|
British Election Study 2001/02 |
|
A
Methodological Note on Forecasting and the "Adjustments" of Polls The Campaign survey results reported here are not intended to constitute a "forecast" of the likely election outcome. We have simply reported our results, unadjusted, as they have been supplied to us by our polling agency, Gallup. Clearly, we could make adjustments to our data, just as many of the commercial polling agencies do, in order to compensate for "shy Tories" or for "tactical voters" or for whatever else we thought might perturb our opinion poll results. Below, we provide three different examples of how our Gallup data could be "adjusted" to produce a much smaller gap between the leading parties. Example 1: Weighting by the Probability of Turning Out to Vote We ask all respondents how likely they are to vote on a 0 to 10 scale. If we assume that only those respondents who score either 9 or 10 on this scale are genuinely likely to vote, we get the following vote prediction:
Example 2: Weighting by recalled vote in1997 In this case we adjust the data so that the vote in 1997 variable reflects the actual (Great Britain) vote shares, rather than the shares given by the respondents.This has the effect of increasing the importance of Conservative and Liberal Democrat voters and decreasing the importance of Labour voters. Weighting by recalled 1997 vote in this way produces the following:
This calculation excludes respondents who reported that they did not vote in 1997, either because they were too young or for other reasons. Example 3: Adjusting by the probability of voting for specific parties We ask all respondents a set of
questions about how likely they are (on 0-10 scales) ever to vote for each of
the main parties. This enables us
to classify respondents according to which party they are most likely to vote
for. The results that we get for the 8-day rolling poll to June 6th (N=1,089)
are:
Assume all Con - Lab ties and all Con - Lib ties are "shy Conservatives". Assume all Lab - Lib ties will result in "Lib Dem protest vote". Assume all 3-way or 4-way ties are non voters. This produces: Con: 23.2 + 3.6 + 2.4 = 29.2% Lab: 43.7% Lib: 13+5.2 = 18.2% In summary:
Clearly it is possible to make a variety of different "adjustments" to reflect different assumptions about what voters' "true preferences" are. We prefer simply to report our "raw" findings and to allow the users of our data to make whatever "adjustments" they see fit.
|
