Analysis of Learning Effects

In experiments of this type, the radiologists see an image at many compression levels during the course of the study. One needs to ascertain whether learning effects are significant. Learning and fatigue are both processes that might change the score of an image depending on when it was seen. For the CT study, the McNemar test [8] was used to examine this possibility.

Suppose one has N observations of paired data. The members of the pair are called sample 1 and sample 2. Each member of the pair can be described as being "A" or "not A." There are clearly four types of pairs: those with both members of type "A," those with neither member of type "A," those where the first member of the pair is of type "A" but the second is not, and those where the second member is of type "A" but the first is not. The last two types are referred to as disparate or "untied" pairs. We denote the number of occurrences of each type of these four pairs by k, m, r, and 5, as shown in a 2 x 2 table, in Table 3.

The proportions of individuals of type "A" in the two samples are k + r , k + 5

and the difference between the two proportions is

TABLE 3 2x2 table of pairs in the McNemar analysis

Sample 2

0 0

Post a comment