Statistical Modelling and MRP
One national survey, an estimate for every area. Multilevel regression and post-stratification (MRP) combines survey responses with census data to produce reliable estimates for places and groups a conventional poll cannot reach on its own: every Westminster constituency, every council area, every customer segment.
What we model
- Constituency and local-authority estimates of opinion, behaviour or intention from a single national sample
- Small-area estimation for any geography you can map to the census: wards, health areas, catchments, regions of another country
- Segment-level estimates where a subgroup is too small to report directly
- Uncertainty for every estimate: credible intervals, not just point values, so you know which differences are real
- Dashboards and maps so the results can be explored, not just read
How it works, briefly
We interview a large, quota-controlled sample (typically 8,000 to 10,000 adults), fit a multilevel model of the outcome on individual and area characteristics, then project it on to a census-based frame of several hundred thousand population cells. Each respondent’s postcode places them in their area. We publish the method and tables for any public release under British Polling Council rules, and our MRP work is developed with Royal Holloway, University of London.
Recent published work
- Scottish Parliament election 2026: our MRP seat projection was the most accurate published, projecting 59 SNP seats against an actual 58, alongside the most accurate final poll
- HOPE not hate: 10,014 adults interviewed in six days, modelled to 631 constituencies with a nine-party model and parallel attitude models, delivered three days after fieldwork closed
- 38 Degrees: the annual Progressive Majority MRP, published with constituency-level results for MPs and campaigners
Beyond elections
The same method estimates rail use, service satisfaction, brand awareness or health behaviours area by area. If you need to know how something varies across 600 places and can only afford one survey, this is the tool.
Talk to us
Tell us what you need to know. We reply the same day, and you will have a price, or a short call booked to get the few facts we need to price it, within two working days: researchteam@survation.com or 020 3818 9661.
< Back
