Dr Grant Blank Prepared for the General Online Research conference, Cologne, Germany, 5 March 2014 Who uses Twitter? Representativeness of Twitter Users
Online population Britain − Richer, better educated and younger − Gender gap disappeared by 2011 − Students and employed − Not disabled USA − Broadly parallels Britain − Younger, richer, better educated
Twitter research Common goal: infer attitudes, preferences or actions Assumption: Twitter users are representative If not true, then Twitter users are a biased sample − A larger sample will not fix bias “Most existing work does not address the sampling bias, simply applying machine learning and data mining algorithms without an understanding of the Twitter user population.” -Mislov et al
Oxford Internet Survey Probability sample of England, Scotland & Wales About 2000 respondents: 14 years and older Face-to-face, in-home interviews − About 45 minutes for Internet users − Collected in February 2013 Waves: 2003, 2005, 2007, 2009, 2011, 2013 Component of World Internet Project (WIP) Sponsors for 2013 from the Nominet Trust, Ofcom, dot.rural
Pew Internet & American Life Probability sample of US adults Landline and cell phone interviews Random Digit Dialling May ,252 adults living in the United States Respondents: 18 years and older ** The Pew Research Center bears no responsibility for the interpretations presented or conclusions reached based on that data.
Variables Demographic – Britain, 9 variables : age, education, income, lifestage, marital status, ethnicity, disability, gender, urban-rural – America, 7 variables : age, education, income, lifestage, marital status, race, gender Make as comparable as possible
Table 2. British Twitter users compared to other groups Twitter users Non-Twitter SNS users Non-Twitter Internet users Full sample Number ,1222,053 % of all Respondents % of all Internet users — Notes: Total SNS users (including Twitter users): 1,267 Total Internet users (including Twitter users): 1,613
British Twitter users (see Table 3) Young University degree or further education Single Students or employed – Not retired High incomes Not disabled No differences for gender, ethnicity, urban-rural
American Twitter users (Table 4) Young Some college or college degree Single or married Employed – Not students, not retired Wealthy – Income differences are smaller than UK Non-white Gender doesn’t matter
Multivariate model (see Table 3) Britain: – Significant effects: Age, income, education & retired America: – Similar: Age, income – Different: non-whites, unemployed, married, not students Conclusion: Countries have different patterns of Twitter use
Twitter biases Twitter users are not representative – Of populations as a whole – Of online populations Twitter users compared to populations – Younger, wealthier – Britain: well educated, students Not interested in politics, not disabled – America: Married (not students), non-white Britain: Elites America: Elite, but non-white
Sampling weights Use percentage tables to construct weights to match population proportions – Important differences between the US and Britain – US: non-white, unmarried, unemployed, no education effect – Implies that British & US are not interchangeable Weights would have to take nationality into account – Few tweets are geo-coded – Some issues are of local interest
Research Questions Why does Twitter use differ so much? Why not use functional alternatives like text messages or instant messaging? What is the appeal of Twitter? What is the appeal of short messages systems?
Thank you. Dr Grant Blank Oxford Internet Surveys Oxford Internet Institute