13.11.2012 Views

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

Introduction to Categorical Data Analysis

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

PROBLEMS 165<br />

c. The concordance index c equals 0.658 for the model with the four main<br />

effects and the six interaction terms, 0.640 for the model with only the four<br />

main effect terms, and 0.568 for the model with only T/F as a predic<strong>to</strong>r.<br />

According <strong>to</strong> this criterion, which model would you choose (i) if you want<br />

<strong>to</strong> maximize sample predictive power (ii) if you think model parsimony is<br />

important?<br />

5.7 From the same survey referred <strong>to</strong> in Problem 4.16, Table 5.11 cross-classifies<br />

whether a person smokes frequently with the four scales of the MBTI personality<br />

test. SAS reports model −2 log likelihood values of 1130.23 with only<br />

an intercept term, 1124.86 with also the main effect predic<strong>to</strong>rs, 1119.87 with<br />

also all the two-fac<strong>to</strong>r interactions, and 1116.47 with also all the three-fac<strong>to</strong>r<br />

interactions.<br />

a. Write the model for each case, and show that the numbers of parameters<br />

are 1, 5, 11, and 15.<br />

b. According <strong>to</strong> AIC, which of these four models is preferable?<br />

c. When a classification table for the model containing the four main effect<br />

terms uses the sample proportion of frequent smokers of 0.23 as the cu<strong>to</strong>ff,<br />

sensitivity = 0.48 and specificity = 0.55. The area under the ROC curve<br />

is c = 0.55. Does knowledge of personality type help you predict well<br />

whether someone is a frequent smoker? Explain.<br />

Table 5.11. <strong>Data</strong> on Smoking Frequently and Four Scales of Myers–Briggs<br />

Personality Test<br />

Extroversion/Introversion E I<br />

Sensing/iNtuitive S N S N<br />

Smoking Frequently<br />

Thinking/ Judging/<br />

Feeling Perceiving Yes No Yes No Yes No Yes No<br />

T J 13 64 6 17 32 108 4 9<br />

P 11 31 4 14 9 43 9 26<br />

F J 16 89 6 25 34 104 4 27<br />

P 19 60 23 57 29 76 22 57<br />

Source: Reproduced with special permission of CPP Inc., Mountain View, CA 94043. Copyright 1996 by<br />

CPP Inc. All rights reserved. Further reproduction is prohibited without the Publisher’s written consent.<br />

5.8 Refer <strong>to</strong> the classification table in Table 5.3 with π0 = 0.50.<br />

a. Explain how this table was constructed.<br />

b. Estimate the sensitivity and specificity, and interpret.<br />

5.9 Problem 4.1 with Table 4.8 used a labeling index (LI) <strong>to</strong> predict π = the<br />

probability of remission in cancer patients.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!