home *** CD-ROM | disk | FTP | other *** search
-
- MCNEMAR'S TEST
-
-
- When the data are correlated in some
-
- way, for example, when two diagnostic
-
- tests are compared by applying them
-
- both to the same people, and then
-
- observing the diagnosis from each
-
- test (positive or negative),
-
- condition 3 for the Chi-square test is
-
- violated.
-
- If we arrange our data as follows,
-
- then it is managable with a test known
-
- as McNemar's test. McNemar's test
-
- uses the Chi-square statistic, and
-
- tests the hypothesis that the two
-
- diagnostic tests are equivalent.
-
- The same hypothesis can be tested by
-
- a straight-forward calculation of a
-
- binomial probability. The program
-
- also does the binomial calculation.
-
- Suppose that two diagnostic tests
-
- are both applied to 75 men. The
-
- results can be tallied as follows:
-
-
- TEST 1
-
- positive negative
- ! ! !
- ---------------------------
- ! ! !
- positive ! 32 ! 10 ! 42
- ! ! !
- TEST 2 ---------------------------
- ! ! !
- negative ! 7 ! 26 ! 33
- ! ! !
- ---------------------------
- ! ! !
- ! 39 ! 36 ! 75
- ! ! !
-
-
- When the tests do not agree, do they
-
- do so in about the same way? That is,
-
- do the cases when TEST 1 is positive
-
- and TEST 2 is negative occur in equal
-
- numbers to the cases when TEST 1 is
-
- negative and TEST 2 is positive?
-
- We see in the example above that
-
- those numbers are 10 and 7,
-
- respectively. Is that a large enough
-
- difference to make the tests
-
- different?
-
- If you want to run LOADSTAR 2X2
- \oad"fisher",8
- STATISTICS now, press "\".
-
- Al Vekovius
-
- ----------< end of article >----------
-
-