 

on
um

VET-’5"

Q

“’3‘-
W‘
‘ ‘4? ..
V 5:.»

,.
'M

4 AV
,5.

3h; ' :‘

 

1131519

I-

x" "(I

J.)

 

LlBRARY
Mich?” State
University

 

 

 

This is to certify that the
dissertation entitled

Partisan Ambivalence: Examining the Causes and
Consequences, 1980-2004

presented by

Judd R. Thornton

has been accepted towards fulﬁllment
of the requirements for the

Ph.D. degree in Political Science

 

 

ﬁlm-am \AQMNN

MajoFProlessior’s Swe
August 24, 2009

 

Date

MSU is an Afﬁrmative Action/Equal Opportunity Employer

 

.. u---.-.-I-.-I”o-n-n—‘--I--_o— —.

 

 

PLACE IN RETURN BOX to remove this checkout from your record.
To AVOID FINES return on or before date due.
MAY BE RECALLED with earlier due date if requested.

 

DATE DUE DATE DUE DATE DUE

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

5/08 Kthrolecc8-Pres/ClRC/DaleDue.indd

 

PARTISAN AMBIVALENCE: EXAMINING THE CAUSES AND
CONSEQUENCES IN PRESIDENTIAL ELECTIONS, 1980—2004

By

Judd R. Thornton

A DISSERTATION
Submitted to
Michigan State University
in partial fulﬁllment of the requirements
for the degree of
DOCTOR OF PHILOSOPHY

Political Science

2009

ABSTRACT

PARTISAN AMBIVALENCE: EXAMINING THE CAUSES AND
CONSEQUENCES IN PRESIDENTIAL ELECTIONS, 1980—2004

By

Judd R. Thornton

This project examines the possible causes and consequence of ambivalence—the state
of experiencing conﬂicted thoughts—toward the American political parties. I intend
to answer three questions. First, I wish to examine the common measures to see
if it is more appropriate to conceptualize ambivalence to the parties separately or
simultaneously. Relatedly, I intend to ﬁnd how to best code the variables. As the
measures have been coded several different ways I hope to offer resolution and offer
guidance as how to do so in future research.

Second, by I look for evidence that prominent campaign messages can serve as
sources of ambivalence. The goal here is to add to existing evidence that one should
look to informational sources. While the idea of value conﬂict was useful in the
formulation of ambivalence in the political science literatures, it now appears to be a
secondary sources of ambivalence. If campaign messages are sources of ambivalence,
this will be further evidence in favor of informational theories.

Finally, I Wish to see to what extent ambivalence inﬂuences the deﬁnitive political
act for most Americans, voting for the president. No previous studies have offered
evidence if ambivalence inﬂuence the most important political act for a large number
Americans: voting. If it does not ultimately alter an one’s vote, this would be evidence
in favor of partisanship as a powerful social identity and psychological attachment
capable of shaping one’s view of politics. At least, powerful enough to overcome a
general sense of conﬂict. On the other hand, if it does inﬂuence one’s choice, there

will be new evidence as to what kinds of voters are likely to defect.

Table of Contents

List of Tables .................................................. v
List of Figures ................................................ vii
1 Background and Theory ......................................... 1
1.1 A History of Ambivalence .................................... 1
1.2 Ambivalence and Politics .................................... 3
1.3 Partisan Ambivalence ....................................... 7
1.4 Campaigns, Ambivalence, and Voting ........................... 9
1.4.1 Campaign Messages as Sources of Ambivalence ............... 10

1.4.2 Partisan Ambivalence and the Voting Decision ............... 11

1.5 The Plan of the Dissertation ................................. 12

2 Measuring Ambivalence ........................................ 13
2.1 Measures of Ambivalence ................................... 15
2.2 Objective Measures of Ambivalence ............................ 17
2.3 Levels of Measurement: Observed and Latent Ambivalence .......... 18
2.3.1 The Zero Responses ...................................... 25
2.4 Empirical Examples ....................................... 29
2.4.1 A Dichotomous Measure of Ambivalence .................... 29

2.4.2 The Possibility of Indifferent Individuals .................... 31

2.5 A Measure of Comparative Ambivalence ........................ 33
2.6 Discussion and Conclusion .................................. 37
Appendix to Chapter 2 ....................................... 39

3 Campaign Messages as Sources of Partisan Ambivalence ................ 41
3.1 Campaigns, Information, and Ambivalence ...................... 42
3.2 Research Design .......................................... 45
3.2.1 Other Measures ...................................... 46

3.2.2 Modeling Ambivalence .................................. 47

3.3 Empirical Results ......................................... 49
3.3.1 Accuracy of Theoretical Model ........................... 49

3.3.2 Sources of Ambivalence ................................. 52

3.3.3 Trends ............................................. ' 53

3.3.4 Alternative Measures of Ambivalence ...................... 59

3.4 Conclusion .............................................. 61
Appendix to Chapter 3 ....................................... 63

iii

4 Partisan Ambivalence and Voting Behavior ......................... 72

4.1 Congressional Polarization .................................. 73
4.2 Polarization and Ambivalence ................................ 75
4.3 Research Design .......................................... 79
4.3.1 Modeling Vote Choice .................................. 79

4.4 Empirical Results ......................................... 80
4.5 Conclusion .............................................. 86
Appendix to Chapter 4 ....................................... 88

5 Ambivalence, Indifference, and Presidential Elections .................. 94
5.1 Partisanship, Ambivalence, and Voting ......................... 94
5.2 Research Design .......................................... 97
5.2.1 Measures ........................................... 97

5.2.2 Modeling Ambivalence and Timing of Vote Decision ........... 98

5.2.3 Ambivalence and Defection ............................. 102

5.3 Conclusion ............................................. 106
Appendix to Chapter 5 ...................................... 109

6 Conclusions ................................................ 121
6.1 Findings ............................................... 121
6.2 Future Directions ........................................ 125
References ................................................... 127

iv

List of Tables

2.1

2.2

2.3

2.4

3.1

3.2

3.3

3.4

3.5

3.6

3.7

3.8

3.9

4.1

4.2

4.3

4.4

4.5

4.6

5.1

R2 before and after optimal scaling ............................. 25
Zero inﬂated count model for likes and dislikes listed ................ 27
Partisan ambivalence ........................................ 30
Partisan ambivalence ........................................ 32
Ambivalence measures and categorization ......................... 46
Predicting Partisan Ambivalence in 1980 ......................... 54
Predicting Partisan Ambivalence in 2004 ......................... 55
Predicting comparative partisan ambivalence ...................... 6O
Predicting Partisan Ambivalence in 1984 ......................... 67
Predicting Partisan Ambivalence in 1988 ......................... 68
Predicting Partisan Ambivalence in 1992 ......................... 69
Predicting Partisan Ambivalence in 1996 ......................... 70
Predicting Partisan Ambivalence in 2000 ......................... 71
Vote choice in 1980 ......................................... 82
Vote choice in 2004 ......................................... 84
Vote choice in 1984 ......................................... 90
Vote choice in 1988 ......................................... 91
Vote choice in 1992 ......................................... 92
Vote choice in 1996 ......................................... 93
Ambivalence and Timing of vote decision in 1980 .................. 100

5.2 Ambivalence and Timing of vote decision in 2004 .................. 101
5.3 Ambivalence and defection in 1980 ............................. 104
5.4 Ambivalence and defection in 2004 ............................. 105
5.5 Ambivalence and Timing of vote decision in 1984 .................. 111
5.6 Ambivalence and Timing of vote decision in 1988 .................. 112
5.7 Ambivalence and Timing of vote decision in 1992 .................. 113
5.8 Ambivalence and Timing of vote decision in 1996 .................. 114
5.9 Ambivalence and Timing of vote decision in 2000 .................. 115
5.10 Ambivalence and defection in 1984 ............................. 116
5.11 Ambivalence and defection in 1988 ............................. 117
5.12 Ambivalence and defection in 1992 ............................. 118
5.13 Ambivalence and defection decision in 1996 ...................... 119
5.14 Ambivalence and defection decision in 2000 ...................... 120

vi

List of Figures

2.1

2.2

2.3

2.4

2.5

2.6

2.7

2.8

2.9

Total number of likes and dislikes toward the parties in 2004 ......... 19
Distribution of comparative partisan ambivalence in 2004 ............ 20

Transformation of original data values for Democratic and Republican
ambivalence, obtained from ALSOS regression analysis in 2004 ......... 24

Transformation of original data values for comparative partisan
ambivalence speciﬁed at the ordinal level .......................... 25

Transformation of original data values for comparative partisan
ambivalence when speciﬁed to be nominal level ..................... 26

Transformation of original data values for comparative partisan

ambivalence excluding those who offered no likes or dislikes, obtained
from ALSOS regression analysis in 2004 ......................... 27

Political knowledge and Pr(listing zero likes or dislikes) from zero-inﬂated
count model ............................................... 28

The impact of partisan intensity on univalence, ambivalence, and
indifference ................................................ 33

The average number of positive and negative statements listed toward the
parties broken down by political party ........................... 35

2.10 Results from factor analysis of positive and negative statements listed

about the parties, as well as partisanship, ideology, and party
thermometer scores ......................................... 36

2.11 Marginal and joint distributions of Pro—Republican and Pro-Democratic

responses for those individuals who list at least one likes or dislike ...... 37
2.12 Republican and Democratic ambivalence in 2004 .................... 40
2.13 Screeplot from factor analysis .................................. 40

3.1

A message likely to be head by many represented by the solid line and one
that is unlikely to be heard by all but the most informed ............. 44

vii

3.2

3.3

3.4

3.5

3.6

3.7

3.8

3.9

4.1

4.2

4.3

4.4

4.5

4.6

4.7

5.1

5.2

5.3

5.4

Testing model assumptions .................................... 50

Concordance correlation coefﬁcient .............................. 52
The impact of campaign messages in 1980 ......................... 54
The impact of campaign messages in 2004 ......................... 56
The impact of foreign policy campaign messages, 1980—2004 ............ 57
The impact of domestic policy campaign messages, 1980-2004 .......... 58
Testing model assumptions ................................... 65
Comparing measures of ambivalence ............................. 66

Congressional polarization and the percent of individuals classiﬁed as

ambivalent toward one or both parties ........................... 74
House polarization and partisan ambivalence ....................... 75
Before and after elite polarization ............................... 77

Before and after elite polarization, with individuals moving with parties . .78
The impact of Democratic ambivalence on vote choice ............... 85
The impact of Republican ambivalence on vote choice ............... 85

Examining for interaction effects between Party ID and Ambivalence. . . . 89

Attitudinal states and intensity of partisan identiﬁcation ............. 96
Attitudinal states and interest in the campaign .................... 96
Attitudinal states and time of decision .......................... 102
Attitudinal states and defection in vote choice ..................... 106

viii

Chapter 1
Background and Theory

Before developing a theory of partisan ambivalence, I ﬁrst explore the current schol-
arly thinking on the subject of ambivalence generally, and partisan ambivalence more
speciﬁcally. In doing show I hope to show three main points: ﬁrst, the concept has
been more rigorously deﬁned over time, which aids in attempts to measure it directly
and objectively. Second, while values have often been at the heart of explaining am-
bivalence, more recently it has been posited that information is the primary source.
It is argued ambivalence should occur among those with a high level of contextual
knowledge, those who are exposed to multiple points of View. Finally, as of now there

is no deﬁnitive evidence that ambivalence alters voting behavior.

1.1 A History of Ambivalence

During the beginning of the behavioral revolution in political science, attitudes were
often assumed to be bipolar and unidimensional. Political scientists were not alone
in this assumption, social psychologists also characterized attitudes in such a manner
(e.g., Allport 1935, Thurstone 1928). While this was certainly the prevailing view,
it was recognized from even the earliest studies of voting and political behavior that
individuals might experience competing and conﬂicting thoughts. The idea of atti-

tudinal conﬂict has existed in the political science literature, in one form or another,

for quite some time. While there is currently a well agreed upon deﬁnition of the
concept—something along the lines of when an individual has internalized competing
arguments—this was not always the case. Indeed, the earliest discussion of what is
now called ambivalence went by an entirely different name. Its origins date back to
a concept termed cross-pressure.

Cross-pressure, as originally formulated in the The People’s Choice (Lazarsfeld,
Berelson and Gaudet 1944) and its followup Voting (Berelson, Lazarsfeld and McPhee
1954) was deﬁned as when one’s “primary environment is internally divided” (1954: 100).
The concept was one that very much showed its sociological roots, focusing directly
on the impact different and competing group identiﬁcations have on one’s voting de-
cision. The primary prediction was that an individual suffering from such pressure
would have difﬁcultly in making up his mind between the candidates and consequently
delay his decision or drop out all together and abstain. The authors of The American
Voter (Campbell et a1. 1960) also examined the roll of external conﬂict on parti-
san choice and one’s decision to vote. Findings suggested individuals with conﬂicted
group identities held less consistent partisan attitudes. The focus is again on social
cross-pressure, in this case in the form of white-collar Catholics.

The matter was investigated in more detail in Chapter 6 of Elections and the
Political Order (Converse 1966) where once again attitudinal conﬂict was found.
However, in the end the ﬁndings suggested “Protestant Democrats were more likely
to behave as Democrats than as Protestants, and Catholic Republicans were more
likely to behave as Republicans than Catholics” (1966:123). Such non-ﬁndings began
to bring an end to the study of cross-pressure (e.g., Davis 1982; Jackson and Curtis
1972; Pool, Abelson and Popkin 1965). Regardless, the concept remained inﬂuential.
Social networks and interaction were long seen as potential catalysts for political be-
havior (Horan 1971; Verba and Nie 1972; Huckfeldt 1979). More recently, it has been

suggested social networks have a greater impact on voting than even the media (Beck

et al. 2002). Recent studies make explicit the connection between social networks and
partisan ambivalence (Huckfeldt, Mendez and Osborn 2004; Mutz 2002). In many
ways, ambivalence is a descendent of the concept of cross-pressure.

The concept is even more closely related to that of attitude conﬂict (Campbell
et al. 1960:81). In fact, Campbell et al. measured the concept with the same items
from the National Election Studies as current measures of partisan ambivalence. Both
use questions asking about likes and dislikes of the two parties. It was shown that
voters who more consistently offered one-sided opinions made their minds up sooner
and were more likely to vote a straight ticket. In this regard, contemporary measures
of ambivalence are more nuanced versions of this measure of attitude conﬂict. Instead
of summing the items, measures of ambivalence attempt to account for both the
intensity and similarity of thoughts.

In the 19503 and 608, social psychologists began to refer to individuals holding in-
consistent beliefs as ambivalent (Brown and Farber 1951; Mowrer 1960). The original
foundations of the deﬁnition now commonly used in both disciplines was set forth by
Scott (1969:262), who suggested that individuals who held both desirable and unde-
sirable beliefs about an object were ambivalent. It was also shown that negative and
positive beliefs were often only moderately correlated (Abelson et al. 1982). Political
scientists also began to more explicitly question the bipolar nature of attitudes, even
more so than the authors of Voting and The American Voter did. For example, it
was posited that conceptualizing of party identiﬁcation as only one dimension might

be incorrect (Weisberg 1980).

1.2 Ambivalence and Politics

Political scientists initially deﬁned the concept of ambivalence rather broadly. Early

deﬁnitions seemed to suggest it existed when there was any inability by an individ-

ual to articulate a position (Hochschild 1981). Such a deﬁnition is problematic as
it deﬁnes the concept based on the observable manifestation of ambivalence with-
out considering the underlying process. There are other processes that can lead to
the same behavior in individuals. Most notably, an individual with a “non-attitude”
(Converse 1964, 1970) would also have a hard time offering an opinion, but the result
is the product of something quite distinct from ambivalence. Zaller (1992:59) adds
nuance, suggesting ambivalence occurs when an individual “internalize[s] many con-
tradictory arguments” which causes him to both favor and oppose the same issue.
Alvarez and Brehm deﬁned the concept somewhat more rigorously, noting the indi-
vidual must have multiple values or predispositions that are recognized as being in
conﬂict and relevant to the decision being made (2002). Simply holding alternative
considerations does not necessarily lead to ambivalence—the considerations must be
equal in importance and recognized as in being in conﬂict. As noted above, that is
why contemporary measures account for the similarity of various attitudinal states.
A rigorous deﬁnition decreases the possibility of confusing ambivalence with re-
lated concepts. For example, there are other known properties of attitudes such as
extremity (Krosnick and Abelson 1992; Krosnick 1988) and intensity (Krosnick et al.
1993) which could conceivably trump ambivalence, or be trumped by it. It has, how-
ever, been demonstrated, through experimentation, that ambivalence is distinct from
these traits (Lavine et al. 1998). Furthermore, those who are ambivalent do not neces-
sarily end up at the midpoint of a bipolar scale and oftentimes have stronger feelings
about an object than those who are not (Eagly and Chaiken 1993). Alvarez and
Brehm (2002) similarly note that one must be careful to avoid conﬂating ambivalence
with indifference and uncertainty. Namely, as as information increases individuals
should become more ambivalent if one is to conclude that ambivalence is the source
of attitude instability. This is compared to, for example, uncertainty, which would

suggest that as information increases attitude stability would increase (Alvarez and

Brehm 1997, 2002).

The above discussion suggests that, regarding political matters, it should be more
likely to happen to those who are relatively sophisticated politically. It is only those
who recognize the conﬂict who will be affected by it. If what appears to be ambiva-
lence occurs among those who are unsophisticated, it is possible it is indifference or
apathy (Sniderman, Brody and Tetlock 1991). Some degree of contextual knowledge
is necessary to recognize the conﬂict and the connections between various predis-
positions and party identiﬁcation. There are other sources of ambivalence, such as
ideology and values. Liberals, for example, are more likely to hold conﬂicted beliefs
toward welfare as two values they rate highly, freedom and equality, are both related
to welfare (Feldman and Zaller 1992). But again, the conﬂict must be recognized.

