Understanding McNemar Sample Size
McNemar studies use paired binary outcomes. Each subject gives two linked responses. The calculator focuses on discordant pairs because concordant pairs do not drive the test statistic. A large difference between p01 and p10 usually needs fewer paired observations. A small difference needs more observations.
Why Discordant Proportions Matter
The values p01 and p10 describe opposite changes. One may represent failure before treatment and success after treatment. The other may represent success before treatment and failure after treatment. Their sum estimates the discordant rate. Their difference estimates the paired effect. Power rises when the difference grows and the discordant rate remains practical.
Planning With Alpha and Power
Alpha controls the false positive risk. Power controls the chance of detecting the planned effect. A two sided design is stricter than a one sided design. It often needs more pairs. The tool uses a normal approximation for planning. It also reports expected discordant counts, adjusted pairs, achieved power, and the discordant odds ratio.
Continuity and Dropout
A continuity correction adds a conservative reserve. It is useful when expected discordant counts are small. Dropout adjustment inflates the final target. That helps protect power after missing visits, unusable records, or incomplete paired measurements. The adjusted total should be treated as the recruitment target.
Good Use Cases
This calculator supports matched case control studies, before after diagnostic studies, crossover screens, and paired classifier comparisons. It is most useful before data collection. Enter realistic values from pilot data, prior studies, or expert assumptions. Then compare several scenarios. Conservative planning avoids underpowered studies and expensive redesigns.
Limits
The result is an approximation. Very rare discordance, very small effects, or strict exact testing may need simulation or specialist software. Always document assumptions. Also keep ethical, clinical, and budget limits visible. A strong sample plan should be statistically clear and operationally possible.
Interpreting Results
The unadjusted total shows the mathematical minimum before loss. The final total includes dropout and rounding. Expected b and c counts help check whether assumptions are believable. If either count is very low, plan a sensitivity check. Review several p01 and p10 pairs. This reveals how fragile the design may be before recruitment begins. It also supports transparent protocol review later.