Power and Effect Size
I factor 4 levels
Testing? Null True Null false Reject Null Retain Null
Null True Null false Reject Null Type I (α) Correct (1-β) Retain Null (1-α) Type II (β)
effect TREATMENT NULL power β 1 - β 1- α α/2 α/2
What determines power? Effect size Sample size Variability Significance level 1 or 2 tail choice Kind of test
effect NULL TREATMENT power β 1 - β 1- α α/2 α/2
Change significance level
1 or 2 tail
Feature Increase power Decrease power Effect size large small Population σ small σ big σ Sample size (n) big significance Lenient (0.05) Strict (0.01) 1 or 2 tail one two
power is in the sampling distributions, whereas effect size is in the population distributions
assumptions
Effect Size The extent to which 2 populations do not overlap d =( μ1 - μ2)/ σ Cohen’s d 0.2 is small effect 0.5 medium 0.8 large
Effect size Small (f =0.1) Medium (f=0.25) Large(f=0.4) 3 groups Approximate number of participants in each group (equal variances) to achieve 80% power for one-way ANOVA at 0.05 significance level Effect size Small (f =0.1) Medium (f=0.25) Large(f=0.4) 3 groups (dfbetween=2) 322 52 21 4 groups (dfbetween =3) 274 45 18 5 groups (dfbetween =4) 240 39 16 How many?
Degrees of freedom 2 factor Factor A has a levels and df= a-1 Factor B has b levels and df = b-1 Interaction df = (a-1)(b-1) Error df = N - ab