe 'berkey98'.
Minor changes
-------------
o Documentation changed accordingly.
==============================================================================
Changes in mvmeta 0.2.1
========================
Minor changes
-------------
o Fixed bug in printing of qtest.mvmeta().
o Documentation changed accordingly.
==============================================================================
Changes in mvmeta 0.2.0
========================
Major changes
-------------
o The arguments 'y' and 'X' has been replaced by 'formula' in mvmeta().
Now the model is specified through a formula, making easier the
inclusion of factors and other variable transformation.
o The objects 'class', 'contrasts' and 'model' have been added to mvmeta
objects, storing info about the fitted model. In particular, method
functions such as predict now use model.matrix() and other functions
to re-build the model.
o The function kXlistmk() has been excluded from the package. Now the
Kronecker expansion is performed directly in the code. The argument
'cen' has been excluded from 'mvmeta().
Minor changes
-------------
o Changes in dependencies: created a new generic for blup(), in order
to avoid the loading of metafor and its initial message. Function
rankMatrix() imported from Matrix to check full-rank of desing matrix.
o Documentation changed accordingly.
==============================================================================
First version in R CRAN mvmeta 0.1.0
==============================================================================