Title: The Science (and Pseudoscience) of Winning Elections Donald Green Yale University
1The Science (and Pseudoscience) of Winning
ElectionsDonald Green Yale University
2What can be done to enhance ones chances of
winning the upcoming election?
- With just a week to go, its too late to recruit
a candidate or alter the political or economic
environment - Too late to fundraise or build coalitions
- At this late date, strategies have been set, and
decisions now center on tactics
3Basic tactical questions
- How will you allocate your remaining resources?
- Which voters will you target
- How will you communicate with them?
4Tactical choices boil down to two approaches to
winning votes
- Mobilization increase net votes for your
candidate by raising the probability that target
voters will cast ballots - Persuasion increase net votes by raising the
probability that target voters will vote for your
candidate if they cast ballots
5Advantages and disadvantages of mobilization and
persuasion strategies
- Mobilization requires accurate measurement or
forecast of voters preferences hence the
importance of microtargeting - Persuasion is, in principle, doubly effective
(subtracting from the opponents vote and adding
to your candidates) but requires an accurate
forecast of voters turnout probabilityOften
causes campaigns to focus on high propensity
voters - Campaigns are reluctant to contact opposing
partisans for fear of mobilizing them
inadvertentlyanother rationale for
microtargeting
6What is microtargeting?
- An exercise in which information in the voter
file and consumer databases are used to forecast
survey responses and turnout - Sometimes targeting data bases are created from
surveys that ID voters candidate preference - More often, however, voters preferences are
merely forecasted based on data base information
7How does microtargeting work? Just ask the
experts
8Does microtargeting work? Just ask the experts
9Does microtargeting work? Just ask the experts
10Does microtargeting work? Just ask the experts
11How accurate is microtargeting in practice?
Example Texas 2006 Contest for Governor Using
microtargeting forecasts to predict survey
responses
Notice that the microtargeting scores do only a
so-so job of predicting who supports the GOP
candidate
12Suppose you conduct a mobilization campaign Will
your tactics work? Just ask the experts
13Heres what proof means in the unregulated
world of campaign services
14Or you could turn to academic experts, who tend
to rely on survey research
- How about comparing those who say they received
calls or visits or mail from a campaign to those
who did not? - Problem 1 Campaigns neither target nor contact
voters at random - Problem 2 People do not report campaign contacts
reliably - Problem 3 Even if survey respondents are
perfectly accurate in their reports, the
questions are vague (e.g., Did someone from a
campaign come around or call you up? When?
How? How often?)
15A new science of voter mobilization using
randomized field experiments to evaluate
mobilization tactics
- Lists of registered voters (or in some cases,
towns or media markets) are randomly assigned to
treatment or control groups - Interventions are directed at the treatment group
(statistical adjustments for failure to contact
the intended voter) - Turnout records from registrars are used to
compare voting rates in treatment and control
groups
16Headlines from 125 experiments conducted between
1998 and 2008
- Quality of interaction matters personal tactics
such as canvassing and volunteer calls work
better than mail, email, robo-calls, or leaflets - Little evidence of synergy between different
types of tactics, but growing evidence that
obtaining commitments to vote follow up contact
works particularly well
17Example Door-to-door canvassing of voters with
Latino surnames in Phoenix, 2003 (ACORN)
18NALEO Volunteer Phone Banks, 2002 2008
19NALEO Pre-Recorded GOTV Calls by A Univision
Anchorwoman, 2002 PICO GOTV Calls by Local
Clergy, 2006
20Example NALEOs direct mail (2-4 pieces) to
low-propensity Latino voters in four states, 2002
21Estimated effectiveness of canvassing has not
changed much with accumulation of experimental
evidence
Estimated effect of canvassing based on a
meta-analysis of 39 canvassing experiments
confirms initial estimates, translates into an
average effect of 7 percentage points per contact
22Does science merely affirm common sense? The
case of vote-by-mail balloting
- Very high rates of turnout among those registered
as vote-by-mail voters - Therefore, campaigns seek to register people to
receive mail in ballots - Experiments were conducted to test whether
encouragement to vote by mail in fact raised
turnout - Treatment group was mailed VBM applications
23Vote by Mail Experiments 2006, 2007
24Scientific methods can be used to evaluate
campaign tactics, but can science-driven
campaigns innovate?
- Drawing inspiration from historical and
comparative instances of high turnout - Public voting
- Election Day festivals
- Block captains
25Science-led Innovation Example of following-up
with yes voters
- Trajectory of four experiments dating back to
2003 - In 2008, Los Angeles group that targets
Asian-Americans conducted an experiment in which
some voters were targeted for an initial call,
and a random subset of the yes group was
targeted for a follow-up call - Initial call increased turnout by 4
percentage-points among those reached and slated
for one round of calls - Follow-up call increased turnout by 13
percentage-points among those reached
26Sometimes personal contact and volunteer calls
are not feasiblethen what?
