SlideShare a Scribd company logo
1 of 27
The PollyVote
Combining forecasts for
U.S. Presidential Elections
Andreas Graefe, Karlsruhe Institute of Technology
J. Scott Armstrong, Wharton School, University of Pennsylvania
Randall Jones, Jr., University of Central Oklahoma
Alfred Cuzán, University of West Florida
The full paper to this talk can be downloaded at: tinyurl.com/combiningelections.
Bucharest Dialogues on
Expert Knowledge, Prediction, Forecasting: A Social Sciences Perspective
November 21, 2010
Background on the PollyVote project
The PollyVote project was begun in 2003 to
demonstrate the value of forecasting principles
by applying them to election forecasting.
The initial focus was on combining forecasts.
Performance of the PollyVote
The PollyVote combined forecasts to obtain highly accurate
forecasts of U.S. Presidential Election outcomes:
– Prospectively for 2004 and 2008 (MAE: 0.4 percentage points)
– Retrospectively for 1992 to 2000
Across these five elections, the PollyVote was on average
more accurate than each of its components:
- Polls
- Prediction markets
- Experts
- Statistical models
Polly achieved this without knowing anything about politics.
Power of combining
Question: What is the ratio of students per teacher in
primary schools in Romania?
Judge Estimate Error
1 18 .5
2 19 1.5
Typical error of individual estimate 1
Combined estimate 18.5 1
Error reduction through combining 0%
Judge Estimate Error
1 18 .5
2 16 1.5
Typical error of individual estimate 1
Combined estimate 17 0.5
Error reduction through combining 50%
Procedure and conditions for combining forecasts
Procedure:
Mechanically combine forecasts equal weights
(unless you have strong evidence for differential weights)
Conditions:
1. Several forecasts available
2. Uncertainty about which forecasts is most accurate
(although combing is often beneficial even when the best
method is known beforehand)
Conditions for when combining is most beneficial:
1. Different forecasting methods are available
2. Forecasts rely upon different data
Benefits of combining
1. Improves accuracy
2. Avoids large errors
3. Provides an additional assessment of
uncertainty
4. Can be used for nearly all forecasting
problems.
5. Simple to describe and apply.
Costs of combining
1. Requires expertise with various methods
2. Higher expenses with more methods
7
Prior research
Meta-analysis of 30 studies on combining: 12% error
reduction vs. error of typical component.
Recommendation: Combine forecasts from
different methods that use different
information
[Armstrong, 2001]
However, few studies have focused on the ex ante
conditions of when combining is most beneficial.
8
9
Polly’s
Components
Polly‘s components
Polls
IEM
prediction
market
Experts
Quantitative
models
10
Polly’s
Components
Polls
Problem:
• Polls often unreliable, especially
early in campaign
• Large differences in results of
individual polls conducted around
the same time
Polls
Within component Combining
11
Polly’s
Components
IEM
prediction
market
Within component Combining
• Polly’s prediction market: Iowa Electronic
Markets (IEM)
• 7-day rolling average of daily market
prices
• Adjust for overreactions of market
such as information cascades
IEM prediction market
12
Polly’s
components Experts
Within component Combining
• Survey of experts
• Assumptions: Experts possess
• Information from polls
• Knowledge about the effect of
debates, campaigns, etc.
Experts
13
Polly’s
components
Quantitative
models
Within combining Combining
 Models focus on 2 to 7 variables,
most often
 Incumbent‘s popularity
 State of economy
 Individual accuracy of models
varies across elections
Quantitative models
14
Mean error reduction