At ﬁrst, political scientists did focus primarily on value conﬂict. Conﬂict that
has been shown to decrease response stability toward policy choices (Alvarez and
Brehm 2002; Craig, Kane and Martinez 2002; Rudolph 2005) and partisan identiﬁca-
tion (Keele and Wolak 2006). Such ﬁndings are consistent with traditional theoretical
reasoning regarding values and their impact on subsequent behavior. Values are said
to provide individuals with a way of orienting themselves in different and varying situ-
ations (Rokeach 1973). The implication that value conﬂict might lead to ambivalence
in the domain of political behavior seems intuitively reasonable.

However, it has been noted that the existence of value conﬂict is likely not common
(Alvarez and Brehm 2002). It has also been posited that the impact of values on
ambivalence is not as direct as often stated (Steenbergen and Brewer 2004). And,
when individuals do experience value conﬂict, it can be resolved by relying on group
affect (Grant and Rudolph 2003; Rudolph 2005) or value hierarchies (Jacoby 2002).
One way value ambivalence might manifest itself empirically is through the existence
of intransitive rankings of values. Studies of such intransitivity in the public show

it is actually not particularly common, and that a lack of political sophistication

and education increases the likelihood of intransitive rankings (J acoby 2006). Such
evidence suggests it is not value conﬂict that leads to ambivalence, at least when it
comes to political issues. The result is not entirely surprising—it has been known
for some time that individuals are not always able to apply broad values to concrete
situations (e.g., McClosky 1967). Consequently, one must look not only at value
conﬂict to other additional sources of partisan ambivalence.

What kinds of stimulus objects might cause ambivalence in the realm of politics?
Individuals might consider both sides of an argument about society in general (e.g.,
Feldman 1988, 2003; Jacoby 2002; McClosky and Zaller 1984; Rokeach 1973). Or,
an object slightly more speciﬁc, such as ambivalence toward government in general
(Cantril and Cantril 1999; Free and Cantril 1967) and speciﬁc institutions (McGraw
and Bartels 2005). Scholars have also looked for eveidence of conﬂicted thoughts
toward speciﬁc policies (Feldman and Zaller 1992), issues (Alvarez and Brehm 2002;
Craig, Kane and Martinez 2002; Craig, Martinez and Kane 2005), as well as toward
candidates and parties (Basinger and Lavine 2005; Lavine 2001; McGraw, Hasecke
and Conger 2003; Meffert, Guge and Lodge 2000).

One issue that has received a good deal of attention is abortion. This is under-
standable as both sides of the controversial issue have been presented to the public
with each party taking a distinct stance in a relatively short period of time (Adams
1997). Craig et al. (2002) use an objective measure to show ambivalence exists for
some people on the matter of abortion. This accords well with Alvarez and Brehm
(1995, 2002) who show as information increases, issue positions become harder to
predict in at least some cases. This coincides well with the deﬁnition above. When
two considerations, in this case values, related to a choice or object, in this case a
abortion policy, are both rated highly and in conﬂict, one should expect ambivalence.
More recently, scholars have turned toward the political parties as objects inspiring

conﬂicted feelings and beliefs.

1 .3 Partisan Ambivalence

Partisan ambivalence is an attempt to measure the conﬂict an individual might have
when thinking about the political parties, either simultaneously using a measure of
comparative ambivalence (Basinger and Lavine 2005) or separately using the stan-
dard objective measure of ambivalence (Thompson, Zanna and Grifﬁn 1995). It has
been demonstrated that an individual suffering from partisan ambivalence alters his
behavior: for example, he seeks out information in ways different from those who
are able to rely on partisan cues. Along these lines there are currently several expla-
nations for the existence of partisan ambivalence. The ﬁrst is value conﬂict. It has
been demonstrated that if two conﬂicting values are both highly rated by an individ-
ual he will show more response instability (Keele and Wolak 2006). The argument
being, the individual is receiving mixed messages and as a consequence suffers from
ambivalence.

More recently, it has been argued that one should look to informational sources
of ambivalence. For example, there is evidence that individuals with a need for
cognition—that is, those who enjoy systematically processing information—are am-
bivalent (Rudolph and Popp 2007).1 There is evidence that when knowledgeable
voters are encouraged to consider information, they hold less stable views (Barker
and Hansen 2005). One should not, however, look only to internal sources. Cam-
paign environments, for example, might induce ambivalence (Keele and Wolak 2008);
similarly, evidence suggests informative campaigns decrease incumbent biases in indi-
vidual’s evaluations and increase open-minded thinking (Kam 2006).2 Additionally,

heterogeneous social networks might encourage deviation from one’s partisan identi-

 

1Borrowing insights from social psycholog regarding systematic processing it is
noted those who engage in such thinking often end up with moderate views (Judd
and Lusk 1984; Linville 1982).

2This is consistent with ﬁndings suggesting campaigns can potentially make per-
ceptions clearer (Franklin 1991) by activating existing political predispositions (Finkel
1993).

ﬁcation when voting (Beck 2002; Beck et al. 2002). All of these explanations suggest
that individuals who are more likely to be exposed to, and subsequently consider
different points of view are more likely to be ambivalent. However, these conditions,
while necessary for the existence of ambivalence, are not themselves direct causes.
They allow for individuals to receive and possibly accept information but they do not
offer evidence as to which messages are being considered.

The study of partisan ambivalence is important due to the primacy of party iden-
tiﬁcation in political behavior research. If an individual is ambivalent toward one
or both of the political parties, his other political beliefs and behavior are possible
going to be affected. While ambivalence is often associated with attitude instability
(Hill and Kriesl 2001; Zaller and Feldman 1992), a moderation of candidate evalua-
tions (Meffert, Guge and Lodge 2000), and an alteration of information processing
(Basinger and Lavine 2005), does it ultimately affect the ﬁnal decision? If voters end
up merely delaying their decision as opposed to changing it, perhaps the study of
ambivalence is not an entirely useful endeavor, and like the study of cross-pressure,
it will fade from the attention of political scientists.

The process of deciding to turn out and for whom is dynamic (Hillygus and Jack-
man 2003; Hillygus 2005). One’s openness to campaign messages is inﬂuenced by if
one has already decided for whom to vote (Fournier et al. 2004). It is known that an
individual suffering from ambivalence toward the political parties alters his behavior:
for example, he seeks out information in ways different from those who are able to
rely on partisan cues (Basinger and Lavine 2005); he may moderate his evaluations of
candidates (Meffert, Guge and Lodge 2000); furthermore, ambivalence inﬂuences the
role issue positions play in the evaluation of candidate (Craig, Martinez and Kane
2005); and, it can inﬂuence patterns of political involvement (Mutz 2002).

If behavior is altered, it needs to be examined whether or not this is dependent

upon the political environment. Elite behavior could inﬂuence the consequences of

ambivalence. Perhaps it makes its existence more or less likely. It has been argued
that as elite polarization increases past a “tipping point” it should lead to an increase
in ambivalence for many (Johnston et al. 2008). The impact of partisanship on voting
has increased since its nadir in the 1980S (Bartels 2000); a phenomenon that has been
attributed to elite polarization and clarity in the parties’ messages (Hetherington
2001). Perhaps this is not entirely beneﬁcial to voters. Although they are now able
to see the difference between the two parties, the difference might now be too large.
Voters might be unhappy with the extreme choices with which they are presented.
Below I introduce a theory of partisan ambivalence and vote choice. As previous
ﬁndings on the subject all suggest it is those who are likely to be receptive to alter-
native arguments, I allow for campaign messages to play a central role. It is those
messages that are likely to be heard by much of the electorate that are possible source
of ambivalence. I then develop a theory as to how the existence of ambivalence will
inﬂuence an individual’s vote choice, allowing for elite behavior to greatly impact
both existence and consequences of ambivalence. I examine ambivalence toward each

party as well as toward the choice between the parties

1.4 Campaigns, Ambivalence, and Voting

I now outline an individual level theory of partisan ambivalence. What might explain
the presence of ambivalent feelings toward the two political parties? From the above
discussion it is clear ambivalence is a manifestation of an internal cognitive process.
But what exactly is this process and when is it likely to result in ambivalent feelings
toward one or both parties? I ﬁrst expand upon informational theories of ambivalence
by explicitly incorporating how campaign messages might increase ambivalence. I
then explain how conﬂicted thoughts might inﬂuence subsequent behavior, both the

actual electoral choice and the timing of the choice.

1.4.1 Campaign Messages as Sources of Ambivalence

In order for an individual to be ambivalent, he must have been exposed to both
negative and positive statements about the parties. The probability of receiving
negative statements about one’s preferred party must be large enough so that the
probability of being ambivalent is non-zero. Already, this should lead one to expect
levels of partisan ambivalence to be low.

I assume individuals who identify with a party are likely to have some “stored”
goodwill to the party, some built-in bias that would likely reduce the effects of negative
messages (e.g., Fiorina 1981, ch. 4; Achen 1992). One way of conceptualizing this
bias is think of partisanship as a “perceptual screen” (Campbell et al. 1960:133).
This is crucial because it means the probability of accepting a negative message
fundamentally different from the probability of accepting a negative message. Not
all received messages are accepted (Zaller 1992), and evidence suggests individuals
do not necessarily update their beliefs in an unbiased manner (Bartels 2002) as new
messages must compete with the opinion the individual already holds (Achen and
Bartels 2006; Bartels 1993). Because individuals may be unreceptive of negative
messages, the focus should be on those messages that will be heard by the greatest
number of voters. It is these messages that are most likely to cause ambivalence as
many will be widely exposed to them.

Regarding causes of ambivalence, we know that awareness, cross-cutting networks,
and a need for cognition certainly play a role as these make individuals more likely
to be exposed to both positive and negative messages. I argue these are indirect
sources of ambivalence—exactly which messages are getting through? Issues poten-
tially directly relate to how individual think about the two parties, and are therefore
potential sources of ambivalence. Issues might cause an individual to momentarily
reconsider his long-standing identiﬁcation. I do not argue that speciﬁc issues will

always cause voters to be ambivalent toward the parties. Instead I expect different

10

issues will cause voters to be ambivalent to the Democrats as opposed to the Repub-
licans. If information is a source of ambivalence, examining campaign messages may

provide evidence as to the kind of information to which voters are being exposed.

1.4.2 Partisan Ambivalence and the Voting Decision

There is, as of yet, no evidence that ambivalence alters the ﬁnal decision of voters.
It could conceivably do so. No matter how one deﬁnes partisanship—as a standing
decision, a psychological attachment, a social identity, eta—conflicted thoughts could
cause a voter to question his partisanship. Therefore, he would be reliant on other
forms of information and might deviate from his previous voting patterns.

Just as the inﬂuence of messages depends on the political climate, so might the
inﬂuence of ambivalence on vote choice. As the parties polarize, a voter might be
forced to choose between one of two extremes. The alternative is even less palatable
than if the parties were more similar. Polarized parties could therefore make ambiva-
lent voters, whose conﬂicted thoughts result from the polarized choices and policy,
more likely to side with their own party.

Of course, the possibility exists that ambivalence will not change a voter’s mind.
Conceptualizing partisan identiﬁcation as a perceptual screen (Campbell et al. 1960)
or a group identity (Green, Palmquist and Schickler 2002), it can be shown one
should expect ambivalence to delay but not change one’s voting decision. If over the
course of a campaign individuals alter their views and perception to align with their
party, conﬂicted thoughts should cause individuals to consider other information and
therefore take longer to decide, but in the end they will not consistently defection is
that, more often than not, they will vote consistently with their partisan identiﬁcation.
I examine the above scenarios in Chapters 4 and 5 for presidential elections in the

years 1980 to 2004.

11

1.5 The Plan of the Dissertation

In the following chapters I have three primary objectives. First, I wish to examine
the common measures. As they have been coded several different ways I hope to offer
resolution and offer guidance as how to do so in future research. Relatedly, I seek to
see if it is more appropriate to conceptualize ambivalence to the parties separately or
simultaneously.

Second, I hope to elaborate on informational theories of ambivalence. One way to
do so is by offering evidence that prominent campaign messages can serve as sources
of ambivalence. The goal here is to add to existing evidence that one should look to
informational sources. While the idea of value conﬂict was useful in the formulation
of ambivalence in the political science literatures, it now appears to be a secondary
sources of ambivalence. If campaign messages are sources of ambivalence, this will be
further evidence in favor of informational theories.“

Finally, I wish to see to what extent ambivalence inﬂuences the deﬁnitive political
act for most Americans, voting in presidential elections. No previous studies have
offered evidence if ambivalence affects vote choice. If it does not ultimately alter
an one’s vote, this would be evidence in favor of partisanship as a powerful social
identity and psychological attachment capable of shaping one’s View of politics. At
least, powerful enough to overcome a general sense of conﬂict. On the other hand, if
it does inﬂuence one’s choice, there will be evidence as to what kinds of voters are

likely to defect.

12

Chapter 2

Measuring Ambivalence

The measurement of key concepts is subject to scrutiny by methodologists and sub-
stantive researchers alike. The spotlight is justiﬁed: different measures of the same
concept often lead to different conclusions; indeed, different coding of the same mea-
sure can lead to different conclusions. From interviewer assessments, and objective
latent measures, to statistical techniques which infer the existence of ambivalence,
there is no shortage of measurement options available to the researcher. There does,
however, seem to be a growing consensus in the study of partisan ambivalence around
the use of objective measures. Consequently, it is important the the properties of
such measures are examined. Therefore, before moving on to empirically examine the
causes and consequences of ambivalence, I empirically examine the properties these
measures.

Ambivalence has been measured inferentially (Alvarez and Brehm 2002), using
objective measures (Craig, Kane and Martinez 2002; Lavine 2001; Rudolph 2005),
and subjective measures (Priester and Petty 1996; Tourangeau et al. 1989). Each of
which has been argued as a superior way of measurement. That debate exists is good:
the quality of measurement is often a yardstick as to the quality of research in a ﬁeld

(Kuhn 1977). Unwilling to settle on a single measure scholars are eager to look for

13

new and better ways to measure the concept.

At its most fundamental, measurement involves the assignment of numbers to ob—
jects according to a rule (Stevens 1951). Measurement is classiﬁcation—classiﬁcation
that is up to the researcher. It is he who takes recorded observations, from a sur-
vey or otherwise, and turns this information into data. This necessarily means that
any observed variance is potentially a result of both the concept being measured and
the process of measurement (Cliff 1993). An additional implication is that there is
no “correct” measurement for any given concept (Jacoby 1991). In fact, the mea-
surement characteristics of a concept may depend on the empirical context (Coombs
1964).

It is then not only up to the researcher to choose what he thinks to be the best
way to measure a concept, but also the appropriate level of measurement (Young
1987; Vellerman and Wilkinson 1993). Lord (1953) famously wrote of numbers—
on football jerseys speciﬁcally, but a point that certainly applies generally— “the
numbers don’t know where they came from.” Speciﬁcally, the data does not decide
how the researcher interacts with it, the researcher does (Guttmann 1977). Because
every decision potentially inﬂuences the ﬁnal results, it is crucial the properties of
commonly used measures are examined.

The results suggest the common objective measures can only place individuals
into one of three categories: ambivalent, univalent (or one-sided), and indifferent.
This is primarily a function of those individuals who offer no positive or negative
statements about the two parties. Such individuals would traditionally be placed
between ambivalence and univalence, but this appears to be inappropriate. The
results also suggest that a measure of ambivalence between the choice of parties is
appropriate and researchers can therefore use it or the more traditional measure of

these ambivalence.

14

2.1 Measures of Ambivalence

While strengths of various measures of ambivalence have been examined elsewhere
(e.g., Martinez et al. 2005), I take some time to compare and contrast each of three
common ways of measuring ambivalence. By doing so, I intend to show that objective
measures are superior. However, I also wish to caution that they must be used
carefully.

The ﬁrst alternative is to infer ambivalence through the relationship of error vari-
ance and a set of explanatory variables associated with ambivalence (Alvarez Alvarez
and Brehm 1995, 2002). A clear strength is that the measure can be used when no
other measures of ambivalence are available. However, there are drawbacks. First, it
is computationally intensive and requires a relatively large sample size. Furthermore,
there is evidence that heteroskedastic probit models suffer from fragile identiﬁcation,
potentially resulting in biased coefficient estimates (Keele and Park 2007). Another
problem is such a measure fails to rule out other possibilities convincingly, which will
necessarily be more likely when using an indirect measure. Also, it is not clear how
this would be interpreted as an explanatory variable. Such a measure fails to rule out
other possibilities convincingly, which will necessarily be more likely when using an
indirect measure. Most important for the purposes of this dissertation, how would it
be used to model ambivalence toward a party or candidate? The dependent variable
could be a dichotomous choice between the two candidates.

In this case it would be possible to model:

I
Vote,- = xz-B + e,-

var(ez') = exp(zgy)2

Where z, is a matrix of independent variables, including education, political sophisti-

cation, interest, and other variables that might cause ambivalence, and y is a matrix

15

of coefﬁcient vectors. The problem however is that Votei would represent a choice
between the two candidates. So, while it would indicate individuals are more diffi-
cult to predict it would not necessarily imply ambivalence.1 For these reasons other
options are preferable when measuring partisan ambivalence; options that go beyond
looking only at theoretically implied consequences of ambivalence.

Subjective measures of ambivalence generally ask individuals whether or not they
have mixed feelings about the stimulus object (e.g., McGraw, Hasecke, & Conger,
2003; Tourangeau et al., 1989). An advantage of such measures is that it demands
less of the respondent and might pick up cases that other measures miss. On the
other hand, a criticism of such measures is that they are more likely to be inﬂuenced
by extraneous factors (Bassili 1996) and therefore might overestimate the existence of
ambivalence. Or, they may conﬂate several sources of uncertainty such as indifference
and equivocation. Interestingly, they are only correlated weakly with other measures
of ambivalence (Priester and Petty 1996; Newbay-Clark, McGregor and Zanna 2002).

