Why the Observer’s opinion polling is wrong

Posted on

This comment was posted by a regular contributor who uses the name DunGroanin' in response to my discussion of the Observer's reported 19% opinion poll lead for the Tories. I tho0ught it worth sharing as it offers what seems like an entirely plausible explanation as to why the lead may be overstated:

I spent a lot of time last election blowing major holes in their balls polls, especially the commissioned ones by Opinium for the Obsessive Groaniad. Also one of the reasons why I done with them.

I called it pretty correctly. By looking at their RAW data. And I say the current tory lead is exactly what it was at the end of 2017. If you believe their raw data.

This time I hadn’t bothered up till now. Drilling down into data is the key. Even as the representation of the data has changed… in my opinion they have manipulated raw data!

The Obsessives splash yesterday that claims a headline of
‘Conservatives open up 19-point lead with 47% share of the vote’

Now understand that there is a catch up period by Labour as the election campaign proceeds and fair coverage in the media allows that. Last time that ended with the overall lead of 800,000 votes over the whole 650 seats by the tories.

‘Despite a drop in the number of seats, the Conservatives actually saw a rise in vote share (up 5.5% points to 42.4%). Labour’s vote share rose at an even greater rate (up 9.5% points) and now sits at 40.0%.’



Remember that percentage — 2.4%.

Now lets drill down into the Opinium poll.

1. They state ‘Source: Opinium poll: 2,003 UK adults surveyed from 20 to 22 November 2019’ — that is 3 days over which the two thousand and three people were interviewed. Things were happening over these 3 days and doesn’t include the Friday debate.
2. In their new presentation of the data — https://www.opinium.co.uk/wp-content/uploads/2019/11/VI-20-11-19-website.xlsx There are now multiple data sheets. The relevant sheet is 5a. It asks what the responders voted in 2017, of the number that are used to derive the headline ‘result’.
3. The actual number of responders who voted in 2017 is a total of 1,368.
4. Off these 46% (508) voted Tory and 36% (421) Labour in 2017 — a difference of 10% not the 2.4% of the actual result remember!
5. The actual 2017 Tory share was 42.4% not 46% as of the responders. The actual Labour share was 40% not 36% of the responders. Self selected responders, many who couldn’t even ‘remember’ how they voted 2 years ago btw (ahem!).
So the Tory share of responders is over represented by 7.6% (4)
6. Then let us look at sheet 3 & 3b which mutates the numbers even further. The first has tory 35% (558) , Lab 23% (372). The second has tory 37% (585) , 24% Lab (379) — both have a total respondents used as 1592.

So the magic trick is pulled! How does my point 6 compare to my point 4?

I put all that up there not expecting most to follow it but to show the ‘data’. It can be ignored and you can just take my opinion below or their opium from Opinium (a firm that should hang it’s head in shame — i believe).
So how many ways to skin this very thin cat?

The raw data adjusted for the 7.6% starting variance would give the Tories a current lead (to the 3 day period ending 22/11) of between 4.4% and 5.4%.

With undecideds of some 16%.

I am sticking to my prediction of a landslide based on how the numbers improved through the campaign last time.

Anyway the 19% HEADLINE current lead is balls!

Do we really need to wonder why such magical thinking is going on in the Obsessive Groaniad?