(93 days prior to
Election Day,
1992 to 2008)
Polly’s
components
Gains from combining within components
Polls IEM Experts Models
Within components Combining Combining Combining Combining
14% 9% 21%18%
Polly’s
components
Combining across components
Polls IEM Experts Models
Within components Combining Combining Combining Combining
Across components
Combining
(unweighted
average)
PollyVote-Prediction
Mean error reduction
(93 days prior to
Election Day,
1992 to 2008)
Polly’s
components
Gains from combining across components
Polls
(combined)
IEM
(combined)
Experts
(combined)
Models
(combined)
PollyVote-Prediction
50% 1% 32%43%
Mean error reduction
(93 days prior to
Election Day,
1992 to 2008)
Polly’s
components
Gains from combining within & across components
Typical
Poll
Original
IEM
Typical
Experts
Typical
Models
PollyVote-Prediction
58% 10% 58%52%
If combining forecasts is so useful,
why is it seldom used?
18
1. Managers do not believe combining helps
In four experiments with MBAs at INSEAD, most
subjects did not realize that the error of the average
forecast would be less than the error of the typical
forecast.
Most subjects thought that averaging forecasts would
yield average performance.
[Larrick & Soll, 2006]
19
2. Some forecasters mistakenly believe
they are combining properly
People often use unaided judgment to assign
differential weights to individual forecasts.
Informal combining is likely to be harmful as people
can select a forecast that suits their biases.
20
3. Managers, forecasters, and researchers are
persuaded by complexity
Simple models often predict complex problems better
than more complex ones.
[Hogarth, in press]
These findings are difficult to believe. There is a strong
belief that complex models are necessary to solve
complex problems.
21
4. Forecasters build reputation with extreme
forecasts
Forecasters do not want to get lost in the crowd.
More extreme forecasts usually gain more
attention and the media is more likely to report
them.
[Batchelor, 2007]
5. People mistakenly believe they can
identify the most accurate forecast
In a series of experiments, when given two
estimates as advice, most people chose one
instead of averaging them – and thereby reduced
accuracy.
[Soll & Larrick, 2009]
Why doesn’t the PollyVote
capture mass media attention?
The PollyVote varies little and, basically, is never
wrong. Thus, no entertainment value.
Instead of accuracy, voters want excitement – and
hope for their candidate.
24
Accuracy problem is solved for
major elections
PollyVote deviation averaged 0.4% for the 2004
and 2008 U.S. presidential elections and
substantial improvements are scheduled for
2012.
Polly is available to researchers and practitioners
for elections in the U.S., as well as in other
countries.
25
Applications of combining
All organizations can benefit from combining.
References
Armstrong, J. S. (2001). Combining forecasts. In: J. S. Armstrong (Ed.),
Principles of Forecasting: A Handbook for Researchers and
Practitioners, Norwell: Kluwer, pp.417-439.
Batchelor, R. (2007). Bias in macroeconomic forecasts, International
Journal of Forecasting, 23, 189-203.
Hogarth, R. (in press). When simple is hard to accept. In P. M. Todd, G.
Gigerenzer, & The ABC Research Group (Eds.), Ecological rationality:
Intelligence in the world. Oxford: Oxford University Press.
Larrick, R. P. & Soll, J. B. (2006). Intuitions about combining opinions:
Misappreciation of the averaging principle. Management Science,
52, 111-127.
Soll, J. B. & Larrick, R. P. (2009). Strategies for revising judgment: How
(and how well) people use others’ opinions, Journal of Experimental
Psychology: Learning, Memory, and Cognition, 35, 780-805.

