Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
I
Andrew Gelman, Department of Statistics and
Department of Political Science, Columbia University
collaborators:
I
I
I
I
I
I
Joseph Bafumi, Dept. of Political Science, Columbia
David Park, Dept. of Political Science, Washington University
Noah Kaplan, Dept. of Political Science, Texas A&M
Gary King, Dept. of Political Science, Harvard
Jonathan Katz, Div. of Social Science, Caltech
Aaron Edlin, Dept. of Economics, UC Berkeley
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Polls and Presidential elections
I
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
I
How are pre-election polls conducted?
How are Presidential elections forecasted?
I
Does the Electoral College favor one party or another?
I
What is the probability that the election is tied?
I
What is the probability that your vote makes a difference?
I
When and why is it rational to vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
First topic:
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Presidential polls fluctuate wildly
I
But the candidates’ vote shares can be accurately forecast
(within a few percentage points) months before the election
I
If voters are so fickle, how can they be predicted?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
First topic:
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Presidential polls fluctuate wildly
I
But the candidates’ vote shares can be accurately forecast
(within a few percentage points) months before the election
I
If voters are so fickle, how can they be predicted?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
First topic:
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Presidential polls fluctuate wildly
I
But the candidates’ vote shares can be accurately forecast
(within a few percentage points) months before the election
I
If voters are so fickle, how can they be predicted?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
First topic:
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Presidential polls fluctuate wildly
I
But the candidates’ vote shares can be accurately forecast
(within a few percentage points) months before the election
I
If voters are so fickle, how can they be predicted?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Pre-election polls
I
ABC, CBS, Gallup, etc.
I
Mostly random-digit-dial surveys (but other countries use
face-to-face interviews)
I
Response rates below 30%
I
Over/undersample by sex, ethnicity, age, education, . . .
I
Weighting to adjust to Census
I
Can estimate state-level opinion from national polls using
Bayesian hierarchical modeling
I
Also state polls, academic polls, internet polls, . . .
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do we trust the polls?
I
A poll is a snapshot, not a forecast
I
Polls taken just before the election are usually pretty accurate
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do we trust the polls?
I
A poll is a snapshot, not a forecast
I
Polls taken just before the election are usually pretty accurate
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Republican share of two-party preference in pre-election polls
-100
-50
-150
-100
-50
0.6
0.4
0.2
Support for Republican candidate
0.6
-200
0
-200
-150
-100
-50
1976
1972
-100
-50
0
-200
-150
-100
-50
0.4
0.2
0.6
0.4
0.2
0.6
-150
0
-200
-150
-100
-50
1968
1964
1960
-50
0
-200
-150
Andrew Gelman
-100
-50
0
0.4
0.2
0.6
0.4
0.2
0.6
0.4
-100
0
0.6
Days before election
Support for Republican candidate
Days before election
Support for Republican candidate
Days before election
-150
0
0.6
1980
Support for Republican candidate
Days before election
Support for Republican candidate
Days before election
0.2
-200
0.4
0
Days before election
0.4
-200
1984
0.2
Support for Republican candidate
0.6
0.4
-150
0.2
Support for Republican candidate
-200
Support for Republican candidate
1988
0.2
Support for Republican candidate
1992
-200
-150
-100
Polls and Presidential Elections
-50
0
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential election forecasting
I
4 years ahead, 2 years ahead, 1 year, 4 months, 2 months, 2
weeks, . . . election night
I
2 months before election: candidates have been chosen, final
campaign still to go
I
Predict election outcome using polls, national economy,
incumbency, candidates’ ideologies, state economic and
political trends, home states, home regions
I
Fit model using elections since 1948, use to predict current
election
I
Errors at state, regional, national levels
I
Can also use this model on election night (Bayesian inference,
combine with exit poll and vote data)
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Forecast for 2004
I
National forecast from Wlezien and Erikson based on polls
and economic indicators: 51.7% (±2.5%) for Bush
I
Take relative positions of states from 2000, correct for
home-state effects
I
(Better state-level forecasting is possible)
I
Add forecasting errors at state, regional, and national levels
Forecast for each state and entire U.S.
