Exam Pattern: Simulation-Based Power for Comparing Proportions
This note abstracts the normal midterm question where power is estimated from repeated simulated studies rather than from a closed-form formula.
Source Question
STA305 Midterm W2026, Question 5
Setup
Suppose a simulation generates $B$ hypothetical studies under a specific alternative, and the hypothesis test is run on each simulated dataset.
If the number of rejections is $R$, then the estimated power is
$$ \widehat{\text{power}} = \frac{R}{B}. $$Workflow
Step 1: Identify the rejection rule
For a two-sided level-$\alpha$ test, reject when
$$ |Z| > z_{1-\alpha/2}. $$Step 2: Count the simulated rejections
Go through the simulation output and count how many of the $B$ studies reject $H_0$.
Step 3: Estimate power
Compute
$$ \widehat{\text{power}} = \frac{R}{B}. $$In the source problem, this was
$$ \frac{20}{30} = 0.6667. $$Step 4: Interpret in context
Interpret the estimate as the approximate probability that the planned study would reject the null under the stated alternative.
R Template
reject <- c(TRUE, FALSE, TRUE) # replace with simulated decisions
mean(reject)
Exam Checklist
- Identify whether the simulation is under the null or under the alternative
- Count rejections, not just extreme statistics
- Divide by the number of simulated studies
- Give a contextual interpretation, not just a decimal