While the results in Figure 2 do demonstrate more success for incumbents in winning late undecided voters, they do not capture the effect of the candidates’ party affiliation. The 2006 midterms were notable for the fact that not a single Democratic incumbent lost a race at the gubernatorial, senatorial, or congressional level. With high presidential disapproval rates placing a drag on GOP candidates, the 2006 campaign became more secure for Democratic incumbents. Of course, it was also a difficult year for Republican incumbents, with the Democrats winning control of both the House and Senate.

Although the effect of partisanship on the overall election outcome is obvious, the effect of party affiliation on undecided voters is less clear. The question in particular is, will a partisan “wave” wash away the general advantages that challengers have in winning undecided voters? In the case of 2006, the answer appears to be yes. As can be seen in Figure 3, both Democratic challengers and incumbents won a majority of undecided voters. This scenario contrasts with 2004, when both Democratic and Republican challengers took a majority of undecided voters.

Another way to examine the allocation of undecided voters is to measure the size of the deviation between preelection polls and the actual election results. In the 2004 general election, this measure followed a pattern that would be expected given previous research on the matter. More specifically, incumbent poll predictions deviated by about three percentage points from actual election results, while challenger predictions were about four and a half points off the actual results. These margins form a ratio of about three to two in terms of undecided voters breaking to challengers. While this margin is much less than is commonly assumed in media accounts of how undecided voters break, the results are in line with what Bowers found in his examination of elections between 2000 and 2004.

Looking more closely at the data in Table 3, it appears that candidate and election characteristics were not important factors in determining the deviation of poll predictions from actual results. Democratic and Republican incumbents had nearly identical levels of deviation—3.08 and 3.10 points, respectively—although Democrats were successful in gaining a larger share of unallocated voters than Republicans. The type of race seemed to have limited effect on the share of undecided voters won by candidates, with only Democratic senatorial incumbents outperforming the overall incumbent average.

While in 2004 undecided voters were more likely to break to challengers over incumbents, in 2006 the split was considerably more even. For incumbents, preelection polls deviated by 2.18 points from actual results, while challengers deviated by 2.34 points. This limited difference underlies the diminished role that incumbency status appears to have played in the 2006 midterms.

However, the aggregate measures mask the interactive effect of incumbency and partisanship. As can be seen in Table 4, Democratic incumbents and challengers outperformed their Republican counterparts by 8 and 14 percent, respectively. Put in another way, preelection polls were more accurate at predicting the final results of candidates challenging Democratic incumbents (1.92) than they were the results of Democratic incumbents themselves (2.29). Such evidence supports the argument that the incumbent rule broke down under the Democratic wave in 2006.