.-
help for ^mlogtest^                                               (STB-58: sg155)
.-

Tests for multinomial logit model
---------------------------------

    ^mlogtest^ [^,^ ^a^ll ^l^r ^w^ald ^c^ombine ^lrc^omb ^h^ausman ^sm^hsiao ^d^etail ^i^ia ^b^ase
        ^s^et^(^varlist [\ varlist]^)^]

^mlogtest^ computes a variety of tests for multinomial logit models.  The user
selects the tests they want by specifying the appropriate options. For each
independent variable, ^mlogtest^ can perform either a likelihood-ratio or 
Wald test of the null hypothesis that the coefficients of the variable equal 
zero across all equations.  ^mlogtest^ can also perform Wald or likelihood-
ratio tests of whether any pair of outcome categories can be combined. In
addition, ^mlogtest^ allows one to compute both Hausman and Small-Hsiao tests
of the assumption of the independence of irrelevance alternatives (IIA) for 
each possible omitted category.

Options
-------

^all^ specifies that all available tests should be performed. 

^lr^ conducts likelihood-ratio tests for each independent variable.

^wald^ performs Wald tests for each independent variable.

^combine^ computes Wald tests of whether two outcomes in the ^mlogit^ can be 
combined.

^lrcomb^ conducts likelihood-ratio tests of whether two outcomes can be 
combined.

^hausman^ computes Hausman tests of the IIA assumption.

^smhsiao^ performs Small-Hsiao tests of the IIA assumption.
  
^iia^ specifies that both tests of the IIA assumption should be performed. 

^detail^ reports -hausman- output from IIA test (default is to provide only a 
summary of results).

^base^ also conducts an IIA test omitting the base category of the original 
^mlogit^ estimation.  This is done by re-estimating the model using the largest 
remaining category as the base category, although the original estimates are 
restored to memory afterward.

^set(^varlist [\ varlist] ...^)^ specifies that a set of variables is to be
considered together for the ^lrtest^ or ^waldtest^.  The slash \ is used to
specify multiple sets of variables.  This option is particularly useful when
one has included a categorical independent variable in the model as a set of
dummy variables, as one can test that the coefficients for all of the dummy
variables are zero across all equations.

Examples
--------

 . ^mlogit whoclass income dad_educ male black hispanic asian^

 . ^mlogtest, all^

 . ^mlogit whoclass income dad_educ male black hispanic asian singlpar stepmom^
   ^stepdad^ 

 . ^mlogtest, lr set(black hispanic asian \ singlpar stepmom stepdad)^

Return matrices
---------------

r(combine): results of Wald tests to combine categories.  Rows represent all 
contrasts among categories; columns indicates the categories contrasted, the 
chisq, df, and p of test.

r(lrcomb): results of LR tests to combine categories.  Rows represent all 
contrasts among categories; columns indicates the categories contrasted, the 
chisq, df, and p of test.

r(hausman): results of Hausman tests of IIA assumption.  Each row is one test.
Columns indicate the omitted category of a given test, the chisq, df, and p.

r(smhsiao): results of Small-Hsiao tests of IIA assumption.  

r(wald): results of Wald test that all coefficients of an independent variable
equals zero.

r(lrtest): results of likelihood-ratio test that all coefficients of an 
independent variable equals zero.

Acknowledgment
--------------
The code used for the routine to calculate the Small-Hsiao test is based 
on a program by Nick Winter.


Authors
-------

         J. Scott Long - jslong@@indiana.edu
         Jeremy Freese - jfreese@@ssc.uwisc.edu
         www.indiana.edu/~jsl650/post/


Also see
--------

    STB: STB-58: sg155