Effect Size in the Two-Factor Mixed Design
Below are the equations for the partial measure for the two-factor design. The overall formula for the partial squared measure is a measure of the effect divided by the measure of the effect plus error variance. The formulas below calculate effect size based on the F statistic for each of the three effects in the two-factor model. Note that (a)(b)(n) in the numerator is the number of levels of factor a times the number of levels of factor a times the number of subjects at each level of the groups formed by crossing factor a and factor b. See material in Kirk (1995 p. 518-519). Note the partial measure ignores other effects in the ANOVA model.
Measures of Population Effect Size
Example Effect Size Calculations
Rehearsal Experiment. The Rehearsal Study conducted on the first
day of class had two levels of rehearsal (primary and secondary) and two
levels of type of word (concrete and not concrete).The interaction F =2.82
was not statistically significant.
ATLAS Study. Here there were three waves of measurement (baseline,
immediate followup and one-year followup) and two groups (ATLAS program
or control). The interaction F=24.46 for nutrition behavior was statistically
significant. I used the average n for the calculations below. The harmonic
mean would be more accurate, however.
Power for the Two-Factor ANOVA Interaction Effect
The power calculations for a Two-factor ANOVA are based on the means for the levels of each factor. Here are the retrospective power calculations for Interaction of Drive and Drug for the monkey study described in Keppel. Below are the means in each of 6 groups that are used in computing . is in the formula for the and f measures. Remember that MS S/AB = 18.33 for the experiment.
a1
a2 a3
mean
b1 3
10 14 9
mean for b1
b2 11
12 10 11
mean for b2
mean 7 11
12 10 grand mean
Computation of Interaction Deviations
-3 = 3 -7 -9 + 10
0 =10 -11 -9 + 10
3 =14 -12 -9 + 10
3 =11 -7 -11 + 10
0 =12 -11 -11 + 10
-3 =10 -12 -11 + 10
=(-3)2+(0)2+(3)2+(3)2+(0)2+(-3)2=36
Pearson-Hartley Tables
Looking up
=1.62 in the Pearson-Hartley tables on page 510 for numerator degrees of
freedom equal to 2, and denominator degrees of freedom equal to 18, we
find power equal to .62.
Cohen's Tables
The f effect size measure is obtained using the following formula:
We also need to calculate the correct sample size for the interaction
given the two main effects in the model. For A X B Interaction the new
n is equal to 7.
Rounding the f = .572 to .6, and sample size of 7, we find that the
power is about .62 reading from the table with numerator degrees of freedom
equal to 2 (page 313).