stat power Flashcards
(24 cards)
Define statistical power.
Probability of correctly rejecting a false null hypothesis (1 − β).
Type II error rate is denoted by (a)_____
(a) β
List the four quantities inter-related in power calculations.
α (significance level), β (Type II error), effect size, sample size.
What is an a priori power analysis?
Power calculation performed during study planning to determine the required sample size.
Post-hoc power analysis is generally discouraged. Why?
A non-significant result does not prove H₀; calculating power afterwards conditions on an unknown effect and is misleading.
Differentiate manipulative and observational experiments.
Manipulative: researcher deliberately alters factors;
Observational: researcher measures variables as they occur naturally.
Give an example of a treatment and block in an observational study.
Treatment: smoking-status categories; Block: participant groups sharing the same treatment level.
Explain why sample sizes that are too large can be unethical in clinical studies.
Participants may face risk unnecessarily when fewer subjects would suffice to detect the effect.
According to Cohen’s convention, what d values correspond to small, medium, and large effect sizes?
0.2 (small), 0.5 (medium), 0.8 (large).
Using G*Power, which three inputs would you supply to compute required N for a t-test?
α (e.g., 0.05), desired power (e.g., 0.80), expected effect size d.
Name two ways to increase power aside from increasing N.
Use a larger α (if justified) or reduce measurement noise (increase precision).
The relationship among α, β, and effect size for fixed N is (a)___ -reducing one inflates (at least) one of the others.
(a) inverse
What graphical tool helps visualise trade-offs among power, N, and effect size?
Power curve (plot of power vs. sample size for a specified effect).
When is a one-tailed test justified in power analysis?
Only when effects in the opposite direction are impossible or biologically meaningless a priori.
Give a rule of thumb for desired power in biomedical research.
0.80 (80 %) is commonly accepted; higher if consequences of Type II error are severe.
What parameter in ANOVA drives the non-centrality parameter used in power calculations?
The ratio of between-group variance to within-group variance (signal-to-noise).
Describe sequential stopping rules in the context of power/ethics.
Interim analyses allow ending a trial early for efficacy or futility, safeguarding participants.
What is the mnemonic “#events ≈ 50 / (d²)” used for?
Quick estimate of required events in survival analysis given effect size d (log-hazard ratio).
Name one assumption common to most power formulas that, if violated, inflates Type I error.
Independence of observations.
Power plot axes:
X-axis: sample size (N); Y-axis: power (1 − β).
Software besides G*Power capable of power analysis?
Minitab (add-on), SPSS (add-on), or R packages (pwr, simr).
In a two-group t-test with d = 0.8, α = 0.05, desired power = 0.80, how many total participants are needed?
≈ 52 participants (26 per group).
True or False: Lowering α from 0.05 to 0.01 increases power (all else equal).
False – lowering α makes it harder to reject H₀, so power decreases.
Power = 1 − (?)
β
