Locating the Voter

• Date: 01 December 1974
• Published date: 01 December 1974
• Author: Joel D. Barkan,James E. Bruno
• DOI: 10.1177/106591297402700410
• Subject Matter: Articles

LOCATING THE VOTER: MATHEMATICAL MODELS
AND THE ANALYSIS OF AGGREGATE DATA
FOR POLITICAL CAMPAIGNS
JOEL D. BARKAN, University of Iowa
JAMES E. BRUNO,
University
of California, Los Angeles
HE
STRONG relationship between a voter’s party identification and his
~ voting decision has been so thoroughly documented in the literature that
-S- it has virtually attained the status of a &dquo;sociological law.&dquo;’- Despite the
widespread acceptance of this finding, the degree to which it has been systematic-
ally incorporated into the making of campaign strategy is surprisingly low.
To activate an electoral majority, campaign strategists must first identify and
geographically locate those segments of the electorate which are most likely to
constitute a base of partisan support, and those which possess the marginal votes on
which the outcome of the election depends. Though campaign managers usually
make intuitive judgments in this regard, detailed maps of the geographical distri-
bution of party identification across a constituency are rarely made because of the
difficulties involved. The result is an inability to target systematically different
campaign activities such as registration drives, appeals to &dquo;independent&dquo; voters,
and get-out-the-vote efforts into those areas where they will have their greatest
effect. This in turn results in a lower level of campaign efficiency, both in terms of
resources spent and prospects for achieving success at the polls.
To map the distribution of party identification across an electorate requires
either a survey of every voter to determine the intensity of his partisanship, or an
analysis of aggregate data which estimates the same. The first is prohibitively ex-
pensive for most campaigns.2 The second is dependent upon the development of
a mathematical model to solve what is essentially a problem of ecological inference
- the estimation of the characteristics of individuals through an examination of
the characteristics of large populations. Several models have been developed to
achieve this goal, but for reasons given below, the degree of precision or practicality
of these approaches is low. None of these models, moreover, has been specifically
developed for the purpose of mapping the distribution of party identification across
a constituency in order to facilitate an optimum allocation of campaign resources
by candidates and their staffs. The discussion which follows proposes one method
by which these obstacles to a more sophisticated level of campaign decision making
1
As is well known, the relationship has been most thoroughly examined in the voting studies
by Angus Cambell and his associates at the Survey Research Center of the University
of Michigan. See The Voter Decides (Evanston: Row Peterson, 1954), chapter 7;
The American Voter (New York: Wiley, 1960) ; Elections and the Political Order
(New York: Wiley, 1966).
2
A survey of every voter in a typical congressional district can cost as much as most candi-
dates spend on an entire campaign. In 1970 such a survey was conducted in the 34th
California, Congressional District by Matt Reese Associates for the campaign of Repre-
sentative Richard Hanna at a cost of $70,000. Despite these costs, many campaigns are
making increasing use of the survey approach to mapping geographical distribution of
political beliefs.
710

