POLLS AND THE ELECTION
SIR,—David Butler in his article last week on Gallup Poll and NOP introduces two tests of their com- petence which strike me, a practising pollster, as being irrelevant and incomplete.
Firstly, Mr. Butler compares the gap between the two major parties shown by the polls with the actual gap. What he does not say is that such a test almost inevitably multiplies by two any error by a poll in the distribution of votes between these parties. How does this multiplication occur? An example will show : if a poll over-estimates the Conservative lead by 2 per cent, it is highly probable that Labour has been underestimated by 2 per cent. Hence, instead of 2 per cent, the difference is now 4 per cent and the gap is affected accordingly. Mr. Butler's yardstick of the gap cannot be accepted, therefore, as a test of statistical accuracy. Prior to the general election the gap reported by the poll is, of course, of con- suming political interest.
Secondly, Mr. Butler suggested checking polls on the distribution of votes between the two major parties. By doing this, he ignores the Liberals' existence. And if one poll consistently under-reports the votes going to the Liberals, Butler's suggested test would prove to be discriminatory and unfair to the other poll. Again, it strikes me, Mr. Butler is confusing statistical accuracy with political signifi- cance.
The test universally accepted throughout the world and the one we have always used ourselves is the divergence of the poll figures from the actual votes as distributed between all the parties, If desired, this can be expressed in a single figure, as the average divergence over all the groups. Such a test is fair, complete, comprehensible and significant, statistically speaking. Can there be any good reason for compli- cating what is basically a straightforward situation?
What I have written above applies whatever the type of sampling used. Since 1955 Gallup have used both random and quota samples for general elec- tion forecasts.
Social Surveys (Gallup Poll) Ltd., 211 Regent Street. WI