8
reflect the candidates’ general liberal-conservative leaning.
3
Thus, we only examine the
distance between candidates on the first dimension.
Independent variables
Ansolabehere, Snyder, and Stewart (2001a) concluded, examining NPAT survey
responses, that incumbents take positions closer to the median voter in their districts than
do challengers. The empirical models in this paper employ a refined measure of valence
advantages, which differentiates between more and less successful incumbents in terms
of “bringing home the bacon.”
The key independent variable of interest is valence advantage. This variable is
measured as the amount of money spent in each district on federal projects for districts in
which incumbents are running (formally “1” if incumbent x project outlays in billions of
dollars). The data on project outlays by congressional district come from Kenneth
Bickers (Stein and Bickers 1996).
4
A second variable is a dummy variable for incumbent
running. This variable is included as we want to separate the effect of the overall
incumbency advantage from the specific valence advantage of interest (differences in
levels of project allocations among incumbents).
The other independent variables specified are from Ansolabehere, Snyder, and
Stewart (2001a). The general ideological leaning of the geographic constituency is
measured as the proportion of two-party Republican presidential vote received by the
Republican presidential candidate in the most recent presidential general election. We
also include a variable for Republican presidential vote
2
. We also include a variable for
3
For other uses of the NPAT survey as a measure of candidates’ public positions, see Erikson and Wright
(1997), Ringquist and Dasse (2000), and Schraufnagel and Mondak (2002).
4
These data are from http://www.polisci.indiana.edu/faad.
Convention