19
scored 1 for freshmen, and the ideological extremity of the incumbent, measured again as the absolute
value of the DW-NOMINATE score from the preceding Congress. As noted above, legislators who are
running for their first reelections may be more vulnerable than other legislators owing to their relative
lack of experience. The same holds for ideological extremity; extreme incumbents, according to Canes-
Wrone, Brady, and Cogan (2002), is generally related to worse electoral performance. Both variables,
while contributing to a lower vote share for the incumbent, may also be related to challenger campaign
strength whereas challengers may be able to mount stronger campaigns against these vulnerable
incumbents. Finally, dummy variables are included for the 2002 and 2004 elections.
[Table 5 here]
Table 5 presents the results of these two models. The first observation to make is how poorly the
model predicts the emergence of an experienced challenger. First, there is no improvement in the
predictive power of the model over the modal category; the proportional reduction in error is in fact
negative. Second, the only variables that significantly affect whether the challenger has held elective
office are the vote share of the incumbent in the last election and the ideological extremity of the
incumbent in the Congress leading up to the election. Increases in both of these variables cause a
decrease in the likelihood of the challenger having held an elective office. While the result for incumbent
vote share needs little explanation, the effect of ideological extremity is, again, unexpected. As we saw
with the models predicting the vote share of the incumbent, it appears that the period under study is
notable for the success of extreme candidates. Whereas in the past, it may have been the case that
candidates had to run to the center during a general election, for the last three elections there may be a
polarizing in the electorate so that it is the firebrand incumbents that are more successful, not the
moderates.
Returning to the model, not surprisingly the coefficient for freshmen, while in the correct
direction, is not significant. It is likely the case that incumbent vote share, which includes some
incumbents that, due to the presence of freshmen in the sample, lost the election, is picking up on the
effects we would observe from the dichotomous variable. The indicator for the 2002 election is also
Convention