It’s very rare indeed for any polling company to admit to making an error that has skewed a published poll. So rare indeed that we might speculate that whenever a polling company itself discovered an error in the past, the company concerned kept quiet and hoped that no-one else had noticed other than to assume that any unusually rogue result was down to sampling error or the like. Better to keep quiet than to suffer reputational damage by admitting an error, so the polling company might reason.
However, in the case of the latest Redfield and Wilton Scottish poll for a UK general election, it might be harder to brazen out what is an apparent error.
At face value, the headline result appears reasonably plausible. SNP 39%, Lab 29%, Con 22%, LD 6%, with fieldwork from 2nd to 5th March. An SNP lead of 10%, fairly unremarkable and consistent with the generally reduced leads that we have seen since February. The SNP must be relieved that their vote share has not fallen further in the midst of a less than convivial leadership election, while Labour will be a tad disappointed. The Conservatives will be buoyed by their apparent recovery to just 3% below 2019 GE levels. But at face value there’s nothing to ring alarm bells.
Dig deeper into the data tables though, and something does seem amiss. The weighted result, before the final turnout filter is applied, has the SNP and Labour level pegging on 30% each, together with 14% of don’t knows and non-voters. Stripping the DKs and DVs out, the voting shares before turnout weighting are 35% SNP, 35% Labour, 19% Con, 6% LD.
Remarkably, the R&W turnout adjustment to the March 2023 poll has turned level pegging into a 10% SNP lead over Labour in the headline result. That’s even more remarkable because the turnout adjustment in the previous November 2022 R&W poll left the SNP lead over Labour unchanged at 10%.
So something doesn’t smell right. And when you look at the detail of R&W’s turnout adjustment, one thing stands out. 34% of those who voted Labour in the 2019 GE are recorded as “probably” or “definitely” not voting in a GE now, compared to just 1% of 2019 Labour GE voters in the November poll. By contrast, the proportion of 2019 SNP, Conservative and LD voters who would probably or definitely not vote now stands at 11%, 10% and 8% respectively.
In the R&W turnout adjustment, the enormous differential between Labour and the other parties is not plausible, nor is the scale of change since the previous R&W poll.
The data is suggesting to me that R&W must have made some sort of error in processing their polling data, something like of a transposed formula in a spreadsheet or a misplaced decimal point perhaps. If not, then R&W need to come up with a plausible explanation for the apparently implausible. And in the meantime, in the absence of such an explanation, it’s reasonable to conclude that Labour may well have drawn level with the SNP in general election polling in Scotland,