Finally there are objective measures of ambivalence. Most common is the Grifﬁn
Index (Thompson, Zanna and Grifﬁn 1995), which is a modiﬁcation of Kaplan’s mea-
sure (1972). The measure uses responses to questions such as “Is there anything you
like about the Democratic party?” to construct a numerical score of ambivalence.
Such a measure meets two conditions that are necessary and sufﬁcient for the exis—
tence of ambivalence. Individuals must have both positive and negative beliefs and
these must be of at least moderate intensity. That voting decisions are related to an
individual’s likes and dislikes toward a candidate (Kelley 1983) suggests that exam-
ining likes and dislikes is a useful endeavor. Objective measures force respondents
to consider the question while also giving them an easy out (Bishop, Tuchfarber and

Oldendick 1986), which is useful since people are willing to answer questions they

 

1An alternative would be to measure the volatility of one’s partisan identiﬁcation
(Keele and Wolak 2006). Of course, once again instability does not necessarily imply
ambivalence.

16

know nothing about (Schuman and Presser 1981). In that sense, objective measures
offer a good balance between a natural setting while not being too reliant on the in-
terviewee to spontaneously display signs ambivalence. The objective measure is less

likely to incorrectly classify an individual as ambivalent.

2.2 Objective Measures of Ambivalence

I ﬁrst focus on the standard, most often used, objective measure of ambivalence. It

deﬁnes ambivalence as:

_ Pj + Nj
Ambivalencej = ——§—— — le — le, (2.1)

where Pj is the number positive feelings and N j is the number of negative feelings
toward object j, which are obtained through open ended responses to such questions
as “Is there anything about the Democratic party you dislike?” If a person lists
ﬁve things they dislike, N would equal ﬁve. Developed by Thompson, Zanna, and
Grifﬁn (1995), and a modiﬁcation of Kaplan’s measure (1972), there seems to be a
growing consensus around the use of the above measure (Armitage and Conner 2005;
Miller and Peterson 2004). It has been shown to have desirable properties among all
objective estimates of ambivalence (Breckler 1994).

I also focus on a modiﬁcation of this measure: comparative partisan ambivalence.
Initially developed to measure ambivalence toward candidates (Lavine 2001), it has
also been used in respect to the political parties (Basinger and Lavine 2005). It is
deﬁned as follows:

D + R
2

 

Ambivalencecomp = — ID — RI, (2.2)

where D = (PD + NR)/2 and R 2 (PR + ND)/2. PD is the number of likes an

17

individual gives for the Democratic party (N R is dislikes for the Republican party,
etc.).

I choose to focus on both measures of ambivalence because each has proved to
be useful. Traditionally, ambivalence is used to refer to attitude conﬂict toward a
stimulus object, not a choice. Therefore combining attitudes about two different
stimulus objects, which is done when measuring comparative ambivalence, could be

seen as problematic. I examine whether or not this is a problem below in Section 2.5.

2.3 Levels of Measurement: Observed and Latent
Ambivalence

In previous research, the appropriate level of measurement is generally assumed and
varies from study to study. It seems reasonable to imagine that several of the scores
could be collapsed: people could be categorized as ambivalent, indifferent, or other-
wise; or, perhaps a simple dichotomy is all that can be garnered from the measure.
The important point is that the appropriate level of measurement has never been
empirically investigated.

There are at least four reasons to suspect an interval level assumption might be
inappropriate. The modal response to these open ended questions is zero (Craig 1985)
suggesting there might be some people who simply do not answer this sort of question.
As can be seen in Figure 2.1, this is the case for 2004. Second, for the comparative
measure, there are multiple ways to achieve many scores, some of which seem to be
contradictory. For example someone who dislikes both parties a great deal could very
well receive the same score as someone who is ambivalent, somewhat, toward both

parties. Are for example, all individuals with a score of 2.5 the same?2

 

2Imagine an individual who lists ﬁve dislikes for both parties. His comparative
score would be 2.5 using the measure. Now imagine an individual who lists three
likes and dislikes for the Republican party and two each for the Democratic party:

18

 

20-

15—

10‘

Percent of Total

 

 

 

 

l l I
10 15 20

Number of likes/dislikes listed

Figure 2.1: Total number of likes and dislikes toward the parties in 2004.

Third, examining the distribution of scores, Figure 2.2 for comparative partisan
ambivalence in 2004, shows that many values cluster around zero, with few in the
tails. About 70% of the observation are in the range of -1 to 1 (the distributions of
Republican and Democratic ambivalence are in the appendix). The scores in the tails
could possibly be measuring little more than noise. For the measures to be interval,
one must assume that each reason listed be given the same weight as all others—is
the fourth truly equal to the ﬁrst?

Finally, the scales resulting from equations (2.1) and (2.2) assume a speciﬁc lin-
ear relationship. As one becomes more ambivalent the score increases, with zero as
an intermediary state. The implication is that indifference is an intermediate state
between one-sided thinking and ambivalence. If this assumption is incorrect, results
treating the scales as linear may be misleading; evidence from the study of value am-
bivalence suggests the linear scales may result in unreliable results (Rudolph 2005).

For these reasons, it is worthwhile to examine the levels of measurement.

 

his score would also be a 2.5. In the ﬁrst case the individual clearl strongly dislikes
both parties, while for the second some ambivalence exists toward oth parties.

19

 

20— )-

Percent of Total

 

 

 

 

 

 

l I l I
—2 0 2 4

Comparative ambivalence

Figure 2.2: Distribution of comparative partisan ambivalence in 2004.

The appropriate level could be empirically examined using an alternating least
squares, optimal scaling (ALSOS) approach (Young 1987). Indeed, the technique
was introduced to the political science literature as a diagnostic tool (Jacoby 1999).
Familiar to psychologists, the technique is rarely used in political science research.
I therefore take some time to, brieﬂy, describe the procedure. Developed by Young
(Young, De Leeuw and Takane 1976; Young 1981), the procedure works as follows:
starting with Y and X, which would contain the original values as determined by the
measurement processes being used for the dependent and independent variiables, two
new matrices, Y* and X*, are constructed.3 In addition to determining the starting
values the researcher sets measurement characteristics of the observed variables as

well. The following model is estimated:

Y* = x*e* + E*. (2.3)

 

3For a more in—depth discussion of ALSOS see Young (1981) and Giﬁ (1990); see
Jacoby (1999) for political science examples.

20

The process is iterative and alternates between estimating coefficients and an
optimal scaling phase that obtains better ﬁtting values for the observed variables,
subject to a function that conforms to the speciﬁed level of measurement for each
variable. In each stage of the process the values from the other are held constant.
For example, when values of the observation matrices are being optimized to improve
the ﬁt the parameter estimates are held constant. The process continues until the
model ﬁt is no longer improved, until the sum of squared residuals can no longer
be minimized any further given the starting values and measurement characteristics
speciﬁed in Y* and X* . The primary difference between this and a standard least
squares regression is that instead of only estimating values of coefﬁcients to maximize
ﬁt, the values of variables are also changed.

Of interest to the problem at hand is that the procedure can be used as a diagnostic
tool to examine the level of measurement. Starting with the assumption of ordinal
data, one can simultaneously test both an interval and ordinal level assumption, and
possibly a nominal level assumption as well. To show why this is the case I brieﬂy, and
for the most part informally, discuss how the new matrices are formed and show how
different restrictions on the transformed variable, depending on the speciﬁed level of
measurement, inﬂuence the new values. Speciﬁcally, how lower levels of measurement
mean fewer restrictions and therefore are more ﬂexible. First, to construct Y* and
X* one sets the initial values, usually the original values in X and Y. Then one
determines the level of measurement of each variable (y; $1,352, ..., 33k). Once this is

done we have for the dependent variable:

v* = fgm. (2.4)

Where f; is an order preserving function and q determines the level of measurement

21

speciﬁed (nominal, ordinal, interval, ratio). For the independent variables:

X* = if8(xo)fi1<x1)f§ (x2)...f,3<xk>1. (2.5)

f; is an order preserving function in that it conforms to restrictions that guarantee
at least the speciﬁed level measurement is met. For example, imagine xj is set to be
an ordinal variable. The new variable would be a monotonic transformation of the
original because of the restrictions in equations (2.6) and (2.7) that are placed on f;
(where q = 0 represents that the function preserves an ordinal level). fJQ(xJ-) maps

. *.
x] to xj.

*_

.rij > xhj => 23:]- 2 3:70. (2.7)

It is the second restriction that forces the transformed variable to be a monotonic
transformation of the original. If the second restriction is relaxed then one would not
necessarily see such a transformation—indeed, only the ﬁrst restriction is used when
a variable is set at the nominal level. The empirical implications are that although
fJQ(xJ-) and f}? (xj) could result in the same transformed value, this is not guaranteed
as the a nominal transformation is actually more ﬂexible.4

Setting the variable to ordinal provides enough ﬂexibility to examine whether
or not the variable can be treated as nominal, ordinal, or interval. When plotting
transformed values against the original values, one could see a line parallel to the

x-axis if there’s no discrimination between values, a 45° line if the variable can be

treated as interval, and anything in between these two extremes. If an ordinal level

 

4Similar mapping functions, with different restrictions, are used depending on the
level of measurement speciﬁed for a given variable (see J acoby 1999); see Young (1981)

for a discussion of how actual values are assigned to transformed variables, including
a discussion of different algorithms.

22

assumption is appropriate, the spacing between different values would be unequal. It
is also possible for several values to collapse to the same value. If such an occurrence
is severe enough it could be the case that the variable can only be considered nominal.
'Ifeating an ordinal level variable as interval is problematic if the overall ﬁt of the
model is found to be quite a bit better after the values have been transformed. If,
on the other hand, overall ﬁt only improves slightly the violation probably can be
viewed as posing little problems for other modeling assumptions.

I use the procedure to test the assumption that objective measures of ambivalence
can be treated as an interval, ordinal, or nominal level variable. I test this on the
standard measure as well as that of comparative ambivalence. I make no claims about
the underlying process, as the procedure examines the empirical manifestation of the
process. Two ALSOS regressions were performed. The ﬁrst was a model of vote
choice in 2004. The following explanatory variables were included: political aware-
ness, education, party identiﬁcation, ideology, issue positions, as well as measures of
Republican ambivalence and Democratic ambivalence. Vote was measured by sub-
tracting an individual’s logged thermometer rating of Kerry from the logged rating
of Bush, as suggested by J acoby (1999), allowing the measure to be treated as a con-
tinuous interval level variable. Results for Republican and Democratic ambivalence
are shown in Figure 2.3.

A similar approach was used examining the properties of comparative ambivalence.
In this case the model of partisan ambivalence in 2004 from Rudolph and Popp (2007)
was replicated, using an ALSOS approach.5 As before, the transformed values of the
measure are compared to the original values. The results are in Figure 2.4.

Both Figures 2.3 and 2.4 suggest the measures carmot be treated as interval level

variables. The measure of comparative partisan ambivalence particularly violates

 

5The variables included are: political knowledge, education, need for cognition
caring about the election, a need to evaluate, the number of evaluative responses and
measures of value conﬂict. The coding is described in the appendix. Similar coding
is used in subsequent analyses, except where noted.

23

 

 

 

 

 

 

 

 

(D 0

s 5‘ ‘ : 5‘ '

.D _ _

E 4— — E 4

(I! <6

.9 a— — r: 3‘ F

a .3

8 2— — T3 2' _

E D. 1_ _

(I) a)

o 1— — o:

'5 ‘5 0- -

S 0‘ ‘ 3 -1- _

s _,_ _ s

s s -2- -

s -2— — s

.9 .9

a a

g I I I I I I I I g I I I I I I I I

l- -2 — 0 1 2 3 4 5 I— -2 —1 1 2 3 4 5
Original value of Democratic a ambivalence Original value of Republican ambivalence

Figure 2.3: Transformation of original data values for Democratic and Republican
ambivalence, obtained from ALSOS regression analysis in 2004. Both measures were
speciﬁed at the ordinal level. Variables were used as explanatory variables.

such an assumption, scores collapse to -1, 0, and 1. While the results for Republican
and Democratic ambivalence are not as striking, they too suggest there is not much
discrimination possible beyond ambivalent or not. It was noted earlier that these
ﬁndings would not be terribly problematic if the R2 obtained from the ALSOS re-
gression is only slightly higher than from the OLS regression. Examining Table 2.1
shows that this is not the case. The R2 is twice as large after the values for com-
parative partisan ambivalence have been optimally scaled. On the other hand, when
Republican and Democratic ambivalence are used as independent variables model ﬁt
is only incrementally improved. This is not a result of the different measures, but
instead appears to be a function of whether or not the measures are used as depen-
dent or independent variables; model ﬁt vastly improves after the ALSOS regression

when the measures are used as dependent variables.6 Before settling on a ﬁnal coding

 

6This suggests including the measures as control variables without recoding should
not pose signiﬁcant problems. However, if they are used as key explanatory variables,
recoding is necessary.

24

 

 

 

 

 

I I I I I I I I
-2-101 2 3 4 5

Original value of comparative ambivalence

Transformed value of comparative ambivalence

Figure 2.4: Transformation of original data values for comparative partisan ambiva—
lence speciﬁed at the ordinal level, obtained from ALSOS regression analysis in 2004.
The variable was modeled as the dependent variable.

scheme one lingering question remains: what about those individuals who offer no

likes or dislikes about either party? This topic is examined in the next section.

Table 2.1: R2 before and after optimal scaling

 

 

OLS ALSOS
Comp. ambiv. .127 .241
Vote choice .380 .417

 

 

2.3.1 The Zero Responses

One of the primary concerns of treating the measures as interval level variables was
that a large number of respondents offer no likes or dislikes for either party. It is
unclear if such individuals can be treated the same as others. In order to see if
this is the case, another ALSOS regression was run, this time specifying the variable

comparative partisan ambivalence as nominal. The transformed values are no longer

25

restricted to be a monotonic transformation of the original values. If the zero re-
sponses are unproblematic then results should be similar to Figure 2.4. If not, the
zeros should be distinctly different from all other values. The results in Figure 2.5 are
unambiguous: zero responses should be treated differently. So long as zero responses
are included, the measure is unquestionably only categorical and treating it at higher

levels could lead to incorrect substantive conclusions.

 

 

 

 

IIIIIIII
~2-1012345

Original value of comparative ambivalence

Transformed value of comparative ambivalence

Figure 2.5: Transformation of original data values for comparative partisan ambiva-
lence when speciﬁed to be nominal level, obtained from ALSOS regression analysis
in 2004. The variable was modeled as the dependent variable.

What if zero responses are excluded? The results, in Figure 2.6, are quite simi-
lar regardless of whether or not comparative ambivalence is originally speciﬁed as a
nominal or an ordinal level variable. This provides some conﬁdence that the measure
is distinguishing between individuals who are ambivalent, those who are one sided in
their evaluations, and those who offer no likes or dislikes.

Another way to gain leverage on such a question would be to model the total
number of items listed using a zero inﬂated count model to see if there is a different

process generating those individuals who list no like or dislikes compared to all oth-

26

 

 

 

 

 

 

 

 

I I | I | I l I

Transformed value of comparative ambivalence
I
|

Transformed value of comparative amblvalence
J
I

I I I I I I | |
—2 -1 0 1 2 3 4 5 — 0 2 3 4 5
Original value of comparative ambivalence Original value of comparative amblvalence

Figure 2.6: Transformation of original data values for comparative partisan ambiva-
lence excluding those who offered no likes or dislikes, obtained from ALSOS regression
analysis in 2004. On the left the measure was speciﬁed to be nominal level, on the
right ordinal. The variable was modeled as the dependent variable.

ers. To do, so I ran a zero inﬂated negative binomial count model (Gurmu 1991) as
a function of the intensity of one’s partisan identiﬁcation, political awareness, educa—
tion, and one’s interest in politics. The results are in Table 2.2 with standard errors

in parentheses.

Table 2.2: Zero inﬂated count model for likes and dislikes listed

 

 

Count equation Binary equation
Party ID intensity 0.034 (0.023) -0.836** (0.117)
Political awareness 0.097** (0. 016) -0.331** (0.077)
Education 0.076** (0.015) —0.163* (0.080)
Interest 0046* (0. 019) - 0.467** (0.084)
N 975
Log-likelihood -2216.316
ﬁlo) 176.076

 

Signiﬁcance levels : f : 10% * : 5% ** : 1%

The results support the idea that two processes are at work. The same variables

that have an impact on how many items are listed also inﬂuences whether or not an

27

individual lists something at all. Political awareness and education increase the prob-
ability of listing more likes and dislikes, and as is indicated by the negative coefficient
for the binary equation model, if someone lists anything at all. Partisan intensity
only inﬂuences the binary portion of the model. Examining the effects graphically is
indicative that across levels of awareness there are two different processes at work. In
Figure 2.7 it is shown that while the two probabilities converge as awareness increases
there are still some even at the highest level with a higher likelihood of never listing

anything.7

 

0.3—

.0
"1’

   
 
 
  

“3°...
Os from blnary equation """" o

Pr(List zero)

 

 

 

0.1 - —
093 from count equation ----- o
‘v“ ...o
~~$~~

o—-_o___o

0.0- —
i I I I I I I
0 3 6

Political knowledge

Figure 2.7: Political knowledge and Pr(listing zero likes or dislikes) from zero-inﬂated
count model.

The above results suggest two possible coding schemes. The ﬁrst suggests indi-
viduals should be classiﬁed indifferent, ambivalent, or univalent. An individual is
indifferent if he offers no likes or dislikes to the open ended questions.8 If his score

from equations (2.1) or (2.2) is less than zero he would be primarily one sided in his

 

7For Figure 2.7, predicted probabilities were calculated by holding all variables at
their means.

8It should be noted that those classiﬁed as indifferent could conceivably be clas-
siﬁed as having a “non-attitude” (Converse 1964), but it is unclear what exactly, in
this case, the empirical distinction would be between the two.

