Mark Pack on his polling archive database

January 5th, 2016


Way back in the 1980s, one day I was watching David Owen being interviewed on television, responding to probing questions about the then Alliance’s opinion polls ratings by saying there was a seasonal pattern in third party support in the polls.

Frustration with the lack of detail in his answers or knowledge in the interviewer’s follow-up questions turned into a fruitless search for sources for answering the question afterwards. Which then led to the birth of my collection of opinion poll data, which with the kind help of many others who have shared their information is now the largest set of national voting intention polls for Great Britain.

Starting in 1943, with a passing reference back to February 1939, the latest update just published takes the data up until the end of December 2015.

I now know there is tentative evidence that David Owen was right, at least about that seasonal trend, though the polling disaster which was May 2015 may well make reasonable people doubt quite what the value of the overall data set is.

The main answer to that is that even in the polling disasters of May 2015, April 1992 and June 1970 the polls were wrong by a handful of percentage points. In a first past the post, two-and-a-bit party system, that margin of error is enough to produce widespread egg on face for pollsters. But it does also mean that even at their worst, polls have been accurate enough to distinguish between, for example, parties languishing in the low thirties or soaring up to the upper forties.

Especially as there is a pattern to the errors. When the polls get it wrong, they usually over-estimate Labour support, and when they get it badly wrong, they’ve always over-estimated Labour’s support.

Knowing that pattern and the level of accuracy with which polling can reasonably be interpreted, there is therefore still plenty of knowledge to be gleaned from voting intention figures.

The polls in the last quarter of 2015 put Labour on average at around 31%. They aren’t accurate enough to say for sure that Labour’s support isn’t really a little higher, and past errors might make you lean to thinking the truth may rest a little lower. But there are enough grounds for confidence that the party’s current support is well short of election winning territory.

Likewise, for the Liberal Democrats – whose support has often varied noticeably over the decades between different polling methodologies, especially back in the days of regular Gallup polling – it is pretty clear the party’s support has not (yet?) recovered substantially from the general election. Perhaps it is a little up, perhaps not, but we can be confident that if there was a general election tomorrow, Tim Farron’s party would not be up there in the 20s.

That is why such a large historical data set is also useful as it allow comparisons with the past. History is not a sure guide to the future, but it does give clues about regular patterns to watch out, such as that Labour’s current poll ratings are well down on where they have been after previous changes of leader.

Knowing just how much history would have to be bucked for Labour to do well from here is useful, even if it does not tell us for sure that it cannot be. And would-be knowledge is at its most reliable when based on data.

Now YouGov – please, no return to daily polling which requires collating…

Mark Pack

Footnote: The full database of national voting intention polls for Great Britain since 1943 is available here.