2.Comparison of conditional probability and unconditional (or marginal)
probability: for example, PrðX¼þjY¼þÞversus PrðX¼þÞ.
3.Comparison of conditional probabilities: for example, PrðX¼þjY¼þÞ
versus PrðX¼þjY¼Þ. The screening example above yields
PrðX¼þjY¼þÞ¼ 0 : 406
whereas
PrðX¼þjY¼Þ¼
362
23 ; 724
¼ 0 : 015
again clearly indicating a statistical relationship. It should also be noted
that we illustrate the concepts using data from a cancer screening test but
that these concepts apply to any cross-classification of two binary factors
or variables. The primary aim is to determine whether a statistical rela-
tionship is present; Exercise 3.1, for example, deals with relationships
between health services and race.
The next two sections present some applications of those simple probability
rules introduced in Section 3.1.2: the problem ofwhento use screening tests and
the problem ofhowto measure agreement.
3.1.4 Using Screening Tests
We have introduced the concept of conditional probability. However, it is
important to distinguish the two conditional probabilities, PrðX¼þjY¼þÞ
and PrðY¼þjX¼þÞ. In Example 1.4, reintroduced in Section 3.1.3, we
have
PrðX¼þjY¼þÞ¼
154
379
¼ 0 : 406
whereas
PrðY¼þjX¼þÞ¼
154
516
¼ 0 : 298
Within the context of screening test evaluation:
PROBABILITY 115