Analysis
Why some polling could be failing to pick up the Jewish shift against Labour
Data scientist Dr Abigail Lebrecht explores how these extraordinary time for the community affects how it will vote
November 21, 2019 12:24
2 min read
Data scientist Dr Abigail Lebrecht explores how these extraordinary time for the community affects how it will vote
November 21, 2019 12:24We like to say ‘Two Jews, three opinions’, so polling is a delicate subject in our community. In 2019 this is highly pertinent — are pollsters representing Jews?
In pre-Corbyn times, this would not have mattered, with the Jewish vote reflecting national trends, but all has changed in these unprecedented times.
“The chances of a Labour majority are as close to zero as it is possible to be,” said political scientist Professor Sir John Curtice last week. Beyond this view, pollsters are reluctant to make firm predictions about this election.
So which polls should we believe and to what extent are they identifying swings in Jewish constituencies against Corbyn’s Labour?
Traditionally constituencies were predicted using local polling. These are expensive and error prone so they cannot predict tight results but they are good at identifying local trends.
For this election, Finchley and Golders Green has been polled twice. In both polls Labour is out of contention in third place, with 21 per cent of the vote from Survation in October and 19 per cent by DeltaPoll in early November.
Recent single-constituency projection has used a new statistical technique, MRP (Multi-Level Regression and Poststratification). Using MRP, YouGov predicted the 2017 hung Parliament. MRP supposes that voters behave more according to their demographics than location.
YouGov’s MRP success, correctly predicting 93 per cent of seats, led to its adoption by tactical voting initiatives Best for Britain and Remain United. However, MRP models are difficult to build and can go wrong.
Lord Ashcroft’s 2017 MRP predicted a large Conservative majority. Factors that make or break MRP predictions include the polling data being reliable and large enough (YouGov asked 50,000 people) and the correct demographics being chosen across the UK population.
Two constituencies mispredicted by YouGov as Labour gains were Hendon and Finchley and Golders Green — apparently ‘Jewish’ was not a chosen demographic.
This isn’t surprising. Jews make up half a per cent of the UK population. The size and dispersion of the community makes it difficult for national polling to represent our views. Remain United (with a poll of just 6,000 people across the UK) and Best for Britain, with 46,000, will not have balanced their models for the Jewish vote.
Their current predictions for the Labour vote in Finchley and Golders Green are higher than both recent constituency polls.
Why use MRP then? The British Election Study suggests this is the most volatile British electorate in modern times. We can not rely on 2017 results to predict 2019 and require statistical techniques to fill in the gaps. However without “Jewish” as a factor, MRP will inflate the Labour vote in Jewish areas. In recent general and local elections, Barnet cast fewer Labour votes than every polling prediction.
Since then, Barnet councillors, canvassers, members and two Jewish MPs have quit Labour. The EHRC is investigating the party for antisemitism.
On December 13 we will see the full extent to which the Jewish community and its neighbours reject Labour. On present evidence MRP-based polls will fail to predict this and must be treated with caution in Jewish areas.
Dr Abigail Lebrecht is a Data Scientist who advised Best for Britain on their tactical voting initiative in 2017
To get more from opinion, click here to sign up for our free Editor's Picks newsletter.