5
$\begingroup$

I have 400 patients and I apply 4 tests to all of them. This gives me a number of positive and negative for each test. I want to compare the tests 2 by 2 to find out if there is any statistically significant difference between them.

Test  Pos   Neg
1     100   300
2     150   250
3     160   240
4     200   200

What statistical test I can use to determine this? I'm very new to the field so I appreciate any pointers.

Improve this question
$\endgroup$
4
$\begingroup$

McNemar's test is the two by two comparison. NB You want to record for tests #1 & #2 how many patients tested +ve in both #1 & #2, how many +ve in #1 but -ve in #2, how many -ve in #1 but +ve in #2, & how many tested -ve in both #1 & #2.

Improve this answer
$\endgroup$
1
$\begingroup$

I was going to suggest the chi-squared test, but I think that your suggestion of McNemar's test would be better. A related topic would be the Fisher's exact test: https://en.wikipedia.org/wiki/Fisher%27s_exact_test.

Improve this answer
$\endgroup$
1
  • 1
    $\begingroup$ From the example it looks like McNemar is what I need, but I'll keep Fisher in my mind too. There are just too many similar tests for a beginner to comprehend :) $\endgroup$
    – rslite
    Apr 8 '13 at 23:25
1
$\begingroup$

Perform a McNemar test for independence in a 2x2 table

Improve this answer
$\endgroup$
0
$\begingroup$

Paired proportions have traditionally been compared using McNemar's test but an exact alternative due to Liddell (1983) is preferable.

Useful links:

  • www.statsdirect.com/help/default.htm#chi_square_tests/mcnemar.htm
  • freesourcecode.net/matlabprojects/68089
  • jech.bmj.com/content/37/1/82.abstract
Improve this answer
$\endgroup$
1
  • $\begingroup$ There is an exact version of McNemar's test. $\endgroup$
    – gung - Reinstate Monica
    May 14 '15 at 4:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.