More Related Content

What's hot

Errors In Spreadsheets Are Common
Errors In  Spreadsheets  Are  CommonErrors In  Spreadsheets  Are  Common
Errors In Spreadsheets Are Common
bgebreyes
 

What's hot (19)

IFPRI- Impact Surveys 1
IFPRI- Impact Surveys 1IFPRI- Impact Surveys 1
IFPRI- Impact Surveys 1
 
Impact Evaluation Overview
Impact Evaluation OverviewImpact Evaluation Overview
Impact Evaluation Overview
 
Impact evaluation an overview
Impact evaluation  an overviewImpact evaluation  an overview
Impact evaluation an overview
 
Or ch1
Or ch1Or ch1
Or ch1
 
How we can use impact evaluation to assure effective use of resources for dev...
How we can use impact evaluation to assure effective use of resources for dev...How we can use impact evaluation to assure effective use of resources for dev...
How we can use impact evaluation to assure effective use of resources for dev...
 
Decision-Making 101: Regulatory Impact Analysis
Decision-Making 101: Regulatory Impact Analysis Decision-Making 101: Regulatory Impact Analysis
Decision-Making 101: Regulatory Impact Analysis
 
What is impact evaluation?
What is impact evaluation?What is impact evaluation?
What is impact evaluation?
 
Expectations and benefits of utilizing social media tools in new product deve...
Expectations and benefits of utilizing social media tools in new product deve...Expectations and benefits of utilizing social media tools in new product deve...
Expectations and benefits of utilizing social media tools in new product deve...
 
The Role of Economic Evaluation and Cost-Effectiveness in Program Science
The Role of Economic Evaluation and Cost-Effectiveness in Program ScienceThe Role of Economic Evaluation and Cost-Effectiveness in Program Science
The Role of Economic Evaluation and Cost-Effectiveness in Program Science
 
Applying Weighted Linear Regression to Stress Testing for J.P. Morgan Chase &...
Applying Weighted Linear Regression to Stress Testing for J.P. Morgan Chase &...Applying Weighted Linear Regression to Stress Testing for J.P. Morgan Chase &...
Applying Weighted Linear Regression to Stress Testing for J.P. Morgan Chase &...
 
Alert 2017 lopreiato - a handover osce to measure skills in transitions of ...
Alert 2017   lopreiato - a handover osce to measure skills in transitions of ...Alert 2017   lopreiato - a handover osce to measure skills in transitions of ...
Alert 2017 lopreiato - a handover osce to measure skills in transitions of ...
 
environmental scanning
environmental scanningenvironmental scanning
environmental scanning
 
M&E for Social Service System Strengthening
M&E for Social Service System Strengthening M&E for Social Service System Strengthening
M&E for Social Service System Strengthening
 
Uop qnt 565 final exam guide 1 new
Uop qnt 565 final exam guide 1 newUop qnt 565 final exam guide 1 new
Uop qnt 565 final exam guide 1 new
 
Errors In Spreadsheets Are Common
Errors In  Spreadsheets  Are  CommonErrors In  Spreadsheets  Are  Common
Errors In Spreadsheets Are Common
 
The Project Impact Pathway
The Project Impact PathwayThe Project Impact Pathway
The Project Impact Pathway
 
How to Evaluate Complex Interventions
How to Evaluate Complex InterventionsHow to Evaluate Complex Interventions
How to Evaluate Complex Interventions
 
Errors Found in National Evaluation of UpwardBound- Postive Re-Analysis Results
Errors Found in National Evaluation of UpwardBound- Postive Re-Analysis ResultsErrors Found in National Evaluation of UpwardBound- Postive Re-Analysis Results
Errors Found in National Evaluation of UpwardBound- Postive Re-Analysis Results
 
Questionnaire2002
Questionnaire2002Questionnaire2002
Questionnaire2002
 

Similar to Forecasting elections from voters' perceptions of candidates' ability to handle issues

Who should be nominated to run in the 2012 U.S. presidential election?
Who should be nominated to run in the 2012 U.S. presidential election?Who should be nominated to run in the 2012 U.S. presidential election?
Who should be nominated to run in the 2012 U.S. presidential election?
agraefe
 
How NOT to Aggregrate Polling Data
How NOT to Aggregrate Polling DataHow NOT to Aggregrate Polling Data
How NOT to Aggregrate Polling Data
DataCards
 
Forecasting Elections from Voters’ Perceptions
Forecasting Elections from Voters’ Perceptions Forecasting Elections from Voters’ Perceptions
Forecasting Elections from Voters’ Perceptions
agraefe
 
