LabourList’s popular vote forecasts: Trends towards May 6th

By Theo Grzegorczyk

Starting today, LabourList will be publishing daily polling analysis to help show the broad changes in public opinion as we approach election day. It will employ a simple technique that has been used by others before, and requires only a quick analysis of readily available data. But what it will hopefully add to the conversation is a simple and visual way of representing the trends that underpin the day-to-day changes in polling.

All too often in Britain, our political culture gets caught up in the daily scuffles and the minor spats in Westminster, while failing to notice the larger trends at work in the country. This is as true for raw numbers as it is for policies and political campaigns.

As the election gets closer, the problem is getting worse. There tends to be a round of commentary every night at about 10pm telling us that Labour has “plummeted” or that the Tories have “reversed the trend”. There are “nosedives” and “bounces” in the course of a week. We hear about two-point leads and twelve-point leads in the course of the same day.

If you look at the polling data since January, it’s possible to substantiate this schizophrenic narrative with visuals. You could look at it like this:

Or, then again, we could be tongue-in-cheek and look at it like this:

Neither version really reflects what is going on. Anthony Wells at UK Polling Report has for a long time helped to resolve this daily volatility by giving a rolling average of the most recent numbers, a “poll of polls”. ConservativeHome has done the same. But this doesn’t pick up on the general direction of the numbers: it just irons away the outliers. One way to demonstrate “direction” would be to apply a linear trend line to the numbers. It would look something like this:

The problem here is that the lines never curve. They won’t bend in one direction and then back in the other. Intuitively, we know this is wrong. Public opinion doesn’t change overnight on the back of a televised debate or a media firestorm – but equally, public opinion isn’t static and linear. It is the gradual culmination of events which slowly bend the opinion of the public in one direction or the other.

To reflect this, you need to take a Polynomial Trend. This has the benefit of both dampening the impact of outliers, but also of taking stock of things like acceleration and deceleration. A polynomial trend line will spot the point at which a series of ‘outliers’ becomes a real ‘trend’. And, above all else, it can be used to sensibly forecast the trajectory of the trend, all the way to May 6th. So it’s these graphs that LabourList will be publishing over the next few weeks. Here is the first:

These trend lines project a final national vote share on May 6th of 39% to 30%. Those numbers will change as the campaign progresses – but what is important to note is how the general direction of travel can help to identify the “state of the race”. From these trends, it looks as though we will enter the short campaign tomorrow with the Conservatives ending – and beginning to slightly reverse – the downward trend that they had seen since the beginning of the year, while Labour enter the race looking as though they are losing the ground that they had gained over the course of January and February.

With thanks to Peter Kellner for his kind advice on this project.