The University of Manchester Get Out the Vote Experiment

1
The University of Manchester Get Out the Vote
Experiment

• Peter John and Tessa Brannan, Institute for
  Political and Economic Governance (IPEG),
  University of Manchester

2
Background

• Declining Voter turnout in the UK 2001, 2004,
  local and European elections
• Explanations marginality, decline in trust,
  poor performance of governments, urban life?
• Less discussed is the absence of mobilisation,
  engagement of the voters
• Decline in the traditional mechanisms political
  party machines
• Can non-partisan campaigns increase turnout?

3
Study Design

• Chose a safe constituency for 2005 Wythenshawe
  and Sale East, had 49 per cent turnout in 2001
• Randomly selected 6,900 from 9,976 voters on
  electoral register who had accessible telephone
  numbers
• Randomised into three groups of 2,300, one
  control, two treatment
• Handed 2,300 to Vision 21, a local survey
  company, who telephoned them with a prepared
  script for two weeks prior to election
• Canvassed the other 2,300 by groups of students
  and Vision 21 fieldworkers following the same
  script

4
Study design continued

• Treatment groups got a letter informing them
  about the study
• Script stressed duty to vote, decline in turnout,
  what would happen if no one turned out,
  importance of influencing outcome
• Questions about intentions to vote and postal
  voting
• Also provided information about voting procedures

5
Issues about canvassing

• Difficulty of communicating with professional
  company not used to this kind of study
• Student workforce were enthusiastic and enjoyed
  the experience (as well as being reasonably well
  paid!)
• Logistics were challenging study much more
  complex than expected planning routes, data
  quality issues, transport
• Most who we contacted liked being canvassed,
  and said they did not see anyone from the
  political parties
• bar the hostile, like the voter who
  threatened to set his dogs on us

6
Data issues

• Poor quality of electoral registration data. No
  of dead people. We ask company to supply list of
  registered deceased. Removed from all three
  groups (but we interviewed some of these dead
  voters!)
• Hard work needed to ensure accuracy of data
  returns
• Overall successful response rate of 47.8 per cent
  for canvassers 43 per cent for telephone survey,
  comparable with US studies

7
(No Transcript)
8
Results

• Still waiting postal voters, so removed from
  current phase of study yielding along with other
  non-eligibles
• Canvassing Telephone
• Treatment 1237 1281
• Control 1273 1273
• Contacted 664 611
• Voted Treatment 68155.1 70455.0
• Voted Control 65551.5 65551.5

9
Statistical issues

• It is possible just to compare these averages and
  run a t or similar type of test, but non
  significant in this case (p.117)
• But Gerber and Green say that the intent to
  treat and the actual treatment rate are
  important to be able understand the impact of the
  experiment
• If a proportion of the population that is
  reachable
• Let pnr be the probability that a non reachable
  person votes without the treatment and Let pr be
  the probability that a reachable person votes
  without the treatment
• Let pr t be the probability that a reachable
  person votes after the experimental treatment
• We want to know what t is

10
Statistical issues

• Probability that a member of the control group
  votes is PC appr (1- a) pnr
• Probability that a member of the treatment group
  votes is PE a(ppr t) (1- a) pnr
• If you combine the two above equations you get
• t PE PC / a
• Sample data can estimate t
• plim VE PE plim VC PC

11
Results

• Canvassing Telephone
• Contact Rate .53 .48
• Treatment Effect 6.7 7.3
• Significance .035 .038
• significant p. lt .05

12
Costs of raising the vote

• Total cost18,687, not including Peter Johns
  time
• Each successful interview costs 10.73
  (telephoning) 7.33 (canvassing)

13
Concluding Points

• The voting experiment worked!
• Both telephone and canvassing have the same
  effects, with telephoning marginally higher
• Why should telephoning get such good results
• personal intervention and monitoring?
• Land-line accessible sample easier to influence
• More research needed on close seats, different
  messages, postal voting