I
I
I
I
66% chance Bush wins Florida
75% chance Bush wins majority of U.S. popular vote
U.S. electoral vote: 320 for Bush (probably between 280 and
360), 80% chance Bush wins election
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Why do Presidential election campaign polls vary so much
when votes are so predictable?
I
Question wording?
I
I
I
I
Some surveys ask “If the election were held tomorrow. . . ”
Some surveys ask “Which candidate will you definitely vote
for?”
Different nonresponse rates but no effect on % support for
each candidate
Undecided voters?
I
No trends during the campaign
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Uniform partisan swing
I
Who’s doing the fluctuating?
I
All groups move together
I
Even during each Democratic and Republican conventions, all
groups move in the same direction (“uniform swing”)
Consider a voter polled before the Republican convention:
I
I
I
“Now I plan to vote for Kerry, but in a week I plan to switch
to Bush.”
Voters do not (on average) discount for expected future events!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Presidential Support by Group
0.6
0.8
-50
0
-200
-150
-100
Race
Region
-50
0
-50
0
0.8
0.6
W
S
E
MW
0.2
0.4
0.6
Bush support
0.8
1.0
Days before election
0.4
0.0
non-wh
-100
-50
0
-200
-150
-100
Days before election
Days before election
Sex
Income
0.8
0.6
Bush support
0.8
1.0
-150
1.0
-200
0.2
high
mid-hi
mid-lo
low
0.0
0.0
0.2
M
F
0.6
Bush support
0.4
Bush support
-100
0.0
0.2
lib
Days before election
1.0
-150
white
0.4
mod
0.0
Dem
-200
Bush support
cons
0.2
Ind
0.4
0.4
0.6
0.8
Rep
0.0
0.2
Bush support
1.0
Ideology
1.0
Party
-200
-150
-100
-50 Gelman
0
Andrew
-200 and Presidential
-150
-100 Elections
-50
Polls
0
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Changes in Presidential campaign polls during
Democratic and Republican conventions, 1964-1992
(conventions in 1988 circled)
0.6
RR
d
R
d
d
0.5
R
d
R
0.4
d
R
d
R
0.3
Republican support
after the convention
R
d
0.3
d
0.4
0.5
0.6
Republican support
before the convention
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Moving toward a predictable outcome
I
I
You can predict a voter’s preference given “demographics”:
sex, ethnicity, age, education, political ideology, party
identification
This prediction improves as the campaign goes on
I
I
I
Fit model to a series of polls before the 2000 election
The coefficients for the predictors increase
The residual error of the model decreases
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Moving toward a predictable outcome
I
I
You can predict a voter’s preference given “demographics”:
sex, ethnicity, age, education, political ideology, party
identification
This prediction improves as the campaign goes on
I
I
I
Fit model to a series of polls before the 2000 election
The coefficients for the predictors increase
The residual error of the model decreases
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Moving toward a predictable outcome
I
I
You can predict a voter’s preference given “demographics”:
sex, ethnicity, age, education, political ideology, party
identification
This prediction improves as the campaign goes on
I
I
I
Fit model to a series of polls before the 2000 election
The coefficients for the predictors increase
The residual error of the model decreases
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Moving toward a predictable outcome
I
I
You can predict a voter’s preference given “demographics”:
sex, ethnicity, age, education, political ideology, party
identification
This prediction improves as the campaign goes on
I
I
I
Fit model to a series of polls before the 2000 election
The coefficients for the predictors increase
The residual error of the model decreases
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Moving toward a predictable outcome
I
I
You can predict a voter’s preference given “demographics”:
sex, ethnicity, age, education, political ideology, party
identification
This prediction improves as the campaign goes on
I
I
I
Fit model to a series of polls before the 2000 election
The coefficients for the predictors increase
The residual error of the model decreases
