What's Happening to the Polls?

Yes, the polls are volatile. All sides are making claims about daily polls, but that is simply optimism. Polls are best viewed over time as trends. The best place I have found that shows the trends is the UK Election Trends blog, This site takes the daily polls and draws a trend line that is calculated as a running average. Here's their description:


Trend graphing will be generated from a moving average taken across the last 100 polls, eliminating polls from the same pollster taken in swift succession down to the latest poll. (ie, when two YouGov polls are issued within 24 hours, the latest one will be taken.) The trends will be taken as exponentials, and the moving average is an area of six polls. Trending data will then also be generated against the vote lead, and an additional short term lead trend will be taken against the last twenty polls.

The first point to make is that this uses a running average rather than a simple mean average. The reason for using a running average is to smooth out the noise and give an indication about the underlying trends. Running averages are not perfect, a big peak or dip within the sample can still have an effect, and potentially a value for today could be affected by a peak or dip 100 days ago, which is nonsense. I would prefer something equivalent to a low bandpass filter used in audio processing, that is, the higher frequency noise is filtered out to show the much lower frequency trend. (I have spent 6 years in spectroscopy so I am used to seeing trends in noisy data, and in my experience a bandpass filter is always better than a running average. After the election I will develop my own model and see if I can more accurately extract the trends.)

The polls have tightened considerably in the last few months, but the election trends help us to see what form this tightening takes and identify the potential cause. First, have a look at the polls for Labour and Conservatives for the last seven months:

There is a clear decline in Conservative support and a clear increase in labour support, The Conservative decline appears to be three-fold, identified by three green lines: Oct-Nov, Dec-Mar and April. The first appears to be a party conference decline, that is, as the public digested the policies that were announced at the Conservative Party Conference their support declined. There is a hiatus for two weeks in December and then from mid-December to the end of March there is a decline. This last decline I call the "policy decline" because it corresponds to the publishing of more details of the Conservative policies, in particular the "draft manifesto" from the beginning of the year. For two weeks from the last week of March to the first week of April there is an increase, then something happens. There is a clear decline from the end of the first week and that decline is continuing right now. This last decline I call the "manifesto-debate" decline.

For Labour there is a clear increase until the beginning of March, the gradient of this increase is equivalent to the Conservative "policy decline" so it could be attributed to a switch from Conservative to Labour due to Conservative policies. Then the Labour rise stops and turns into a decline for a month. The start of this decline corresponds to Brown's appearance at the Chilcott Enquiry. I call this the "Chilcott decline". There is an increase in support for the first week of April and then a debate decline. There is no Labour manifesto decline.

The following shows the last three months month in more detail.

The Chilcott Enquiry is marked with a C, and there is a clear decline in Labour support after this date. From the beginning of April until the first leaders' debate (#1) Labour supports starts to increase. This increase is due to the hype for the manifesto, but it continues after the Labour manifesto launch. The leaders' debate is clearly the cause of Labour's decline in the latter half of April ("debate decline").

The Conservative support shows the "policy decline" until the last week of March and then there is a rise to the end of the first week of April, again this is hype for the manifesto launch. In fact, the peak in support occurs at the end of the first week which appears to coincide with the National Youth Service announcement (BS = Big Society) and the announcement of the celebrity endorsement from Sir Michael Caine. After the BS point there is a decline (the "manifesto decline") until the leaders' debate when the Conservatives support declines further ("debate decline").

It is interesting that for the first  week or so of the "debate decline" the rate is the same for Conservatives and Labour.

My conclusion is that the Conservatives are losing on policies and that the "Big Society" is a massive turn-off for candidates and the public alike. Labour's support has increased in line with the Conservative decline in general, most likely because the public is neutral to their policies and are switching support from the Conservatives. If it had not been for the leader's debate then the Labour support would be steadily increasing and the Conservative support steadily declining. The leaders' debate caused declines in both party's support equally. This means that for over a week the gap between Labour and Conservatives have remained the same at about 5%.

However, I think more worryingly for Labour is the "Chilcott-decline". This is entirely self inflicted. Brown did not have to have the enquiry before the election and he did not have to give evidence before the election. If this month of decline had not happened then Labour's support would now be much stronger and even with the debate decline, the gap between Conservatives and Labour would be less (perhaps down to 3%) than they are now. If the Conservatives get a majority then I think it will be due to the "Chilcott decline". That enquiry was a tactical mistake.