A Journalist’s Guide to Survey Research and Election Polls by Cliff Zuskin
A 61 million-person experiment in social influence and political
1. A 61-million-person experiment in
social influence and political
mobilization
Robert M. Bond, Christopher J. Fariss, Janson J.
Jones, Jaime E. Settle and James H. Fowler
UCSD
Adam D. I. Kramer and Cameron Marlow
Facebook
3. Main Results
• A randomized controlled trial
• > 18 years old, U.S.
• Nov. 2nd, 2010 – U.S. congressional election.
4. Main Results
• Three groups of users
▫ “social message”
n = 60,055,176 (99%)
▫ “informational”
n = 611,096 (1%)
▫ “no message”
n = 611,044 (1%)
32. Main Results
• Some context
▫ US midterm elections
36.3% (2002) to 37.2%
(2006) To 37.8% (2010)
▫ 60,000 voters (d)
▫ 280,000 voters (i)
▫ 0.14% of 236M in 2010
33. Main Results
• Claims
▫ Online political mobilization works.
▫ Generate real validate votes.
Contradict previous research.
▫ Social mobilization is significantly effective than
informational mobilization alone.
▫ Close friends have about four times more influence.
▫ Online social networks influence offline behaviors.
38. Technical Details
• Matching to Voting Records
▫ First names + last Names + full birthdates
▫ Cost $0 -> $1500 per state
▫ Exclude Texas
▫ Include
Arkansas, California, Connecticut, Florida, Kansas,
Kentucky, Missouri, Nevada, New Jersey, New
York, Oklahoma, Pennsylvania and Rhode Island
▫ 40% of all registered votes -> 6,338,882 voters
and abstainers.
40. Technical Details
• Overreporting and Underreporting
▫ 3.8% self-reported -> abstained
▫ 50.1% declined to report -> voted
• Social message vs. informational message
▫ 0.99% more likely to overreport to voting
▫ 4.19% less likely to underreport voting
41. Technical Details
• Determination of “Close” Friends
▫ Media multiplexity
▫ # interactions of f/ total # of interactions with all
▫ Divide by decile.
▫ Each decile is a subset of the previous one.
▫ Validated via survey.
42. Technical Details
• Network permutation
▫ Keep the network topology fixed
▫ Randomly permute the assignment to treatment
▫ Repeat 1,000 times
▫ Theoretical null distribution