What statistical test should you first perform prior to determining agreement between two tests, and why?

• A. McNemar's chi-square test to test if significant exists between test results because if there is significance no reason to quantify level of agreement
• B. Pearson correlation to assess association
• C. ANOVA to compare means
• D. Fisher's exact test to compare proportions

Answer: (A)

When you’re comparing two diagnostic tests on the same subjects, you need a test that uses the pairing of results. Build a 2x2 table with the counts where both tests are positive, both negative, and the two discordant cases where they disagree. McNemar’s test focuses on those discordant pairs to ask whether the two tests have the same probability of giving a positive result. It tests the null hypothesis that the off-diagonal counts are equal, i.e., there’s no systematic difference between the tests.

If this test is significant, there’s evidence of a bias or difference between the tests, which means that a single overall measure of agreement could be misleading or require special interpretation. If it’s not significant, the tests don’t show a detectable bias, and you can proceed to interpret agreement measures like kappa with more confidence.

Why the other options aren’t fit as the first step: Pearson correlation is for continuous data and captures association, not agreement on binary classifications. ANOVA compares means of continuous outcomes, not agreement between paired binary tests. Fisher’s exact test looks at independence in a 2x2 table for unpaired data and doesn’t account for the paired nature of the test results.