G. Conditional algorithm estimatesbs but nota
(nuisance parameter).
H. Matched data: unconditional gives biased
estimates, whereas conditional gives unbiased
estimates.
V. Overview on statistical inferences for logistic
regression(pages 117–121)
A. Two types of inferences: testing hypotheses and
confidence interval estimation.
B. Three items obtained from computer output for
inferences:
i. Maximized likelihood value L(^u)
ii. Estimated variance–covariance matrixV^ð^uÞ:
variances on diagonal and covariances on the
off-diagonal;
iii. Variable listing with ML estimates and
standard errors.
C. Two testing procedures:
i. Likelihood ratio test: a chi-square statistic
using2lnL^.
ii. Wald test:aZtest (or equivalentw^2 test) using
standard errors listed with each variable.
D. Both testing procedures give approximately same
results with large samples; with small samples,
different results are possible; likelihood ratio test is
preferred.
E. Confidence intervals: use large sample formulae
that involve variances and covariances from
variance–covariance matrix.
Detailed Outline 123