Who should be nominated to run in the 2012 U.S. Presidential Election?
Who should be nominated to run in the 2012 U.S. Presidential Election?Who should be nominated to run in the 2012 U.S. Presidential Election?
Who should be nominated to run in the 2012 U.S. Presidential Election?
agraefe
 
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrimeePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
Cindy Howry, MS
 
ePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
ePRO_Presentation_BYOD Webinar_10Mar2016_FINALePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
ePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
jencrager
 
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docxRunning Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
toltonkendal
 
Spf overview(1)
Spf overview(1)Spf overview(1)
Spf overview(1)
progroup
 
Proposal writing resource the logframe approach
Proposal writing  resource   the logframe approachProposal writing  resource   the logframe approach
Proposal writing resource the logframe approach
tccafrica
 

Similar to Forecasting elections from voters' perceptions of candidates' ability to handle issues (20)

Who should be nominated to run in the 2012 U.S. presidential election?
Who should be nominated to run in the 2012 U.S. presidential election?Who should be nominated to run in the 2012 U.S. presidential election?
Who should be nominated to run in the 2012 U.S. presidential election?
 
How NOT to Aggregrate Polling Data
How NOT to Aggregrate Polling DataHow NOT to Aggregrate Polling Data
How NOT to Aggregrate Polling Data
 
Forecasting Elections from Voters’ Perceptions
Forecasting Elections from Voters’ Perceptions Forecasting Elections from Voters’ Perceptions
Forecasting Elections from Voters’ Perceptions
 
Evaluation: Lessons Learned for the Global Health Initiative
Evaluation: Lessons Learned for the Global Health InitiativeEvaluation: Lessons Learned for the Global Health Initiative
Evaluation: Lessons Learned for the Global Health Initiative
 
Who should be nominated to run in the 2012 U.S. Presidential Election?
Who should be nominated to run in the 2012 U.S. Presidential Election?Who should be nominated to run in the 2012 U.S. Presidential Election?
Who should be nominated to run in the 2012 U.S. Presidential Election?
 
A Primer For Applying Propensity-Score Matching
A Primer For Applying Propensity-Score MatchingA Primer For Applying Propensity-Score Matching
A Primer For Applying Propensity-Score Matching
 
first-batch-me-training.pptx
first-batch-me-training.pptxfirst-batch-me-training.pptx
first-batch-me-training.pptx
 
Improving inferences from poor quality samples
Improving inferences from poor quality samplesImproving inferences from poor quality samples
Improving inferences from poor quality samples
 
How Do We Evaluate That? Evaluation in the Uncontrolled World
How Do We Evaluate That? Evaluation in the Uncontrolled WorldHow Do We Evaluate That? Evaluation in the Uncontrolled World
How Do We Evaluate That? Evaluation in the Uncontrolled World
 
Key Issues in Impact Evaluation: A MEET and GEMNet-Health Virtual Event
Key Issues in Impact Evaluation: A MEET and GEMNet-Health Virtual EventKey Issues in Impact Evaluation: A MEET and GEMNet-Health Virtual Event
Key Issues in Impact Evaluation: A MEET and GEMNet-Health Virtual Event
 
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrimeePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
ePRO_Presentation_BYOD Webinar_5 Final 9 March 2016 YPrime
 
April Webinar: Sample Balancing in 2012
April Webinar: Sample Balancing in 2012April Webinar: Sample Balancing in 2012
April Webinar: Sample Balancing in 2012
 
Harnessing Mobile Technology to Draw Insights from Health Care Professionals ...
Harnessing Mobile Technology to Draw Insights from Health Care Professionals ...Harnessing Mobile Technology to Draw Insights from Health Care Professionals ...
Harnessing Mobile Technology to Draw Insights from Health Care Professionals ...
 
ePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
ePRO_Presentation_BYOD Webinar_10Mar2016_FINALePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
ePRO_Presentation_BYOD Webinar_10Mar2016_FINAL
 
