Share this post
FaceBook  Twitter  


Polly Says “No”

November 5th, 2006 Update

Alfred G. Cuzán

With only two days before voters go to the polls, Polly’s final forecast is that the Republicans will lose 23 seats in Tuesday’s elections to the U.S. House of Representatives and, hence, their majority. Their 12-year reign in the chamber, their longest since the 1920s (see Figure 1), is about to end.

Since Polly’s first report some three weeks ago, the president’s party’s “stock” has slumped in both the Iowa Electronic Markets and Tradesports.com. These markets trade in contracts specifying that the Republicans will hold on to their majority in the House. The resulting prices are interpreted as probabilities that they will do so. Averaging the two values yields the Pollybility, displayed in Figure2. The very latest calculation (computed with the closing prices of November 4) shows the Republicans with only a 0.25 chance of retaining control. (For two skeptical views, see Kastellec et al. 2006 and McTague 2006.)

Also, election watchers Stuart Rothenberg and Charlie Cook, and political scientist Larry Sabato, all have raised their estimate of the number of seats the Republicans stand to lose, an opinion shared by traders in Tradesports.com contracts on the number of seats the Democrats will win. These forecasts have been incorporated into a revised Table 1.

As explained previously, to arrive at her own prediction Polly performs simple computations on the forecasts of others, namely averaging. Our parrot first computes the average within the several methods by which forecasts are made, and then the average of these averages, i.e., the average across methods. This procedure has been shown to increase forecasting accuracy. (See Armstrong 2001. For an application to the 2004 presidential election, see Cuzán, Armstrong, and Jones 2005.)

According to Polly’s latest calculations, on Tuesday the Republicans will win 209 seats, or 23 fewer than they did in 2004. In sum, the will lose control of the U.S. House of Representatives to the Democrats.

REFERENCES

Armstrong, J. Scott (2001) Combining Forecasts, in Principles of Forecasting , Kluwer, pp. 417-439.

Cuzán, A. G., J. Scott Armstrong, and R. Jones, Jr. 2004. Combining Methods to Forecast the 2004 Presidential Election. Paper presented at the 2005 meeting of the Southern Political Science Association.

Kastellec, J. P., A. Gelman, and J. P. Chandler. 2006. Seeking 50% of the Seats, Needing More Than, 50% of the Votes: Predicting the Seats-Votes Curve in the 2006 Elections.

McTague, Jim. 2006. Survivor! The GOP Victory. Barron’s On-Line. October 23rd.


Table1. Forecasts for Republican Seat Losses 1 in the 2006
Elections for the U. S. House of Representatives

Author
(approximate date of prediction)

Method

Forecast
range2

Forecast
point2

Formal Models Not Driven by Late Polls
N3

Klarner & Buchanan**
(late April)

16

14-31

22*

Brandt & Brunell**
(late August)

30

6-19

12

Campbell
(early September)

14-30

10-16

13

Cuzán & Bundrick
(mid-October)

46

16-25

20*

Abramowitz 6
(mid October)

28

22-28

29*

Within-method average

19.2*

Polls and Polls-Driven
Models

Ansolabehere4
(early August)

8.0

Bafumi, Erikson & Wlezien
(late October)

15

32*

Within-method average

20*

Model Aided by
Judgment

Erikson4
(early August)

21*

Jacobson4
(late August)

20*

Within-method average

20.5*

Polls Aided by
Judgment

Cook
(30 October)

20-35

28*

Mann
(July)

20-30

25*

Rothenberg
(2 November)

34-40

37*

Sabato
(26 October)

23-30

27*

Within-method average

29.3*

Prediction Market

Tradesports.com5
(5 November)

25*

Pollyseat (average of the within-method averages)

22.8*

1 The base from which the loss is calculated is 232, the number of seats won in 2004. Any seat gain for the Democrats amounts to a seat loss for the Republicans.
2 The range refers to the lowest and the highest forecast obtained with alternative specifications or different models estimated by the author, or to his own educated guess based on a subjective estimate, if he relied on judgment. The forecast point was either specified by the author or calculated as the midpoint of the range.
3 N is the number of historical elections on which the model was based, although two models simulated additional “elections”.
4 Participants in Boston Review forecasting contest.
5 The contracts are specified in five-seat increments, i.e., more than 19.5 seats, more than 24.5 seats, and so on. The forecast was obtained by regressing seats on the closing prices (omitting those at either end of the spectrum) of 5 November GMT, and solving for 50%, a method suggested by Kesten Green.
6 Abramowitz’s range and point forecast do not match because the former is what was published in the October issue of PS: Politics and Political Science, but the latter was conveyed to Polly by him via e-mail on October 13th. Based on “information about challenger quality from Gary Jacobson,” Abramowitz’s “new prediction, based on previous Republican seat % of 53.5, midterm dummy variable of +1 (GOP presidential midterm), net presidential approval in first September Gallup Poll of -17, generic ballot margin in first September Gallup Poll of -12 (based on registered voters), 2% Democratic advantage in open seats and 3% Democratic advantage in challenger quality is a Democratic gain of 29 seats in the House.”

* Forecast calls for a turnover in control of the House of Representatives in favor of Democrats.
** Pollyprize entrant.

Figure1

Figure1

Return to Polly's Page