28

thinking and would be coded as univalent. If the score is greater than zero, the in-
dividual should be coded as ambivalent. If one’s substantive theory suggests there is
no difference between univalent and indifferent individuals, coding individuals simply

as ambivalent or not is appropriate.

2.4 Empirical Examples

The above ﬁndings are especially important if different measurement schemes lead
to different substantive conclusions. To see if ﬁndings are robust to assumptions of
the level of measurement, I model partisan ambivalence with the recoding suggested
by the ALSOS regressions above. For the empirical examples I focus on comparative
ambivalence as the dependent variable.

Rudolph and Popp (2007) argue that it is not only value conﬂict (e.g., Alvarez
and Brehm 2002; Keele and Wolak 2006) that causes ambivalence; instead, one should
also look to other, informational, sources of ambivalence. For example, an individual’s
ability and interest in seeking out multiple sources of information should inﬂuence the
likelihood of experiencing partisan ambivalence: as one has more sources of informa-
tion and therefore more points of view, one is more likely to experience ambivalence.
Their analysis ﬁnds evidence for both explanations, I examine if such ﬁndings are

identical under ways of coding the variable.

2.4.1 A Dichotomous Measure of Ambivalence

I ﬁrst model comparative partisan ambivalence by coding individuals as ambivalent
or not (indifferent and one sided thinking individuals are placed in the same cate-
gory) using logistic regression. Included are variables associated with ambivalence:

measures of value conﬂict, speciﬁcally conﬂict between egalitarianism and limited

government and egalitarianism and moral traditionalism; one’s need for cognition,

29

measured using two questions from the N ES; a measure of one’s need to evaluate;
a measure of political knowledge using factual questions from the NES; education;
the intensity with which one holds one’s partisan identiﬁcation; and an interaction
between partisan intensity and political knowledge; and ﬁnally, a measure of the num-
ber of items listed. The results are in Table 2.3, where entries represent coefﬁcient

estimates with standard errors in parentheses.

Table 2.3: Partisan ambivalence

 

 

 

 

Variable Coefficient (Std. Err.)
Political knowledge 0.369n (0.109)
Education 0.033 (0.057)
Need for cognition 0.221I (0.121)
PID intensity 0.228 (0.184)
Care who wins -0.383 (0.270)
PID Intensityxknowledge -0.160** (0.049)
Egalitarianism -3.145* (1.480)
Limited government -0.838 (0.844)
Moral traditionalism -2.087 (1.473)
Egalitarianismxlimited government 2.044 (1.363)
Egalitarianismxmoral traditionalism 3.201 (2.356)
Need to evaluate 0.095 (0.425)
# of evaluative responses 0.254“ (0.026)
Intercept -1.141 (0.975)
N 970
Log-likelihood -496.315
X%13) 200.798
Pseudo R2 0.1683

 

Signiﬁcance levels: 1 : 10% * : 5% ** : 1%
Estimated using logistic regression

How one ”codes the variable affects substantive conclusions. For example, once
coding the variable as dichotomous, value conﬂict plays no role in explaining partisan
ambivalence in 2004. It is only measures that proxy contextual knowledge, political
knowledge, education, and a need for cognition, that now best explain whether or
not an individual is ambivalent or not. Whereas before the two explanations were

viewed as both offering insight as to who will be ambivalent toward the parties,

30

now only one adds explanatory power. However, such a model treats indifferent and
univalent individuals as the same, while in Section 2.3.1, it was clearly demonstrated
that individuals who offer no likes or dislikes could, if the theory dictates, be treated
differently.

In the case of information, value conﬂict, and ambivalence there is reason to sus-
pect indifferent and univalent individuals should be treated differently. An individual
who has an interest in seeking information is not likely to be indifferent; he may,
on the other hand, be univalent or ambivalent depending on how he processes the
information. Similarly, an individual experiencing value conﬂict probably will not be

univalent, but the conﬂict might lead to ambivalence or indifference.

2.4.2 The Possibility of Indifferent Individuals

Based on coding in Basinger and Lavine (2005), those who offer no likes or dislikes
are coded as indifferent, those with a comparative ambivalence score greater than
zero are coded as ambivalent, and those with a score less than zero—individuals
with primarily negative or positive reactions—are coded as univalent. Therefore, the
appropriate model would be a multinomial regression, in this case I use a multinomial
logistic regression. In Table 2.4, the columns represent indifferent and ambivalent,
the base category is univalent.9 The entries are coeﬂicients showing the inﬂuence the
variable has on leaving the base category, with standard errors in parentheses.-

The ﬁndings strongly support the claim that the three categories are distinct.
Political knowledge has an inﬂuence on whether or not one is indifferent, but not
ambivalent. Similarly, education inﬂuences ambivalence but not indifference.10 Most

interesting are the coefﬁcients for the variables representing motivational goals: inten-

 

91n this model I remove the variable # of items listed as it perfectly predicts
indifference.

10The coefﬁcient for education between the categories indifference and ambivalence
is greater than zero and statistically signiﬁcant.

31

Table 2.4: Partisan ambivalence

 

 

Variable Indifferent Ambivalent
Political knowledge -0.216 (0.186) 0410* (0.161)
Education -0.210* (0.100) 0.073 (0.067)
Need for cognition 0.102 (0.209) 0.222 (0.149)
PID intensity -0.955** (0.306) -0050 (0.277)
Care who wins -1.182* (0.479) -0.713 (0.448)
PID Intensityxknowledge —0.013 (0.095) -0.111 (0.069)
Egalitarianism -0.893 (2.838) -1513 (1.766)
Limited government -1.001 (1.430) -1.249 (0.977)
Moral traditionalism 1.240 (2.657) -0.621 (1.752)

Egalitarianismxlimited government 0.932 (2.432) 2.860l (1.594)
Egalitarianismxmoral traditionalism -1.147 (4.289) 0.012 (2.776)

 

Need to evaluate -1.500* (0.761) 0.586 (0.520)
Intercept 5.438** (1.901) 0.696 (1.300)
N 975

X?28) 198.605

Pseudo R2 0.1624

 

Signiﬁcance levels : Jr : 10% >1: : 5% an: : 1%
Estimated using multinomial logistic regression, base category is univalence

sity of partisan identity and if the individual cares who wins. Both variables predict
indifference but not ambivalence. Another interesting ﬁnding is that value conﬂict
now does predict ambivalence; furthermore, it does not predict a difference between
indifferent and univalent individuals.

The relationship between partisan intensity and the probability of being classiﬁed
in each of the three categories is examined graphically in Figure 2.8.11 It has been
argued that the intensity of one’s partisanship should reduce one’s likelihood of expe-
riencing ambivalence. This is shown not to be the case in Figure 2.8; instead, it has
no clear effect on ambivalence. However, it does reduce the likelihood of indifference,
which makes intuitive sense: those individuals who are care strongly about the par-

ties are unlikely to be indifferent. The ﬁnding is identical to the count model where

 

11Figure 2.7 was constructed by holding all other variables at their means except
PID intensityxknowledge, which varies with the value of PID intensity while holding
knowledge at its mean.

32

intensity was signiﬁcant in the binary equations, but not the count equation.

 

 

 

Univalence Ambivalence Indifference
I I l I I I I l I
0.5— 0.6- 0.6-
5 _ _ _
g 0.4- 0.4- 0.4- \\
Q _ _ vvvvvvv _ \~
9
0- 0.2- o.2- 0.2—
_ _ _ I. _ _

 

 

 

 

 

 

 

 

 

I I l I
Intensity of partisan
identification

Figure 2.8: The impact of partisan intensity on univalence, ambivalence, and indif-
ference.

Clearly, substantive conclusions depend on how the variable is coded. Rudolph
and Popp (2007) treated the variable as continuous and found partisan intensity to
be a signiﬁcant predictor of ambivalence. However, once coded as dichotomous the
ﬁnding disappeared. This appears to be the case because intensity actually inﬂuences
indifference, which was shown earlier to be qualitatively different from ambivalence.
One could examine other substantive conclusions—for example the role value conﬂict
plays, have depending on the coding scheme—the important point is they differ from

one scheme to the next.

2.5 A Measure of Comparative Ambivalence

Tfaditionally, ambivalence is used to refer to attitude conﬂict toward a stimulus ob-
ject, not a choice. Therefore combining attitudes about two different stimulus ob—
jects, which is done when measuring comparative ambivalence, could be problematic.
Speciﬁcally, I examine whether or not it is appropriate to measure ambivalence com-
paratively, which instead of looking at two stimulus objects independent of one an-

other comparative ambivalence measures to attitudinal conﬂict regarding the choice

33

between the two objects. Such a measure assumes feelings between the two parties
are related. Only if negative feelings about one party are generated from the same
process as positive feelings toward the other is such a measure appropraite. Many
question the extent to which attitudes toward the two parties related (e.g., Alvarez
1990; Weisberg 1980). It has been argued once controlling for measurement error
(Craig, Martinez and Kane 1999; Green 1988) or when using more psychologically
grounded measurements (Greene 2005) attitudes about the two. parties are related.
While there is evidence positive and negative evaluations of the same object are of-
ten independent of one another (Abelson et al. 1982; Cacioppo and Bernston 1994),
it has not been ﬁrmly established that positive and negative reactions toward two
competing objects are related.

If the process that generates negative feelings toward party generates negative
feelings toward the other, one would, for example, expect Democrats on average to
list zero negative statements about the Democratic party and zero positive statements
about the Republican party. Statements offered about each party, broken down by
partisanship, are shown in Figure 2.9, where the plotted points corresponds to the
average number of statements listed and the horizontal bars represent plus and minus
one standard deviation. The patterns is as one would expect: each individual offers
about 1.5 positive statements for his preferred party and 1.5 negative statements for
the opposing party. However, at the 95% conﬁdence level these are distinguishable
neither from how many statements are listed about the opposing party nor from zero.
So, while the picture looks as one might expect, it does not offer support for a measure
of comparative ambivalence (it does not refute the notion either).

Another way to examine if listing positive statements about one party and negative
statement of hte other are empirical manifestations of the same process is to perform
a factor analysis. If this is the case, a two factor solution with negative Democratic

statements and positive Republican statements onione factor and negative Republican

34

 

 

 

 

 

 

 

 

 

 

 

 

Positive Democrat O . 0 Republicans
0 Democrats
Negative Democrat _._3
Positive Republican __._ 0
Negative Republican 0 ‘
I m I i
0 1 2 3

Average number of responses

Figure 2.9: The average number of positive and negative statements listed toward
the parties broken down by political party. The plotted points represent the mean
number of statements and the horizontal bars represent one standard deviation.

statemtns and positive Democratic statements on the other should be found. The
reasoning is as follows: in equation (2.2), D and R should be distinct in order for
the ﬁrst term on the right-hand side to measure intensity and the second term to
measure similarity, as the Grifﬁn index does for a single stimulus object (Thompson,
Zanna and Griﬂin 1995); as each is made up negative statements toward one party,
and positive toward another, these should be related. When performing the factor
analysis included in addition to the likes and dislikes toward the parties were: partisan
identiﬁcation, ideology, and thermometer scores for the two parties. A two-factor
solution was found (a scree plot is in the appendix). The results are presented below in
Figure 2.10, after performing a varimax rotation on the solution. Positive statements
about the Democratic party and negative statements about the Republican party have
high values on factor 2 and are nearly zero on factor 1. Similarly positive statements
about the Republican party and negative statements about the Democratic party

have high values on factor 2 and are nearly zero on factor 1.

35

 

1 _
SH
oF’D
OParty ID

0'5 q OTherm Rep.
‘1‘ Ideology
o
g ONFI
“- o

0.0 7 PD

—0. 5 ‘ TherméDem.

 

 

 

—1'.0 -0.5 0:0 0:5 1:0
Factor1

Figure 2.10: Results from factor analysis of positive and negative statements listed
about the parties, as well as partisanship, ideology, and party thermometer scores.

Finally, I see how well each items scales with the others. If one takes the neg-
ative and positive responses for each party to form a scale, the scale in unreliable
with a Cronbach’s a of 0.124 for the Democratic Party and 0.121 for the Republi-
can Party. However, if one forms a scale from the negative responses toward one
party and the positive responses for the other—as the factor analysis above suggests
appropriate—the resulting scales are indeed quite reliable: Cronbach’s a=0.755 when
including negative Democratic responses and favorable Republican; 0.742 for nega-
tive Republican and favorable Democratic responses. Each of the resultant scales
correlate with party identiﬁcation relatively highly (-0.4025 for positive Democratic
and negative Republican responses; 0.4633 for negative Democratic and positive Re-
publican responses). The two resulting indices seem to be unrelated, the correlation
between the two 0.166. The marginal and joint distributions for pro-Republican and
pro-Democratic responses for those who listed at least one statement are displayed in

Figure 2.11.

36

 

 

 

 

 

 

Pro-Democract
O N -k 0) co

 

 

   

 

I I I I I I
0246810 0246810 02468
Pro-Republican Pro-Democrat Pro-Republican

Figure 2.11: Marginal and joint distributions of Pro-Republican and Pro-Democratic
responses for those individuals who list at least one likes or dislike.

As can be seen in the far right portion of Figure 2.11—which is a. two-dimensional
histogram where darker shades indicate a greater number of respondents—there is
no discernible relationship between pro-Democrat and pro-Republican responses. It
therefore can be concluded that while one’s thoughts about the two parties are related
to one another, this is only the case in that pro-responses toward one party are related
to negative responses toward the other party, the D and R terms in equation (2.2).
And, because these two terms can be thought of as independent of one another, a
measure of comparative ambivalence is valid. Therefore, measures toward each party
as well as a measure of choice ambivalence can be used depending on what one’s

theory suggests is more useful.

2.6 Discussion and Conclusion

While the proceeding sections could be interpreted negatively, there is good news.
Even though the best coding scheme appears to discard variance, at least some of the
variance is a result of measurement error. Also, there is a clear difference between

those who are ambivalent and those who are not. The measures are certainly able to

37

capture a difference between these two kinds of individuals. Furthermore those who
offer no likes and dislikes are also clearly different. This is useful because such ﬁndings
are tied to well developed theories and an established literature. Second, overall ﬁt of
the models are not considerably better when using the optimally scaled variables as
explanatory variables suggesting including the measures of control variables should
not be problematic. On the other hand, when using the measures as an dependent
variable or the key explanatory variable, one should be skeptical.

This leads to some practical advice. The results of the ALSOS regressions offers
empirical support for Basinger and Lavine’s (2005) coding of individuals as ambiva-
lent, univalent, or indifferent, with those offering no responses being indifferent. This
means that depending on the research question, when modeling the variable as the
dependent variable one should categorize individuals as ambivalent or not. Or, if
the question of interest suggests that there should be a difference between univalent,
and indifferent individuals then a trichotomous nominal dependent variable should
be used. This suggests that binary or multinomial methods should be used when
ambivalence is the dependent variable. If the variable is being used simply as a con-
trol variable including it as is, as a pseudo-interval level measure, without recoding
will probably not result in biased or misleading results. If, on the other hand, it is
a variable of interest in explaining some political phenomenon and will be used to
calculate such things as predicted probabilities more reliable results will be obtained
if the variable is recoded accordingly.

The above results also offer support for a measure of comparative ambivalence.
If positive feelings about one party are related to negative feelings of another, the
idea of choice ambivalence is appropriate. Therefore, depending on the question of
interest either measure could be used, but one must be careful how it is coded. In
conclusion, objective measures of ambivalence are useful in distinguishing between

those with conﬂicted thoughts, but less so in measuring the intensity of such feelings.

38

Appendix to Chapter 2

Measures

Political knowledge was measured by asking a series of factual questions to create an
index. For 2004 the questions were: knowledge of Tony Blair (v045164), of William
Rehnquist’s position (v045165), of Dennis Hastret’s position (v045162), and the ofﬁce
of Dick Cheney (v045163), of which party controlled the House (v045089), and the
Senate (v045090). For this index, Cronbach’s a=0.801.

To measure need for cognition in 2004, I followed the coding in Rudolph and
Popp (2007) using the following variables: “Which type of problem do you prefer to
solve: SIMPLE or COMPLEX?” (v045221); and how much does the individual like
responsibility for thinking (v045220a).

For value conﬂict, I used the coding in Keele and Wolak (2006) and included
egalitarianism, moral traditionalism, limited government; and, the conﬂict between
egalitarianism and moral traditionalism, and between egalitarianism and limited gov-
ernment.

To measure party identiﬁcation intensity, I modiﬁed the standard measures of
party identiﬁcation by taking the absolute value of the the median subtracted from
the respondents response, or lx, — 3|, resulting in a variable that is coded 0 to 3, with
3 being most intense.

Education is measured on a 7 point scale.

39

Republican and Democratic Ambivalence

 

 

Percent of Total
Percent of Total

 

 

 

 

 

 

 

 

Republican ambivalence Democratic ambivalence

Figure 2.12: Republican and Democratic ambivalence in 2004.

Factor Analyis

 

 

 

 

3_ _

8

22— —

(U

>

C

d)

F.”

Lu

1_ _
o_ _
I I 1 I I r I I
1 3 5 7

Number

Figure 2.13: Screeplot from factor analysis.

40

Chapter 3

Campaign Messages as Sources of

Partisan Ambivalence

By the early fall of 1980, a year after the Iran hostage crisis began, inﬂation and
unemployment were rising with no signs of slowing. Against this backdrop, Ronald
Reagan asked the American people a simple, powerful question: “Are you better
off now than you were four years ago?” While this question may have crystalized
the doubts many people had, it was the culmination of months of negative coverage
regarding the economy and foreign affairs. These major campaign themes compelled
many Americans to pause and think about the direction of the country. Even long
time Democrats may now have had negative thoughts about their party.

Recently, scholars of ambivalence have moved away from value conﬂict as the pri-
mary source of ambivalence and have begun to look to other, informational, sources.
It could be those individuals who are subject to two-sided information ﬂows who are
likely to be ambivalent. To expand upon the example above, I argue the nature of
campaign messages are such that they might cause an individual, and the electorate,
to question long-standing identiﬁcations, even if only brieﬂy. So long as messages

are able to spread through the electorate sufﬁciently, they might consistently pre-

41

dict ambivalence. In this chapter I intend to elaborate on informational theories of
ambivalence in two ways. First, I hope to show that one of the primary implicit as-
sumptions in the theory is true: that accepting positive and negative messages about
the parties leads to ambivalence. Second, I wish to offer some evidence as to what
precise issues are leading to ambivalent individuals. The ﬁndings provide support
for an informational theory of ambivalence. Campaign messages play a modest but
consistent role in predicting ambivalence. Therefore, for the ﬁrst time direct evidence

of what messages might be causing ambivalence is found.

3.1 Campaigns, Information, and Ambivalence

Focusing only on political knowledge and reception for the moment gives insight as to
what kind of messages are likely to be sources of ambivalence. As knowledge tends to
be a prominent explanation of both receiving political messages and ambivalence, it
seems a natural starting point. Furthermore, a simple model can quickly demonstrate
the sort of messages that one should look to as potentially inﬂuencing ambivalence.
Before moving on, I wish to explain more precisely what I mean by campaign mes-
sages. In this case, I am referring to the major campaign themes, those issues that are
brought up again and again. How these messages, or themes, are dispersed through
the electorate is not the primary goal of this chapter. Instead, I simply look to see
if the kind of message—or theme, or issue—that might be considered prominent is
associated with ambivalence.

Consider the following simple model of reception:
Pr(Rm = 1) = F (am + ﬁmKnowledge),

where a is function of how easy a message is to hear regardless of political knowledge,

6 is a discrimination parameter, m is the message in question (which can be classiﬁed

42

as either positive, p, or negative n), and F is a link function. It is those messages
that would have low levels of a that would be most likely to be received by the
greatest number of people. As am decreases, the probability that the message will
be received is higher for all values of political knowledge. As it decreases enough,
then one need not even worry about the value of 5m as Pr(R/nﬁ will be nearly the
same for all individuals and therefore will not vary over values of political knowledge.
Now, incorporating knowledge into the discussion, one would look to messages with
high values of 6 because a small increase in knowledge would lead to a signiﬁcantly
greater chance of receiving the message.

Therefore, the focus should be on those messages that will be heard by the greatest
number of voters: those messages most widely dispersed. This is especially true
because most of the discrimination is probably happening at the reception stage
(Dobrynska and Blais 2008), not at the acceptance stage. As an example to show the
kind of messages I am referring to, Figure 3.1 displays a message likely to be heard
by many (low values of a and high value of 6) represented by the solid line, and
one what is unlikely to be heard by all but the most informed, the dashed line. The
picture makes clear how important the role of am is. Imagine if the solid line had a
higher value of a, it would shift to the right, and although political knowledge would
still be helpful, only a few individuals would hear the message.

Implicit in an informational theory of ambivalence is that individuals who are
exposed to more information are more likely to accumulate both positive and nega-
tive beliefs, or whichever they do not already have. Regardless if it is the campaign
environment, discussion networks, or some other source of information, the argument
necessarily implies that individuals who are exposed to information must be hearing
both positive and negative messages about the parties in order to be ambivalent. It
follows that each individual has some latent probability of accepting and receiving

positive and negative messages about the parties, and therefore some latent proba-

43

 

Pr(Receive)

 

 

 

Political knowledge

Figure 3.1: A message likely to be head by many represented by the solid line and
one what is unlikely to be heard by all but the most informed, the dashed line.

bility of experiencing ambivalence. A theoretical model of the probability that an

individual is ambivalent could be then expressed as:
Pr(Ambivalence) = Pr(An & Rn) X Pr(Ap & Rp). (3.1)

Rather obviously, it is clear that for ambivalence to exist, Pr(An & Rn) x Pr(Ap & Rp)
> 0, meaning the value of each term on the right-hand side must be greater than zero.
Equation (3.1) is not itself a measure, but a theoretical account of the underlying
processes that empirical measures of ambivalence attempt to capture: the listing of
likes and dislikes about the parties. In that regard, although model looks very familiar
to measures that deﬁne ambivalence as the product of conﬂicting and thoughts, P x N,
this is actually not the case. It is just as related to the measure of ambivalence
provided in equation (2.1).

In order for equation (3.1) to be an accurate representation of ambivalence, then

Pr(Aann) must be independent of Pr(Alep). There is empirical evidence to sup-

44

port such an assumption: for example Kaplan (1972) found only an average cor-
relation of -.05 between positive and negative beliefs; and, even when considering
measurement error, there still appears to be only a modest relationship between pos-
itive and negative thoughts (Thompson, Zarma, and Grifﬁn 1995). I also empirically
examine this assumption below.

Regarding causes of ambivalence, we know that knowledge, cross-cutting networks,
and a need for cognition certainly play a role as these make individuals more likely to
be exposed to bout positive and negative messages. I argue these are indirect sources
of ambivalence—which messages are getting through? Issues potentially directly re-
late to how individual think about the two parties, and are therefore potential sources
of ambivalence.

Issues, or messages, or themes of a campaign might cause an individual to mo-
mentarily reconsider his long-standing identiﬁcation. Regarding various issues I have
only general expectations: different issues will cause voters to be ambivalent to the
parties depending on context. That is, different issues will be related to the two
parties in different ways. I do not argue that speciﬁc issues will always cause voters
to be ambivalent toward the parties. Instead I expect different issues to cause voters
to be ambivalent to the Democrats as opposed to the Republicans. I use the measure

of choice ambivalence to test the robustness of the ﬁndings.

3.2 Research Design

To examine the above theoretical account of ambivalence and campaigns, I ﬁrst clas—
sify individuals as holding ambivalent feelings toward one of the two parties, both,
or neither political party. Such a classiﬁcation scheme allows the use of multinomial
regression, which permits a direct test of whether or not topics important in cam-

paigns can have different a impact on the probability of being ambivalent toward

45

the Republican and Democratic parties. Additionally, I test the hypotheses in two
presidential elections, 1980 and 2004. I choose these elections as 2004 represents one
of relative polarization of the two parties while in 1980 the two parties were relatively
close. Results suggest campaign messages are sources of ambivalence, and inﬂuence
the likelihood of being ambivalent toward the two parties in different ways.

In order to classify individuals as ambivalent I make use of a measurement scheme
as suggested in Chapter 2 and equation (2.1). I dichotomize the variable so that any
individual with a score greater than zero is classiﬁed as ambivalent toward the party
in question. Those individuals who have a score greater than zero for both parties
are categorized as ambivalent toward both parties. Coding is summarized in Table

1.1 I use the measure comparative ambivalence to test the robustness of the ﬁndings.

Table 3.1: Ambivalence measures and categorization.

 

 

 

Amivaep. Amivaem_ Categorization
>0 30 Republican ambivalence
SO >0 Democratic ambivalence
>0 >0 Both
_<_0 30 None

 

3.2.1 Other Measures

I include other potential sources of ambivalence in the model. Political knowledge is
measured using questions asking factual questions about the political system. Addi-
tionally, education is included using a 7 point scale from the NES. Similarly, interest
is included using a question asking respondent’s interest in the campaign. In all three

cases we should expect increased contextual knowledge and an ability to assimilate

 

1For the few cases where a Republican would be classiﬁed as ambivalent toward
the Democratic parties, I recoded these individuals as not experiencing ambivalence
(I recoded Democrats similarly). This makes interpreting the results more straight
forward. Very few cases need to be recoded and substantive results remain unchanged.

46

such knowledge to increase the likelihood of ambivalence. Because individuals who
have a greater need for cognition and need to evaluate are more likely to experience
partisan ambivalence a scale measuring such need is included (see Rudolph & Popp
2007 for more detail). Also included are measures of value conﬂict between egalitari-
anism and limited government, as well as egalitarianism and moral values (see Keele
& Wolak 2006 for more detail). In 2000, I also include measures of discussion network
heterogeneity (Huckfeldt, Mendez and Osborn 2004).

Included are measures of individuals’ responses to prominent campaign messages.
Included in each year are a domestic issue and a foreign policy issue, each which
could be considered salient for the year in question. To reiterate, I hope to capture
explicitly what other accounts of ambivalence imply: a source of ambivalence, not
just the type of environment that makes being exposed to such a source likely. For
1980, the two issues are the individual’s opinion on Carter’s handling of inﬂation
and his handling of the hostage situation. In 2004, Bush’s taxcuts and one’s opinion
of the Iraq war are included. While there might be an issue with including items
related directly to a speciﬁc president’s performance, I do so for two reasons. The
ﬁrst is practical, in both years the only measures related to some of the prominent
campaign issues ask about the president’s performance. Second, it is not unreasonable
to suppose an incumbent’s performance will be a major point of discussion during a
campaign, individuals might update their opinion based on the evaluation of previous

performance (Fiorina 1981; Lodge, Steenbergen and Brau 1995).

3.2.2 Modeling Ambivalence

In the multinomial regression, the base category is holding no ambivalent feelings; the
other categories are ambivalent toward the Democratic Party, ambivalent toward the
Republican Party, and ambivalent toward both parties. This makes the interpretation

of coefﬁcients relatively straight forward: they represent the effect a variable has on

47

changing the probability that one holds no ambivalent feelings toward entering one of
the other categories. Importantly, this allows for variables to have different effects for
different categories. One could imagine that a different set of determinants inﬂuences
the probability an individual feels ambivalent about both parties as compared to just
one. It also seems possible there are different reasons why one might feel ambivalent
about Republicans as opposed to Democrats.

A strength of modeling partisan ambivalence using multinomial regression is that
it reﬂects the theoretical model presented above. It is possible for an individual to be
ambivalent toward neither, one, or both parties, depending on which messages he has
heard and accepted. Meaning, each term in equation (3.1) can be expressed as its own
equation as a function of a set of explanatory variables. Those with high probabilities
of accepting both positive and negative messages will have a high probability of being

ambivalent.

Hypothesis 3.1a: The predicted probability of listing a positive statement about
a political party is unrelated to the predicted probability of listing a negative

statement about the same party.

Hypothesis 3.1b: The predicted probability of listing a positive statement about a
political party multiplied by the predicted probability of listing a negative state-
ment about the same party should approximately equal the predicted probability

of being classiﬁed as ambivalent.

The main question posed in this chapter is whether or not including these mes-
sages and individual’s opinions on them in the model adds any explanatory power.
Additionally, including these should give insight as to which issues are likely to cause
an individual to experience ambivalence. So, issues will predict ambivalence but co-
efﬁcients will not be equal across outcomes. That is to say, the expectation is that

ﬂkt,Rep. will be different from ﬁkt,Dem. for at least some of the campaign messages,

48

where t is the year being examined, and k is the issue, where ﬂkt j represents the
impact variable k in year t has on entering category j from the base category, no

ambivalence.

Hypothesis 3.2: Different issues will impact the probability an individual is am-

bivalent toward Democrats, Republicans, or both parties.

3.3 Empirical Results

3.3.1 Accuracy of Theoretical Model

To demonstrate the accuracy of the theoretical model, I run separate a logistic regres-
sion for each term in equation (3.1) for the both parties. Therefore, positive messages

for the Republican party, for example, could be modeled as:

Pr(ARep.p,i&RRep.p,i = llxi) = A(xéB).

Where ARep.p,i&RRep.p,i is coded as 1 if the respondent offers at least one posi-
tive statement about the Republican party and A a function linking the underlying
tendency to a probability, and is the cdf of the logistic distribution.

After running the model for positive and negative messages for both parties, I
calculate predicted probabilities. If the assumption that the two probabilities are
independent is true, the correlation between the two obtained probabilities should be
low and there should be no discernible relationship between the two values. .As can
be seen in the left of Figure 3.2, which is for the Republican party in presidential
election years 1980-2004, the two probabilities are unrelated.

If the theoretical model is accurate, not only should these two predicted proba-

bilities be unrelated, when they are multiplied by one another the result should be

49

 

IIIIII
2004

33?:53'1-‘2 F” ‘t .

 

      
     

 

99999
waomo

IIIII

 

 

l I | I |

ppppe

N A O) on O
Pr(Republican Ambivalence)

 

Pr(Recall negative)

 

 

99999
nammo
Llll

 

 

 

 

 

 

 

' 3231.91.84;
I II I I I T
0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.8
Pr(Recall positive) Pr(RecaIl negative)xPr(Recal| positive)

Figure 3.2: Testing assumption that Pr(APIRp) is independent of Pr(Aann). Addi-
tionally examining the extent to which the theoretical model is empirically true, for
presidential years 1980 to 2004.

approximately equal to the predicted probability obtained from a model of
Pr(Rep. Ambiv.,- = llxi),

as this is the relationship suggested in equation (3.1).2 This is shown in the right-
hand side of Figure 1. As can be seen, the relationship is linear and appears to be
quite strong. As noted earlier, the model in equation in (3.1) looks quite similar to
measures of the product of conﬂicting and thoughts, P x N. However, it states the

latent tendency to be ambivalent is related to the latent tendency to receive and then

 

2We should expect Pr(An&Rn) x Pr(Ap&Rp) to slightly overestimate
Pr(Ambivalence) because the model of receiving and accepting a message is ac-
tually of recalling at least one message. Some individuals who have recalled at
least one ositive and one negative item ma have recalled more of one category
and therefgre are not ambivalent. This actu ly makes seeing a correlation between
Pr(An&Rn) x Pr(Ap&Rp) and Pr(Ambivalence) less likely.

50

measures of the product of conﬂicting and thoughts, P x N. However, it states the
latent tendency to be ambivalent is related to the latent tendency to receive and then
recall positive and negative messages about the parties. The cross-product measure
is based on the actual number of items listed. Equation (3.1) is theoretical model
of objective measures of ambivalence, not itself a measure. Also, while it has been
shown that the cross-product model is correlated with the Grifﬁn index (Priester and
Petty 1996), this is somewhat misleading as the relationship between the two will
almost always be non-linear: individuals with negative scores on the Grifﬁn index
will often have a zero on the cross-product measure—I show that this is the case for
Republicans and Democrats in 2004 in the appendix. This demonstrates that the
linear relationship shown in Figure 3.2 is distinct from correlations between different
objective measures of ambivalence. Instead, the ﬁgure is evidence that the propensity
to be ambivalent is a function of the product of the propensities of being able to recall
both negative and positive messages about a party.

More important than simply a strong linear relationship between the two predicted
probabilities, the loess curve shows that the two values are not a bad approximation
for one another. This ﬁnding generalizes to the Democratic party, which is shown in
the appendix. As an additional test to examine how similar the two probabilities are
to each other, I calculated a concordance correlation coefﬁcient (Lin 1989). The coef-
ﬁcient, rc, looks not only for a linear relationship between two variables but how much
agreement there is between observations across the two variables, that is does vari-
able 23 equal variable z?3 The coefﬁcients, displayed in Figure 3.3, demonstrate that
two probabilities are good approximations for one another for both parties. Because
the two key assumptions of the theoretical model are met, running a multinomial

regression is an appropriate method to investigate such a model.

 

3The coefﬁcient is bounded between -1 and 1 'ust like Pearson’s r; for example,
if 582:2,- for all i, re would equal 1. For a detai ed discussion of the concordance

correlation coefﬁcient and some applications see Wittenberg (2008).

51

Republicans Democrats

 

 

 

 

 

 

 

 

I I I I I I I I L I
2004 —o— --o-
2000 «WWW... -, _.__O_.. .._ _nr..-,_ .0
1996 M" ~ --O —-O-
1992 o - -~4-<>
1984 2..-. _. ._ _ _. ”n.0,. . _. __. _ ...O
1980 —- ~ ~<> - - ---O

I I I I I I I I I

0.8 0.9 0.8 0.9

Concordance correlation coefficient

Figure 3.3: Concordance correlation coefﬁcient, with 95% conﬁdence intervals, for
presidential years 1980 to 2004, between the relationship of Pr(Alep) x Pr(AnIRn)
and Pr(Ambiv.).

Support is found for hypotheses 1a and 1b. Moreover, the support is found in
all presidential election years from 1980 to 2004. Beyond the practical importance of
these ﬁndings for the empirical analysis below, the fact that these two probabilities
are found to be statistically independent of one another seems to have its own intrinsic
importance. That individuals positive and negative thoughts about the two parties
can be thought of as independent of one another has potentially large implications
for other areas of research in political behavior. This ﬁnding coupled with those from
Section 2.5 suggest examining in more detail how expressions of thoughts about the

two parties are related is a fruitful area for further research.

3.3.2 Sources of Ambivalence

The results of the multinomial regression are in Tables 3.2 and 3.3, one for each year

of analysis.4 The columns in both tables represent each of three outcomes (Demo-

 

4Overall levels of ambivalence for all years are in the appendix.

52

cratic ambivalence, Republican ambivalence, and both). The entries are coefﬁcients
showing the impact the variable has on leaving the base category (no ambivalence),
with standard errors in parentheses.5 The control variables by and large behave as
expected. Political information and interest predict ambivalence. Similarly, a need
for cognition and a need to evaluate predict ambivalence.

The results in Tables 3.2 and 3.3 sprovide support for Hypothesis 3.2. One’s opin-
ion of Carter’s handling of foreign affairs and inﬂation inﬂuences if one is ambivalent
toward the Republican party, but not the Democratic party. As individuals take a
view more favorable to the Republican party, they are less likely to be ambivalent
to Republicans. All issue variables were coded so that higher values correspond with
pro-Republican responses. The ﬁndings are displayed graphically in Figure 3.4 and
provide the same conclusion.6 '

Table 3.3 shows the results for the analysis performed on 2004. Looking ﬁrst at
one’s support for the tax cuts, as one moves from strongly approving to strongly dis-
approving the tax cuts, he becomes less likely to be ambivalent toward the Democratic
party and being ambivalent toward both parties. A clearer picture of this can be seen
in Figure 3.5. This suggests, perhaps, that an individual who normally might be sup-
portive of the Democratic party is unhappy with the extent to which the Democrats
acquiesced. The ﬁgure also shows that as one becomes more approving of the war
he is less likely to ambivalent of the Republican part. In both years, each issue has
a different inﬂuence on Republican and Democratic ambivalence offering support for

Hypothesis 3.2.

 

5In both years, both Hausman (1984) and Small— Hsiao (1985) tests suggest the
assumption of IIA is not violated.

6Predicted probabilities for Figures 3 and 4 were constructed using SPOS T (Long
and Freese 2005).

53

Table 3.2: Predicting Partisan Ambivalence in 1980

 

 

 

 

 

Dem. Rep. Both

Party ID 0.049 (0.059) 0.114 (0.075) -0113 (0.070)
Ideology -0004 (0.079) 0.076 (0.102) 0.127 (0.095)
Issues
Inﬂation -0100 (0.135) -0.310I (0.181) -0.304I (0.172)
Hostage —0.074 (0.074) -0.216* (0.089) -0.012 (0.090)

situation
Sophistication
Education 0.018 (0.016) -0.004 (0.030) 0.020 (0.017)
Knowledge 0175* (0.085) 0.364** (0.106) 0.499” (0.100)
Interest 0.291I (0.160) 0.011 (0.196) 0509* (0.199)
Intercept -2.399** (0.702) 2145* (0.898) -4.210** (0.877)
N 821
Log-likelihood -589.375
x(39) 141.754
Pseudo R2 0.112
Signiﬁcance levels: T: 10% *: 5% *z 1%

Pr(Ambivalent)

 

0.10-

0.08-

0.06-

0.04-

 

 

 

I I I
Approve to Disapprove

Carter’s handlng of inflation

Pr(Ambivalent)

 

0.15-

0.10—

 

0.05“

 

 

I r I
Approve to Disa

Carter’s handling 0

 

—0— Democratic

- -O - - Republican

 

 

Figure 3.4: The impact of campaign messages in 1980.

54

prove

f) hostages

Table 3.3: Predicting Partisan Ambivalence in 2004

 

 

 

 

 

Democratic Republican Both
Party ID 0218* (0.107) -0052 (0.114) 0252* (0.105)
Ideology —0.221I (0.134) 0309* (0.155) 0.241I (0.130)
Issues
Iraq war 0.063 (0.503) -0.868l (0.494) -0.685 (0.466)
Tax cuts -0.273* (0.122) 0.136 (0.128) -0.327** (0.114)
Sophistication
Knowledge 0.293** (0.106) 0.308** (0.114) 0.113 (0.100)
Education 0045 (0.095) 0238* (0.100) 0.075 (0.092)
Interest -0042 (0.120) 0279* (0.135) -0154 (0.118)
Cognitive style
Need for cognition 0.305 (0.213) 0.015 (0.206) 0.391I (0.201)
Need to evaluate 1668* (0.703) 0.038 (0.749) 0.490 (0.678)
Value conflict
Moral trad. 0.220 (1.546) 0.789 (1.621) -1.860 (1.568)
Limited gov. —1.245 (1.319) -O.619 (1.359) 0.765 (1.251)
Egalitarianism -0031 (0.050) -0.015 (0.050) -0.060 (0.047)
Egalitarianismx -0.439 (2.174) -0.889 (2.442) 1.860 (2.296)
Moral trad.
Egalitarianismx 1.672 (2.128) 2.858 (2.258) -0.198 (2.038)
Limited gov.
Intercept -2.167l (1.258) -5.077** (1.340) 2432* (1.206)
N 556
Log-likelihood -589.375
Xf39) 141.754
Pseudo R2 0.112
Signiﬁcance levels : I : 10% * : 5% ** : 1%

55

 

 

 

 

 

 

 

 

 

0.30— -
016*
93 0.20— — e
(U (U
.2 .2
g g 0.12—
< .1 , _ <1:
* —. .— ‘.
0 00_ _ 0.08~
' I I I I I , I I I I I I
Strongly Oppose to Strongly-Favor DIsapprove to Approve
Tax Cuts War in Iraq

 

—0— Democratic --O-- Republican

 

 

 

Figure 3.5: The impact of campaign messages in 2004.

3.3.3 Trends

Figures 3.6 and 3.7 show the inﬂuence of prominent campaign messages on the prob-
ability of being classiﬁed as ambivalent for presidential election years 1980 through
2004. As before, the x-axis ranges from the most liberal to most conservative re-
sponse. In each year at least one policy predicts ambivalent toward at least one
party. As can be seen, in some years the impact is small and in others, nearly non-
existent. Importantly, as the theory suggested, the issues inﬂuence the probability
of being ambivalent to the two parties in different ways in most years. The theory
also suggested that foreign and domestic policies will have different effects in different
years which can be seen in the ﬁgures.

The impact of these messages on ambivalence is modest. The change in probability
is never much greater than 0.10 moving from the most liberal to most conservative
position. These messages are necessarily measured with signiﬁcant error; for example,
only when scales are created of many issues can a strong relationship be found between

issues and vote choice (Ansolabehere, Rodden and Snyder, Jr. 2008), something much

56

1 992

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

H I
or o
o o
l - I-
II I
I I
I I
I l
I
gl I_§ / _
1 l j g
N o N
o o :5 a“:
_ -3
I \I D.
I \
I \
I I
I \
I \
a ..g I._
l l j I
m o 0! o
o o o o
I“ "
/ \
I
I
I I
o / co
$3 I/ 3 ll
I l T l
N. O. 91 0.
o o o o

(lueIeAlqu)Jd

Figure 3.6: The impact of foreign policy campaign messages. The solid line represents
Democratic ambivalence, the dashed Republican. The :r-axis moves from most liberal
to most conservative position.

57

 

1 992

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

 

I l
“I o
O O
\ .. _
l . ‘
\ \
I \
I \
co I <1- \
§ I I ._ § \ _.
I l I I
N O N O
o o' o o :9
_ c '5
I \ 0.
I
I
I \
I \
I o \
g I _ § \ _
I I I I
N q N O
O O O O
I " \ "
I \
I
I
I
8 ’ 8
93 I I _ 93 I _
‘I I I T
N O N O
o' d o' o'
(IUGIBA!QwV)Jd

Figure 3.7: The impact of domestic policy campaign messages. The solid line repre-
sents Democratic ambivalence, the dashed Republican. The :r-axis moves from most
liberal to most conservative position.

58

easier to measure than ambivalence. It is then not that surprising that these messages
do not have the same impact as other variables such as political knowledge. Indeed, in
some respects when considering the small inﬂuence a single issue has on vote choice,
it is impressive the effects are as large as found here.

The evidence supports an informational theory of ambivalence. Other informal-
tional explanations suggest the experience of hearing multiple sides of an argument
increases ambivalence. What I have shown is evidence of the kinds of arguments
are being considered. Certainly, I am not making the case these are the only, or
perhaps even the primary, arguments. Instead, I suggest they are pervasive enough
they should be considered as starting points. More than likely, while an issue that
is very important to an individual is a more likely cause ambivalence, testing such a
hypothesis would require specially collected data. The primary conclusion that can
be drawn is that those issues which are brought up again and again in a campaign,
what I refer to as campaign themes, are potential sources of ambivalence. I have also
offered the ﬁrst speciﬁc evidence of what takes place in environments that lead to

ambivalence.

3.3.4 Alternative Measures of Ambivalence

To test the robustness of the above ﬁndings, I replicate the model above using the
measure of comparative ambivalence. I dichotomize the measure so those ”with a
score greater than zero are modeled as ambivalent and all others as not (there are
no substantive differences for the variables of interest if I account for indifference).
Several measures used above were altered for the study of comparative ambivalence:
partisanship, ideology, and issue opinions are recoded as the intensity with which one
holds the identiﬁcation and attitudes. The results for 2004 are in Table 5, where the
entries represent coefﬁcient estimates, with standard errors in parentheses.

The results provide support for the theory. The intensity with which a respondent

59

Table 3.4: Predicting comparative partisan ambivalence

 

 

 

 

 

 

Variable Coefficient (Std. Err.)
PID intensity -0.276** (0.105)
Ideological intensity -0.023 (0.1 12)
Issues
Iraq opinion intensity -1.392JI (0.722)
Tax-cuts intensity -0.024 (0.188)
Political sophistication
Political knowledge 0169* (0.073)
Education 0.125I (0.067)
Interest 0.258 (0.174)
Cognitive style
Need for cognition 0.221 (0.145)
Opinion 0.517 (0.510)
Value conﬂict
Mortal traditionalism -2.379 (1.707)
Limited government -0.621 (0.940)
Egalitarianism -2.755 (1.744)
Egalitarianismx 2.957 (2.727)
moral traditionalism
Egalitarianismx 1.960 (1.536)
limited government
Intercept -0.296 (1.118)
N 555
Log-likelihood -337.202
X%14) 51.12
Signiﬁcance levels : I : 10% >I< : 5% ** : 1%

.60

holds his opinion on the Iraq war are associated with being less ambivalent. In this
model, such a variable represents a motivational goal; as one cares more about the
issue, one is less likely to be ambivalent between the two parties. Such a ﬁnding
strengthens the argument that campaign messages are sources of ambivalence, as the

ﬁnding is robust to how ambivalence is measured.

3.4 Conclusion

The analysis set out to examine the extent to which campaign messages, those topics
widely discussed and covered, could be possible sources of ambivalence toward the
American political parties. Speciﬁcally, in a given year issues impact the likelihood
an individual is ambivalent toward the Republican party differently than it does the
Democratic party.

The inﬂuence is modest. It certainly does not match that of political knowledge.
In light of the crudeness of the measures, as well as measurement error, the ﬁndings
presented can be thought of as adding to base of evidence that information is the
source of ambivalence. Additionally, the ﬁndings offer strong support for the idea
that the process driving ambivalence is a result of cognition and not affect. The
strongest predictor is political knowledge which proxies the concept of contextual
knowledge about the political system. Furthermore, that Pr(Alep) and Pr(AnIRn)
are independent of one another suggests that positive messages messages are not
received at the expense of receiving negative messages.

In conclusion, the ﬁndings suggest the theoretical formulation of partisan ambiva-
lence is accurate. That is, conceptualizing ambivalence as the probability of receiving
and accepting positive and negative messages is well represented by the empirical ﬁnd-
ings. Furthermore, the notion that issues are the messages receives empirical support

as well. That the ﬁndings are conditional on both the party and the year conforms

61

with a general understanding of how issues impact the political system, suggesting
they are related to the parties in different ways and over time the importance of a
given issue changes over time. The ﬁndings in this chapter add to existing evidence
that one should expect to see ambivalence among those most in tune with the political
system as one must have a certain base of contextual knowledge to understand one’s

own inconsistent thoughts, should they exist.

62

Appendix to Chapter 3

Variables

Political knowledge was measured by asking a series of factual questions to create an
index. For 1980 the questions were: did the individual express an opinion regarding
the tax cut proposal (v800322)?; did the individual know the names and parties of
Congressional candidates in his or her district (v800826, v800829)?, did the individ-
ual know which party was more likely to support a stronger national government
(v801131)?, has the individual been paying attention to polls (v800129)? For 1980
the knowledge measure has a Cronbach’s a of 0.5622.

Included in the 1984 scale were: Does respondent konw the names of candidates
running for Congress (v840741)?; which party has house majority before (v841006)
after election (v841007)?; and for Senate before (v841008) and after (v841009)? The
scale has a Cronbach’s a of 0.743.

For 1988: knowledge Rehnquist’s position (v880873), Gorbachev’s job (v880874),
Jim Wright (v880877), and which party controlled the House (v880878) and which
the Senate (v880879), and the names of the candidates runnig for congress (v880569).
The scale has a Cronbach’s a of 0.599.

The questions included for the 1992 scale were: does the respondent know which
party ismore conservative (v9205915)?; what ofﬁce Dan Qualye (v9205916), Rehnquist
(v9205917), Yeltsin (v9205918), and Foley (v9205919) hold; if the respondent knows
who is resopnsble for determining if a law is unconstitutional (v9205920); and if the
respondent know which party controlled the House before the election 11 (v9205951).
The Cronbach’s a is 0.764.

Included for 1996: Al Gore’s position (V961189), Rehnquist’s position (V961190),
Boris Yeltsin’s position (V961191), Newt Gingirch’s position (V961192), and and the
who conttrolled House before election (V961072). Cronbach’s a =0.827.

63

For 2000 the questions were: What job or political oﬂice does he NOW hold?
(v001450); What job or political ofﬁce does he NOW hold? (v001453); What job
or political ofﬁce does he NOW hold? (v00147); What job or political office does
she NOW hold? (v001456); the names of the Candidates in running for office in
the respondent’s district (v000347,v000351). For 2000 the knowledge measure has a
Cronbach’s a of 0.680.

For 2004 the questions were the same as those used in the proceeding chapter.

To measure need for cognition in 2000, the same questions were used as in 2004.

For, I used the coding in Keele and Wolak (2006); moral traditionalism was not
available until 1992. Similarly, for discussion networks, I followed the coding in Huck-

feldt et al. (2004).

Campaign Messages

Included in each year were two issues, one domestic and one foreign, that were promi-
nently involved in the campaign. For 1984 the issues were the two parties handling
of war (created from a scale of three variables; V840247—840249 with a Cronbach’s
a of 0.651), and one’s opinion about women’s rights. For 1988, the included issues
were crime—Bush made a large deal of Dukakiss handling of crime as governor of
Masaschussetes, and Reagan’s handling of the Contra situation. In 1992, the foreign
policy question was whether or not we continue the war, the domestic was Bushs
handling of economy. In 1996 Clinton’s foreign policy and his handling of healthcare.
For 2000, the two issues are if the budget surplus should be used to protecting social
security and Clinton’s handling of foreign affairs, that is if one approves of an active

role for the US. in world affairs.

64

Testing Assumption for Democratic Ambivalence

   

1.0 0.8
3% A 0.6
04 8 0-4
0.2 g 0.2
’g 0.0 a 0.0
-.= .2
§’ 1.0 .2
0.8
c 0.6 g
= 0.4 .5
§ 0.2 93
E 0.0 8
E 1.0 CE) 0.8
0.8 o 06
06 T: '
0'4 0. 0.4
0:2 0.2 _ -
0.0 0.0
0.0 0.4 0.8 0.0 0.4 0.8 0.0 0.4 0.0 0.4
Pr(RecaIl positive) Pr(RecaIl negative)xPr(RecaIl positive)

Figure 3.8: Testing assumption that Pr(AplAp) is independent of Pr(AnlAn). Addi-
tionally examining the extent to which the theoretical model is empirically true, for
Presidential years 1980 to 2004.

65

Comparing Measures of Ambivalence

To demonstrate that the ﬁndings in Figure 3.2 are distinct from high correlations
found between the cross-product measure of ambivalence and the Grifﬁn index, I
show the empirical relationship below, which is clearly non-linear (the plotted points
are jittered). Intuitively, it is easy to see that this will generally be the case. Many
individuals found to be univalent with the Grifﬁn index will have scores of zero for

the cross-product measure, which is what leads to the non-linear relationship.

 

 

   

 

 

 

 

 

 

Democratic Republican
I I l I I l I I
25 - «f - 25 - t—
1-5 20 - -- - *5 20 - ' —
3 3 4A...
915‘ .. - 315- -
g 10 - - ‘3 10 a -
2 9
o 5 - — o 5 - -
0-;7'. - o-I"'“‘ -
I I I I I I l l
-2 o 2 4 -2 o 2 4
Grifﬁn index Grifﬁn index

Figure 3.9: Comparing measures of ambivalence.

66

Model Results: 1984-2000

Tables for results from 1984 to 2000 to be inserted here.

