social’pressure,’social’media,’and’ voter’turnout · 2015-11-30 ·...
TRANSCRIPT
Social Pressure, Social Media, and Voter Turnout
Kevin Collins Director of Research, Analyst Ins>tute
Overview
• Diffusion of social pressure language in direct mail voter mobiliza>on shows strength of social pressure for GOTV
• Tests on social media suggest that messages from one medium don’t necessarily work in another
DIFFUSION OF SOCIAL PRESSURE VOTER MOBILIZATION
&RQJUDWXODWLRQV«RQ«\RXU«¿UVW«9RWHU«5HSRUW«&DUG�Who you vote for is your secret. But
whether you vote is public record.The Voter Report Card on the back shows
your voting attendance record. It’s designed to help you keep track of how you’re doing.
All citizens should have a voice in their government. Our democracy works best
when everyone is a voter—including you.
That’s why we’ve selected you for this important research project. In the future,
we hope to send you an updated Voter Re-port Card before each election.
If you’re not sure where to vote, visit Vote411.org, an independent website. And
ask about early voting options in your area.
As a voter this year, you’ll be joined by millions of fellow citizens. Many con-
tests, though, will be so close that just a tiny number of votes will determine the outcome.
One of those crucial votes could be yours—but only if you’re a voter.
%H«D«YRWHU«RQ«7XHVGD\�«1RY�«� — or ask about early voting.
Source: hKps://twiKer.com/kbenker/status/527627603373735936
VOTER MOBILIZATION THROUGH FACEBOOK ADS Joint work with Josh Kalla (UC-‐Berkeley) and Lauren Keane
Research Ques>ons
• What is the average treatment effect of online GOTV adver>sing at mobilizing voters?
• Are there par>cular kinds of voters that are more responsive to online ads, based on behavioral trace data?
• How do these effects compare to other GOTV strategies in the same electoral context
2012 Experimental Condi>ons
• Voters were randomly assigned to one of two experimental condi>ons:
– A control group that was not assigned to receive GOTV Facebook adver>sements (n = 365,668)
– A treatment group that was assigned to receive GOTV Facebook adver>sements (n = 365,900)
• Used “Custom Audiences” tool to upload targeted list
Examples of Ads
02
46
8
Trea
tmen
t Effe
ct(P
erce
ntage
Poin
ts)
Full Sample Registered Online Online Ad Facebook Ad
2012 Condi>onal Average Treatment Effects
(N=731,568) (N=152,059) (N=26,948) (N=1,839)
2013 Experimental Condi>ons
• Voters were randomly assigned to one of two experimental condi>ons: – A control group that was not assigned to receive GOTV Facebook adver>sements (n = 46,540)
– A treatment group that was assigned to receive GOTV Facebook adver>sements (n = 46,513)
• In Virginia only, this experiment was orthogonally assigned with another experiment on volunteer phone banks
2013 Condi>onal Average Treatment Effects
(N=93,053) Control Group
Turnout = 14.56%
(N=90,786) (N=12,539) (N=1,410)
Virginia Overlapping Experiments
• Only Virginia • Volunteer phone calls made through online call tool
• Used modified “rolling” protocol (Nickerson 2005)
• Contact rate in volunteer phone condi>on was 23%
No FB Ad
FB Ad
No Volunteer
Call
9,458 9,335
Volunteer Call
2,444 2,537
Intent to Treat Effects (VA)
Key Takeaways
• Poli sci research adopted and adapted • Context maKers more than content • Match the message to the medium
• More tes>ng needed
Thank You!
Kevin Collins Email: kcollins@analys>ns>tute.org
TwiKer: @kwcollins