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
Increasing coefficients for fundamental predictors
Female
Education
Age 30−44
50
150
100
50
0
150
100
50
0.5
−0.5
Logistic Coefficient
200
−1.5
0.2
−0.2
Logistic Coefficient
200
−0.6
0.0
0
0
200
150
100
50
Age 65+
Income
Party ID
50
0
200
150
100
50
Ideology
South
50
0
100
50
1.0
0.0
0.2
150
0
Days Before Election
0.5
0.0
Logistic Coefficient
100
200
−0.5
0.8
0.4
150
Days Before Election
0
1.0
Days Before Election
−0.2
Logistic Coefficient
−0.5
Logistic Coefficient
100
−1.5
0.5
150
Days Before Election
200
150
100
50
0
Days Before Election
Andrew Gelman
0
2.0
Age 45−64
Logistic Coefficient
Days Before Election
0.5
Days Before Election
1.5
Days Before Election
0.0
200
−0.5
Logistic Coefficient
100
1.2
200
Logistic Coefficient
150
Days Before Election
−1.5 −0.5
Logistic Coefficient
200
−1.0
0
−2
−4
Logistic Coefficient
0.6
Black
Polls and Presidential Elections
200
150
100
50
Days Before Election
0
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
0.5
0.4
Deviance per Observation
0.6
0.7
Decreasing residual error of model of individual vote
200
150
100
50
Days Before Election
Andrew Gelman
Polls and Presidential Elections
0
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
0.85
0.80
0.75
0.70
0.65
0.60
0.55
Coefficient of 1996 Fundamentals Predicting 2000 Vote Choice
0.90
Increasing predictive power for new data
200
150
100
50
Days Before Election
Andrew Gelman
Polls and Presidential Elections
0
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
The random-walk and mean-reversion models
I
Random-walk model: voters are bounced around by campaign
events, then the election comes
I
Mean-reversion model: voters will mostly end up where
predicted. It just takes them awhile to get there
I
Mean-reversion model fits the data better, also explains why
polls vary so much when elections are so predictable
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
The random-walk and mean-reversion models
I
Random-walk model: voters are bounced around by campaign
events, then the election comes
I
Mean-reversion model: voters will mostly end up where
predicted. It just takes them awhile to get there
I
Mean-reversion model fits the data better, also explains why
polls vary so much when elections are so predictable
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
How are pre-election polls conducted?
How are Presidential elections forecasted?
Why do Presidential election campaign polls vary so much . . .
The random-walk and mean-reversion models
I
Random-walk model: voters are bounced around by campaign
events, then the election comes
I
Mean-reversion model: voters will mostly end up where
predicted. It just takes them awhile to get there
I
Mean-reversion model fits the data better, also explains why
polls vary so much when elections are so predictable
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
Does the Electoral College favor one party or the other?
I
To find out, use state-by-state forecasts for each election year
I
Forecasts are uncertain (probabilistic)
I
What is Pr(Democrats win in electoral college), if they receive
X% of the popular vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
Does the Electoral College favor one party or the other?
I
To find out, use state-by-state forecasts for each election year
I
Forecasts are uncertain (probabilistic)
I
What is Pr(Democrats win in electoral college), if they receive
X% of the popular vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
Does the Electoral College favor one party or the other?
I
To find out, use state-by-state forecasts for each election year
I
Forecasts are uncertain (probabilistic)
I
What is Pr(Democrats win in electoral college), if they receive
X% of the popular vote?
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Total vote share
0.50
0.52
Percentage of the popular vote required for Democrats to
have a given chance of winning the Electoral College
95% chance
50% chance
0.48
5% chance
1950
1960
1970
1980
year
Andrew Gelman
1990
Polls and Presidential Elections
2000
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that the Electoral College will be tied?