---

711
might be removed, and in so doing might fully incorporate one of the basic findings
of voting research into the art of campaign management.
I. THE APPROACH
Since the distribution of party identification in an area is but a series of
individual opinions which can only be precisely determined through a direct ques-
tioning of the area’s inhabitants, estimates of this distribution which are based on
analyses of aggregate data are necessarily surrogate measures of what lies inside
the voters’ heads.
One surrogate measure of party identification which can be obtained from
aggregate data in those states where voters indicate their party preference when
they register is suggested by the typology of electoral performance in Table 1.3 As
shown by the matrix, voters in a geographical unit can be classified in terms of
their party registration on the one hand, and the party they supported in a given
election on the other. Registrants with either of the two major parties can thus
be classified as &dquo;loyalists&dquo; or &dquo;defectors&dquo; depending on the candidate they sup-
ported. One can also account for those voters who did not participate in the elec-
tion, an important consideration as shall be shown below.
TABLE 1
TYPOLOGY OF ELECTORAL PERFORMANCE
Key to symbols :
Dd Registered Democrats who voted Democratic
Id
Registered Independents who voted Democratic
Dr Registered Democrats who voted Republican
Ir
Registered Independents who voted Republican
Dnv Registered Democrats who did not vote
Inv Registered Independents who did not vote
A
Total number of registered Democrats
C
Total number of registered Independents
Rd Registered Republicans who voted Democratic
Ta
d
Total vote for Democratic candidate
Rr Registered Republicans who voted Republican
Tr
Total vote for Republican candidate
Rnv Registered Republicans who did note vote
Tnv Total number of non-voters
B
Total number of registered Republicans
When operationalized, this simple typology of electoral loyalty provides a
surrogate measure of party identification, because of the strong relationship between
the level of a voter’s electoral loyalty to his party of registration, and the intensity
of his party identification. By estimating through an analysis of aggregate data the
3
Data to operationalize the typology of electoral performance is available in such large states
as California, New York, Ohio, and Pennsylvania, but not in Texas or Illinois.

---

712
number of loyal party registrants in an area, and the number of defectors, one can
in essence determine the distribution of party identification in that area. When
these estimates are computed for a population of geographical units, such as pre-
cincts, an efficient mapping of the distribution of party identification across the
electorate can be achieved. The units can then be ordinally ranked on the basis
of these estimates to provide lists of priority areas for different campaign tasks.
The assumption that there is a close relationship between the level of a voter’s
electoral loyalty to his party of registration, and the intensity of his party identifi-
cation is but a logical and obvious correlate of the positive relationship between
party identification and the way people vote. People who identify strongly with a
given party are more likely to register with the same party, and in turn vote for
the party’s nominees than people whose partisan feelings are less pronounced.
Evidence for the existence of this relationship, and hence the propriety of using a
measure of voter loyalty as a surrogate for party identification, is presented in
Table 2. The evidence is drawn from the results of a recent survey of voters in
Southern California.
TABLE 2
ELECTORAL LOYALTY TO PARTY OF
REGISTRATION BY PARTY IDENTIFICATION*
’~ Data are from a stratified sample of voters in Los Angeles and Orange Counties interviewed in Novem-
ber 1970. A stratified sample was employed in order to obtain the opinions of a statistically significant number
of voters who had defected from their party of registration at the polls. The method of drawing the sample
is based on the mathematical model discussed in this article. &dquo;Loyal Democrats&dquo; are registered Democrats who
voted for Hubert Humphrey in the 1968 presidential election. &dquo;Defectors&dquo; are registered Democrats who voted
for either Nixon or Wallace or registered Republicans who voted for Humphrey. Registered Republicans who
voted for Wallace were exc uded from this category, because the authors are primarily interested in isolating
those defectors who position themselves in the center of the political spectrum. &dquo;Loyal Republicans&dquo; are
registered Republicans who voted for Richard Nixon.
Estimates of the level of electoral loyalty to party registration also provide a
far better surrogate measure of party identification than registration figures alone.
While the latter are closely correlated with party identification, they are normally
not a useful indicator of where independent and weak party identifiers reside.
Because most states which register by party hold closed primaries, voters registering
as independents may not participate in these elections. Many independents and
weak partisans thus register with one of the major parties in order to avoid being
partially excluded from the electoral process. A.s a result, the number of registered
voters who refuse to state a party preference is often substantially less than the
number ~~ho regard themselves as independents. In California, for example, only
3 percent of the registered voters are classified as &dquo;declined to state,&dquo; even though
public opinion surveys consistently show that almost a third of the state’s electorate
does not identify with any party.

---

713
Finally, estimates of electoral loyalty may be more accurate predictors of vot-
ing behavior than party identification. Whereas party identification is an attitude
which correlates closely with the voting decision, estimates of electoral loyalty based
on aggregate data are measures of that behavior itself. Though we have hereto-
fore defined our problem as one of...