Changes since version 0-1.0

	* robust inference is provided with meat and estfunc methods
	defined for mlogit models.

	* subset argument is added to mlogit so that the model may be
	estimated on a subset of alternatives.

	* reflevel argument is added to mlogit which defines the base
	alternative.

	* hmftest implements the Hausman McFadden test for the IIA
	hypothesis.

	* mlogit.data function has been rewriten. It now use the reshape function.

	* logitform class is provided to describe a logit model: update,
	model.matrix and model.frame methods are available.

	
Changes since version 0.1-2

	* major change, most of the package has been rewriten

	* it is now possible to estimate heteroscedastic, nested and mixed
	effects logit model

	* the package doesn't depend any more on maxLik but a specific
	optimization function is provided for efficiency reason

Changes since version 0.1-3

	* mlogit didn't work when the dependent variable was an ordered
	factor in a "wide-shaped" data.frame.

	* the reflevel argument didn't work any more in version 0.1-3.

Changes since version 0.1-4

	* if the choice variable is not an ordered factor, use as.factor()
	instead of class() <- "factor"

	* cov.mlogit, cor.mlogit, rpar , med, rg, stdev, mean functions
	are added to extract and analyse random coefficients.
	
	* a panel argument is added to mlogit so that mixed models with
	repeated observation can be estimated using panel methods or not

	* a problem with the weights argument is fixed

	* the estimation of nested logit models with a unique elasticity
	is now possible using un.nest.el = TRUE

	* the estimation of nested logit models can now be done with or
	without normalization depending on the value of the argument
	unscaled

Changes since version 0.1-5

	* a third part of the formula is added : it concerns alternative
	specific variables with alternative specific coefficients

	* improved presentation for the Fishing dataset.

	* a bug (forgotten drop = FALSE) corrected in
	model.matrix.mFormula

	* Electricity and ModeCanada datasets are added
	
	
Changes since version 0.1-6

	* a bug in mFormula (effects vs variable) is fixed
	
Changes since version 0.1-7

	* mFormula modified so that models can be updated

	* likelihood has been rewriten for the heteroscedastic logit
	model, the computation is now much faster

	* nested logit models with overlapping nests are now supported;
	nests = "pcl" enables the estimation of the pair combinatorial
	logit model

	* the norm argument is added to rpar

	* the logLik argument is now of class logLik

	* mlogit.data is modified so that an id argument can be used with
	data in long shape

	* the argument of mlogit.data used to define longitudinal data is
	now called id.var

	* mlogit.lnls is corrected so that the estimation of multinomial
	models can handle unbalanced data (pb with Reduce)
 
	* the three tests are temporary removed
	
Changes since version 0.1-8

	* all the models could normally be estimated on unbalanced data

	* the three tests are added, i.e. a new scoretest function and
	specific methods for waldtest and lrtest from the lmtest package

	* the model.matrix method for mlogit objects is now exported


Changes since version 0.2-0

	* all the rda files are now compressed

Changes since version 0.2-1

	* ranked-order models can be now estimated ; a new argument called
	'ranked' is introduced in mlogit.data which performs the relevant
	transformation of the data.frame. The estimated model is then a
	standard multinomial logit model

	* multinomial probit model is now estimated by setting the new
	probit arguments to TRUE

	* for the mixed logit model, different draws are now used for each
	observation

	* a predict method is now available for mlogit objects

	* a coef method is added which removes the fixed argument

	* constPar can now be a named numeric vector. In this case,
	default starting values are changed according to constPar

	* the vcov method for mlogit objects is greatly enhanced.

	* mlogit objects now have two elements which indicate the fitted
	probabilities : fitted is the estimated probability for the
	outcome and probabilities contains the fitted probabilities for
	all the alternatives

	* mentions to 'alt' in the names of the effects is canceled ;
	moreover, the intercepts are now called altname:('intercept')

	* a 'choice' attribute is added to mlogit.data objects

	* an effects method is provided, which computes the marginal
	effects of a covariate

Changes since version 0.2-2
	
	* some sys.frame() changed to parent.frame()