Why were the opinion polls so wrong in predicting the result of the British General Election of May 2015?

This week, a report by Professor John Curtice for research agency NatCen was published which essentially concluded that the pollsters failed to obtain a genuinely random sample. As a result, they underestimated support for the Conservative Party.

The problem is that some voters are easier to contact than others.

Those who are easier to contact are:

  • those with a landline
  • those who answer a phone call from a stranger
  • those who have an Internet connection
  • those who are willing to sign up to online polling
  • those who are at home and not too busy
  • those who are politically engaged

Polling is becoming more difficult because fewer homes have a landline, many people do not like to answer unsolicited calls, not everyone is on the Net, volunteers for online polling are somewhat self-selecting, older people are less likely to be contacted by pollsters but more likely to vote, and those who are busy with work are less likely to be available but often more conservative.

You can read the BBC coverage of the report here and the text of the full report here.

I would make two observations:

  1. The problems for UK political pollsters in obtaining a genuinely random sample is one faced by such organisations in most developed nations. We have seen political pollsters get it wrong in cases in the USA and Israel and I fear that the practical and statistical problems are only going to become greater and the accuracy of political polling is going to become weaker.
  2. The difficulty that pollsters have in engaging with harder to reach citizens and consumers affects the reliability of polling more generally including that by companies and campaigns. Citizens and consumers who are older, poorer, have a disability, live in isolated circumstances, or do not have the vernacular as a spoken language tend to be under-represented in such polling.

 




XHTML: You can use these tags: <a href="" title=""> <abbr title=""> <acronym title=""> <b> <blockquote cite=""> <cite> <code> <del datetime=""> <em> <i> <q cite=""> <s> <strike> <strong>