Use the state-by-state forecast for 2004
Estimated probability is 0.05 (1 in 200)
Combinatorics is not an issue
I
I
#states is large
Central Limit Theorem takes over
Probability
0.004
0.008
I
I
0.000
I
100
200
300
Electoral votes for Bush
Andrew Gelman
400
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that the Electoral College will be tied?
Use the state-by-state forecast for 2004
Estimated probability is 0.05 (1 in 200)
Combinatorics is not an issue
I
I
#states is large
Central Limit Theorem takes over
Probability
0.004
0.008
I
I
0.000
I
100
200
300
Electoral votes for Bush
Andrew Gelman
400
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that the Electoral College will be tied?
Use the state-by-state forecast for 2004
Estimated probability is 0.05 (1 in 200)
Combinatorics is not an issue
I
I
#states is large
Central Limit Theorem takes over
Probability
0.004
0.008
I
I
0.000
I
100
200
300
Electoral votes for Bush
Andrew Gelman
400
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that the Electoral College will be tied?
Use the state-by-state forecast for 2004
Estimated probability is 0.05 (1 in 200)
Combinatorics is not an issue
I
I
#states is large
Central Limit Theorem takes over
Probability
0.004
0.008
I
I
0.000
I
100
200
300
Electoral votes for Bush
Andrew Gelman
400
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that the Electoral College will be tied?
Use the state-by-state forecast for 2004
Estimated probability is 0.05 (1 in 200)
Combinatorics is not an issue
I
I
#states is large
Central Limit Theorem takes over
Probability
0.004
0.008
I
I
0.000
I
100
200
300
Electoral votes for Bush
Andrew Gelman
400
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Next topic:
What is the chance that your vote will be decisive?
I
I
Pr(your state is tied) × Pr(your state’s electoral votes are
needed, given that your state is tied)
A state with N voters and E electoral votes
I
I
I
I
approx. Pr(your state is tied) ∝ 1/N
approx. Pr(your state’s electoral votes are needed) ∝ E
approx. Pr(your vote is decisive) ∝ E /N
Higher for small states and states closer to the national
average
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
What is the chance that your vote will be decisive?
Pr (individual voter is decisive)
4*10^-8
10^-7
NM
WV
AR
AZ
VTNV
NH
IA
ME
DE
OH
TN
OR
FL
PA
MO
WI
MI
MN
LA
WA
VA
CO
HI
SD
GA
NC
MS
SC
KYAL
ND
AK
MT
CA
NJ
CT MD
WYRI
IL
IN
KS OK
NE
0
ID
DC UT
0
TXNY
MA
10
20
30
40
Number of electoral votes for the state
Andrew Gelman
Polls and Presidential Elections
50
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
A mathematical digression
I
I
We said Pr(your state is tied) ∝ 1/N
Simple “binomial model” of random votes
I
I
√
Mean proportion
√ of votes for Democrat is 0.5, sd is 0.5/ N
Pr(tie) ∝ 1/ N
I
Binomial model implies that elections in large states are much
closer than in small states
I
Binomial model does not fit actual election data!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
0.0
Proportional vote differential
0.2
0.4
0.6
0.8
Historical Pres. elections by state: vote margins vs. N
.
.
.
.
.
.
. ..
.
.
.
.
.
... .. ...
.
... .. .. . . .. .
.
.
.
..... .. . . . .
.
.
.
. . .. ... .
.
.
.
.
... . . . . .
.. .
.
. ..... . .. .... . . .
.
.
. ........... . .. . .. . . . ...
.
.. .... . .. .
.
.
.
.
........ .. .. .. .
.
.
.
.
.
.
.
.
.
.
.
.
...... . ... . ... . .
.
.
.
.
.
..
.... .. . .. .. . .... . . . . ..
. .
.. .. . ... .. .. .... ..
.
.
......... .. . . .... ... . .. . . . .
.
.
.
.. . .
..... . . ...... . .
.. ...... . ... ... ..... .. . .. .
