2011-02-23  Sren Hjsgaard <sorenh@agrsci.dk>

	* In geeglm is checked if the model matrix is rank deficient; if
	so then geeglm exits. Thanks to Jason D Thorpe for pointing this out.

	* Bug in anova fixed. Thanks to Stefan Boehringer for pointing
	this out.
	
	* Version 1.0-18 uploaded.
	

2010-01-26  Sren Hjsgaard <sorenh@agrsci.dk>

	* fixed2Zcor function has been added. This function makes it
	easier to work with a fixed correlation matrix (in particular when
	cluster sizes are not equal).

	* A vignette on models with unequal cluster size, fixed
	correlation matrices etc. has been added. 
	
	* Version 1.0-17 uploaded

2008-12-08  Sren Hjsgaard <sorenh@agrsci.dk>

	* Bugfix in 1.0.15 caused problem in binomial case using
	cbind(pos,neg). Thanks to Tobias Verbeke. Fixed now. 

	* Bugfix in 1.0.15 caused problem when using variables derived
	from gam package call to geeglm. Thanks to Eric Rexstad. Fixed now.
	
	* Using scale.fix argument gave problem in geeglm. Thanks to
	Tobias Verbeke. Fixed now.  

	* Fitting large data set with binomial model and 3 measurements
	per unit using unstructured correlation crashes R. Thanks to
	Tobias Verbeke. Problem has *NOT* been solved but a remark has
	been added to doc file. 
	
	* Version 1.0-16 uploaded

2008-12-01  Sren Hjsgaard <sorenh@agrsci.dk>

	* When a factor has unused levels, the underlying C code
	fails. This is now catured in geeglm. Thanks to Janet Young

	* tests directory added to package
	
	* Version 1.0-15 uploaded

2007-07-13  Sren Hjsgaard <sorenh@agrsci.dk>

	* Smaller 'standardization' changes implemented. Thanks to Achim

	* Version 1.0-14 uploaded

2007-03-21  Sren Hjsgaard <sorenh@agrsci.dk>

	* A remark on fixed correlation structures has been added to the
	geeglm.Rd doc file.

	* A Wishlist file has been added
	
	* Version 1.0-13 uploaded

2007-03-21  Sren Hjsgaard <sorenh@agrsci.dk>

	* Bug in geeglm fixed so that it now works with a fixed
	correlation structure. Thanks to Ulrich Halekoh for the fix.


2006-02-13  Jun Yan  <jyan@stat.uiowa.edu>

	* Bug in genZcor and genZodds fixed. 
	Thanks to Chongzhi Di <cdi@jhsph.edu>. 
	Previously, the matrice generated by genZcor/genZodds are 
	wrong when the number of unique waves is greater than or
	equal to 10.

2006-01-24 Sren Hjsgaard <sorenh@agrsci.dk>

	* Bug in anova.geeglm fixed (showed up when there was only one
	term on the right hand side of ~). 

	* geeglm only works on complete data. This has been pointed out in
	the man pages. (At some point of time, a proper na.action should be
	taken). 

	* All datasets are saved as .txt files (previously some were saved
	as .rda files but that caused problems in building the package on
	windows xp).

	
2005-08-13  Jun Yan  <jyan@stat.uiowa.edu>

	* The working correlation structure can now be "fixed".


2005-06-13 Sren Hjsgaard <sorenh@agrsci.dk>

	* Sren Hjsgaard has modified the geeglm function so that it can
	take the 'waves' argument which is used for explicitely specifying
	the ordering of repeated measurements on the same unit. geeglm can
	also take the 'zcor' argument for a user defined working
	correlation structure. 

	* A documentation file for the genZcor
	function has been added. 

	* All data sets provided are now saved as
	text files.

	* ordgee has been modified to check if the response variable is an
	ordered factor.

	* geeglm can now take the argument std.err which specifies the
	type of variance estimate to be calculated.
	
2005-05-11 Sren Hjsgaard <sorenh@agrsci.dk>

	* Sren Hjsgaard has modified the anova function to make it
	calculate the correct degrees of freedom.


2005-04-11  Jun Yan  <jyan@stat.uiowa.edu>

	* Sren Hjsgaard made some minor changes to geeglm - 
	basically, that geese.control did not work. Now it does.
	

2005-02-09  Jun Yan  <jyan@stat.uiowa.edu>

	* Sren Hjsgaard joined the development with geeglm, which 
	"works like" glm and returns an object which is similar to 
	a glm object. Residuals and predicted values can be extracted
	using the generic functions as with a glm object. 

	* An important feature of geeglm, is that an anova method 
	exists for these models.
	
