Did people do what they said? Martin Boon, ICM Research John Curtice, University of Strathclyde

1
Did people do what they said?Martin Boon, ICM
ResearchJohn Curtice, University of Strathclyde
2
Recap The ICM method

• All ICM Vote Intention polls employ a telephone
  method.
• Representative samples of around 1,000 people
  each time, although the final prediction poll in
  2010 interviewed just over 2,000 people.
• Various weighting and adjustment procedures are
  employed, including
• Weighting by likelihood to turnout.
• Weighting by recall of past voting, with target
  weights being 20 to the actual result of the
  previous General Election, and 80 to the average
  recall generated across (up to) the previous 25
  ICM vote intention polls.
• Adjustment procedure reallocating 50 of the Dks
  and Partial Refusers back to the party they voted
  for in the previous General Election.

3
Recap The ICM performance
Con Lab LD Other Average error
ICM 36 28 26 10 1.25
Harris 35 29 27 10 1.5
Ipsos-MORI 36 29 27 8 1.75
Populus 37 28 27 11 1.75
ComRes 37 28 28 7 2.25
You Gov 35 28 28 9 2.25
Result 37 30 24 10 -
4
ICM Polls campaign tracking
1. ICM cannot escape Lib Dem surge issues. 2.
Penultimate poll would have been pretty poor (AE
2.75) 3. All adds up to a need to test the
efficacy of our methods.
5
The Recall Poll

• ICM has conducted a Recall Poll after every
  election since 1997. It has been used to evaluate
  various conditions that affect past vote recall,
  including the tendency of LD voters to be
  somewhat more forgetful.
• Certainly, the behaviour of Liberal Democrat
  supporters is in the firing line of our efforts
  on this occasion.
• On this occasion we called back 4,662 people who
  had previously been interviewed on one of our
  campaign polls up to and including (original)
  poll fieldwork on April 14th 2010.

14-16th Apr 17-23rd Apr 24th-30th Apr 1st-3rd May 4th May Prediction Poll Total
814 1,018 1,104 594 1,132 4,662
6
What we can examine

• How well did people forecast whether they would
  vote or not?
• What happened to the Dont Knows and Wont
  Says?
• Was there a late swing?

7
Potential Pitfalls

• Those who participated in the recall exercise may
  not be representative of all those who
  participated (sample is skewed towards older and
  middle class people)
• In particular they may have been particularly
  likely to have voted (84 did so)
• Some of the Dont Knows and Wont Says
  wouldnt say afterwards either! (14 of voters
  didnt declare their party id)
• Not all respondents were originally interviewed
  close to polling day. The swing amongst those
  whose reported vote did not match their vote
  intention may not have been late (46
  interviewed before the last week)

8
Some Decisions

• Data weighted to be demographically
  representative (but not past vote weighted)
• In analysis of reported vote, those who failed to
  say for whom they voted are excluded.
• Some analyses based only on those interviewed in
  the last week.

9
Anticipated and Actual Turnout
10
Estimating The Party Differential
Vote Intention Estimated Turnout Actual Turnout Gap
Conservative 95 92 3
Labour 93 86 7
Liberal Democrat 92 90 2
Other 88 75 13
11
Alternative Estimates
Voting Intention 7-10 Only Gap 3 point scale Gap
Conservative 95 3 90 -2
Labour 90 4 88 2
Liberal Democrat 90 0 87 -3
Other 84 9 84 9
12
Too big to ignore?

• 68 of those who did not declare a vote intention
  reported having voted.
• These constituted 29 of all those who reported
  having voted
• Including 22 of those who voted who revealed for
  whom they voted.

13
A Distinctive Group
14
Their Recall of 2005
15
How Recall 2005 Vote Anticipates Reported 2010
Vote
Con Lab Lib Dem Other DNV
Con 66 14 15 17 7
Lab 7 48 12 15 13
Lib Dem 10 14 54 24 9
Other 4 3 8 28 2
DNV 13 20 11 15 70
16
Campaign and After
17
The Last Week and After
18
The Last Poll and After
19
The (Last Week) Crossbreak
Con Lab Lib Dem Other
Con 87 1 4 7
Lab 2 80 7 4
Lib Dem 3 3 78 8
Other 1 3 54
DNV 8 15 9 27
20
Conclusion

• Some differential abstention that was only partly
  captured by turnout weighting.
• Labour profited including relative to the Lib
  Dems - from the shy Tory phenomenon though
  this was (adequately?) allowed for in some
  polls.
• But no convincing evidence of a significant late
  swing away from the Lib Dems.
• So also need to look to possibility of sampling
  bias or weighting error?