.-
help for ^pre^ - 1.2 - 21 Feb 2006
.-
Calculate the Proportional Reduction in Errors achieved by a model
.-
^pre^ [^, ^c^utoff(#) ^ct^type(measure of central tendency) Maxcut]
Description
-----------
^pre^ is a post-estimation command that calculates the proportional reduction in
errors achieved by the model. pre is a measure of the precictive power of a model.
This routine is simpler and is applicable to more models than lstat.
The approach of this procedure has three steps
1. find the errors when guessing without the model = E1
(the guess is a measure of central tendency)
2. find the errors when guessing what the model predicts = E2
3. PRE = (E1-E2)/E1
If the central tendency is mode, the errors are the number of cases
not predicted exactly.
If the central tendency is median, the errors are the sum of the absolute value of
difference between the value and the median.
If the central tendency is mean, the errors are the sum of the squares of
difference between the value and the mean.
Options
-------
^c^utoff(real) is the cutoff probability for logistic regression, i.e. the
value above which the prediction is 1. The default is 0.5.
^ct^type(string) is the measure of central tendency (mean, median or mode).
The default for logit, probit, mlogit and gologit2 is mode.
The default for ologit, plogit, streg is median.
The default for all other models is mean.
^m^axcut specifies using a cutoff that maximizes the number
of correctly predicted outcomes. Used with logistic, logit, probit.
Examples
--------
. sysuse auto, clear
. logit foreign rep78 price turn
. ^pre^
. ^pre^, maxcut
. sysuse citytemp, clear
. nbreg heatdd tempjan region division cooldd tempjuly
. ^pre^
Author: Paul Millar
www.paulmillar.ca
paulmi@@nipissingu.ca
See also:
---------
Online: help for @lstat@ (if installed)