US Presidential elections, rather like American football, have always provided a rich seam of data for anyone who likes number crunching and with candidates separated by less than the margin of error in almost every poll, this cycle is no exception. What is different this time is a meme emerging on the internet and in sections of the media that the market research industry and its clients are deliberately producing biased results as a result of their sample weighting.
At the heart of this is a debate about what whether the turnout will most resemble the 2008 Presidential election or the 2010 mid-terms. In 2008 President Obama combined high turnouts from African Americans and Hispanics with the mobilisation of traditionally low-turnout groups such as students to win by a considerable margin. In 2010 many of these voters stayed at home to hand the Republicans a significant victory.
Most pollsters believe that things will look broadly similar to 2008, with Rasmussen as the main hold out until last weekend. However this hasn’t stopped the growth of a cottage industry of re-weighting samples to produce the desired result while decrying bias in the original polling. Here’s a link to my favourite, UnskewedPolls.com.
So what can researchers learn from this? Well a good start would be to stop clients producing headlines totally unsupported by research data. The highlight this week was ABC’s dramatic announcement of an eleven point Obama lead in four Swing States which turned out to be based on 161 respondents.
It’s hard enough convincing the lay-person that sample of 1,000 can represent the population, but when a major media outlet runs a story based on numbers which are barely significant at the 80% confidence level we’re just asking for trouble.