forked from cran/mlogit
-
Notifications
You must be signed in to change notification settings - Fork 0
/
NEWS
executable file
·206 lines (122 loc) · 6 KB
/
NEWS
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
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()
Changes since version 0.2-3
* the list of primes used to generate halton sequences was too
short, its length has been increased
* halton sequences where used to estimate mixed logit even for
the default value of halton (NULL), this has been fixed
* the contribution of each observation to the gradient is not
returned as the 'gradient' element of mlogit objects
* the distributions are now checked for rpar and an error is
returned in case of unknown distribution
Changes since version 0.2-4
* the id series (one observation per choice situation) was
badly constructed, it is now fixed
* the levels of the choice variable are now equalized to the
those of the alt variable, allowing the case were some
alternatives are never chosen
* mlogit is now able to estimate models with singular matrix
of covariates. At the end of model.matrix.mformula, the
linear dependent columns of X are removed
* group-hetheroscedastic model can be estimated by setting the
relevant covariates in the 4th part of the formula
* correlation can still be a boolean, but can also be a
character vector if one wants that a subset of the random
parameters being correlated
* zbu and zbt distributions are added : these are
one-parameter distributions for which the lower bond is 0.
* there was a bug in the triangular distribution which is now
fixed
* bug in the effects method fixed
* a new iv function is provided
* the linear predictor is now returned