How Have The Polls Changed Since 2015?

How Have The Polls
Changed Since 2015?
John Curtice
British Polling Council
Performance of Final Polls
38
40
35
34
34
31
30
25
20
13 13
15
9
10
8
5
5
4
6
6
0
Con
Lab
UKIP
Poll Average
Lib Dem
Green
Others
Result
Based on 10 polls whose fieldwork did not end before 5.5; Polls by Opinium; YouGov; Survation; ComRes; Populus;
Ashcroft, Ipsos MORI: BMG Research, Panelbase; ICM Research
Inquiry’s Key Recommendations
• Turnout
• ascertain who has voted by post;
• identify better ways of who is more likely to vote
• Don’t Knows
• review ways of imputing likely voting behaviour
• Sampling/Representativeness
• improve representativeness of who is interviewed;
• in setting quotas and weighting targets investigate use
new variables associated with responding to polls and
vote choice
• benchmark against random probability surveys
Overview
• Nearly every company has made at least some
change to their methods.
• Most have addressed to some extent all of the
principal recommendations in the Sturgis Report,
albeit not necessarily in the same way
• Some have also made changes that do not flow
directly from the report
% voted
Age Gap in Turnout
100
90
80
70
60
50
40
30
20
10
0
55 58 54
56 57
64
68
75 78 72
80 83
84
77
88
78
64
43
18-24
25-34
BSA
35-44
BES
45-54
MORI Estimate
55-64
65+
Anticipated Age Gap in Turnout
Estimated Gap 18-24 vs 65+
45
40
35
30
25
20
15
10
5
0
39
35 35
34
27
22
24
21
18
15
12
9
10
4
MORI Post Ashcroft
Elec
ComRes
Certain to Vote
* 18-24 vs 60+
ICM
Ipsos MORI Populus
Average Propensity
YouGov*
The Widened Age Gap in Party
Support
Change in % vote since 2010
20
17
15
10
10
8
6
5
3
4
1
0
18-24
-5
-10
Source: BES 2010 and 2015
25-34
-3
35-44
-3
45-54
55-64
-2
65+
-5
Con
Lab
-7
Sampling/Representativeness - 1
• Secure greater inclusion of those who are less
engaged in politics (YouGov)
• Changes of mode
• ICM from phone to online
• ComRes all polls online
• Quota/Weight data by (BES) interest in politics
(ICM; YouGov) or overall estimated turnout (Kantar)
or newspaper readership (Ipsos MORI)
• Weight data so that % in each demographic (esp.
age) group represents % of voters (BES) rather than
% population (ORB; Panelbase?)
Sampling/Representativeness - 2
• Quotas/weighting by education (Ipsos MORI;
Kantar; ORB)
• Change quota/weighting age bands to secure more
older voters (Kantar; YouGov)
• Changes to quotas/weighting by past vote/party
identification etc. (ICM; Opinium; YouGov)
• Weighting by 2016 EU Referendum vote (Opinium,
Panelbase; YouGov)
Identifying Who Will Vote - 1
• Weight data so that estimated turnout in different
demographic groups is in line with modelling of
turnout in 2015; drop asking people how likely they
are to vote (ComRes)
• Use reported likelihood to vote in a model of the
relationship between likelihood and turnout in
2015 (Kantar)
• Weight data so that estimated turnout in each
demographic group matches a variety of estimates
for 2015 (ICM)
Identifying Who Will Vote - 2
• Ask/use additional questions apart from reported
likelihood of voting (ICM, Ipsos MORI; YouGov)
• Change use/wording of likelihood of voting
question (Ipsos MORI; Opinium)
• Not ask postal voters likelihood of voting (Kantar)
Shy Voters?
• Introduce ‘squeeze’ question (Kantar)
• Introduce/extend imputation of Don’t Knows etc
(ICM; Kantar?)
Conclusion - 1
• Biggest change is the application of greater effort to
ensuring a more accurate estimate of key
differences in turnout
• Some effort to contact the less politically engaged
• Much greater attention to weighting/modelling data to
reflect patterns of turnout/engagement at previous
elections, sometimes with reference to the 2015
random probability BES
• Somewhat less reliance on a likelihood to vote question
alone
Conclusion - 2
• Also some changes designed (inter alia) to improve
representativeness of those who do vote
• Greater use of past vote weighting
• Some use of weighting by education and EU referendum
vote
• Many of the changes clearly informed by an
understanding of what went wrong in 2015 (thanks,
not least, to the Sturgis report)
• But inevitably whether they will prove sufficient
will only be known on June 9!