Everyday News Update

UK election 2015: The Guardian poll projection methodology – The Guardian

No model is perfect, and of course the election is three months away. What the Guardian’s model does is take the average of the polls as they stand today and suggest how that might translate into seats. It’s one projection among many but it should reflect how the changing polls between now and election day might translate into change in the House of Commons.

It’s important to note this is not a prediction of what will happen on voting day. Instead, it’s a model of how parliament might look if the nation votes exactly like the average of the latest polls. It does that by looking at national polls, sub-UK data, and constituency polling, plus some historical context.

The UK votes on 7 May in what is expected to be the most uncertain election in recent memory. Over the course of the past five years Britain’s electorate has become increasingly fragmented.

This has led to a shift from a two-and-a-half-party system to one comprising six parties. Modelling what might happen on a playing field of six parties is extremely complicated, and a world away from the days of the old “swingometer”. Anybody who says they can apply the patterns of the past to today is being speculative.

May’s election is also the first time in generations that British voters expect a coalition, although that was the unexpected outcome of the 2010 election. How this will influence behaviour such as tactical voting is impossible to anticipate.

The Guardian’s average of polls aims to best represent a snapshot of the polls as they stand. The model aims to translate what polling figures would mean in terms of seats if an election were to take place today (and, of course, if the polls were right).

Here’s how the model works.

To reach a projection of seats we take a three-step approach to the many polls that are released on an almost daily basis.

1. Constituency-level polling

Firstly, we look at constituency polls. These are the ideal source of data, but there are several limitations with the data at this local level. There aren’t many companies polling constituencies, because it’s so expensive and surveys aren’t carried out regularly. Because of this, where there is limited and relatively old data, the figures are adjusted based on the change in the national average of polls since the constituency poll was carried out. For example if we have a poll for Pendle, but it was conducted in October 2014, we continue to use that poll as a starting point, but impose the swing in this seat which we register in our rolling average of polls since that date.

This approach is obviously only applied to project seats where constituency polls are available.

2. Polling in the nations and English regions

Secondly, we look at Scottish, Welsh or other polls conducted in the nations or English regions. For those seats where we have both constituency and sub-UK polls, the former is adjusted and bought up to date in light of an average of the latter.

These polls therefore serve two purposes: to both project seats where there isn’t constituency-level polling, and to adjust, where applicable, constituency polls.

3. National-level polling

Finally, for all remaining seats we use an average of polls based on about 20 days of figures. Because different firms publish results with varying frequency, for those pollsters that release figures more regularly we apply an average over the course of the analysed period. This means that no single pollster is given a greater weight compared to others.

We then impose the change in party support in each seat that we recorded in the relevant tier of data – constituency if it’s available, sub-UK if that option is there instead, or UK-wide polling – for that constituency.

We have considered additional adjustments and different transition models, which look at proportional rather than absolute swings, but felt that these wouldn’t provide a noticeable systematic advantage to the approach we have adopted, while instead adding considerable complexity and clutter to our model.

We only use polls carried out by Lord Ashcroft and by members of the British Polling Council.

Ours is just one model for analysing the polls. There are many other different and valid approaches, which you can check out here.

UK election 2015: The Guardian poll projection methodology – The Guardian

Leave a Reply

Featured Links

    Search Archive

    Search by Date
    Search by Category
    Search with Google

    Photo Gallery

    @2012 Designed By Quick News Update