4
which voters are influenced by three factors: their proximities to the candidates’ ideologi-
cal positions; their partisan loyalties; a random variable representing unmeasured influ-
ences on the vote. Thus a voter i’s utilities for D and R, denoted U
i
(D) and U
i
(R), respec-
tively, are given by:
U
i
(D) = - a(x
i
- D)
2
+ bp
iD
+
ε
iD =
)
,
(
D
x
V
i
+
ε
iD
U
i
(R) = - a(x
i
- R)
2
+ bp
iR
+
ε
iR =
)
,
(
R
x
V
i
+
ε
iR
where a is the salience of ideology, x
i
is voter i’s ideological position, p
iD
and p
iR
are
dummy variables that equal 1 if i identifies with the candidate’s party and zero otherwise, b
represents the salience of party identification and
ε
iD
and
ε
iR
are random disturbance terms.
We further represent the deterministic components of utility as
)
,
(
D
x
V
i
and
)
,
(
R
x
V
i
.
We
specify that a>0 and b
≥
0, i.e. that voter utilities decline with the ideological distance to the
candidate, and that partisanship cannot bias the voter against his party’s candidate. This
latter specification encompasses both the standard spatial model in which partisanship does
not influence voters (b=0) and the more general case in which partisanship biases the voter
in favor of his party’s candidate (b>0).
Our specification, that the voter’s party identification influences her evaluation of
the candidate independently of her ideological position, is supported by empirical research
on presidential elections (Markus and Converse, 1979; Alvarez and Nagler, 1995, 1998),
Congressional elections (Erikson and Wright, 1993, 1997; Ansolabehere et al., 2001; Bur-
den, forthcoming; Krasno, 1990), and gubernatorial elections (Lacy and Paolino, 1999).
This finding is consistent with the “Michigan model” of voting (Campbell et al., 1960), in
which partisanship is conceptualized as a long-term, affective orientation towards one’s
preferred party – one that grows out of early socialization experiences and positive evalua-
tions of the party’s past performance – and which is largely independent of the candidates’
positions in the current election (see Fiorina, 1981; Jennings and Niemi, 1975, 1981; Green
et al., 2002). Previous spatial modeling work by Erikson and Romero (1990) and by Ad-
ams and Merrill (2003) analyzes American candidates’ strategies in situations where voters
display such partisan biases.
Convention