- Conveying the effects of social norms through
direct mail and phone calls - Direct Mail experiments
- Michigan 2006 and 2007
- Boston 2007
- Iowa 2008
- Iowa Michigan 2007
- Robo calls
- Michigan 2008
27Design of 2006 Michigan Study (Gerber, Green,
and Larimer 2008 APSR)
- Sample 180,002 households in Michigan,
registered voters who voted in 2004 - Setting August primary election, contested only
on the Republican side - Assignment 10,000 clusters of 18 households
each in each cluster, households assigned at
random to one of five groups - Outcome Voting in the primary election, as
indicated by official records for each individual
28Results from the 2006 Michigan Social Pressure
Experiment
29Do presidential campaign ads mobilize? Probably
not, judging from natural experiments
30What about persuasion?
- Seems to depend on who the candidates are known
or unknown? - Unknown powerful effects of increasing
name-recognition - Known much weaker or transitory effects
31Texas Television and Radio Experiment
- Stage set for experimentation in December, but
late filing deadline leaves primary contest up in
the air - Randomized the timing and volume of television
ads in 18 of 20 media markets - Varied the timing (not volume) of radio ads,
mapped signal propagation zones and attempted to
confine radio messages to DMA boundaries - Daily tracking poll gauged whether support for
Perry moved as media was introduced in or
withdrawn from each market
32Rollout of Experimental TV and Radio Campaign
33Illustration of TV Effects Differenced Weekly
Data, with Fixed Effects for Week
34TV ads show big effects, but rapid decay
TVs effect
TVs effect a week later
35Conditions under which persuasion occurs?
- Novel information may account for name
recognition effect (which in turn has interesting
policy implications) - Reference groups, but evidence on the
effectiveness of endorsements (from credible
sources) is mixed - Emotional content, but evidence of decay
36Which is more cost-effective, mobilization or
persuasion?
- Perhaps the shift from persuasion-oriented
campaigns to mobilization-oriented campaigns
reflects the fact that the latter has a more
predictable payoff - Persuasion campaigning still attracts the bulk of
campaign resources, but changes in the media
environment make it increasingly difficult - The technology for voter mobilization, on the
other hand, has arguably made mobilization easier
37The Scientific Study of Political Campaigns
Transforming Political Practiceand Political
Science
- Political campaigns although beset by
conflicts-of-interest, there are some incentives
for learning particularly among enduring
institutions (e.g., parties, unions, foundations) - Political science slow to embrace field
experimental methods, but research on campaign
effects has shifted fundamentally in ten years
38The Scientific Study of Political Campaigns A
Work in Progress
- Broader message of using scientific methods to
guide the allocation of scarce resources - The challenge of applying science to campaign
craft is balancing the desire to learn with the
desire to win - What if political scientists ran their own
(experimental) campaigns?
39The political scientist as politician? (running
for office and losing in the name of science?)
40Where to learn more about the scientific study of
campaigns?
- GOTV II includes a mini-course on experimental
design and an open invitation to replicate or
challenge existing findings - ICPSR Summer Workshop on experiments
41Acknowledgements
- Foundation support Irvine, Carnegie, JEHT,
Boston, Pew Charitable Trusts, CIRCLE, Smith
Richardson - Partners in the field nonpartisan, partisan,
campaigns, and interest groups - Dozens of experimental researchers nationally and
internationally
42For Further Information
- Donald P. Green
- Director, Institution for Social and Policy
Studies A. Whitney Griswold Professor of
Political Science - Yale University
- 77 Prospect St.
- New Haven, CT 06520-8209
- 203-432-3237
- Email donald.green_at_yale.edu
- Web research.yale.edu/vote
43The algebra of expected net votes
- Definition expected number of votes added to
your candidates total minus the expected number
added to the opponents total - Example A voter has an 80 chance of voting for
McCain if she votes and 20 chance of voting for
Obama. If this person has a 60 chance of
voting, her expected net vote for McCain is
(.80)(.60)-(.20)(.60).36 votes
44The algebra of increasing net votes through
mobilization (Levels dont matter you care about
your effect on the outcome)
- Mobilization Example A voter has an 80 chance
of voting for McCain if she votes and 20 chance
of voting for Obama. Suppose GOTV activity
raises the probability of voting from 60 to 70,
the expected net vote for McCain rises from
(.80)(.60)-(.20)(.60).36 votes to
(.80)(.70)-(.20)(.70).42 votes, or a gain of .06
votes. - Note that the stronger the McCain support, the
bigger the returns from mobilization. (Returns
are negative if you mobilize your opponents!)
45The algebra of increasing net votes through
persuasion
- Persuasion Example A voter has an 80 chance of
voting for McCain if she votes and 20 chance of
voting for Obama. If this persons probability
of McCain support rises from 80 to 90, the
expected net vote for McCain rises from
(.80)(.60)-(.20)(.60).36 votes to
(.90)(.60)-(.10)(.60).48 votes, or a gain of .12
votes. - Note that the higher the probability of voting,
the bigger the returns from persuasion