For our most recent MRP study Focaldata collected data from 28,272 respondents between 4th of November and the 5th of December 2019 using a range of online panel providers, including Opinium, BMG, and Focaldata’s own panel. Focaldata uses data from a range of suppliers to adjust for the ‘house effects’ between data providers. Focaldata's MRP model uses a range of individual and constituency level variables, these include (but are not limited to) age, gender, education, VoteGE2017, VoteEURef, population density, % long term unemployed, % leave 2016, GE2017 vote share, Deprivation index and EU parliament 2019 vote share. They use a bayesian exploded logit model, which is fit using Hamiltonian Monte Carlo with the open-source software Stan. The models are trained on Google Cloud Platform.