h1

Indian pollster wins the 2010 polling race

May 9th, 2010
Pollster CON LAB LD Error
RNB India: phone 37 (+0.1) 28 (-1.7) 26 (+2.4) 4.2
ICM phone/past vote weighted 36 (-0.9) 28 (-1.7) 26 (+2.4) 5
Populus: phone/past vote weighted 37 (+0.1) 28 (-1.7) 27 (+3.4) 5.2
Ipsos-MORI: phone 36 (-0.9) 29 (-0.7) 27 (+3.4) 5
Harris: Online 35 (-1.9) 29 (-0.7) 27 (+3.4) 6
ComRes: phone/past vote weighted 37 (+0.1) 28 (-1.7) 28 (+4.4) 6.2
Opinium: online 35 (-1.9) 27 (-2.7) 26 (+2.4) 7
YouGov: online 35 (-1.9) 28 (-1.7) 28 (+4.4) 8
Angus Reid: online 36 (-0.9) 24 (-5.7) 29 (+5.4) 12
BPIX: online 34 (-2.9) 27 (-2.7) 30 (+6.4) 12
TNS-BMRB: face to face 33 (-3.9) 27 (-2.7) 29 (-5.4) 12
OnePoll: online 30 (-6.9) 21 (+8.7) 32 (+8.4) 24
Actual GB share 36.9 29.7 23.6 -

What does this say about online versus phone polling

I’ve been trying to find a slot for this post since Friday but the ongoing post-election developments have dominated just about everything.

Before we move on too far from the election itself I want to publish, as a matter of record, the PB polling accuracy table for 2010 based on the aggregate differences for the main parties between the firms’ final polls and what actually happened.

Clearly the big upset was the size of the Lib Dem vote and how none of the final polls picked up the scale of the change.

The big picture, though, is that the telephone firms seems to have come out better than the online ones and I’m sure that we will hear a lot more about that in the future.

They seem to have picked up the decline in the Lib Dem share much more than the others. YouGov must feel particularly annoyed. On the Tuesday evening before the election it reported a four point drop in the LD share down to 24%. By Wednesday evening this had moved back to 28%.

Angus Reid, who polled for PB as well as the Sunday Express and the Economist, had a particularly disappointing outcome which does raise questions over its weighting approach. Controversially they past voted weighted without taking into account an element for false recall – something that kept its Labour shares much smaller than the others.

Congratulations to RNB of India which only produced two polls during the campaign and came top.

RIP – the “Golden Polling Rule”. For the first time since 1983 the most accurate polls were NOT those with Labour in the least favourable position.

There will be a separate betting aftermath in the next day or two.

Mike Smithson