Total N
-
Power
-
Effect (Cohen's d)
-
Min N for 80% power
-
⚠ Cohort-anchored requires at least 2 control-phase assessments (nDE ≥ 2). With only one baseline observation, the model cannot separate natural change over time from the treatment effect. If you never observe how participants change during the control phase, you can't attribute post-crossover improvement to treatment rather than time. A second pre-crossover assessment anchors the time slope and is essential for this design. Power at nDE=1 collapses to near the alpha level regardless of sample size. Either add a second control-phase assessment, or switch to Calendar-anchored: No (rolling), which drops the time covariate and works fine at nDE=1.
⚠ Extreme event rates detected. The linearized normal approximation used for binary power is most accurate for event rates between 0.10 and 0.90. At extreme rates, the approximation may over- or underestimate power. For validation, use the exported R simulation script which uses exact logistic mixed models.
Design matrix
CSV design matrix — one row per cluster, one column per period. Cell values (case-insensitive):
Control (pre-treatment):
Treatment:
Excluded (cell not observed):
Follow-up:
Follow-up coded as control:
Example row:
Control (pre-treatment):
DE / 0 / C / controlTreatment:
TX / 1 / T / treatmentExcluded (cell not observed):
X / - / . / blankFollow-up:
FU / followup / follow-upFollow-up coded as control:
FU-ctrl / FUCExample row:
DE,DE,TX,TX. Accepted separators: commas, tabs, semicolons, pipes. Lines starting with # are treated as comments and skipped.Design info
Design comparison
Power visualized
Sensitivity analysis min N across ICC × effect
Summary paragraph
Statistical model
Export report
Parameters, justification, design matrix, and sensitivity table.
Share configuration