On the eve of the 2015 General Election, one thing is certain: this has been the most predicted General Election in history. And psephology is evolving. From opinion polls, to polls of polls, to polling models, and here, now, to “meta-psephology”. The first represents a single survey of voters, the second an average of multiple surveys of voters, and the third an analysis of multiple polls plus an analysis of how polls have performed against actual results in previous elections and assumptions about how various parties might perform differently to polling data. In the latter case, this might include an allowance for “Shy Tories” who are embarrassed to admit that they vote Conservative, and for LibDem incumbency in which LibDem localism bucks the national trend. The fourth and final approach is my analysis of what the models might collectively tell us.
Each of the above techniques are based on something economists call “expressed preference”, that is, they rely on what people say they will do. However, economists argue that a more reliable indicator of preference is “revealed preference” in which what people actually do is measured. Of course, actual voting behavior can only be measured in the outcome of the election itself. Nevertheless, a proxy measure of revealed preference is available in an analysis of bookmakers odds because these are set according to the behaviours of people who have put their money where their mouth is in betting on the election outcome. Although, this is a measure of what people think the national outcome will be, rather than a measure of the way in which they will vote themselves. As such, someone may bet on a Conservative victory, but vote LibDem.
So how might these various predictions be made sense of? On the day of polling, five polling “models” were available (three academic and two “nowcasts”) and, for fun, we can add Ladbrokes odds into the mix. The models are each underpinned by different assumptions, and the spread of predictions across the five models and Ladbrokes odds (eg, from 272 to 284 seats for the Conservatives) is smaller than the statistical margins of error of the models. Given that the models are largely based on the same data, the differences lie in the assumptions that are being made about “trend bucking” effects and the relevance of historical data.
Taking a meta-psephological approach to analysing these models and the Ladbrokes odds, my prediction is as follows:
|Conservative||278 (+/-6)||268 (+/-3)||Labour|
|LibDem||26 (+/-1)||53 (+/-2)||SNP|
|UKIP||2 (+/- 1)||3 (+/- 1)||Plaid Cymru|
|DUP||9 (+/- 0)||1 (+/- 0)||Green|
|3 (+/- 0)||SDLP|
|Possible Cons-led||315||328||Possible Lab-led|
|Others||7 (+/- 1)|
The spread is the adjusted average of the three lowest predictions set against the adjusted average of the three highest predictions. As such, it is a spread of predictions not a margin of error. So here are the headlines: