Transcript Forecasting Elections

Forecasting Elections
POL 3070
26.11.2007
Forecasting Models
• Aim to accurately predict the results of an election, before the
election is held, identifying the winning candidate and the share
of the vote he or she received.
• Models can be very similar in the variables they use to
forecast, although they are different either by the type of
variables they use or the way they interpret those variables.
• Some models have been better at forecasting than others;
somewhere between 0-3 percentage points is seen as a good
level of accuracy.
• Nearly all presidential models have been based on presidential
elections from 1948 onwards.
Abramowitz Model (1996)
Abramowitz Model (1996), uses three predictors:
1. Incumbent president’s approval rating mid-year
2. Annual rate of growth of real GDP during the first 2 quarters of the
election year
3. Length of time the president’s party has held office in the White House
• Time factor distinguishes his model from others (according to himself)
• Believes his model is able to forecast the result in early August with
greater accuracy than most final pre-election polls
• Some models have attempted to predict the popular vote in each
state, then combine the state results (Rosenstone, 1983; Campbell,
1992). He is looking at the national popular vote. Believes that this is
a very accurate model to use when predicting. <1% error
• “The presidential outcome is fundamentally a referendum
on the performance of the incumbent president”
(Abramowitz, 1996)
• State of the economy is crucial, albeit the economic
conditions have a delayed influence
• Positive value on alternation of power – “Time for a
change”
• Often the result of the election is predetermined before the
official campaign gets under way
• Campaigns do not play much of a role
• Independent and 3rd party candidates ignored (n.b. Ralph
Nader 2000)
Key Findings:
• 1 point increase in president’s mid-year approval = 0.2 point increase in the vote
• 1 point increase in real GDP growth in first 2 quarters = 0.8
• A party that has been in power for 8+ years suffers a penalty of 4.4 points
compared to a party that has been in power for 4 years
• Only 2 attempts in 7 has a party held onto the presidency after 2 successive terms
(1948 & 1988)
• In 5 attempts where a party was seeking a second term in office, 4 were successful,
1980 being the exception
• His model is fairly accurate from 0.2 points to 3 points. Absolute error of 1.4 points
compared to Gallup 3 points
• Questions the influence of campaigning and money
• Natural tendency towards alternation in power
• Predicted that the Republicans had a good chance of winning in 2000.
2000 Presidential Election – (Campbell, 2001)
•
2000 election was not good for forecasting models (Campbell, 2001)
•
Most models from the 1996 election remained largely unchanged
•
Most predicted a win for Gore (popular vote). In reality none were
within 2.5 points, some more than 5 points off
•
Why so bad?
1. Little experience of open-seat presidential elections. Only 4 elections
since 1948 has this happened (‘52, ‘60, ‘68 & ‘88)
2. Economy was sound
3. Time for a change – Republicans united, Democrats took power for
granted
4. Campaigns of the candidates
2000 Presidential Elections – (Wlezien, 2001)
• Everyone was expecting Bush to win
• Outcome becomes clearer in the polls as the election cycle evolves
• Economic performance features in almost every model, although the form
this takes can vary (i.e. Lewis-Beck & Tien, 2005). Other variables can also
include primary performance.
• August – forecasters thought Gore would win 52.8 – 60.3. On average
many were out by 5.8
• Time for a change argument
• Nader factor
• Gore not getting the full credit for economic performance
• Gore’s campaign; moving towards the left
Conclusions
• No magic model for forecasting elections
• Models need updating
• Unaccountable factors – 2000 election in particular
• Difficult to establish how much of an influence each variable really has
• Role of money to spend on campaigning?
• Significance of election models used on data from 1948 onwards
• Can you accurately forecast an election?
•What variable is the most significant?
• Who will win the 2008 presidential election?
• Are these models applicable to the UK?