PROC LOGISTIC: Effect coding

bncoxuk — Thu, 24 Nov 2011 19:36:05 GMT

Hi Sir, sorry to bother. I really like to understand the default coding method used in the LOGISTIC procedure. SAS manual said the default one is 'Effect coding'. I used this default setting and got the results as below. Note the model has an interaction term age*use.

My questions are:

1) how to interpret the coefficents of the interaction term?

2) how to calculate the coefficents for the other groups of the interaction term? Age has the other group 'M', and use has the other group 'T'. So I want to get the coefficents for age='D' & use='T', age='G' & use='T', age='O' & use='T', age='M' & use='T'.

I spent a lot of time searching on the internet about 'effect coding', interaction, but cannot find any useful learning resources. Any recommendation of textbooks?

Many thanks a lot for help.

The LOGISTIC Procedure

Analysis of Maximum Likelihood Estimates

Standard Wald
Parameter DF Estimate Error Chi-Square Pr > ChiSq

Intercept       1     -3.5735      0.0138    66945.2041        <.0001
age     D       1     -0.1990      0.0171      136.2082        <.0001
age     G       1      0.0993      0.0245       16.3795        <.0001
age     O       1      0.2356      0.0284       68.8867        <.0001
use     B       1      0.0716      0.0162       19.6340        <.0001
use     C       1     -0.4513      0.0245      338.7862        <.0001
age*use D B     1     0.00122      0.0200        0.0037        0.9514
age*use D C     1     -0.0115      0.0301        0.1474        0.7010
age*use G B     1      0.0727      0.0287        6.4342        0.0112
age*use G C     1     -0.0459      0.0444        1.0681        0.3014
age*use O B     1     -0.0953      0.0333        8.2016        0.0042
age*use O C     1      0.0834      0.0500        2.7795        0.0955

PROC LOGISTIC: Effect coding

Reeza — Fri, 25 Nov 2011 22:56:42 GMT

Design and Analysis of Experiments by Montgomery is a good one.

Another good exercise is to write out your model with the dummy variables as SAS has coded them and see if its what you need.

PROC LOGISTIC: Effect coding

Doc_Duke — Mon, 28 Nov 2011 15:18:45 GMT

Personally, I find the coefficients easier to interpret if I use "reference cell" coding.

A good reference that uses SAS for the examples is Paul Allison's "Logistic Regression Using the SAS System" that is a SAS BBU.

topic PROC LOGISTIC: Effect coding in Statistical Procedures

PROC LOGISTIC: Effect coding

PROC LOGISTIC: Effect coding

PROC LOGISTIC: Effect coding