How we do our research

Data collection methods

Depending on the study, we draw on one or more of the following collection methods, chosen for the population and question at hand:

• CATI (phone interviews): structured telephone surveys for broad geographic reach
• CAPI (face-to-face interviews): in-person interviews for areas where phone penetration is lower or where nuance matters
• SMS / USSD polling: low-cost, high-reach digital polling for time-sensitive questions
• Social listening: automated monitoring of public conversation across X, TikTok, Facebook and Instagram for sentiment and trend signal
• Online panels: for consumer and brand studies where a representative online population is appropriate

Sampling and sample size

National surveys typically field a minimum of 2,000 respondents, stratified by region, age, gender and urban/rural split to reflect the population being studied. Regional or single-race studies (e.g. a county governor poll) are sized to the population of that specific race, with a target margin of error disclosed in every release.

We use multi-stage random sampling for national studies: first stratifying by region, then randomly selecting respondents within each stratum, weighting results back to known population proportions where necessary.

Quality control

Every study goes through the same checks before publication:

• Enumerator training and field supervision for CATI/CAPI studies
• Back-checks on a random subset of completed interviews
• Duplicate and bot detection on digital/online responses
• Data cleaning and outlier review before any figures are calculated
• Internal review of topline numbers against prior waves for consistency

How we calculate findings

Toplines are calculated as simple proportions of valid responses, weighted where the sample requires it to reflect the underlying population. Margin of error is calculated at a 95% confidence level and published alongside every topline figure. For tracking indices (like the Rapid Political Pulse), we show trend lines across waves rather than single snapshots, so readers can see direction and momentum, not just a single moment.

Disclosure, every time

Every published report includes: sample size, margin of error, fieldwork dates, geographic coverage, and who commissioned the study, if anyone. This is non-negotiable — it's what lets journalists, researchers and the public evaluate our findings rather than take them on faith.