Internet Experiments in the 2005 BES

1
Internet Experiments in the 2005 BES

• David Sanders
• Harold Clarke
• Marianne Stewart
• Paul Whiteley

2
Preferences and Party Choice

• Two traditions
• Downs and spatial modellers Preferences are
  exogenous
• Campbell and the social psychologists
  Preferences are endogenous, determined by a
  variety of contextual and conditioning factors

3
Preferences are Exogenous

• Voters maximise utility by selecting the party
  that is closest to them in a left-right
  ideological space. Voter is utility for party k
  is given by  
• Ui(k) (xi sk)2
• where xi is is preferred ideological position
  and sk is is estimate of the position of party
  k.
• Note here that xi is assumed to be time invariant.

4
Preferences are Endogenous

• What happens if measures of ideological position
  cannot be anchored in what voters say at time t
  because their self locations are not exogenous?
• Suppose that xi are affected by the campaign
  messages to which they are exposed
• xit g0 g1xit-1 gkzkt
• where gk is a vector of effect parameters and zkt
  is a vector of campaign variables.
• Formal theory implications How do we obtain an
  equilibrium in this situation? Can only do so if
  we make further assumptions about zkt
• Empirical implications we need to know what t is
  and what the relationship is between xit and zkt

5
How can this question be addressed?

• Experimental work an obvious approach do voters
  shift their ideological positions if they are
  given new information analogous to the
  information they receive during campaigns?
• Considerable advantages in conducting experiments
  on representative national samples much more
  likely to produce generalisable conclusions.
• One way of assessing whether preferences differ
  at two points in time is to give feedback to
  respondents about their initial ideological
  positions.
• In conjunction with these initial positions
  the xit-1 in our previous slides we can also
  give campaign information the zkt in our
  previous slides for example about the
  positions of the parties
• The internet represents an ideal vehicle for
  providing feedback about the xi and for
  manipulating the zkt

6
Procedure the Internet Survey Experiments

• Ask respondent to self-locate on two 0-10 scales
  tax/spend and liberal/authoritarian
• Later in survey show respondent where s/he is
  located in 2-d space defined by earlier responses
• Ask if respondent wishes to re-locate self
• Split sample on cues provided (eight experimental
  groups plus control see next slide)
• Explore differences in patterns of response in
  control and test groups

7
What are the zkt?

• Two general sets of zkt
• Parties
• Leaders
• and combinations
• Experiments
• 1 Control just feed back xit-1 for Respondent
  (R)
• 2 R average voter
• 3 R party supporters for Lab, Con LD
• 4 R named leaders (Blair, Howard, Kennedy)
• 5 R leader party label
• 6 R parties 1983 scenario
• 7 R parties 1964 scenario
• 8 R parties 2005 scenario
• 9 R leaders supporters (5)

8
(No Transcript)
9
Note the difference between the control and all
test groups
1 Control 4 R leaders 7 R parties 1964
scenario 2 R average voter 5 R leader
party label 8 R parties 2005 scenario 3 R
party supporters 6 R parties 1983 scenario 9
R leaders supporters (5)
10
1 Control 4 R leaders 7 R parties 1964
scenario 2 R average voter 5 R leader
party label 8 R parties 2005 scenario 3 R
party supporters 6 R parties 1983 scenario 9
R leaders supporters (5)
Differences between control and each test group
almost all statistically significant
11
1 Control 4 R leaders 7 R parties 1964
scenario 2 R average voter 5 R leader
party label 8 R parties 2005 scenario 3 R
party supporters 6 R parties 1983 scenario 9
R leaders supporters (5)
Differences between control and each test group
almost all statistically significant
12
Modelling absolute changes in self-locations

• With multivariate controls, it is the
  party-based effects than continue to be
  statistically significant respondents change
  their tax-spend ideological positions in response
  primarily to party cues.
• Same result with crime-rights scale and with
  Euclidean distance model based on the two scales
  in a 2-d space.

13
Modelling directional changes in self-locations

• Same pattern of effects observed with separate
  models of cues relative to Rs Personal
  Pre-experimental positioning of self and of the
  parties.
• Conclusion positive coefficients indicate that
  PARTY CUES ATTRACT

14
Conclusions

• Spatial model of voter utility assumes that
  voters ideological/policy preferences are fixed.
  
• When people are invited to adjust their recently
  made self-placements on two ideological scales, a
  non-trivial proportion of them elects to do so.
• Voters xit-1 values differ from their xit
  values even within the space of a few minutes.
• The type of positional cues to which people are
  exposed the zij affects the extent to which
  they wish to adjust their self-placements.
• Information about the positions of named party
  leaders seems to have little effect on
  adjustment.
• Information about parties, party supporters
  or leaders with party labels does affect the
  extent to which people wish to adjust.

15
Conclusions.

• Analyses of respondents directional movement
  suggests that voters are attracted to party cues
  (of whatever sort), rather than repelled by them.
  
• Party cues help to persuade people to shift
  their ideological/policy positions.
• The internet allows relatively sophisticated
  survey experiments to be conducted with
  representative samples of the electorate, rather
  than with small and unrepresentative groups of
  (e.g.) undergraduate students.
• Future experiments need to simplify the
  stimuli, e.g. single or two-party stimuli
  allowing respondents to vary parties positions
  as well as their own.

16
(No Transcript)
17
Party-based comparisons with control tend to give
higher eta values than leader-based equivalents
18
Again, party-based comparisons with control tend
to give higher eta values than leader-based
equivalents