Methodology
A poll is an estimate, and an estimate is only as good as the way it was made. This is how we make ours, and what we tell you about it every time.
How a poll gets made
Start with registered voters, not all adults
Public polls draw from the state voter file with cell numbers appended. Results describe registered voters, not all adults, and every report says so.
Draw a sample sized to be honest
Typical samples are 400 people (±4.9 points) or 600 (±4.0). Each person gets a text with a personal link, plus at most one reminder, and chooses whether to answer.
Weight to the population
People choose whether to answer, so the raw sample never matches the population. We weight to the voter file on age, sex, party registration and geography, and publish weighted and unweighted counts and the design effect.
Report the range, and everything else
Every result carries its margin of error. Every release lists sponsor, field dates, mode, frame, weighting variables, margins for each group, and every question word for word, in order.
Why the number of people matters
The margin of error shrinks with the square root of the number of people who answer. Going from 100 to 400 respondents cuts it in half. Going from 400 to 600 takes off less than a point.
Example: one simulated poll of 600 people whose true value is 58%. Margins are the ones we publish, at 95% confidence and assuming a 50/50 split. Weighting widens the margin a little, so we publish the weighted figure.
What every release includes
- Who paid for the poll and who conducted it.
- Field dates and mode (text to web).
- The frame and how the sample was drawn.
- Weighted and unweighted number of respondents, the weighting variables, and the design effect.
- The margin of error overall and for each reported group.
- Every question word for word, in the order asked.
- A link to the full topline.
What a poll can't tell you
- Results describe registered voters, not all adults.
- People choose whether to answer. Weighting corrects for what we can measure in the voter file, not for what we can't.
- Only people with a cell number on the voter file can be reached by text.
- We publish no result for a group smaller than 80 respondents, and none for an area the sample is too small to describe.
- A gap smaller than the margin of error is called too close to call.
Questions about our method are welcome: hello@aequida.com.