Calculates change in coefficients between logistic models ----------------------------
^lrchg^ , store(n) or ^lrchg^ , contr(n m)
Description -----------
^lrchg^ displays the coefficients in two logistic models, and calculates the proportion of change in coefficients between two models.
This is useful in logistic regression modelling in epidemiology.
Use ^lrchg^ , store(1) ^lrchg^ , store(2) ^lrchg^ , contr(1 2)
to store two models, then contrast them.
Options -------
store(n) Stores logistic regression results in a variable coeffn Use before contrasting two models.
contr(n m) Contrasts two models previously stored.
nocoeff Drop the calculations of changes in coefficients. Examples --------
. xi: logistic response i.agek6 i.gend i.agek6*i.gend i.regio . lrchg, store(0) . xi: logistic response i.agek6 i.gend i.agek6*i.gend . lrchg, store(1) . lrchg, contr(0 1)
Likelihood ratio tests and change in coefficients between logistic regression models -------------------------------------------------- Diff. in LL = 99.855 Diff. in df = 1 P-value = 0.0000
Change in common coefficients between models 0 and 1:
Coefficient Model 0 Model 1 %change -------------------------------------------------- Iage_1 0.9293 0.8923 4.0 Iage_2 1.2648 1.2064 4.6 Igend_2 1.6332 1.6133 1.2 IaXg_1_2 0.9801 0.9834 -0.3 IaXg_2_2 0.9768 0.9848 -0.8 -------------------------------------------------- Iregio_2 0.9293 0.0000 uniq
Remarks ------- Presently the coefficients are stored in matrices named coeff`n'. This shouldn't cause any problems, but it is not elegant.
It might be better to use -estimates hold- but will have to get back to this.
The author wonders what to do if there are zero degrees of freedom ?
Inspired by the program lrdrop1 by Zhiqiang Wang.
Author ------- Jan Brogger, University of Bergen, Norway jan.brogger@@med.uib.no (21 May 2000)
Shameless plug :) -------------- I am looking for internship & residency in internal medicine/pulmonary medicine US/EU. Am doing PhD in respiratory epidemiology now.