Forecasting Presidential Elections

1
Forecasting Presidential Elections

• How can we best model presidential elections?
• Fun models
• Serious models

2
Forecasting Presidential elections

• Can citizens forecast the winner?
• National Election Study question, 1956-84, about
  69 pct of respondents correctly forecast the
  winner (median survey date Oct. 7 of election
  year)
• model of these predictions respondents tend to
  predict that their candidate will win strength
  of party ID, attention to news coverage not
  significantly related to forecasting success

3
Fun models

• Red Sox Model Red Sox success/failure in the
  World Series in presidential election year
  exactly predicts Democratic success/failure in
  the election
• 1904 Boston Pilgrims won AL pennant, but no
  World Series was held when NL champs (NY Giants)
  refused to play the AL champs. The Pilgrims had
  won the first official World Series in 1903.
  Teddy Roosevelt wins the presidency
• 1912 the Red Sox (renamed in 1907) beat the
  Giants Woodrow Wilson wins presidency
• 1916 Red Sox beat the Brooklyn Robins Wilson
  reelected
• low-powered test. Not enough obs.

4
The AL/Republican Alliance

• American League victory in the World Series is
  associated with Republican victory, National
  League victory with Democratic triumphs
  (1940-2000 data). Results are worse before 1940

Repub winner
Dem. winner
5
2
Am. League
?2 2.38, Pr 0.13
3
6
Natl League
5
Redskins Model

• Results of the last Redskins home game preceding
  the election predict the success/failure of the
  incumbent party
• since 1936, every time the Redskins won, so did
  the incumbent partys candidate every time they
  lost, so did the incumbent partys candidate.
  Perfect fit.

6
Campbells model

• Forecasting state Dem. 2-party vote shares,
  1948-88
• Key explanatory vars
• September polling data
• 2nd-quarter GNP growth X incumbent party
• Incumbent running for reelection
• State vote history (4- and 8-year lags of
  deviation from natl avg)
• 1st-quarter state GNP growth X incumbent party
• Avg error -- 3 pct points

7
This years models

• PS, October 2004 issue
• Earliest prediction Jan. 29, 2004 (Norpoth)
  Bush 54.7 pct of 2-party vote, 95 pct probable gt
  50 pct
• Median forecast (7 models) Bush 53.8 pct,
  ranging from 49.9 (Aug. 27, 2004) to 57.6 pct
  (May 21, 2004)
• Final pre-election Gallup poll, 1948-2000 has
  been /- 2.1 pct from actual outcome
• Current Electoral College predictors based on
  state polls range from high-probability Bush
  victory to high-probability Kerry victory.