Polled Over
In the 10 days since the UK general election there has been a veritable gnashing of teeth in the media as anger is channelled towards the pollsters for getting it wrong. Whilst it’s understandable that paying clients are frustrated at being supplied with inaccurate data, there are wider implications for how the polls impact on media coverage and the subsequent influence on voter behaviour. Who is to say that voters may have voted differently if newspaper front pages, which were largely dedicated to examining coalition building, instead provided greater discourse around party policy?
Whilst we look on in sympathy as our political polling cousins are held to account, any high profile criticism of the research industry, which includes accusations such as unrepresentative consumer panels and systematic methodological bias, should make any researcher sit up and listen.
Consequently I’ve been reading with interest the different explanations that the pollsters are giving for the error in their results. Whilst there are many methodological issues being discussed, the finger is frequently being pointed at party voter biases. ‘Shy Tories’ and ‘Soft Labour’ are being reproached in many a newspaper column for not voting as they said they would in the polls. What’s interesting is that for many commentators the gap between intention and behaviour seems to come as a surprise whilst we as researchers face this challenge on a day to day basis. Our behaviour change practice provides many examples of how behaviour is influenced by external cues and internal biases, of which consumers are largely unaware.
Despite the public derision of the pollsters there were polls that correctly predicted a Tory majority. The Conservative party claim their internal polls predicted a Conservative majority for the entirety of their 6 week campaign. Whilst there were some methodological differences between the large pollsters and the Tory run polls what is particularly interesting is that the Tory polls used some derived measures of behaviour to predict voter preferences.
Over the past week analysts within the Tory party have attributed some of their success to moving away from a traditional, direct questioning approach towards using derived measures of voter behaviour. They used an approach of modelling the attitudes and values that would drive a Tory vote and included questions measuring these attitudes in polls conducted in marginal seats. The data were analysed to measure the extent to which these vote-driving attitudes and values, which formed the basis of Tory campaign messaging, were being adopted by swing voters.
Reading this, it not only struck me that this seems a more sophisticated approach compared to the direct questioning method adopted by pollsters but it’s also far more in line with how we would go about predicting behaviour for our commercial clients. Consumer purchase decisions often require prioritisation and trade off of a range of influencing factors (brand, price, perceived product benefits) and consumers are typically poor at expressing how these factors interact in their decision making. Consequently we use research techniques like conjoint and driver analysis (to name but two) implicitly to identify and measure which factors are the most important at driving purchase behaviour.
Who is to say that the undecided voter doesn’t go through a similar decision-making process as someone purchasing a consumer product, having to prioritise and trade off a range of factors covering party policies on health, economy, immigration against the softer brand factors of candidate personalities as well as the cost of their choice in the form of taxation or welfare payments? Using an approach that accommodated this complexity could be the reason why the Tory polls predicted the final result more accurately than the large political pollsters.
Whilst no doubt the pollsters are examining their approach from a methodological perspective, I hope there is also some acknowledgement that there are learnings to made from behavioural science and by adopting some of the techniques that we use in commercial research to provide more accurate predictions of voter behaviour in the future.