Appendix A - About the Autumn 2022 MRP

This note has been provided by Focaldata:

Focaldata is a polling company that specialises in using machine learning to generate more accurate results. In particular, we use a technique called Multilevel Regression with Post-stratification (MRP) to map opinion poll data onto smaller geographic areas.


Data was collected from a representative sample of 10,010 adults (18+) living in Britain between 20-26 October 2022. About 75% of the fieldwork was completed before Liz Truss resigned. In response to this, Focaldata ran a top-up poll of 2,000 respondents between 28-30 October. The modelling uses data across all 12,000 respondents. Users filled out the surveys in real-time across mobile, desktop, and tablet devices on the Focaldata platform.

These respondents are sourced from different panels, which we then apply additional quality checks to in order to ensure the responses are of the highest quality. MRP weights these respondents to be representative of both national and sub-national (e.g. constituency) demographics.


Inputs: The MRP model uses a range of individual and constituency level variables, these include (but are not limited to) age, gender, education, votes at previous elections (2019 General and EU referendum), population density, % long term unemployed, % leave 2016, GE2019 vote share, Deprivation index, MP Incumbency_2018. We include several interactions to the above as well.

All data is sourced from the Office for National Statistics (Annual Population Survey and Census) where possible, plus the Electoral Commission for election data,  and estimated by Focaldata otherwise. 

Accuracy and uncertainty

There are sources of uncertainty throughout the MRP model, with uncertainty highest at the most granular predictions (e.g. constituency). A typical margin of error at a national level is ±4%, and at constituency level is ±7%. These estimates are based on historical performance, namely the 2019 general election.

Due to the way MRP works, there are higher levels of uncertainty where there are strong local factors, for example if there is a strong local candidate standing for election. MRP draws inferences from similar voters and constituencies, but there are too few observations at individual constituencies to pick up these effects well.

British Polling Council

Focaldata is a member of the British Polling Council and abides by its rules.