Plots logit of a variable by categories of another variable ----------------------------
^grlogit^ depvar indepvar [if] [, gropt(...) ]
^grlogit^ plots the logit of one variable against categories of another variabl > e.
This may be of some use in the beginning of logistic regression modelling. You could use this program to confirm visually that there isn't a non-linear relationship of the logit of the dependent variable to the predictor variable. Thus, you could screen your explanatory variables during the univariate analysis stage.
It is not for use in assessing incorrectly specified link functions in multivariable models.
The logit of a variable x is defined as ^logit^(x) = x / (1-x)
^gropt(...)^ specifies a string with options that are passed along to the graph command.
Remarks ------- This program does not accept weights yet. It should be easily modified for this. Just fiddle around with the "table" statement.
For a discussion goodness-of-fit of multivariable models, see Hosmer & Lemeshow: Applied Logistic Regression, 1989 (John Wiley & Sons)
. ^grlogit^ response age . ^grlogit^ response age, gropt("xlab(15(10)70)")
Author ------- Jan Brogger, University of Bergen, Norway jan.brogger@@med.uib.no
Shameless plug :) -------------- I am looking for internship & residency in internal medicine/pulmonary medicine US/EU. Am doing PhD in respiratory epidemiology now.