Research instrument · Study design

Study Planner.

Before collecting a single observation, understand the scale of the question. Explore effect size, power and the sample your design requires.

Define the question

Study assumptions

Live calculation

Two independent groupsContinuous outcome

Two-sided, pooled-variance t test. Equal outcome variance and independent observations are assumed.

Meaningful differenceendpoint unitsThe difference you plan to detect

Common standard deviationsame unitsUse a defensible prior estimate

Standardized effect · Cohen’s d0.50

Target statistical power 80%

50%99%

Significance levelα0.001–0.10; two-sided

Allocation B : Aratio1 is equal group allocation

Expected dropout%Recruitment inflated per group

Starting values are an illustrative design, not parameters from a clinical study.

Your design, made visible

Participants to recruit

152total

80.1%Achieved power

Group A76to recruit64 evaluable

Group B76to recruit64 evaluable

128 evaluable participants · 15% expected dropout · 152 to recruit.

Power × evaluable sampleRecommended sample · 64 in group A · 80.1%

Evaluable participants in group A · allocation held constant. Marker matches the recommended design. Drag to explore alternatives.

Power is the probability of detecting the assumed effect with this design. It does not estimate whether a treatment works. Multiplicity, clustering and unequal variance require a different analysis.

Noncentral t modelBoth rejection tails

Method & interpretation

Design first. Interpretation follows.

A defined statistical design

Two independent groups, a continuous outcome, a common standard deviation and a two-sided pooled-variance t test. Both rejection tails are included. Power is computed by numerical integration of the noncentral t distribution.

R stats · power.t.test reference ↗

Recruitment, not just analysis

The minimum whole-number evaluable sample is calculated for each group. Recruitment is then increased by the assumed dropout fraction and rounded up separately. Allocation is B:A; 1 means equal groups.

Effect size is the meaningful mean difference divided by the common standard deviation.

Know when to change the model

Paired data, cluster randomization, repeated measures, binary outcomes, multiple comparisons and unequal variances need different methods. These controls cannot represent those designs. A statistician should review an actual study protocol.

A planning estimate, not a protocol approval or evidence of treatment efficacy.

Open Institute of Age and Aging

Published methods. Traceable sources. Questions worth pursuing.

Collaborate with the laboratory