GALLUP NEWS SERVICE
PRINCETON, NJ -- For the past three months, CNN/USA Today/Gallup Polls have shown the Democrats leading the Republicans among registered voters in the race for majority control of the U.S. House of Representatives. The lead is small in each poll, but an aggregate of all four conducted during that period suggests the lead is statistically significant. The results are based on a "generic ballot," which asks respondents to name the party of the congressional candidate for whom they would vote if the election were held today.
| Vote for Congress If the elections for Congress were being held today, which party's candidate would you vote for in your Congressional district? |
 |
A year ago last June, Gallup's first reading showed Democrats with a slight advantage, 49% to 45%. Following the Sept. 11 terrorist attacks, voters shifted their preferences to the Republican Party, giving the GOP small leads over the next several months. But in March, as news broke about the business accounting scandals, and the general rally effect caused by 9/11 began to dissipate, the Democratic Party moved ahead again. However, during the summer, the lead changed parties four times, suggesting a conflicted electorate. Now, Democrats have maintained a small advantage for the past three months, with the Republicans attracting about 42% to 43% of the vote, and Democrats getting from 46% to 50% of the vote. The current results are consistent with the historical pattern -- the president's party usually suffers Congressional seat losses in midterm elections (1998 was a rare exception). However, as the nation appears to be shifting its focus from domestic issues back to international issues and a potential war with Iraq, Republicans may draw closer to the Democrats in coming weeks.
The current poll includes questions that help to identify most likely voters. Among this sub-sample of respondents, Democrats still lead Republicans in the generic ballot, by a margin of 50% to 46%.
The Two-Party Vote Predicts Vote Seats With 96% Accuracy
The generic ballot is intended to measure the aggregate party vote for Congress across the country, though in practice it is not the national party vote that counts, but the vote in each of 435 congressional districts. In theory, a party could win 100% of the congressional seats by winning just over 50% of the two-party vote in each district. In the actual elections, however, when a party wins about 50% of the national two-party vote, it also wins somewhere around half of the seats. The exact percentage varies over time.
A special Gallup analysis of the past 14 off-year elections, from the post-World War II period until now (1946-1998), shows that the accuracy of the two-party vote depends, at least in part, on which party has the White House and which party controls Congress leading up to the election. The incumbent Congressional party enjoys a slight advantage over the "out" party, which means that the incumbent party typically can win more than 50% of the seats even though it may win less than 50% of the national two-party vote. Such appears to be the case for Republicans this year.
Taking the incumbency factors into account, an analysis of the data finds that, over the past half-century, the two-party vote has predicted the number of seats won by the two parties with an average accuracy of better than 96%.
|
Accuracy of Predicting House Seats Using The National Two-Party Vote -- Controlling for Incumbency |
|||
|
Year |
Democratic Seats |
Difference |
|
|
Predicted |
Actual |
(Over/Under Prediction) |
|
|
1946 |
190 |
188 |
2 |
|
1950 |
235 |
234 |
1 |
|
1954 |
232 |
232 |
0 |
|
1958 |
272 |
283 |
-11 |
|
1962 |
259 |
258 |
1 |
|
1966 |
247 |
248 |
-1 |
|
1970 |
260 |
255 |
5 |
|
1974 |
292 |
291 |
1 |
|
1978 |
277 |
277 |
0 |
|
1982 |
275 |
269 |
6 |
|
1986 |
264 |
258 |
6 |
|
1990 |
260 |
267 |
-7 |
|
1994 |
201 |
204 |
-3 |
|
1998 |
211 |
211 |
0 |
|
Average Error |
3.1 seats |
||
The largest error was in 1958, when the actual vote for the Democrats (taking into account the party of the president and the party of the incumbent party in Congress) resulted in 11 more seats than what was predicted. In the most recent mid-term election, the prediction was exact -- 211 seats for the Democrats, the rest for the Republicans. In the past six mid-term elections, 1978-1998, two elections have resulted in more seats for the Democrats than predicted, two in fewer seats, and two have been exactly correct. The average error over the 14 elections is 3.1 seats.
Two-Party Vote Prediction for 2002 -- Bias in Favor of Republicans
Given the fact that the Republicans currently enjoy majority control in the House, and that there is a Republican president, the analysis predicts that the Democratic Party will probably have to win more than 51% of the two-party vote in order to win control of the House. If the Democrats actually win the two-party national vote by the percentage found in this poll (a 4-point margin), the analysis predicts they would win 226 seats, eight more than needed to obtain majority control. With an actual 3-point winning margin, the analysis predicts the Democrats to win 221 seats, three more than needed. But given the average error of just over three seats, a Democratic victory would not be assured.