.. . .
.
.... .. . ..... .... ..... . ... . . . . .
. .. . ..
. .
.... ....... ... ....... ... . . . ..
.
.. .
.
.
.
.
.
.
.
. . .. ..
.
.
.
. ..
.. . .
.
.............. ... .......... ........... . ... . .. .. . . . .. .. .... . .
.. .
. . .. . .
. . . .
0
.
.
.
.
.
.
.
.
lowess fit
alpha = -0.16
alpha = -0.5
2*10^6
6*10^6
10^7
Total vote for the two leading candidates
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Tie in the electoral vote
Tie in the popular vote
Other electoral systems: vote margins vs. N
.
.
.
alpha = -0.23
lowess fit
. alpha = -0.5
0
4*10^6
8*10^6
1.2*10^7
Total vote for the two leading candidates
Proportional vote differential
0.2
0.4
0.6
0.8
.
0.0
Proportional vote differential
0.2
0.4
0.6
0.8
0.0
.
..
..
. .
.
. ..
. . ..
...
.
.
... . . . .
.
.. .
.
.... .....
.
. . ..
. .
.
. .
... . . .. .... .
.
.
.
......
..
.. .
.
.
.
.. . .. .
.
.
.. . .
. .. .... . .. . .
.
.
..
.. . . . . .
.
.
....... .. ... . . ..
.. .
. . .... . . ..... .
..... .. ..... . ...... . . .
.
.
.
.
. . ... ... .
lowess fit
.
0.0
alpha = -0.5
Proportional vote differential
0.2
0.4
0.6
0.8
alpha = 0.06
.
........... .. . .. .. . . . .
...............
........................................ ... ......... ... .. . . .
................................ ......... .. . . ....
.
......... ... ............... . .. . . .
.......................
............. ........................................... ..... .... .. .. . .. .. . .
.................................................................... ....... ... ...... .. .. . . . .
.
............................................ ....................... .. . ..... .
................. ........................................... .. . .. .. .. . . .
................................................................................................................................. ... . .
.
.. ... .. ... . .. . . . . .. ..
.............................................................................................................................................. .. . .. . .
.
.
...................................................................................................................................................................... .. . .
................... . .. . .
....................................................................................................................................
. ..... .. ... ... . ............. . . .... ......... ..... ..
............................................................................................................................................................................................................... .. . .
..................................................................................................................................................................................... ..................... . . ..
............................................................................................................................................................................. ....
......................................................................................................................................
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
............................................................................................................
...... ...... . . .. .
. ...................................................................
..... ............................ .. . . .
..................................
....... .......................................................................................................... . . .. ... .
..
..................................................................................................................................................
.
........................................................................... ... .. .
...............
.......................................................................................................................................................................................
............... . . .... . .
.......................................
.
.........................
..........
......................................................................................................................................................... .... .
.................
.
.
.
..
.
.
.
..
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
................ ......... . .. . .
......................................................................................................
..
.
...................... . ............................................................................. ... ...
............................................................................................................................................................................................................................................................................... ....... .. . .. .
................................................................................................................................................................. ........ .. .. ..
........ ........ ......... .............................................. .. . .
.................................................
..... .. . .... ....... . .. .. . .
............................................................................................................................................................................................
.
. ..................................................................................................................
..................................
.........
........................
..................
.. ..
. . .. . . . .. .
... ...................
.....................
...
........
.............
...................................................................................................................................... ... .... .
.............................................................
.....................
..........
... ................................................. ........ . .. .. .
...............................................
..................................................................
.............................
...................................................... . ...
..................................
.............
.......
..... .... ...........
.....
...................
..
............................................................................................................................................................. ..... . .
.......
lowess fit
alpha = -0.11
. .
alpha = -0.5
.
0 50000
150000
250000
Total vote for the two leading candidates
0
200000
400000
600000
Total vote for the two leading candidates
U.S. statewide offices
European national elections
U.S. Senate elections
.