Research Report
Research ReportResearch Report
Research Report
 
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docxRunning Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
Running Head REPLY TO OPINION 5.1 FOR KIMBRILEE SCHMITZ 1REPLY.docx
 
q method research (1).pptx
q method research (1).pptxq method research (1).pptx
q method research (1).pptx
 
Spf overview(1)
Spf overview(1)Spf overview(1)
Spf overview(1)
 
Proposal writing resource the logframe approach
Proposal writing  resource   the logframe approachProposal writing  resource   the logframe approach
Proposal writing resource the logframe approach
 
Prediciting happiness from mobile app survey data
Prediciting happiness from mobile app survey dataPrediciting happiness from mobile app survey data
Prediciting happiness from mobile app survey data
 

Recently uploaded

Salient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functionsSalient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functions
KarakKing
 
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
ZurliaSoop
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
heathfieldcps1
 

Recently uploaded (20)

Salient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functionsSalient Features of India constitution especially power and functions
Salient Features of India constitution especially power and functions
 
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
Jual Obat Aborsi Hongkong ( Asli No.1 ) 085657271886 Obat Penggugur Kandungan...
 
Graduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - EnglishGraduate Outcomes Presentation Slides - English
Graduate Outcomes Presentation Slides - English
 
Sociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning ExhibitSociology 101 Demonstration of Learning Exhibit
Sociology 101 Demonstration of Learning Exhibit
 
Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...Making communications land - Are they received and understood as intended? we...
Making communications land - Are they received and understood as intended? we...
 
FSB Advising Checklist - Orientation 2024
FSB Advising Checklist - Orientation 2024FSB Advising Checklist - Orientation 2024
FSB Advising Checklist - Orientation 2024
 
How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17How to Give a Domain for a Field in Odoo 17
How to Give a Domain for a Field in Odoo 17
 
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
Beyond_Borders_Understanding_Anime_and_Manga_Fandom_A_Comprehensive_Audience_...
 
Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024Mehran University Newsletter Vol-X, Issue-I, 2024
Mehran University Newsletter Vol-X, Issue-I, 2024
 
ICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptxICT Role in 21st Century Education & its Challenges.pptx
ICT Role in 21st Century Education & its Challenges.pptx
 
Interdisciplinary_Insights_Data_Collection_Methods.pptx
Interdisciplinary_Insights_Data_Collection_Methods.pptxInterdisciplinary_Insights_Data_Collection_Methods.pptx
Interdisciplinary_Insights_Data_Collection_Methods.pptx
 
Jamworks pilot and AI at Jisc (20/03/2024)
Jamworks pilot and AI at Jisc (20/03/2024)Jamworks pilot and AI at Jisc (20/03/2024)
Jamworks pilot and AI at Jisc (20/03/2024)
 
General Principles of Intellectual Property: Concepts of Intellectual Proper...
General Principles of Intellectual Property: Concepts of Intellectual  Proper...General Principles of Intellectual Property: Concepts of Intellectual  Proper...
General Principles of Intellectual Property: Concepts of Intellectual Proper...
 
Towards a code of practice for AI in AT.pptx
Towards a code of practice for AI in AT.pptxTowards a code of practice for AI in AT.pptx
Towards a code of practice for AI in AT.pptx
 
Unit 3 Emotional Intelligence and Spiritual Intelligence.pdf
Unit 3 Emotional Intelligence and Spiritual Intelligence.pdfUnit 3 Emotional Intelligence and Spiritual Intelligence.pdf
Unit 3 Emotional Intelligence and Spiritual Intelligence.pdf
 
Fostering Friendships - Enhancing Social Bonds in the Classroom
Fostering Friendships - Enhancing Social Bonds  in the ClassroomFostering Friendships - Enhancing Social Bonds  in the Classroom
Fostering Friendships - Enhancing Social Bonds in the Classroom
 
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
TỔNG ÔN TẬP THI VÀO LỚP 10 MÔN TIẾNG ANH NĂM HỌC 2023 - 2024 CÓ ĐÁP ÁN (NGỮ Â...
 
