Example 74.3 Logistic Regression

In this example, a series of people are asked whether or not they would subscribe to a new newspaper. For each person, the variables sex (Female, Male), age, and subs (1=yes,0=no) are recorded. The PROBIT procedure is used to fit a logistic regression model to the probability of a positive response (subscribing) as a function of the variables sex and age. Specifically, the probability of subscribing is modeled as

     

where is the cumulative logistic distribution function.

By default, the PROBIT procedure models the probability of the lower response level for binary data. One way to model is to format the response variable so that the formatted value corresponding to is the lower level. The following statements format the values of subs as 1 = ’accept’ and 0 = ’reject’, so that PROBIT models . They produce Output 74.3.1.

data news;
   input sex $ age subs @@;
   datalines;
Female     35    0   Male       44    0
Male       45    1   Female     47    1
Female     51    0   Female     47    0
Male       54    1   Male       47    1
Female     35    0   Female     34    0
Female     48    0   Female     56    1
Male       46    1   Female     59    1
Female     46    1   Male       59    1
Male       38    1   Female     39    0
Male       49    1   Male       42    1
Male       50    1   Female     45    0
Female     47    0   Female     30    1
Female     39    0   Female     51    0
Female     45    0   Female     43    1
Male       39    1   Male       31    0
Female     39    0   Male       34    0
Female     52    1   Female     46    0
Male       58    1   Female     50    1
Female     32    0   Female     52    1
Female     35    0   Female     51    0
;
 
proc format;
   value subscrib 1 = 'accept' 0 = 'reject';
run;
proc probit data=news;
   class subs sex;
   model subs=sex age / d=logistic itprint;
   format subs subscrib.;
run;

Output 74.3.1 Logistic Regression of Subscription Status
The Probit Procedure

Iteration History for Parameter Estimates
Iter Ridge Loglikelihood Intercept sexFemale age
0 0 -27.725887 0 0 0
1 0 -20.142659 -3.634567629 -1.648455751 0.1051634384
2 0 -19.52245 -5.254865196 -2.234724956 0.1506493473
3 0 -19.490439 -5.728485385 -2.409827238 0.1639621828
4 0 -19.490303 -5.76187293 -2.422349862 0.1649007124
5 0 -19.490303 -5.7620267 -2.422407743 0.1649050312
6 0 -19.490303 -5.7620267 -2.422407743 0.1649050312

Model Information
Data Set WORK.NEWS
Dependent Variable subs
Number of Observations 40
Name of Distribution Logistic
Log Likelihood -19.49030281

Class Level Information
Name Levels Values
subs 2 accept reject
sex 2 Female Male

Last Evaluation of the Negative of the Gradient
Intercept sexFemale age
-5.95557E-12 8.768324E-10 -1.6367E-8

Last Evaluation of the Negative of the Hessian
  Intercept sexFemale age
Intercept 6.4597397447 4.6042218284 292.04051848
sexFemale 4.6042218284 4.6042218284 216.20829515
age 292.04051848 216.20829515 13487.329973

Analysis of Maximum Likelihood Parameter Estimates
Parameter   DF Estimate Standard Error 95% Confidence Limits Chi-Square Pr > ChiSq
Intercept   1 -5.7620 2.7635 -11.1783 -0.3458 4.35 0.0371
sex Female 1 -2.4224 0.9559 -4.2959 -0.5489 6.42 0.0113
sex Male 0 0.0000 . . . . .
age   1 0.1649 0.0652 0.0371 0.2927 6.40 0.0114

Output 74.3.1 shows that there appears to be an effect due to both the variables sex and age. The positive coefficient for age indicates that older people are more likely to subscribe than younger people. The negative coefficient for sex indicates that females are less likely to subscribe than males.