.
.
U.S. Congressional elections
.
.
.
. ..
.
..
. .. . . .
..
... .
.. .. .
.
. . .. . .... .
... ... .
.. .
. . . .. . . .
.
.
..
.... ........
... . . .....
... . . . ..
..... .. ....... .. . ....
.. . .
.
. . ... ... .. ..
......... .......... . .. ..
.. ... . . .
. .
...... . .
.
.
.
..
. .
.
. .
.
.. ..
.
.
.. ..
.
. ..
.
.. .
.
.
.
alpha = -0.01
.
.
lowess fit
= -0.5
Proportional vote differential
0.2
0.4
0.6
0.8
0
20000
60000
100000
Total vote for the two leading candidates
. . . ..
.
..
.
... ..... . . . .. . . . ..
.. .
.... . .. . . .
. .
. .........
. ..
. .. .
.. ......... . .... . . .
....... ..... ... .. .. . .. .... ..
. .. ....... . . . . .
. . .
.
... .
. ............. .... .. ... . ....
. .
. ............ ... ..... .
.. ....... ...... . .. . . . . . . . ..
. . ...... . ...... .... . .
......................... . . . ...... . . . .. . .. . . . .
.
..
.
............................................ .... .. . . .. . . . ... . ..
.. .
. . .. ...... ... . .. . .
....................................... ... ............ ..... . . . . .. .. .
................... ..... .. ..... . .... .. .... . . ...... .
. ............................................................. ......... ......... . . ..........
..... .................................... . .... .... ... . . .
. .
................................... .. . ... . . ..
.... ..... . .. .. . . . . . .. . ..
........................................................... ..... ........... .. ... ..... ... .......
. .
.................................. ....... .. . .....
.. . . .
............................... . ... ........ . . .. ...... .
.............................................. ... .... ...... .. . . .. . . . .
.......................................................................... .......... .. . . .. . .. . ..
. .
...................... .............. .. .. .... . . ... . . . .
............................ . .... ...... ... . . . ..
.........................
.
.
.
.
.
.
.
.
.
.
.
...................................... ... . . . . . .
.....................................
........ . ....... . .. .... . .. . . . .
.
.........................................
. .
.
. .. ...... .. . .. .... . . . . . . .. . .
.....................
.......................................................... .. ..... .. .... ..... . .... . ..
.
.. .. ... .. . .
..
............................................................................... ...... ..... ..... .... . .
. . alpha
0
4*10^6
8*10^6
1.4*10^7
Total vote for the two leading candidates
Andrew Gelman
0.0
Proportional vote differential
0.2
0.4
0.6
0.8
U.S. state senate elections
0.0
0.0
Proportional vote differential
0.2
0.4
0.6
0.8
U.S. state house elections
. ... .... . .. .. .. . . .. .. . . ... . ..
. ..
. . .
.... ............ .. .. ........... . .. . .... . .
. . . . .. . . .
.
.. .... . .. . ... . .... . . .
..... ... ... ....... .. .................. ... . .. ..
.
.. . .. .
...... ..... .... ................ ..... .. ..... . . .. . .
.. ... ... . . ... .. ..... .. .
.
.
..................................... .................... ..... . .. . . .. . . .
.
.
. .
.
.
.
.
.
.
.
.
................ .............. .... .. ... . . ..
.
. .. .. .. .. . . .. .. ..
................. ................. ... ...
.
......................................................... ..................................... .... ...... ....... .. .. .. . .
. . ......... ........ ......... ....... . .
.. . .
................................................................................................................................. .... .. .... .. . .... .. .
. ..
. .. .. . .. .... . . . . .
.
........................................................................................................... ............ ............... .............. .... ..... .
................................................................................... ...... .......... ...... ........... ..
...............................................................
. . . .. . . . .. . ... . . .
.................................................................................................................... .... .. . .. .. .
........................................................................................................................................................ ........................ ....... .... ....... . .
