20 January 2010 Internet polling: a more hesitant approach Andrew Cooper, Founder, Populus
| Choosing not to be an online pollster (2003) NCRM seminar2
| 3 ALLMENWOMENABC1C2DE %67%52%70%72%57%31%75%80%70%71%41%22% 1. Low (and unrepresentative) internet penetration Source: Ofcom, 2003
| 2. Panel issues PROFESSIONAL PANELLISTS Panels dominated by small minority of all internet-users, most of whom are members of several panels, raising further concerns about data quality and representativeness (research suggests that people who are members of several panels are relatively unrepresentative of all internet users – more likely to be female and lower socio-economic status). ‘HAPPY CLICKING’ Absence of a human interviewer means there is no sanction for giving random responses to complete surveys as quickly as possible – and the primary motivation for respondents is to get the financial reward. PANEL CONDITIONING Respondents are polled multiple times – becoming more and more atypical. 4NCRM seminar
| 5 EDD Group in the EP & U.K. Independence Party Press Digest Wednesday 10th July 2002 UKIP & YouGov - Urgent action required by members/supporters Dear Member/Supporter, I would recommend that all members visit the website, and register to participate in polls. YouGov suffer from a shortage of persons in the older age groups (55+) on low incomes, and consequently are forced to weight the polling data from those groups accordingly. I would urge all members/supporters to register as retired persons on low incomes, former Labour voters who are now undecided. Do not mention UKIP! The result will be a disproportionate response in favour of UKIP (or 'Other Party' if they continue to miss us out). 3. Panel infiltration
| 4. Political sample bias 6NCRM seminar PAST VOTEYouGovPopulus2001 result Labour58%47%42% Conservative28%30%33% Lib Dem11%16%19% Others4%7%6% DNV17%31%41% Source: Populus & YouGov data tables, November 2004
| 5. Under-representing the don’t knows & the disaffected (and over-representing the informed & interested) 7NCRM seminar Source: Populus & YouGov data tables, December 2007 & January 2010 YouGovPopulus Dec-07Jan-10Dec-07Jan-10 WOULD NOT VOTE7% 11%17% DON'T KNOW16%13%18%10% REFUSE0% 7%5% TOTAL23%20%36%32%
| 6. The theological questions What exactly can internet poll data be said to be representative of? If you get the answer right, does it matter how you got to it? “Respondents for this survey were selected from among those who have agreed to participate in Harris Interactive surveys. The data have been weighted to reflect the composition of the adult population. Because the sample is based on those who agreed to participate in the Harris Interactive panel, no estimates of theoretical sampling error can be calculated.” 8NCRM seminar
| Looking again at online polling (2010) NCRM seminar9
| 1.Internet penetration has improved. Nielsen estimates 80% of UK adults now have internet access (up from 64% in 2007 and 60% in 2005) 2.Best practice deals with ‘happy clickers’ & professional panellists. Consistency checks, time stamping, digital fingerprinting & data cleaning 3.Larger panels reduce over-sampling, panel-conditioning & onerous weighting factors. Many panels now have hundreds of thousands of members, rather than tens of thousands 4.‘River sampling’ offers the prospect of moving away from panels altogether, enabling quasi-random online polls and reducing the infrastructural cost of panel maintenance. 10NCRM seminar
| 5.Fixed line telephony is in decline. 87% of households have a fixed-line telephone, down from 90% in 2007 & 93% in In the US, in-home internet penetration has just overtaken fixed-line penetration. 6.13% now have only a mobile phone, not a landline. 26% of year-olds, 21% of year-olds and 22% of DEs 7.Market forces, technological change and cultural trends are making telephone polling more expensive. 8.YouGov has shown that the challenges of gauging voting intention by internet polling can be overcome! 11NCRM seminar
| But... challenges remain 12NCRM seminar Recalled past voteTelephoneOnlineActual result Labour44.7%42.6%36.2% Conservative31.8%31.6%33.2% Liberal Democrat16.4%18.8%22.7% Others7.1%7.0%7.9% Vote intentionTelephoneOnline v.1Online v.2 Labour28%23%25% Conservative40%46%45% Liberal Democrat18% 16% Others13% 14% Net lead12%23%20% Source: Populus data tables, October as do the theological questions