Table 3.5: Predicting Partisan Ambivalence in 1984

 

 

 

 

 

 

Democratic Republican Both
Party ID 0.000 (0.044) -0.009 (0.051) -0.026 (0.048)
Ideology -0.013 (0.044) 0.042 (0.051) 0.060 (0.049)
Issues
war_defense 0.031 (0.078) 0.064 (0.090) 0.094 (0.085)
Women’s rights -0.074Jr (0.044) -0.044 (0.050) 0.020 (0.045)
Sophistication
Knowledge 0.151** (0.043) 0.174** (0.049) 0.216** (0.046)
Education 0.247** (0.052) 0.256** (0.060) 0.373** (0.056)
Interest 0.549** (0.126) 0282* (0.141) 0.377** (0.135)
Intercept -4.247** (0.430) -4.461** (0.481) -5.434** (0.470)
N 1800
Log-likelihood -1520.714
Xf21) 232.259
Signiﬁcance levels: 1’: 10% *2 5% lot: 1%

67

Table 3.6: Predicting Partisan Ambivalence in 1988

 

 

 

 

 

Democratic Republican Both
Party ID 0.089 (0.068) 0.079l (0.044) 0.087 (0.065)
Ideology -0.086 (0.070) 0.025 (0.042) 0.000 (0.065)
Issues
crime 0.336I (0.173) 0.022 (0.108) 0.124 (0.162)
contra 0.124 (0.151) -0.036 (0.100) -0.080 (0.143)
Sophistication
Knowledge 0.364** (0.100) 0.313** (0.065) 0.538” (0.092)
Education 0276** (0.077) 0.255** (0.050) 0.470** (0.073)
Interest -0.316** (0.106) -0.203** (0.065) -0.052 (0.093)
Value conﬂict
Egalitarianism 1214* (0.572) 0.423 (0.333) 0.450 (0.514)
Limited gov. 0652* (0.292) 0.197 (0.168) 0.196 (0.266)
Egalitarianism -0.282* (0.122) -0.105 (0.072) -0.084 (0.109)
x Limited gov.
Intercept -7.810** (1.487) -3.796** (0.829) -6.912** (1.336)
N 1319
Log-likelihood -1094.78
Xfm) 266.36
Pseudo R.2 0.109
Signiﬁcance levels: I: 10% a: : 5% **I 1%

68

Table 3.7: Predicting Partisan Ambivalence in 1992

 

 

 

 

 

Democratic Republican Both
Party ID -0.126l (0.066) 0170* (0.068) 0152* (0.073)
Ideology 0.075 (0.089) 0233* (0.098) 0.073 (0.103)
Issues
continuewar 0.226 (0.212) -0.021 (0.214) 0.128 (0.223)
economy -0.105 (0.138) -0.372** (0.129) -0.401** (0.140)
Sophistication
Knowledge 0.282** (0.063) 0227** (0.066) 0.370** (0.074)
Education 0.124l (0.068) 0.217** (0.070) 0.402“ (0.075)
Interest 0064 (0.083) -0.138 (0.088) —0.145 (0.092)
Value conﬂict
Egalitarianism -0.322 (0.366) -0.514 (0.354) -1.132** (0.380)
limited -1.600* (0.808) -1190 (0.842) -1.800* (0.829)
Moral trad. 0.160 (0.319) -0.281 (0.364) -0.692l (0.361)
Egalitarianism 0.586lL (0.308) 0.338 (0.304) 0.536l (0.315)
x Limited gov.
Egalitarianism -0073 (0.131) 0.087 (0.137) 0.187 (0.144)
x Moral trad.
Intercept -2.753* (1.208) -2.553* (1.291) -1.382 (1.288)
N 1350
Log-likelihood -1092.627
st1) 266.992
Pseudo R2 0.108
Signiﬁcance levels: I:10% *: 5% ** 1%

69

Table 3.8: Predicting Partisan Ambivalence in 1996

 

 

 

 

Democratic Republican Both
Party ID 0.086 (0.085) 0.010 (0.082) 0.022 (0.081)
Ideology 0220i (0.127) 0.069 (0.129) 0.109 (0.126)
Issues
domestic 0.363 (0.456) 0913* (0.462) 0.535 (0.451)
clinton_foreign 0.105 (0.126) 0282* (0.124) 0.136 (0.123)
Sophistication
Knowledge 0.211I (0.121) 0.199I (0.120) 0327* (0.129)
Education 0.020 (0.084) 0.117 (0.081) 0172* (0.082)
Interest 0064 (0.187) -0.138 (0.187) -0.189 (0.184)
Value conﬂict
Egalitarianism 0.692 (0.693) -0.565 (0.727) -1093 (0.722)
Limited gov. 2.460I (1.458) -1.981 (1.398) -0.127 (1.367)
Moral trad. -0.634 (0.535) 0.361 (0.517) 4301* (0.535)
Egalitarianism -0.558 (0.402) 0.618 (0.414) 0.083 (0.396)
x Limited gov.
Egalitarianism 0.180 (0.151) -0015 (0.155) 0381* (0.160)
x Moral trad.
Intercept -6.282* (2.659) 4320 (2.627) -1.833 (2.596)
N 1164
Log-likelihood —779.283
3%,) 98.542
Pseudo R2 0.060

 

Signiﬁcance levels : I : 10% * : 5% ** : 1%

70

 

 

 

 

 

Table 3.9: Predicting Partisan Ambivalence in 2000
Democratic Republican Both
Party ID —0228** (0.064) -0.020 (0.066) -0.063 (0.068)
Ideology 0.051 (0.070) 0.144I (0.078) 0.117 (0.081)
surplus -0003 (0.100) 0235* (0.095) 0.141 (0.102)
Foreign affairs
-0.060 (0.102) 0.178I (0.101) 0.020 (.109)

Sophistication
Education 0.225“ (0.069) 0173* (0.072) 0.392** (0.078)
Political knowledge 0.251** (0.063) 0.329“ (0.067) 0.329** (0.069)
Interest 0037 (0.072) 0.139I (0.077) 0.013 (0.080)
Need for cognition 0.113 (0.294) 0.001 (0.308) 0720* (0.337)
Need to evaluate 0.441** (0.122) 0.385** (0.126) 0.233I (0.133)
Value conﬂict
Moral trad. -0.871 (1.417) -1.831 (1.408) -3.021* (1.524)
Limited gov. 1.647I (0.849) 0.127 (0.827) 0.219 (0.913)
Egalitarianism -0734 (1.295) 3593* (1.522) -3.462* (1.478)
Egalitarianismx 2.672 (2.003) 4586* (2.210) 6.166** (2.257)

Moral trad.
Egalitarianismx -1.634 (1.259 0.042 (1.297) -0.317 (1.367)

Limited gov.
Discussion network
# Gore supporters 0.153 (0.120) 0291* (0.135) 0.000 (0.145)
# Bush supporters 0.349** (0.105) 0.181Jr (0.107) 0.043 (0.122)
# Gore supportersx -O.178 (0.116) -0.044 (0.117) 0274* (0.119)

#Bush supporters
Intercept —4.555** (1.068) -4.769** (1.141) -4.648** (1.191)
N 1225
Log—likelihood -1139.8
Xf51) 366.965
Pseudo R2 0.139
Significance levels : I : 10% : 5% ** : 1%

71

Chapter 4

Partisan Ambivalence and Voting

Behavior

Imagine a typical Democrat in the fall of 1980. It is once again time for him to consider
politics. Will he rely, as he often has, on his partisanship to make sense of the political
landscape? Recently, the country has suffered through the embarrassment of the Iran
hostage crisis while experiencing both high inﬂation and unemployment. Such a voter
might now be ambivalent toward his preferred party and consider sitting this election
out, or possibly defecting to vote for the Republicans.

Now, consider a Democrat in 2004. Her party lost a close election four years
ago, only to acquiesce to Republicans on several major issues as though a mandate
for such policies existed. She too might be unhappy with her party. However, in a
time of political polarization sitting this election out or defecting seems unimaginable.
Her conﬂicted thoughts result from an understanding of the polarized parties and she
might now be more likely to vote Democratic.

An individual experiencing ambivalence toward his preferred party may be inﬂu-
enced in different ways in different eras. When both parties are perceived as relatively

similar, individuals may be more likely to vote based on competence and advantages

72

on valence issues. On the other hand, if one recognizes the large difference between
the two parties, one is aware that the alternative is considerably further from his
own preferred position than his party is. I argue that the increasing importance of
partisanship on vote choice hides, for many voters, mixed feelings about the current
political climate.

Examining trends from 1980 to 2004, limited evidence is provided that while
ambivalence once predicted abstention, in 2004 it predicts the opposite. Although
polarization may make differences between the two parties clearer, to some voters,
this is not necessarily entirely good news. However, for the most part, partisan
ambivalence toward a speciﬁc party does not predict either abstention or defection.
In fact, it is only in 1980 and 2004 that ambivalence is related to voting, and only for

Democrats.

4.1 Congressional Polarization

Recently, scholars began to question the consensus existing around the notion of the
declining importance of parties to the mass electorate (e.g., Nie, Verba and Petrocik
1979; Wattenberg 1998; Niemi and Weisberg 1976). The impact of partisanship on
voting has increased since its nadir in the 19808 (Bartels 2000); a phenomenon that has
been attributed to elite polarization and clarity in the parties’ messages (Hetherington
2001). As the two parties polarized, an increasing number of survey respondents
began offering “one sided” opinions about the two parties. However, such a coding
scheme ignores information. An individual who offers four likes and three dislikes
would be classiﬁed the same as someone who offers ﬁve likes and no dislikes. This
is precisely what measures of ambivalence try to capture. While individuals are now
better able to see important differences between the two parties, many might view

this as a mixed blessing. And though it is normatively desirable for voters to be

73

able to hold parties responsible (Rarmey 1954), in a polarized enviromnent, voters
are forced to decide between extremes. This is particularly troubling to those voters
who are not polarizing with elites (Baldassarri and Gelman 2008).

Though split-ticket voting is on the decline (Hetherington 2001), and the impact
of party identiﬁcation on vote choice is on the rise (Bartels 2000), there might also
be an increase in negative feelings as well if the parties have polarized too much.
Even assuming no change in the opinions of voters, polarization can make it appear
as though they are stronger partisans (Fiorina, Abrams and Pope 2005). It has been
argued that as elite polarization increases past a “tipping point” it should lead to
an increase in ambivalence for many (Johnston et al. 2008). The trend is shown in
Figure 4.1: as polarization increases, ambivalence seems to as well, once polarization
increases enough. Individuals may be more apt to choose one side over the other, but

that does mean they are entirely thrilled with their choice.

 

 

 

' 30°/o

"20%:

 

 

 

1960 1 19'70 ' 19'80ﬁ 1390 ' 2000
Year

Figure 4.1: Congressional polarization and the percent of individuals classiﬁed as
ambivalent toward one or both parties.

The non-linearity of this relationship is shown by plotting house polarization

74

against the percent of individuals who are ambivalent in Figure 4.2, the line is a
loess curve (Cleveland 1979). As can be seen, initially elite polarization seems to
have a slight negative impact on the percent of individuals who are offering conﬂicted
thoughts on the parties. However, this quickly reverses itself and as the parties con-
tinue to polarize, the percentage of individuals experiencing partisan ambivalence
increases. It is therefore quite possible the impact of partisan ambivalence on vote
choice is conditional on the positions of the two parties relative to one another. If
they are viewed as relatively similar, perhaps abstention and defection are seen as
viable options. However, in an era where the choices are clear, ambivalent voters may

be forced to choose the closest alternative.

 

 

Percent ambivalent

 

 

 

I r I I
0.6 0.7 0.8 0.9

House polarization

Figure 4.2: House polarization and partisan ambivalence.

4.2 Polarization and Ambivalence

Is an ambivalent voter’s ﬁnal choice altered, or does it simply take him longer to

make the decision he was likely to make anyway? Defection, abstention, or voting for

75

a third party candidate all seem like reasonable options for an ambivalent voter. For
example, if an individual has negative feelings because of an issue position, perhaps
the other party better represents him on this issue; if the issue is salient enough
perhaps the individual might cross party lines. Similar logic would apply for voting
for a third party: this option, long viewed as unrealistic to the voter, might now seem
viable because neither major party represents his views, or he is tired of politics as
usual. What if neither party represents the voter and there is no viable third party?
In this case, one would expect the voter to be more likely to abstain.

While these three outcomes seem fairly intuitive, there is a fourth that is less
so. I argue if the two parties are polarized enough, an ambivalent individual prefers
his party over the other so much that he will turn out to vote. If ambivalence is
related to political knowledge and interest (Rudolph and Popp 2007), it follows that
an ambivalent individual is aware of his own views as well as the positions of both
parties. He would know that although his preferred party is more extreme than he
would like, it is considerably better than the alternative. To show why this is the case,
ﬁrst ignoring the importance of partisanship on issues and perceptions, I start with
a simple, standard model of voter utility (e.g., Enelow and Hinich 1984). Each voter,
i, can be represented by a single point 0,- which can be characterized as a summation

of his issue positions:
n .
62- = Z Issueik, (4.1)
10:1
and receives utility Uz- j for party j, represented by (153- (05]- : 22:1 Issuejk):

Uij = - (0i — ¢j)2

76

Therefore in a two party system, j E (A, B}, voter i votes for party A if Uz- A > Uz- 3.1

How might polarization inﬂuence ambivalence and voting behavior? First, imagine
parties A and B where IA — ml=lB — ml, in which 777. represents the median voter,
as seen in the top potion of Figure 4.3. In this case, for individual i, Uz- A < UiB' If
parties A and B polarize in the same manner: lA’ — ml = lB’ — ml and IA — ml <
lA’ — ml, the difference between Uz- A and Uz- B will necessarily increase, as seen in the
bottom portion of Figure 4.3. While it remains the case that Uz- A < UiB’ it is also
true that le- A — UiBI < le. A’ — UiB’l; although individual i is less pleased with his
preferred party than he used to be, it is now relatively an even better choice. When
both parties move from the center at the same time, partisanship’s inﬂuence on voting
decision increases even if individuals are now less pleased with their party than they
once were. Meaning, the party differential (the distance between the two parties)
should inﬂuence the behavior of the electorate, as predicted and demonstrated by

Fiorina (1976).

Figure 4.3: Before and after elite polarization.

The above argument could be strengthened without adding much complexity.
Most theories of opinion change suggest a topdown process (e.g., Carmines and Stim-
son 1989, Zaller 1992): as elites polarize, their own party identiﬁcation sends a clearer
signal to the electorate, perhaps even causing individuals to alter their own issue po-

sitions.2 There is considerable evidence that partisanship can inﬂuence one’s percep—

 

1Evidence suggests a single dimension is adequate for capturing preferences for
voters (Bafumi and Herron 2007).

2“Classic” views of partisanship (e.g., Campbell et al. 1960) are consistent with

77

tion of presidential performance (Bafumi 2004), the economy (Parker-Stephen 2007)
and issue positions (Layman and Carsey 2002a; Carsey and Layman 2006); this last
relationship holds even when accounting for the non-recipricol relationship between
issues and partisanship (J acoby 1988). Perceptions track together by party identiﬁca-
tion but never converge, as implied by updating models (Bartels 2002). Furthermore,
movement seems to be largest among those who are most aware of the parties’ posi-
tions, but only for issues not held strongly (Carsey and Layman 2006); this ﬁnding
suggests those paying closet attention are most likely to be aware that for at least
some issues their party is now more polarized than they are.

If one’s issue positions are determined at least somewhat by factors other than
their partisanship, individuals will not polarize as quickly or as far as the parties.
As before, if we assume parties are equidistant from the center and polarize at the
same rate then le- A — UiBl < le. A’ — UiB’l' The difference between the two
parties relative to individual i are now even greater after polarization, making the
choice even clearer—even if individual i would prefer for his party to be closer. Such
a story is depicted in Figure 4.4, where i’ represents individuals i’s position after
polarization. The alternative is now even farther from i than before. He is forced to
choose between one of two extremes, and the alternative is even more so than if his

position had remained constant.

Figure 4.4: Before and after elite polarization, with individuals moving with parties.

 

such a story. The notion is, however, at odds with revisionist models of partisanship
(Franklin and Jackson 1983; Fiorina 1981) and Bayesian updating models (Green,
Palmquist and Schickler 2002; Achen 1992).

78

4.3 Research Design

In order to see when and if ambivalence inﬂuences vote choice, including abstention,
a model of vote choice as a function of partisan ambivalence and control variables
is modeled using multinomial regression. I run the same model and present detailed
results of the 1980 and 2004 elections, which allows examination of how polarization
impacts the inﬂuence of ambivalence. Additional evidence from all elections from 1980
to 2004 is provided. I use similar measures and coding as in Chapter 3. I replace
the speciﬁc issue variables with a scale of all 7 point domestic issue questions in each
year.3 Additionally, I include perceptions of the economy and personal ﬁnances. I

also include demographic controls, including: gender, race, age, income, and interest.

4.3.1 Modeling Vote Choice

In order to examine the inﬂuence of ambivalence on voter behavior, I model vote
choice as a function of partisan ambivalence and control variables. I make use of
multinomial regression to account for abstention, as suggested by Lacy and Burden
(1999), as it seems a likely consequence of ambivalence.4 Such a speciﬁcation makes
it possible to see how ambivalence inﬂuences the decision to turn out and the choice
between the two parties.

As suggested by the theory presented above, so long as an individual is, not ter-

ribly concerned with which party is in power, defection or abstention is seen as an

 

3In each ylear a factor analysis suggested one dimension explained the variance,
and Cronbac ’s a were all above 0.60.

4The analysis presented below is a multinomial logit. Although it has been sug-
gested multinomial logit is undesirable for models of vote choice because the as-
sumption of independence of irrelevant alternatives is restrictive (Alvarez and Nagler
1995), others have argued that the primary advantage of multinomial probit—that

the choices need not be inde endent across alternatives—is outweighed by its own
problem, namely weak identi cation; and, that the problem of IIA Is not often vio-

lated in practice (Dow and Endersby 2004). In both years, both Hausman (Hausman
and McFadden 1984) and Small—Hsiao (Small 1985) tests suggest the assumption of
IIA is not violated.

79

option. As polarization increases ambivalent individuals will be more likely to vote
for their preferred party. Therefore, for a Democrat, the choice between abstention
and Republicans in 1980, one should expect a negative coeﬂicient for the variable Re-
publican ambivalence. However, if the choices are polarized, ambivalence will predict

voting for one’s preferred party.

Hypothesis 4a: In periods of elite moderation, ambivalence should predict absten-

tion or defection.

Hypothesis 4b: As elites polarize, ambivalent partisans should be more likely to

vote for their preferred party.

For example, in 1980 the expectation is ﬁk, abstainl m should be negative, when k
and 777. both represent the same party; in the above example It represents Republican
ambivalence and m represents voting for the Republican candidate.5 In 2004, we
would expect ﬁle, abstainl m to be positive when m and k represent the same party.
For example, an individual experiencing Democratic ambivalence is more likely to

vote for the Democratic candidate.

4.4 Empirical Results

The results of the multinomial regression are in Tables 4.1 and 4.2, one for each year of
the analysis. The columns in both tables represent voting for the Democratic and the
Republican candidates, the base category is abstention. The entries are coefﬁcients
showing the impact the variable has on leaving the base category, with standard errors

in parentheses.

 

5 ﬁle, abs tainl m represents the impact variable It has on entering category m from

the base category, abstention; m re resents the two choices other than the base cat-
egory, either the Republican candi ate or the Democratic candidate.

80

The control variables behave as expected. Partisanship predicts voting (although
not for Reagan in 1980). Interest in politics is positively related to voting, although
awareness and sophistication are not. As one believes the economy is improving, one
is more likely to vote for the incumbent. Similarly, issues also predict voting as one
expects: for example, positive thoughts about the Iraq war are associated with not
voting for Kerry.

