Propensity Score Matching Power Calculator

Size your matched observational study before launching data pulls. Enter the treated and control counts you expect to keep after propensity score matching, the effect size you need to detect, and your alpha threshold. Add an optional post-match standard deviation to convert your effect into Cohen's d while the calculator reports statistical power and control counts required to hit 80% power with the current treated cohort.

Count of treated observations retained after matching.
Count of control observations retained after matching.
Effect size in outcome units after matching (e.g., cost savings or rate change).
Two-sided Type I error rate to test the treatment effect.
Defaults to 1. Use the pooled post-match standard deviation of the outcome.

Applies large-sample normal approximations. For small matched cohorts or binary outcomes, use exact or simulation-based power analyses.

Examples

  • 420 treated, 630 controls, ATE 0.18, alpha 5%, SD blank ⇒ With 420 treated and 630 controls, the detectable effect of 0.18 units equals 0.18 SD after matching. Two-sided power at 5.00% alpha is 81.53%. Your current sample meets or exceeds the 80% power target for this effect size.
  • 220 treated, 220 controls, ATE 0.25, alpha 1%, SD 1.2 ⇒ With 220 treated and 220 controls, the detectable effect of 0.25 units equals 0.21 SD after matching. Two-sided power at 1.00% alpha is 34.80%. Even with unlimited controls, you cannot hit 80% power without adding more treated cases.

FAQ

How do caliper widths impact these results?

Tighter calipers reduce bias but can shrink your matched sample. Recalculate power after each matching iteration to confirm you still meet your detectable effect goals.

Can I model 2:1 or variable-ratio matching?

Yes. Enter the projected treated and control counts after applying your preferred ratio. The calculator adjusts power automatically.

Do I need to adjust alpha for multiple outcomes?

If you plan to test several endpoints, lower the alpha input to reflect your multiplicity strategy, such as Bonferroni or Holm adjustments.

What if the outcome is binary instead of continuous?

Use the calculator for a quick directional read, then validate with logistic regression power simulations or exact methods tailored to binary endpoints.

Additional Information

  • Power uses a two-sided z-test approximation, appropriate for large matched samples.
  • The optional standard deviation converts your treatment effect into Cohen's d for easy benchmarking.
  • Required controls for 80% power assume the treated sample remains fixed and matching quality holds as you add controls.
  • If your alpha is more conservative than 5%, expect the minimum control count for 80% power to rise noticeably.

Related calculators