.-
help for ^mlogplot^ - 03Nov2005
.-
Odds ratio and discrete change plots for multinomial logit
----------------------------------------------------------
For odds ratio plots, the results of ^mlogit^ must be in memory; for discrete
change plots or odds ratio plots using discrete change the results of ^prchange^
must be in memory. The syntax is:
^mlogplot^ [varlist] [^, or^atio ^dc^hange ^s^td^(^[^s^|^u^|^0^]...[^s^|^u^|^0^]^)^
^min(^#^) max(^#^) packed labels p^rob^(^#^) b^asecategory^(^#^)^
nt^ics^(^#^) n^ote^(^string^)^ ^val^ues ^dcadd(^#^) ^saving(^filename[^,replace^]^)^]
To create odds ratio plots with coefficients that you have placed into
matrices (e.g.,to plot coefficients from a published paper):
^mlogplot,^ ^matrix v^ars^(^varlist^)^ ^s^td^(^[^s^|^u^|^0^]...[^s^|^u^|^0^]^)^
[^min(^#^) max(^#^) packed labels b^asecategory^(^#^) nt^ics^(^#^) n^ote^(^string^)^
^saving(^filename[^,replace^]^)^]
The matrices you must create are discussed below.
Description
-----------
^mlogplot^ takes coefficients either from ^mlogit^ or from other programs that
estimate the multinomial logit model. When estimates from other programs are
used they must be passed to ^mlogplot^ as matrices. ^mlogplot^ then plots the
odds ratios (i.e., exp(b)) and/or the discrete change coefficients. You can
list the same variable more than once if you want both standardized and
unstandardized coefficients to be plotted in the same graph. A variety of
options control the way the final graph looks. The dialog box program ^mlogview^
provides a convenient front end that allows you to use this program
interactively.
^IMPORTANT^: When using coefficients from ^mlogit^, you must be sure that the
value labels for the dependent variable begin with different letters or
numbers. Otherwise, the plot might be misleading.
Options for Plotting Estimates from ^mlogit^
---------------------------------------------
^oratio^ indicates that you want to plot the odds ratios (i.e., exp(b)).
^dchange^ requests plots involving discrete change. To use this option, you
must have first run ^prchange^. This option does not work with the ^matrix^
option.
If ^oratio^ has been specified, this options plots the size of the letters
proportional to the size of the corresponding discrete change coefficient.
See ^dcadd^ and ^dcbase^ for further details.
If ^oratio^ is not specified, ^dchange^ indicates that you want to plot the
discrete change coefficients.
^basecategory^ is used for an odds ratio plot to specify which category of
the outcome measure is to be used as the reference point.
^labels^ uses variable labels to label each row of the plot. You might need to
revise your variable labels to make them fit the graph. This option does
not work with the ^matrix^ option.
^min(^#^) and max(^#^)^ specify the minimum value and the maximum value on which
the coefficients are plotted. This is useful if you want to compare
coefficients from different logits, or produces several plots from the
same logit.
^ntics^ sets the number of tic marks to show on the axes. Used along with ^min^
and ^max^, this allows you to determine the numbering on the axes and the
location of tic marks.
^packed^ removes the vertical spacing among the outcome categories. This allows
up to 11 variables on a single graph. Otherwise, the maximum is 5.
^prob^ is used for an odds ratio plot to specify that if a coefficient
contrasting two outcomes is not signficant at this level, a line is to
be drawn connecting the letters.
^std(^[^s^|^u^|^0^]...[^s^|^u^|^0^]^)^ specifies the type of coefficient to plot. s specifies
standardized coefficients; u specifies unstandardized coefficients; 0
specifies changes from 0 to 1 in discrete change plots. For example,
std(u0su) indicates that the first variable is unstandardized, the second
is binary, the third standardized, and the last is unstandardized.
^dcadd(^#^)^ is rarely used. In odds ratio plots where
the ^dchange^ option is specified, the size of the letter corresponds to
the square root of the size of the discrete change coefficient. ^dcadd^ adds
an amount to each discrete change making the size of all letters larger,
making it easier to see the letters for small discrete changes. By
default this quantity is 0. If your letters print too small, you might
want to increase this by a small amount, say ^dcadd(.03)^.
Plotting Estimates from Matrices
--------------------------------
The ^matrix^ option indicates that estimates are to be obtained from global
matrices, which are described below. When this option is used, the options
^prob^ and ^dchange^ are not allowed. All other options can be used.
^vars(varlist)^ contains the names of the variables whose coefficients you
want to plot. The names must be in the order you want to plot them and
must be included in the global ^mnlname^.
Globals for Plotting Matrices
-----------------------------
^mnlbeta^ contains the betas in a matrix where element (i,j) is the j-th
variable for comparison i relative to the reference category. That is,
columns are for variables; rows for different contrasts. Note that
constants are NOT included.
^mnlname^ contains the names of the variables corresponding to the columns of
mnlbeta.
^mnlcatnm^ is a string with labels for the outcome categories. The
1st category corresponds to the 1st column of mnlbeta, the 2nd to the
2nd, etc. The label for the reference category should be last. Thus,
mnlcatnm provides labels for the rows of mnlbeta. The option
^basecategory^ references which letter in this string represents the
base category.
^mnlrefn^ is the number of the category in mnlcatnm that is the reference
category for the contrasts contained in mnlbeta.
^mnldepnm^ contains name of dependent variable.
^mnlsd^ contains the standard deviations for the variables that correspond
to columns of mnlbeta
Example using mlogit
--------------------
. ^mlogit occ white ed exper,basecategory(5)^
. ^prchange^
. ^mlogplot white ed exper, dc std(0ss) min(-2.75) max(.55)^
Example using matrices
----------------------
. ^matrix mnlsd = (2.946427, 13.95936, 2.946427, 13.95936)^
. ^global mnlname = "W_Educ W_Exper NW_Educ NW_Exper"^
. ^global mnlrefn = 5^
. ^global mnlcatnm = "Menial BlueC Craft WhiteC Prof"^
. ^matrix mnlbeta = (-.83075, -.92255, -.68761, -.41964 \ /*^
> ^*/ -.03380, -.03145, -.00026, .00085 \ /*^
> ^*/ -.70126, -.56070, -.88250, -.53115 \ /*^
> ^*/ -.11084, -.02611, -.15979, -.05209 )^
. ^matrix mnlbeta = mnlbeta'^
. ^mlogplot, vars(W_Educ NW_Educ W_Exper NW_Exper) matrix /*^
> ^*/ std(ssss) note("Effects of Education")
.-
Authors: J. Scott Long and Jeremy Freese
www.indiana.edu/~jsl650/spost.htm
spostsup@@indiana.edu