This data set contains estimates of average public opinion on several topics for 632 British parliamentary constituencies. These estimates are derived from existing primary data sources, including national survey data, UK Census data and geodata on constituency boundaries. Information from these data sources is combined using 'multilevel regression and post-stratification' (MRP) techniques which are detailed in 'data collection method' section and the technical report accompanying the data. This project generated new estimates of public opinion in Westminster constituencies. We utilised multilevel regression and post-stratification (MRP) techniques for estimating public opinion in small areas, combining information from several existing data sources: ESRC-funded national surveys, Census data, and geographic information from the Ordinance Survey.

The data from this project is not primary data. Rather, it is made up of estimates of constituency opinion derived from a variety of primary data sources. The data contains estimates of average public opinion on a variety of issue topics for each of 632 British parliamentary constituencies (Northern Irish constituencies were excluded due to survey data unavailability). These estimates are generated using multilevel regression and post-stratification (MRP), a technique for estimating opinion within small areas using large national survey samples in conjunction with auxiliary information about the small area characteristics. Briefly, the MRP procedure works as follows. We start with a national survey sample which measures opinion on the topic of interest, as well as respondents’ demographic characteristics and constituency location. We then use a multilevel regression to model respondent opinion as a function of their individual demographic characteristics and the characteristics of the constituency in which they live. We include in this multilevel model a spatially structured constituency random effect (which allows for greater partial pooling of information across respondents whose constituencies are closer together geographically) and a non-spatially structured random effect (which allows for partial pooling of information across constituency sub-samples more broadly). We then take the resulting regression estimates and post-stratify to constituency population characteristics to obtain estimates of constituency opinion. For each opinion topic, our data contains a point estimate of constituency opinion and the lower and upper bounds of corresponding 95 per cent confidence intervals. Our survey data comes predominantly from the 2010 and 2015 British Election Studies. Our data on population characteristics comes from the UK Census. Our geodata on constituency boundaries comes from the Ordnance Survey Boundary-Line Data Service. For further technical details on the MRP specification and primary data that we use, please see the Technical Report which is included in the documentation accompanying this data set. For further information on the MRP method in general, see: Park, D., Gelman, A., and Bafumi, J. (2004). Bayesian multilevel estimation with post-stratification: State-level estimates from national polls. Political Analysis, 12(4):375-85.