🚨 Long thread incoming!
Why I believe the polls are (again) overstating the Conservative lead over Labour.
Polls weight their data to a variety of factors so the poll sample broadly matches the national population, such as age, education, 2017 vote and 2016 EU referendum vote.
Why I believe the polls are (again) overstating the Conservative lead over Labour.
Polls weight their data to a variety of factors so the poll sample broadly matches the national population, such as age, education, 2017 vote and 2016 EU referendum vote.
It is the latter demographic on which I want to focus. YouGov, for example, weight their samples to a target % of 38.3% Remain voters, 41.4% Leave voters and 20.3% did not vote. To begin with, this poses a problem given that it assumes a 2016 turnout of ~80%...
when estimates of eligible voter turnout indicate a turnout of around 70–75%. This may be YouGov adjusting for turnout to take account of the fact 2016 non-voters are less likely to vote in future, but this is not my issue.
The problem, I believe, lies in the fact YouGov are weighting to the 2016 result and not accounting for changes in the electorate since then. In the past, when voter choice was arguably less correlated with age, weighting to past election results would not cause...
much of a problem, as the % of the population who voted a certain way would remain relatively stable over a number of years. However, in recent elections, age has been a major factor in determining voter choice, including in the EU referendum.
We also know that age is highly correlated with mortality rate. Therefore, would it be correct to say that EU referendum vote is highly correlated with mortality rate? In my work for 326 Politics, I have analysed and utilised a number of large datasets including...
population estimates by age, mortality rates by age, the International Passenger Survey (to determine migration levels by nationality) and census data on voter nationalities (to determine eligibility to vote).
In doing so, I have tracked the way the VEP (voter eligible population) has changed since the referendum. In 2016, I estimate that 33.5% of the VEP voted Remain, 36.5% voted Leave and 30.0% did not vote. DNV percentage could be slightly lower given academic studies on turnout.
However, what I am more interested in is the change in VEP between 2016 and now. Taking into account mortality by age and incoming and outgoing migration, I estimate that in mid-2019 (three years after the referendum), the VEP figures had changed to 32.6% Remain (⬇️ 0.9),...
35.1% Leave (⬇️ 1.4) and 32.3% DNV (⬆️ 2.3). This may not seem like much of a shift, but an increase in 2016 non-voters in the 2019 electorate excluded from polling samples means polling results are likely to be less accurate. If I were to weight 326 Politics' polling averages...
by VEP in 2016 (taking into account turnout of Remain, Leave and non-voters), we would see a predicted national voting intention of 40.2% for the Conservatives and 29.2% for Labour, a lead of 11.0% (almost identical to what @britainelects poll tracker indicates).
However, adjusting for VEP in 2019 with fewer 2016 Leave and Remain voters, we get a national voting intention of 39.3% for the Conservatives and 29.9% for Labour — a lead of 9.4%, 1.6 pts lower than the polls currently suggest. Now this may not make much of a difference...
to the overall result — both of these leads would indicate a Conservative majority, but if the polls get tighter, that 1.6% change in margin could be the difference between a Conservative majority and a hung parliament, which the polls (mostly) failed to predict in 2017. (13/13)
To add: all this assumes the same turnout by 2016 referendum vote as the 2017 election, when my analysis suggests Leave turnout is down slightly, and turnout among 2016 non-voters is up around 8% (much of this is caused by those too young to vote in 2016).
