Logistic Regression: A Self-learning Text, Third Edition (Statistics in the Health Sciences)

(vip2019) #1

  1. No. The model is not fully parameterized, since it
    contains five parameters whereas the number of
    covariate patterns is 186.

  2. No. The model does not perfectly predict the case/
    noncase status of each of the 289 subjects in the data.

  3. a. The deviance value of 159.2017 is not calculated
    using the deviance formula
    Devðb^Þ¼ 2 lnðL^c=L^maxÞ.
    In particular 2 lnL^c¼ 279 : 317 and 2 lnL^max¼ 0 ,
    so Devð^bÞ¼ 279 : 317.
    b. The other logistic model, i.e., Model 2, is the model
    that is defined by the 186 covariate patterns that
    comprise the fully parameterized model derived
    from the four independent variables PREVHOSP,
    AGE, GENDER, and PAMU. This model will
    contain 185 predictor variables plus an intercept
    term.
    c. 159 : 2017 ¼ 2 lnL^Model 1 ð 2 lnL^Model 2 Þ,
    where 2 lnL^Model 1 ¼ 279 : 3170
    and 2 lnL^Model 2 ¼ 279 : 3170  159 : 2017 ¼ 120 : 1153.
    d. G¼#of covariate patterns¼186 is large relative
    ton¼289.

  4. a. The HL test has aP‐value of 0.4553, which is highly
    nonsignificant. Therefore, the HL test indicates
    that the model does not have lack of fit.
    b. Models 1 and 2 as described in question 5b.
    c. Choose Model 1 since it provides adequate fit and is
    more parsimonious than Model 2. However, a LR
    test cannot be performed since the deviance for
    Model 1 is based on a large number of covariate
    patterns.
    d. Neither model perfectly fits the data, since neither
    model is a saturated model.

  5. Consider the information shown in the output under
    the heading “Partition for the Hosmer and Lemeshow
    Test.”
    a. The 10 groups are formed by partitioning the 289
    predicted risks into 10 deciles of risk, where, for
    example, the highest (i.e., 10th) decile contains
    approximately the highest 10% of the predicted
    risks.
    b. Because subjects with the same value for the
    predicted risk cannot be separated into different
    deciles.
    c. For group 5:
    Expected number of cases¼sum of predicted risks
    for all 29 subjects in group 5.
    Expected number of noncases¼ 29 expected
    number of cases


680 Test Answers

Free download pdf