How to setup Pycharm environment for Odoo 17.pptx
How to setup Pycharm environment for Odoo 17.pptxHow to setup Pycharm environment for Odoo 17.pptx
How to setup Pycharm environment for Odoo 17.pptx
 
REMIFENTANIL: An Ultra short acting opioid.pptx
REMIFENTANIL: An Ultra short acting opioid.pptxREMIFENTANIL: An Ultra short acting opioid.pptx
REMIFENTANIL: An Ultra short acting opioid.pptx
 
The basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptxThe basics of sentences session 3pptx.pptx
The basics of sentences session 3pptx.pptx
 

Forecasting elections from voters' perceptions of candidates' ability to handle issues

  • 1. The PollyVote Combining forecasts for U.S. Presidential Elections Andreas Graefe, Karlsruhe Institute of Technology J. Scott Armstrong, Wharton School, University of Pennsylvania Randall Jones, Jr., University of Central Oklahoma Alfred Cuzán, University of West Florida The full paper to this talk can be downloaded at: tinyurl.com/combiningelections. Bucharest Dialogues on Expert Knowledge, Prediction, Forecasting: A Social Sciences Perspective November 21, 2010
  • 2. Background on the PollyVote project The PollyVote project was begun in 2003 to demonstrate the value of forecasting principles by applying them to election forecasting. The initial focus was on combining forecasts.
  • 3. Performance of the PollyVote The PollyVote combined forecasts to obtain highly accurate forecasts of U.S. Presidential Election outcomes: – Prospectively for 2004 and 2008 (MAE: 0.4 percentage points) – Retrospectively for 1992 to 2000 Across these five elections, the PollyVote was on average more accurate than each of its components: - Polls - Prediction markets - Experts - Statistical models Polly achieved this without knowing anything about politics.
  • 4. Power of combining Question: What is the ratio of students per teacher in primary schools in Romania? Judge Estimate Error 1 18 .5 2 19 1.5 Typical error of individual estimate 1 Combined estimate 18.5 1 Error reduction through combining 0% Judge Estimate Error 1 18 .5 2 16 1.5 Typical error of individual estimate 1 Combined estimate 17 0.5 Error reduction through combining 50%
  • 5. Procedure and conditions for combining forecasts Procedure: Mechanically combine forecasts equal weights (unless you have strong evidence for differential weights) Conditions: 1. Several forecasts available 2. Uncertainty about which forecasts is most accurate (although combing is often beneficial even when the best method is known beforehand) Conditions for when combining is most beneficial: 1. Different forecasting methods are available 2. Forecasts rely upon different data
  • 6. Benefits of combining 1. Improves accuracy 2. Avoids large errors 3. Provides an additional assessment of uncertainty 4. Can be used for nearly all forecasting problems. 5. Simple to describe and apply.
  • 7. Costs of combining 1. Requires expertise with various methods 2. Higher expenses with more methods 7
  • 8. Prior research Meta-analysis of 30 studies on combining: 12% error reduction vs. error of typical component. Recommendation: Combine forecasts from different methods that use different information [Armstrong, 2001] However, few studies have focused on the ex ante conditions of when combining is most beneficial. 8
  • 10. 10 Polly’s Components Polls Problem: • Polls often unreliable, especially early in campaign • Large differences in results of individual polls conducted around the same time Polls Within component Combining
  • 11. 11 Polly’s Components IEM prediction market Within component Combining • Polly’s prediction market: Iowa Electronic Markets (IEM) • 7-day rolling average of daily market prices • Adjust for overreactions of market such as information cascades IEM prediction market
  • 12. 12 Polly’s components Experts Within component Combining • Survey of experts • Assumptions: Experts possess • Information from polls • Knowledge about the effect of debates, campaigns, etc. Experts
