Power Calculation
Design studies that can truly detect meaningful effects. We scope your hypotheses, select the right tests, set defensible assumptions, and compute sample size or power with sensitivity/robustness checks delivered with reproducible scripts and a clear write-up.
What you get
- Assumption sheet (effect size logic, α, tails, design factors)
- Sample size & power tables for candidate tests/designs
- Sensitivity analyses (dropouts, unequal groups, ICC/cluster effects)
- Reproducible scripts (G*Power settings and/or R code)
- Methods write-up suitable for protocol/IRB/grant/manuscript
- Two revision rounds to match supervisor/journal feedback
Outcome: a defensible, reviewer-ready sample size justification aligned to your aims, design, and constraints.
Why Power Analysis Matters
- Reduces false negatives (Type II): avoids underpowered, inconclusive studies.
- Efficient resource use: right-sized samples save time, funds, and participant burden.
- Reviewer compliance: many IRBs/journals require explicit sample size justifications.
- Transparent assumptions: makes design choices explicit and reproducible.
- Ethical design: balances detection capability with minimal risk/exposure.
Core Inputs We Specify
Effect size
Clinically/academically meaningful difference; standardized metrics (d, r, OR/RR), or raw deltas with SD/variance.
α & tails
Significance level (e.g., .05) and one- vs two-sided hypotheses tied to prior evidence and risks.
Desired power (1–β)
Typical targets 0.80–0.90; higher for critical decisions or multiple endpoints.
Design factors
Group allocation ratio, pairing/repeated measures correlation, blocking/stratification, covariates.
Variance & ICC
Heterogeneity, clustering (multisite/classroom/clinic), design effect for cluster trials.
Outcome type & test
Continuous, binary, count, time-to-event; appropriate tests/models matched to aim.
Common Tests & Designs We Cover
t-tests
One-sample, two-sample (equal/unequal n), paired/repeated measures.
ANOVA/ANCOVA
One-way/Factorial, repeated measures/mixed; covariate-adjusted designs.
Regression/GLM
Linear, logistic, Poisson/negative binomial; R² or predictor effect targets.
Proportions/Chi-square
Two-proportion tests, goodness-of-fit, RxC tables with expected counts.
Survival/Time-to-event
Log-rank/Cox; events needed given hazard ratios, accrual, follow-up, censoring.
Cluster & multilevel
Design effect via ICC; cluster-randomized or classroom/clinic designs.
Non-parametric
Rank-based approximations and robust alternatives when assumptions fail.
Tools we use include G*Power and R (pwr/pwr2ppl/sim-based checks) with shared settings for full reproducibility.
Our Process
1) Define aims & endpoints
Clarify primary/secondary outcomes, hypotheses, and minimal meaningful effects grounded in literature/practice.
2) Choose tests/models
Map outcomes and design to appropriate tests (two-sample, ANCOVA, Cox, GLMM, etc.), including tails and α.
3) Set assumptions
Variance/SD, ICC, correlations, allocation ratio, covariates, drop-out rates; document sources and rationale.
4) Compute & compare
Generate sample size/power tables across plausible ranges; highlight feasible scenarios and trade-offs.
5) Sensitivity & robustness
Stress-test assumptions (e.g., higher variance, unequal groups, attrition) and show implication on n and power.
6) Deliver write-up & scripts
Provide a method statement for protocol/IRB/manuscript plus G*Power screenshots or R code for replication.
Typical Deliverables
- Sample size justification (PDF/DOCX) with assumptions, tables, and interpretation
- G*Power files/settings and/or R scripts (annotated) to reproduce results
- Sensitivity analysis appendix (dropouts, ICC, unequal allocation, alt. effect sizes)
- Protocol/IRB/manuscript paragraph templates describing power calculation
What to Share
- Research question(s), primary endpoint, expected/meaningful effect, and hypothesis direction
- Design details (parallel/paired/repeated/cluster), allocation ratio, and anticipated drop-out
- Any pilot data or literature for SD/variance, ICC, or baseline rates
- Constraints (max feasible n, time, budget) and target power/α
FAQ
We use prior literature, pilot data, or a “minimal important difference” from domain guidance, and run sensitivity tables across plausible values.
Yes calculations account for k:n ratios and the efficiency loss with imbalance; we recommend feasible ratios given constraints.
We incorporate ICC to compute the design effect and adjust sample size for the number/size of clusters.
Within-subject correlation increases efficiency; we use estimated correlations to adjust n and show sensitivity.
Two-tailed is standard unless a strong a-priori directional claim and no interest in the opposite effect; we document the rationale.
Yes expected R² or baseline-outcome correlation can reduce required n; we show both adjusted and unadjusted scenarios.
We inflate n by anticipated attrition and advise strategies (follow-up windows, ITT principles) to mitigate loss of power.
We use approximations/transformations or simulation-based checks where parametric assumptions are doubtful.
For designs beyond closed-form solutions, we use R-based simulations to approximate power under your data-generating assumptions.
A concise paragraph with test, α, tails, effect size definition, power target, assumptions/sources, resulting n, and citations, plus scripts/screens.
Need a defensible sample size fast?
Send your aim, endpoint, and any pilot numbers. We’ll return a scoped plan, fixed quote, and delivery timeline.