If the Democrats win 51% of the vote (and Republicans win just 49% of the vote), the analysis predicts that the GOP would retain majority control of the House with 219 seats, one more than needed. Again, this margin is too close to make a firm prediction, but the numbers reflect the statistical advantage that Republicans enjoy because of their incumbency. If the Republicans and Democrats split the national two-party vote evenly, the analysis predicts Republicans would win 230 seats, 12 more than they need to maintain majority control -- and seven more than they have now.
Generic Ballot Accurate in Predicting National Two-Party Vote Within Two Percentage Points
The preceding results show that the national two-party vote is a fairly accurate predictor of the number of House seats that the parties win. The table below shows how accurate the Gallup generic ballot is in predicting the two-party vote. Over the 12 elections since 1950 when the generic ballot has been asked (it was not asked in 1986), the average error is 1.3 percentage points. In 1994, when Gallup predicted the GOP to regain majority control of the House for the first time in four decades, the generic ballot was exactly correct. However, in 1998, the generic ballot showed the Democrats winning with 52.1% of the vote, and in fact they barely lost by a margin of 0.6 of one percent (50.3% for Republicans to 49.7% for Democrats).
|
Accuracy of Gallup Generic Ballot in Predicting the National Two-Party Vote |
|||
|
Year |
Generic Ballot |
National Two-Party Vote |
Error |
|
% Democratic |
% Democratic |
Percentage Points |
|
|
1950 |
51.0 |
50.0 |
1.0 |
|
1954 |
51.5 |
52.6 |
-1.1 |
|
1958 |
57.0 |
56.0 |
1.0 |
|
1962 |
55.5 |
52.5 |
3.0 |
|
1966 |
52.5 |
51.3 |
1.2 |
|
1970 |
53.0 |
54.4 |
-1.4 |
|
1974 |
60.0 |
58.5 |
1.5 |
|
1978 |
55.0 |
54.4 |
0.6 |
|
1982 |
55.0 |
56.4 |
-1.4 |
|
1990 |
54.0 |
54.5 |
-0.5 |
|
1994 |
46.5 |
46.5 |
0.0 |
|
1998 |
52.1 |
49.7 |
2.4 |
|
Average Error |
1.3 |
||
These results point to the caution that is necessary in using the generic ballot for predicting the national two-party vote, and in turn predicting the number of seats. The theoretical margins of error for each of the surveys listed above were typically about three percentage points, though the average accuracy has been much closer than the margins of error would suggest. Still, with the Democrats enjoying just a four-point advantage over the Republicans in the generic ballot, the race for majority control of Congress is still too close to call.
Survey Methods
The most recent results reported here are based on telephone interviews with a randomly selected national sample of 1,010 adults, 18 years and older, conducted Sept. 20-22, 2002. Among these respondents, there are 865 respondents who indicated they were registered to vote or didn't need to register. For results based on the sample of registered voters, one can say with 95% confidence that the margin of sampling error is +/- 4 percentage points. For results based on the sample of --508-- likely voters, the maximum margin of sampling error is ±5 percentage points.
If the elections for Congress were being held today, which party's candidate would you vote for in your Congressional district -- [ROTATE: 1) The Democratic Party's candidate or 2) The Republican Party's candidate]?
As of today, do you lean more toward -- [ROTATE: 1) The Democratic Party's candidate or 2) The Republican Party's candidate]?
|
Democratic |
Republican |
Undecided/ |
|
|
% |
% |
% |
|
|
Likely Voters |
|||
|
2002 Sep 20-22 |
50 |
46 |
4 |
|
Registered Voters |
|||
|
2002 Sep 20-22 |
48 |
43 |
9 |
|
2002 Sep 2-4 |
46 |
43 |
11 |
|
2002 Aug 19-21 |
50 |
42 |
8 |
|
2002 Jul 26-28 |
48 |
42 |
10 |
|
2002 Jun 28-30 |
44 |
49 |
7 |
|
2002 Jun 21-23 |
50 |
42 |
8 |
|
2002 May 28-29 |
45 |
46 |
9 |
|
2002 Apr 29-May 1 |
48 |
44 |
8 |
|
2002 Apr 5-7 |
50 |
43 |
7 |
|
2002 Mar 22-24 |
46 |
46 |
8 |
|
2002 Feb 8-10 |
43 |
47 |
10 |
|
2002 Jan 25-27 |
44 |
46 |
10 |
|
2002 Jan 11-14 |
43 |
46 |
11 |
|
2001 Dec 14-16 |
43 |
48 |
9 |
|
2001 Nov 2-4 |
45 |
45 |
10 |
|
2001 Jun 8-10 |
49 |
45 |
6 |