  • 13. 13 Polly’s components Quantitative models Within combining Combining  Models focus on 2 to 7 variables, most often  Incumbent‘s popularity  State of economy  Individual accuracy of models varies across elections Quantitative models
  • 14. 14 Mean error reduction (93 days prior to Election Day, 1992 to 2008) Polly’s components Gains from combining within components Polls IEM Experts Models Within components Combining Combining Combining Combining 14% 9% 21%18%
  • 15. Polly’s components Combining across components Polls IEM Experts Models Within components Combining Combining Combining Combining Across components Combining (unweighted average) PollyVote-Prediction
  • 16. Mean error reduction (93 days prior to Election Day, 1992 to 2008) Polly’s components Gains from combining across components Polls (combined) IEM (combined) Experts (combined) Models (combined) PollyVote-Prediction 50% 1% 32%43%
  • 17. Mean error reduction (93 days prior to Election Day, 1992 to 2008) Polly’s components Gains from combining within & across components Typical Poll Original IEM Typical Experts Typical Models PollyVote-Prediction 58% 10% 58%52%
  • 18. If combining forecasts is so useful, why is it seldom used? 18
  • 19. 1. Managers do not believe combining helps In four experiments with MBAs at INSEAD, most subjects did not realize that the error of the average forecast would be less than the error of the typical forecast. Most subjects thought that averaging forecasts would yield average performance. [Larrick & Soll, 2006] 19
  • 20. 2. Some forecasters mistakenly believe they are combining properly People often use unaided judgment to assign differential weights to individual forecasts. Informal combining is likely to be harmful as people can select a forecast that suits their biases. 20
  • 21. 3. Managers, forecasters, and researchers are persuaded by complexity Simple models often predict complex problems better than more complex ones. [Hogarth, in press] These findings are difficult to believe. There is a strong belief that complex models are necessary to solve complex problems. 21
  • 22. 4. Forecasters build reputation with extreme forecasts Forecasters do not want to get lost in the crowd. More extreme forecasts usually gain more attention and the media is more likely to report them. [Batchelor, 2007]
  • 23. 5. People mistakenly believe they can identify the most accurate forecast In a series of experiments, when given two estimates as advice, most people chose one instead of averaging them – and thereby reduced accuracy. [Soll & Larrick, 2009]
  • 24. Why doesn’t the PollyVote capture mass media attention? The PollyVote varies little and, basically, is never wrong. Thus, no entertainment value. Instead of accuracy, voters want excitement – and hope for their candidate. 24
  • 25. Accuracy problem is solved for major elections PollyVote deviation averaged 0.4% for the 2004 and 2008 U.S. presidential elections and substantial improvements are scheduled for 2012. Polly is available to researchers and practitioners for elections in the U.S., as well as in other countries. 25
  • 26. Applications of combining All organizations can benefit from combining.
  • 27. References Armstrong, J. S. (2001). Combining forecasts. In: J. S. Armstrong (Ed.), Principles of Forecasting: A Handbook for Researchers and Practitioners, Norwell: Kluwer, pp.417-439. Batchelor, R. (2007). Bias in macroeconomic forecasts, International Journal of Forecasting, 23, 189-203. Hogarth, R. (in press). When simple is hard to accept. In P. M. Todd, G. Gigerenzer, & The ABC Research Group (Eds.), Ecological rationality: Intelligence in the world. Oxford: Oxford University Press. Larrick, R. P. & Soll, J. B. (2006). Intuitions about combining opinions: Misappreciation of the averaging principle. Management Science, 52, 111-127. Soll, J. B. & Larrick, R. P. (2009). Strategies for revising judgment: How (and how well) people use others’ opinions, Journal of Experimental Psychology: Learning, Memory, and Cognition, 35, 780-805.