.
........................................................................................................................................ .............. . ....... .
........................................
................................................................................................................. ................ . . . .
............................................. ............ .. . . . .. ..
.............................................................................................................
alpha = 0.08
.............................................................................................................................................................. ..... ........... ........... .. . .
...................................
........................................................ ............. . . .
. . . . ..
........
... .. .. ............
.. .. .. .. . .. . . . . . . ..
............................................................................................................................................................................. ................... ............. ... . .
.. . . . . . .
........ .. .. .. .
lowess fit
.
.............................................................................................................................................................................. ............... .. .. .....
.
.
.......................
. .............................................. ................. .. .. .. .... . . .. . .
........
.. .
.. ..... .. .... ... . . .... . ..
................................................
....................................................................................................................................................................... . .. . ... . ..
.............................
................................................................................................ .... .. . ...... ..... .. .
... .
.
....... . . ... . .... . . .... . ... ...
...............................................
. . ..
........................................................................... ....... .. . .. . .. .. .
......................................................................
alpha = -0.5
....................................................................................................................................................... .. ... .......... . . . . .
................
.
....................................................................................................................................................................................... .. .... .. . ... ..... . ..
.
. ..
................ .......... .. .... .... . . .. . . ..
.......
. .
..
.... .
. . ..
. .... . .
.
. ... ..
. .
..
.. .. .. ... .
. .. .
.. .... .. ..
. . ..
.. ..... .. .
. . .
. .... ..... .. ...
.. ....... ... . .
.... . .. .
. ..
....... .. ....... ....
.. . .. .
..... ..... . .
.
.
.
.
. .
.
.
. . ..
.
.
..
.
.
.
.
. .
.. .
.. .
..
.
. . ..
. .
.
.
. . .. ..
..
.. . .
.
.
. ..
.
alpha = -0.07
.
.
lowess fit
alpha = -0.5
0
10^7
3*10^7
Total vote for the two leading candidates
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Next topic:
When and why is it rational to vote?
I
I
The probability of your vote being decisive is about 1 in 10
million, so why vote?
Utility calculation: ∆U = pB − c
I
I
I
I
I
∆U = utility from voting
p = probability that your vote changes the election outcome
B = your benefit from your preferred candidate winning
c = net cost of voting
Suppose p = 10−7 and B = $1000. Then pB = 1/100 of a
cent!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: typical explanations
I
I
(Utility calculation: ∆U = pB − c)
Maybe p is overestimated
I
I
But even if p = 10−3 , the product pB is only $1!
Maybe the net “cost” c is negative
I
I
I
Voting is fun!
Or a “civic duty”: you feel bad if you didn’t vote
But then you have no motivation to increase the chance that
your candidate will win!
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Rationality and voting: our explanation
I
I
(Utility calculation: ∆U = pB − c)
B = Bself + αNBsoc
I
I
I
I
Bself = individual benefit of your candidate winning
Bsoc = your view of the average social benefit of your
candidate winning
α < 1 implying that you care less about other people than
yourself
α > 0 implying that you are not completely selfish
I
Now B is proportional to N, and so it can be rational to act
so as to improve your candidate’s chance of winning
I
Feedback keeps voter turnout stable: if turnout declines, then
p increases, and it becomes more compelling to vote
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections
Pre-election polls and forecasting
Partisan bias in the electoral college?
Probability the election is tied
When and why is it rational to vote?
Summary
I
Short-term political tactics aren’t so important—even if the
polls jump around
I
Elections are predictable given fundamental variables and
campaign resources
I
The Electoral College does not favor either party
I
The Electoral College favors voters in small states (so does
the U.S. Senate!)
I
National opinion moves in synch
I
Pr(your vote is decisive) ≈ 10−7
I
But voting is rational if you think your guy can make the
country a better place!
I
The rational reason for voting is altruistic
Andrew Gelman
Polls and Presidential Elections