Minimum detectable effect size for main outcomes (accounting for sample
design and clustering)
2,500 observations per experience treatment, a group of 4,000 without any experience in the survey. A maximum of 60 subjects per treatment in the lab (a total of 360). To look into the effect of information, we pool together experience treatment (5,000 respondents, 180 subjects per information treatment).
Examples of two computations are provided below for equivalence trials, with a 80% power and a 5% threshold:
- In a continuous case, say with the short-run inflation point forecast, in percentage points, with an average of 4 in Tr. A vs. 4.5 in Tr. B, and a standard deviation close to 2, so effect size of 0.5/2= 0.25, we need about 250 respondents per treatment.
- Dichotomous outcomes (e.g. judging likely or unlikely that inflation will go back close to target in two years): with an effect size of 5 p.p.between two independent groups, say from 20 to 25%, effect size: [(0.2*0.8+0.25*0.75)/sqrt(0.05^2)]*7.9 = 1100 per group.