Skip to main content

Weighting the BES Post-Election Random Probability Survey

The British Election Study Team

The 2019 BES Post-Election Random Probability Survey uses gold-standard random probability methods. Still, it is important to ensure that it is representative of the target population. In this case, we take as our target all adults aged 18 or over who live in Great Britain and who are also eligible to vote.

We provide three corrections to our data to make them representative. First, we apply a set of design weights. These help to account for respondents having unequal selection probabilities. Second, we apply a set of demographic weights. These account for different groups in the population having different levels of response. Third, we apply a set of results weights. These account for voters and non-voters being more or less likely to take part in the survey itself and any remaining imbalances in levels party support.

The turnout weights are especially important if we wish to infer who did and did not vote. Questions like this have obvious social and political value. Turnout is not just a vital statistic in its own right. A high turnout also implies a vibrant civil society and an engaged electorate. Furthermore, encouraging people to take part in elections is much easier if we know who to target in the first place.

Unfortunately, many countries mismeasure aggregate turnout. This is true of Britain, making weighting to turnout much more difficult. In its simplest form, turnout is the number of votes counted divided by the size of the population. Yet there are many different ways to define what counts as “the population”.

Most often, we want to measure the population in one of two ways:

  1. Registered voters (RV): those people registered to vote on the day of the election
  2. Voting eligible population (VEP): those people permitted by law to vote on the day of the election

Yet most studies tend not to use either definition. Instead, they use the total number of register entries (RE). The logic behind using RE turnout is not always clear. Even so, it is almost always treated either as a proxy for VEP turnout or as a measure of RV turnout.

The problem is that the number of register entries is not likely to equal either the eligible population or the number of unique people registered to vote. For example, some entries might be inaccurate, duplicated, or outdated; some voters (e.g. students) might be registered at more than one address; those who are eligible but not registered to vote will not be included on the register; and eligible voters who happen to live overseas might not be included at all.

To overcome this problem, we estimate Britain’s RV turnout since 1979 and VEP turnout since 2001. Our estimates show that official turnout estimates underestimate RV turnout by up to 22 percentage points. Official turnout is, however, similar to VEP turnout, overestimating it by up to 9 percentage points and underestimating it by up to 1 percentage point. We use these new estimates of 2019 VEP turnout to produce our survey weights. This ensures that our target population (the voting eligible population) also matches the population that we use to measure turnout. Note that our estimates rely on data from the Office for National Statistics (ONS). Similar VEP estimates can also be calculated using data from the Electoral Commission (EC). If we were to use the EC data to estimate VEP turnout, it would imply an aggregate turnout lower than the ONS data that we use here. Still, the ordering of elections in terms of turnout would remain the same and any gaps between them would also be similar.

The estimate that we use suggests that VEP turnout in 2019 was 66.2%. This is 1.7 percentage points lower than the official RE turnout estimate of 67.9%. We used an earlier version of these estimates for the 2015 and 2017 BES Face-to-Face Surveys. We weight the 2015 data to a VEP turnout estimate of 65.1%. Our most recent estimate now puts this figure at 64.8%. Likewise, we weight the 2017 data to a VEP turnout estimate of 68.2%. Again, our most recent estimate now puts this figure at 67.3%. As these differences are only very small, we remain confident that our 2015 and 2017 weights remain reliable.

Version 1.0.0 of the 2019 BES Post-Election Random Probability Survey contains 7 weighting variables:

  • wt_sel_wt: Selection weight including capping (all modes)
  • wt_demog: VEP demographic weight (all modes)
  • wt_vote: VEP self-reported vote weight (all modes)
  • wt_demog_f2f: VEP demographic weight (face-to-face only)
  • wt_vote_f2f: VEP Self-reported vote weight (face-to-face only)
  • wt_demog_cses: VEP demographic weight (CSES only)
  • wt_vote_cses: VEP self-reported vote weight (CSES only)

We recommend using the “wt_vote” variables for most analyses. These variables include both the combined demographic and result weights. Or, more specifically, the capped selection plus capped demographic weights targeted to the voting eligible population and weighted to vote choice and VEP turnout in Britain in 2019. This weight is most useful when answering political questions as it accounts for selection bias and biases that may lead some groups to be over-represented in the raw, unweighted data.

The “wt_sel_wt” and “wt_demog” weighting variables should be used only in certain circumstances. “wt_sel_wt” includes the selection weights (including capping). These are useful only in very specific cases. As such, you should not use them to weight the data for most common analyses. “wt_demog” includes the demographic and selection weights. That is, the capped selection weights plus uncapped demographic weights targeted to the voting eligible population. Thus, it does not account for vote choice or VEP turnout in 2019. Again, you should not use them to weight the data for most common analyses. They may, however, be useful in comparing recent BES data to historic BES data that did not include result weights.