tracker sample planner
Before you commission the next wave, find out how many respondents it takes to see a move that is real.
The score is the percent who say yes (aware, consider, would buy). Power is how often a real move of that size gets caught. Leave design effect at 1 if you do not know it. Scanning many metrics raises the bar, so a bigger sample is needed to keep the same power.
What each size of move costs to see
| Move to see | Respondents per wave | Cost per wave | Can you see it? |
|---|
Halving the move you want to see roughly quadruples the sample.
What each sample size can see
| Respondents per wave | Smallest move seen | Margin of error on one score |
|---|
How it works
The calculation
You compare two waves with the same number of respondents. The score moves from p1 to p2, which is p1 plus the move. With confidence z and power zb:
The smallest move a sample can see comes from solving the same equation for the move. Wave to wave noise is the thing being sized, so a metric measured on a small sample can swing several points with nothing changing.
Read this before you set the sample
This assumes a simple random sample. Quotas, panels and weighting add noise. The design effect box is a rough patch.
Sampling error is only one source of noise. A changed panel provider, a rewritten question or a different field month can move a score by more than this.
Score levels matter. Scores near 50% need the biggest samples. Scores near 0% or 100% need fewer, and the normal approximation gets shaky.