We use cookies to ensure that we give you the best experience on our website.If you continue without changing your settings, we'll assume that you are happy to receive all cookies on the Ipsos MORI website. However, if you would like to , you can change your cookie settings at any time.

General Election 2001 : Constituency Polls — How Not To Do It

Date:17 May 2001
Category:Comment & Analysis
Specialism:Social Research
Keywords:2001 British general election, Scotland
(Click on keywords to find related News Items)

Today's Daily Record carries polling results from six marginal seats in Scotland, and compares the results with those that it published in the same six marginal seats last week; both polls were conducted by Scottish Opinion. All very well, except that the first poll interviewed only 744 respondents in total (an average of 124 per constituency), and today's interviewed 911 (average about 152).

Such tiny sample sizes mean that the margin of error is far too large for the findings in the individual constituencies to be of any practical use. Nor has the presentation of the figures been ideal: percentages have been calculated including don't knows (normal polling practice in Britain is to exclude don't knows, allowing direct comparison with actual election results, as shown in the first section of the table below [see How Does The Voting Intention Question Work? from an earlier commentary, which describes how the figures are calculated], and figures given to one decimal place. (Use of decimals implies an entirely spurious level of accuracy: it is plainly ludicrous to report figures to the nearest tenth of a percentage point when the margin of error is eight points!) And why on earth is the column presenting findings from a poll on 10-14 May, published on 17 May, labelled as '15/05'? In the table below, the figures as given in the Record have been recalculated to exclude don't knows, and are given to the nearest whole number.

The low value that should be put on such findings is evident when we come to look at the differences in results between the two polls (I hesitate to refer to it as "change"). In Eastwood, Labour's share has apparently risen by twelve points in a week. In Edinburgh Pentlands, where last week's poll put the Tories (whose candidate is the former Foreign Secretary, Sir Malcolm Rifkind) on 16%, half his share at the last election; but that has doubled in six days. In Stirling, meanwhile, there has apparently been a 22.5% swing from SNP to Labour. In Perth (not Perth and Kinross as the Record calls it - Kinross has been in the neighbouring Ochil constituency since 1997), the Lib Dems seem to have had an impressive week, more than quadrupling their support. Or not.

As the Record notes of Stirling, with massive understatement, "last week's SNP lead must have been a freak result, because they are now third". Well, no, not really a 'freak result'; exactly the sort of result that will inevitably occur far too frequently when polls are based on inadequate sample sizes.

  Ayr Eastwood Ed. Pent's Perth Stirling Tayside
Overall (approx)
  % % % % % % %
Result 1997
Con 34 34 32 29 33 36 33
Lab 48 40 43 25 47 11 36
LD 5 12 10 8 6 8 8
SNP 13 13 13 36 13 45 22
Other * 2 2 1 1 - 1
4-7 May
Con 25 25 16 25 26 16 22
Lab 47 52 56 25 31 19 38
LD 6 9 2 2 1 0 4
SNP 24 12 12 45 41 50 31
Other 0 2 0 1 1 1 1
10-14 May
Con 27 19 33 24 21 22 24
Lab 50 65 50 19 53 35 45
LD 5 12 8 9 6 2 7
SNP 17 14 14 36 18 38 23
Other 2 0 4 1 2 1 2
Con 2 -5 16 -1 -5 6  
Lab 2 12 -6 -6 22 16  
LD -1 3 5 6 4 2  
SNP -7 1 1 -9 -23 -12  
Other 2 -2 4 0 1 0  

Source: Recalculated from published figures of Scottish Opinion/Daily Record polls

Suppose that instead of deriving six constituency 'results' from each poll the Record had been content to present the overall findings from voters in six key marginals — as calculated approximately in the final column of the table. Last week's polls found the Conservatives in a hopeless position, having lost a third of their support in these key seats; Labour was a little stronger than in 1997 and the SNP much stronger, up nine points. Today's poll shows no significant change in the position for the Tories, but a 7.5% swing over the week from SNP to Labour. That is a conclusion that the sample sizes should be big enough to stand up; and isn't it a better story than constituency figures that nobody will take seriously, especially after even the Record had to admit that last week's most startling figure was 'a statistical blip'?

Add to My Ipsos MORI Bookmark & Share Print this page

Contact Us

Sir Robert Worcester
Sir Robert Worcester

Founder of MORI