Inference
Robustness lab
15 min
Learning goals
- •You can explain why pre-tests for normality before a t-test are obsolete.
- •You can justify why Welch and Yuen are preferable to Student's t for real-world data.
- •You can interpret the concepts of 20% robustness and actual power from simulation results.
Population distributions
Kernel density estimate (Gaussian, Silverman bandwidth) from 15,000 simulated observations per group.
σ₁ / σ₂ = 1.00 · δ = 0.00
Pick a scenario on the left (or customise one) and run the simulation to see how the different test strategies perform.
Each run computes the actual error rate (α̂) and – if δ > 0 – the empirical power at the same time.