Forecasting House Seats from
Generic Congressional Polls: The
2010 Midterm Election
Joseph Bafumi
Robert S. Erikson
Christopher Wlezien
The Model
• Building on successful prediction of the 2006
House seat division:
– Predict the midterm vote division from generic
poll result and party of the president.
– Forecast winner of each House race with
statistical models for open seat and incumbent
races.
– Incorporate uncertainty with 1000 simulations.
The Model: Step 1
• Predict the midterm vote division:
– As the election approaches, poll respondents
increasingly take into account the party of the
president when reporting their generic “vote.”
The Model: Step 1
66
94
02
46
0
10
10
5
20
50
98
-10
66
78
62
94
0
10
20
61-120 Days Before Election
02
46
-10
62
94
0
10
5
90
54
50
98
0
5
0
98
78
74
20
82
58
86
78
06 70
62
66
94
02
46
-5
10
82
58
0670
54
66
50
10
121-180 Days Before Election
Vote (% Dem. minus 50%)
Generic Vote (% Dem. minus 50%)
86
90
-10
0
10
20
31-60 Days Before Election
1-30 Days Before Election
0
54
98
94
50
62
66
02
46
0
10
Generic Vote (% Dem. minus 50%)
20
5
0
90
74
98
82
58
86
78
90 7006
54
62
66
50
94
02
46
-5
10
5
82 58
86
78
06 70
10
Generic Vote (% Dem. minus 50%)
Vote (% Dem. minus 50%)
Generic Vote (% Dem. minus 50%)
74
-5
54
02
46
58
06 90 70
Generic Vote (% Dem. minus 50%)
74
-10
82
86
0
62
Vote (% Dem. minus 50%)
78
50
98
-5
Vote (% Dem. minus 50%)
82 58
74
-5
10
0
5
86
709006
54
-10
Vote (% Dem. minus 50%)
181-240 Days Before Election
74
-5
Vote (% Dem. minus 50%)
241-300 Days Before Election
-10
0
10
Generic Vote (% Dem. minus 50%)
20
The Model: Step 1
• Predict the midterm vote division:
– Forecasts from these two variables are about
equally predictive regardless of when in the
election year the generic poll results are
measured.
The Model: Step 1
Table 4. CAMPAIGN DYNAMICS OVER VARYING TIME INTERVALS. Predicting Midterm
Congressional Vote from the Generic Ballot Poll Results at different times plus Presidential Party, 16
midterm elections 1946-2006
Dependent Variable = Democratic % of Actual Two–Party Vote in
November, minus 50%
(18)
241-300
(19)
181-240
(20)
121-180
(21)
61-120
(22)
31-60
(23)
1-30
Days Out
(Feb.)
Days Out
(April)
Days Out
(June)
Days Out
(August)
Days Out
(Sept.)
Days Out
(Oct.)
Generic Poll
Results
(% Dem. minus
50%)
0.46***
(0.08)
0.49***
(0.10)
0.50***
(0.10)
0.50***
(0.09)
0.52***
(0.07)
0.59***
(0.10)
Current
Presidential Party
(R=-1, D=+1)
-2.58***
(0.46)
-2.04**
(0.49)
-1.99***
(0.49)
-1.44**
(0.46)
-1.15**
(0.40)
-0.82
(0.50)
Intercept
-0.66
(0.68)
-0.92
(0.79)
-0.50
(0.73)
-0.46
(0.64)
-0.19
(0.51)
0.20
(0.56)
Adj. R2
.77
73
.73
.77
.83
.77
Root MSE
1.79
1.94
1.94
1.77
1.51
1.77
Note: Generic poll results and the vote are measured as the Democratic percent of the two-party vote minus
50 percent. Equation 18 is a repeat of equation 13 in Table 3.
* p < .05, ** p < .01, *** p < .001
The Model: Step 1
• Predict the midterm vote division:
– The pooled generic polls conducted 121-180 days
in advance of the 2010 election show a very close
division—49.1% Democratic and 50.9%
Republican.
– Once you take into account the “out-party” drift,
our election forecast is that the Democrats will
win 47.1% of the two-party vote and the
Republicans the remaining 52.9%.
The Model: Step 2
• Translate the vote into seats:
– For each simulated value of the national vote, we
need to simulate the outcome in 435
congressional districts.
– We use different equations for open and
incumbent seats.
The Model: Step 2
• Translate the vote into seats:
– Open seats:
Djk = -3.92 + 0.82 %Obamak + P + ej + uk_open
– Incumbent seats:
Djk = -5.79 + 0.98 Dem Vote (2008)k + 3.85 Froshk + P + ej
+ uk_incumbent
Where constants are adjusted to neutralize the 2008 vote swing and where P is the specific
national vote prediction measured with error (ej) around it.
0
10
Frequency
20
30
40
Results
160
180
200
220
240
Number of Dem Seats Won (1000 simulations)
Republican
260
Democrat
The Republicans win the majority of seats in 79% of the trials. On average, the Republicans
win 229 seats, twenty-three more than the Democrats and eleven more than the 218
needed for a majority. But the simulations yield considerable variation, with a 95 percent
confidence interval of 176 to 236.
Note of Caution
• Applying our model of the vote to 2010
assumes that the forces at work in 2010 are
unchanged from past midterm elections.
– “Out-party” trend was not evident in first six
months of election year.
– Has picked-up since then.
• Seat estimate for the 61 to 120 days out interval is
about the same as the estimate for the 121 to 180 days
out interval reported here.
0
179
Seats Won by Democrats
192
205
218
231
244
.25
.5
.75
Probability Democrats Keep House
1
257
Forecasting from October Polls
40
45
50
55
Dem % of Two-Party Vote in the Generic Polls
Seats
Prob Dems Keep House
60
Conclusion
• At this point, it looks like Republicans have
reason to be very optimistic about their
electoral prospects in November.