1. Gregory Holyk, Langer Research Associates
Seth Brohinsky, Abt-SRBI
Dean Williams, Abt-SRBI
Damla Ergun, Langer Research Associates
Gary Langer, Langer Research Associates
Julie Phelan, Langer Research Associates
Polling in the Midst of a Natural Disaster:
The ABC News/Washington Post 2012
Election Tracking Poll and Hurricane Sandy
2. What Could Possibly Go Wrong?
Sandy made landfall on the NY/NJ coast the night of Oct. 29,
2012 (night 12 of our tracking poll in the presidential election)
Presented an unprecedented situation in our experience
Intense internal discussion:
Empirical question: Can we continue to obtain a sufficiently reliable
national sample?
Ethical question: Should we continue to call affected areas in the
aftermath of a natural disaster?
If we proceed, how will we diagnose our sample is acceptable and
determine whether or not to proceed?
2
3. How We Proceeded
Some said the storm made it impossible to continue sampling
and obtain a representative national estimate.
Our approach: Test it and see.
Steps we implemented:
Changed our intro language in the affected areas to be sensitive
to the situation:
“…. We know it’s a difficult time in much of the Northeast because
of the storm damage. But we’re doing an opinion poll on interesting
subjects in the news and would appreciate your participating. Would
that be OK?”
Evaluate sample dispositions.
Compare each night’s results (demographics, attitudes) vs. their
pre-Sandy averages and ranges.
3
4. Field Procedures and Affected Areas
On the night the storm hit the final NE interview was completed
at 7:56 p.m. – before landfall.
Call centers (Ft. Myers, FL and Huntington, WV) not affected.
Received hourly status updates from field site staff to project
management team during every interviewing shift.
Project managers debriefed nightly by field staff.
Northeast region = 18% of the country’s LVs.
Census divisions:
Middle Atlantic (NJ, NY, PA) = 13%
New England (CT, ME, MA, NH, RI, VT) = 5%
South Atlantic (DE, DC, FL, GA, MD, NC, SC, VA, WV) = 20%
Hardest hit states (NY and NJ) = 8%
But impact was coastal, not statewide
4
21. A National “Sandy bump” for Obama?
50%50%
47%47%
20%
30%
40%
50%
60%
70%
80%
Obama Romney
Obama vs. Romney, 2012
Among all likely voters
ABC News/Washington Post polls
Hurricane Sandy
10/21 and later: four-day rolling average
21
23. Sample Facts
Field Period: Oct. 18-Nov. 4, 2012
Non-overlapping RDD and CPO design
Max of 6 call attempts per record
Mix of fresh and live sample released every day
Nights 1-14: 440 total x’s per night
Nights 15-18: 800 total x’s per night
LL sample:
66,414 unique numbers dialed
224,922 call attempts made to complete n=7,874 interviews
Cell sample:
22,750 unique numbers dialed
70,665 call attempts made to complete n=1,193 interviews
All respondents were cell-phone only
All phone numbers were dialed manually
23
30. Landline Sample Cont’d
Percentage of busy numbers by night:
Average percentage of busy numbers: 3.2% of total dialed
Night 12: 4.3%; night 13: 9.4%; night 14: 5.3%
0%
5%
10%
15%
20%
25%
30%
35%
40%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
Busy Numbers as % of Total Sample
Hurricane Sandy
30
31. Landline Sample – By Census Division
0%
5%
10%
15%
20%
25%
30%
35%
40%
45%
50%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
Busy as a Percentage of Total Numbers
Dialed by Census Division
Censdiv 1 Censdiv 2 Censdiv 3
Censdiv 4 Censdiv 5 Censdiv 6
Censdiv 7 Censdiv 8 Censdiv 9
Hurricane Sandy
31
35. Landline Sample – Respondent Participation
0%
5%
10%
15%
20%
25%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
Respondent Participation by Night
Census Division 2 Total
Hurricane Sandy
35
37. Cell Phone vs. Landline Samples
Main differences:
LL sample uses autodialer, cell phone does not.
Cell sample dispositions are manually entered by a live
interviewer based on his/her interpretation of the outcome.
The following disposition codes were used to determine
“busy” equivalent in cell sample:
Answering machine/voicemail – unknown eligibility
Busy
Incomplete call/line problems (temporary)
No answer
Not In service/disconnected
37
39. Cell Phone Sample
No significant increase in total #s dialed / complete:
Burn rate for the entire tracking poll: 36:1, same as night 13
0
10
20
30
40
50
60
70
Night 1Night 2Night 3Night 4Night 5Night 6Night 7Night 8Night 9Night 10Night 11Night 12Night 13Night 14Night 15Night 16Night 17Night 18
Total Numbers Dialed per Complete
Hurricane Sandy
39
40. Cell Phone Sample – By Region
Total #’s dialed / complete: cell average 36:1
Night 13: no outliers
Night 14: NE jumps to highest burn rate of any region 88:1
0
10
20
30
40
50
60
70
80
90
100
Night 1Night 2Night 3Night 4Night 5Night 6Night 7Night 8Night 9Night 10Night 11Night 12Night 13Night 14Night 15Night 16Night 17Night 18
Numbers Dialed per Complete by Region
Northwest
Midwest
South
West
Hurricane Sandy
40
42. Cell Phone Sample
Percentage of NIS/VM/busy numbers by night:
Average percentage of NIS/VM/busy: 81% of total dialed
Consistent % across all nights
50%
55%
60%
65%
70%
75%
80%
85%
90%
95%
100%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
NIS/VM/Busy as a % of Total Dialed
Hurricane Sandy
42
43. Cell Phone – By Region
Increase in NIS/VM/busy percentage on night 14
Night 14: NE hits 85% (749 NIS/VM/busy is tops for any night)
Night 13: 237 and 80% of total
60%
65%
70%
75%
80%
85%
90%
95%
100%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
Percent of #'s NIS/unknown
Northwest
Midwest
South
WestHurricane Sandy
43
44. Cell Phone Sample – Participation Analysis
0%
5%
10%
15%
20%
Night
1
Night
2
Night
3
Night
4
Night
5
Night
6
Night
7
Night
8
Night
9
Night
10
Night
11
Night
12
Night
13
Night
14
Night
15
Night
16
Night
17
Night
18
Cell Phone Respondent Participation
Total Northeast
Hurricane Sandy
44
45. Conclusions
Post-Sandy dialings-to-completes in the affected areas were
within the normal range.
Demographic differences were generally within range.
No evidence of a “Sandy bump” in the NE or nationally.
Continuing tracking provided an important source of accurate
ongoing information about the campaign that otherwise
would have been absent.
45