There is some support for the hypotheses laid out above. In 1980 partisan ambiva-
lence, generally speaking, predicts abstention. The coefﬁcients for all three of dummy
variables are statistically signiﬁcant from zero with a negative sign. Individuals who
are experiencing mixed feelings about either or both parties are dropping out of the
process, evidence that dovetails well with what was originally termed “cross-pressure”
and its impact on abstention (Berelson, Lazarsfeld and McPhee 1954). Voters who
are not happy with one party or the other are deciding to forego voting, but not in the
exact way the theory suggests. It is not only the case, for example, that Democratic
ambivalence predicts defection or abstention away from the Democratic candidate, it
has a similar effect on voting for Reagan. As presented in Table 4.1, the coefﬁcients
represent the impact on abstaining or voting for the party, ﬁk,abstainlDem. for col-
umn 1 and ﬁlaabstaianep. for column 2. What is the impact on choosing between
the two parties, 'Bk,DemlRep.? After altering the base category to one of the two
candidates, the variables have no inﬂuences in 1980: in all cases the coefﬁcients for
the variables representing partisan ambivalence are not statistically signiﬁcant from
zero between the two parties. In 1980, ambivalence inﬂuences the choice to turn out,
not the choice between candidates.

A different pattern emerges in 2004. In this case none of the measures of parti-
san ambivalence inﬂuence the choice between abstention and either candidates. The
theory predicted individuals would be less likely to abstain if they were ambivalent,

because it is only those individuals paying attention who are ambivalent, and it is

81

Tege4l: Vote choice in 1980

 

 

 

 

Variable Carter Reagan
Party ID -0.606*I (0.166) 0.211 (0.162)
Ideology 0.233 (0.245) 0.438I (0.245)
Issues

Issue attitudes -0.597I (0.327) -0.066 (0.333)
Personal ﬁnances 0.018 (0.151 0.189 (0.150)
perception

National economy 0.036 (0.450) -l.059* (0.489)
perception

Sophistication

Political knowledge 0.332 (0.267) 0.440I (0.264)
Education 0.000 (0.067) -0.005 (0.070)
Interest 0.772I (0.412) 0859* (0.410)
Partisan Ambivalence

Republican -0.718 (0.884) -0.957 (0.886)
Democratic —1.477* (0.732) -1.502* (0.737)
Both -1.701* (0.808) -1.326l (0.785)
Intercept 3.6461 (2.066) -1212 (2.127)
N 517

Log-likelihood -248.611

XI22) 339.622

Pseudo R2 0.4058

Signiﬁcance levels: I: 10% : 5% **: 1%

82

these same individuals who are likely to recognize the large difference between the
two parties. It was also predicted that ambivalence would inﬂuence the choice be-
tween candidates, as those who recognized the polarized environment would choose
the side closest. Switching the base category, to see if ambivalence is inﬂuencing turn
out or vote choice, shows that Democratic ambivalence is a signiﬁcant predictor of
the choice between parties, not between voting or staying home. The coefﬁcient for
ﬁDem. Ambiv, Kerrleush is 1.727 with a standard error of 0.552.

The ﬁndings offer support for the claim that polarizing parties are forcing indi-
viduals to choose between two extremes. In 1980, ambivalence is associated with not
voting; while in 2004, those ambivalent about the Democratic are more likely to vote
Democratic. In order to make sure these ﬁndings generalize beyond the years exam-
ined above, the analysis was also performed on all presidential elections from 1976 to
2004.6 Until 2004, partisan ambivalence has little inﬂuence on vote choice, or predicts
defection. In Figure 4.5, the coefﬁcient for the variable Democratic ambivalence is
plotted with 95% conﬁdence intervals, when the variable has a negative coefﬁcient,
those experiencing Democratic ambivalence chose vote for the Democratic candidate.

The ﬁgure is evidence that, at least for those experiencing Democratic ambiva-
lence, the position of elites affects the inﬂuence of partisan ambivalence on vote choice.
When an individual is unhappy with his options, he will vote for the closest in a polar-
ized environment. We would especially expect this from supporters of the party not
in power: policy output is extreme and contrary to their preferences. When examin-
ing Republican ambivalence over time, the variable has no impact on vote choice. It
is possible Republicans are more resistant to internal disagreements. But, any such
conclusion would be speculation.

The ﬁndings for the theory are mixed. While the theory seems to explain the

 

. 6Only variables common to each year were included. There are: partisanship,
ldeglo , personal ﬁnances, awareness, education, interest, and measures of partisan
am 1 ence.

83

Table 4.2: Vote choice in 2004

 

 

 

 

Variable Kerry Bush

Party ID 0530” (0.130) 0.632** (0.122)

Ideology -0.039 (0.169) 0.159 (0.164)

Issues

Issue attitudes -1.054** (0.271) -0.320 (0.238)

Personal ﬁnance -0.005 (0.149) —0.252 (0.155)
perceptions

National economy 0.732** (0.205) -0.310 (0.192)
perceptions

Sophistication

Political knowledge 0.507** (0.150) 0.432** (0.143)

Education 0.256I (0.151) 3.826 (4.885)

Interest -0.56I** (0.151) -0.519** (0.145)

Values

Egalitarianism 0.406 (4.961) 3.826 (4.885)

Limited government 1.510 (2.304) 0.745 (2.003)

Moral traditionalism -5.499 (4.987) 2.721 (4.427)

Egalitarianismx -2.689 (3.930) -0.613 (3.502)
limited government

Egalitarianismx 6.620 (8.460) -1.639 (7.761)
moral traditionalism

Partisan Ambivalence

Republican 0.927 (0.860) 0.339 (0.848)

Democratic 1.038 (0.679) -0.690 (0.654)

Both -1.657 (1.199) -0.926 (1.148)

Intercept 3.897 (3.189) -2.288 (3.091)

N 724

Log-likelihood -255.383

X02) 804.807

Pseudo R2 0.6118

Signiﬁcance levels : I : 10% * : 5% ** : 1%

84

 

 

Coeffecient
Coeffecient

   

 

 

 

 

 

 

193019'8419'881992139620002004 198019841988199219'9620002004
Year Year

Figure 4.5: The impact of Democratic ambivalence on vote choice. The plotted points
represent the coefﬁcient from a model of vote choice with 95% conﬁdence intervals.

Negative values indicate an increase in the likelihood of voting for the Democratic
candidate.

 

 

 

 

   

 

 

 

 

 

Abstention Defection
E E
9 .9
8 8
’5 ‘5
O O
o o
1380 1384 1388 1992 1936 2000 2034 1980 1934 1388 1392 1396 2000 2004
Year Year

Figure 4.6: The impact of Republican ambivalence on vote choice. The plotted points
represent the coefﬁcient from a model of vote choice with 95% conﬁdence intervals.

Negative values indicate an increase in the likelihood of voting for the Republican
candidate.

behavior of those individuals experiencing ambivalence toward the Democratic party,
it does not do so for those experiencing it toward Republicans. Interestingly, in the
years examined, ambivalence only inﬂuences vote choice in 1980 and 2004. This
could be consistent with those suggesting much the time examined was in an era of
“dealignment.” While not direct evidence of the theory, it is indirect in the sense
that the theory suggests the inﬂuence of ambivalence on vote choice is conditional on
elites; if elites are not views as particularly partisan, partisanship, and by extension

partisan ambivalence, will not exert much inﬂuence on vote choice.

85

4.5 Conclusion

The analysis set out to see how and when partisan ambivalence might have an im-
pact on one’s vote choice. Speciﬁcally, it was posited that in times of polarization,
ambivalent individuals will actually be more likely to vote for their preferred party.
When the two parties are both moderate, ambivalent individuals will be more likely
to defect or simply abstain; if the other side is viewed as relatively similar, there is
no harm in sitting an election out, or even helping that party to power. Polarized
parties force voters to choose between two extremes.

Evidence was somewhat supportive of the theory. In 1980, when the two parties
were relatively close to one another, partisan ambivalence predicted abstention—
the decision was between turning out or staying home. In 2004, however, those
experiencing Democratic ambivalence were more likely to vote for the Democratic
party. Moreover, in 2004 partisan ambivalence is unrelated to the decision to turn
out. The inﬂuence of ambivalence on vote choice is dependent on the political climate.
The results offer support for a period of realignment in the 19808, with partisanship
strengthening in the 19908.

Another implication of the above ﬁndings is that strong signals to voters are not
necessarily only good news, so long as they result mostly from polarization. While
voters are able to understand the differences between the two parties, this is of little
use to voters who prefer more centrist policies, for which there is considerable evidence
that many do (e.g., Stimson 2004). Individuals are now no longer able to vote for
policies they want, only for their side. While the results might be seem reasonable for
those whose party is in power, those on the other side are likely to have have negative
view of the political system, as manifested by the increasing number of ambivalent
individuals. That partisan ambivalence was once associated with abstention, but now
predicts voting for one’s party suggests polarization might lead to an artiﬁcial inﬂation

of the relationship between voting and partisanship. Particularly, among those who

86

are not as quick to follow elites. Individuals who are aware of the changing landscape
of American politics but who are not moving fully in step with elites must make a
choice; and, although they are more inclined to vote with their party than before,
masked behind such an inclination may be conﬂicted thoughts and mixed feelings.
Perhaps the most intriguing ﬁnding is that for the most part ambivalence toward
the two parties rarely inﬂuences one’s decision. Only in 1980 and 2004, and only for
Democrats, does ambivalence affect one’s choices. This suggests a few avenues for
further analysis. Perhaps it is those experiencing choice ambivalence who are more
likely to defect. Ambivalence toward one’s preferred party may not be enough to
alter vote choice, but perhaps conﬂicted thoughts about the choice is. Additionally,
maybe the strength of partisanship is enough to overcome defection, but not so strong
that it ambivalence might cause one to take longer to make up his mind. And ﬁnally,
perhaps the methodology used in the above analysis did not appropriately take into
account the recursive relationship between ambivalence and vote choice. I address all

of these concerns in the following chapter.

87

Appendix to Chapter 4

Interaction Between Partisanship and Partisan Ambivalence

Is there an interaction between partisanship and ambivalence? Are stronger partisans
more or less affected by ambivalence? It has been established that even if the coefﬁ-
cient is not statistically signiﬁcant there could be an interaction in non-linear models
along some portion of the variable (Brambor, Clark and Golder 2006) or for some
observations (Ai and Norton 2003). The magnitude of interaction effect depends on
all the other included variables and therefore when looking for the presence of an
interaction effect it one should look the t-statistic for each observation as well as im-
pact of the interaction for each observation to see patterns. In Figure 4.7, I plot the
t-statistic for each observation against the predicted probability for PIDX Republican
Ambivalence and PIDxDemocratic Ambivalence in 1980 and 2004. The results sug-
gest that there is no consistent interactive effect between ambivalence and partisan

identiﬁcation.

88

000204060810
III

I I l I I I l l I
2004 Democrat 004 Republica

 

 

 

 

 

 

'7 — 6
0
- 0% o "" 4
r, is). 2
_ D . “WV— 0
o ' i. '2
'5 'I ' —-4
E 1980 Democrat 1980 Republican
(I)

 

 

 

 

 

 

 

0.0 0.2 0.4 0.6 0.8 1.0
de=1)

Figure 4.7: Examining for interaction effects between Party ID and Ambivalence.

89

Model Results: 1984-2000

Table 4.3: Vote choice in 1984

 

 

 

Variable Mondale Reagan

Party ID -0.613** (0.069) 0.246** (0.050)

Ideology -0.087 (0.055) 0.078 (0.050)

Issues

Issue attitudes -0.484* (0.191) 0.306I (0.170)

Personal ﬁnancial 0.078 (0.063) -0.013 (0.058)
perception

National economic 0.216** (0.072) -0.158* (0.068)
perception

Sophistication

Political knowledge 0.359** (0.066) 0.286” (0.057)

Education 0.269** (0.078) 0.271** (0.070)

Interest 0735’” (0.160) 0.426** (0.142)
Value conﬂict

Egalitarianism 0.552 (0.417) 0.195 (0.409)

Limited government 0.483 (0.334) 0.414 (0.286)

Egalitarianism -0.098 (0.112) -0.067 (0.099)
xlimited government

Ambivalence

Democratic 0.084 (0.281) 0.300 (0.246)

Republican -0.215 (0.288) -0.108 (0.253)

Intercept -2.868* (1.338) -5.334** (1.264)

N 1203

Log-likelihood -770.496

x2222) 955.871

Pseudo R2 0.406

 

Signiﬁcance levels : I : 10% >1: : 5% ** : 1%

90

Table 4.4: Vote choice in 1988

 

 

 

Variable Dukakis Bush
Party ID 0471“ (0.050) 0.368W(0.046)
Ideology 0.001 (0.036) 0.163** (0.040)
Issues

Issue attitudes -0.239* (0.117) 0.409** (0.125)
Personal ﬁnancial 0.134Jr (0.075) 0.016 (0.080)
National economic 0.025 (0.059) -0.036 (0.062)
perceptions

Sophistication

Political knowledge 0.423" (0.075) 0.454** (0.076)
Education 0.249” (0.059) 0.280** (0.059)
Interest -0.493** (0.064) -0.417** (0.066)
Value conﬂict

Egalitarianism 0.498 (0.312) 0975* (0.381)
Limited government 0.454** (0.170) 0.386I (0.204)
Egalitarianism -0.173* (0.074) -0.160l (0.082)
x limited government

Ambivalence

Democratic 0.140 (0.312) 0.300 (0.300)
Republican 0.259 (0.230) 0.092 (0.225)
Intercept -0.091 (0.747) -6.290** (1.001)
N 1440

Log-likelihood 4019.477

Xf22) 1017.579

Pseudo R2 0.352

 

Signiﬁcance levels : I : 10% >I< : 5% ** : 1%

91

Table 4.5: Vote choice in 1992

 

 

 

Variable Clinton Bush
Party ID -0.570** (0.092) 0.408** (0.101)
Ideology -0.008 (0.122) 0.530** (0.143)
Issues

Issue attitudes 0.099 (0.186) 0.287 (0.208)
Personal ﬁances 0.134 (0.141) -0.210 (0.155)
National economic 0.108 (0.185) -0.415* (0.195)
perceptions

Sophistication

Political knowledge 0.095 (0.084) 0.003 (0.095)
Education 0234* (0.108) 0.495** (0.118)
Interest -0.517** (0.107) -0.440** (0.120)
Value conﬂict

Egalitarianism —0.691I (0.420) -0215 (0.408)
Limited government -1559 (1.363) 2.496” (1.464)
Egalitarianism 0.626 (0.508) 0.814 (0.530)
xlimited government

Ambivalence

Democratic 0856* (0.434) 0.776I (0.465)
Republican 0.163 (0.444) 0.432 (0.452)
Intercept 3251* (1.647) -1.820 (1.731)
N 782

Log-likelihood -349.996

Xf22) 751.023

Pseudo R2 0.518

 

Signiﬁcance levels :

I:10% *:5% **:

92

1%

Table 4.6: Vote choice in 1996

 

 

 

 

Variable Clinton Dole

Party ID -0.443** (0.121) 0.452** (0.134)

Ideology -0137 (0.177) 0.325 (0.203)

Issues

Issue attitudes 0.081 (0.306) 0.904** (0.348)

Personal ﬁnancial -0.075 (0.214) -0.027 (0.239)
perceptions

National economy -0.656** (0.230) 0.656** (0.253)
perceptions

Sophistication

Political knowledge 0.252I (0.138) 0.213 (0.160)

Education 0.362** (0.138) 0.505** (0.154)

Interest -1.150** (0.318) -0.991** (0.346)

Value conﬂict

Limited government -6.502** (2.436) -0.261 (2.514)

Moral traditionalism 0.744 (1.055) 0.387 (1.060)

Egalitarianism -l .950 (1.400) 0.062 (1.500)

Egalitarianism 1762* (0.719 0.098 (0.759))
xlimited government

Egalitarianism -0.156 (0.314) -0.073 (0.332)
xmoral traditionalism

Ambivalence

Democratic 0.091 (0.534) 0.377 (0.564)

Republican -0.573 (0.538) -0.386 (0.574)

Intercept 12.054* (5.303) -8.874 (5.596)

N 502

Log-likelihood -214.508

Xf22) 498.245

Pseudo R2 0.537

Signiﬁcance levels : I : 10% >1: : 5% *4: : 1%

93

Chapter 5

Ambivalence, Indifference, and

Presidential Elections

What if an individual has conﬂicted thoughts not about just a party, but about the
choice between the two parties? Extending the previous chapter’s analysis, I seek
to see if comparative ambivalence alters ones’ choice. Furthermore, I examine the
possibility that an ambivalent individual may be inclined to consider more informa-
tion, and therefore take a longer time to decide. Using methods appropriate for the
recursive relationship between ambivalence, timing of vote decision, and defection it
is shown that while ambivalence between the choice of parties predicts individuals
will delay their decision, it does not predict defection. Instead, it is indifferent indi-
viduals who are more likely to defect. The ﬁnding that ambivalence plays little role in
presidential vote choice appears to be robust to different measurement of the concept

and to different model speciﬁcations

5.1 Partisanship, Ambivalence, and Voting

Why might ambivalent voters delay their vote? The process of deciding to turn out

and for whom is dynamic (Hillygus and Jackman 2003; Hillygus 2005). While it is

94

well documented that those who decide early are more interested in the campaigns
(Chaffee and Choe 1980; Chaffee and Rimal 1996; Gopoian and Hadjiharalambous
1994), it is also known that one’s openness to campaign messages is inﬂuenced by
if one has already decided for whom to vote (Fournier et al. 2004)——Chapter 3 as
well as does other research (Rudolph and Popp 2007) suggests ambivalent individuals
are more receptive. While Nir (2005) shows cross-pressure is not associated with
abstaining, it has been established that candidate ambivalence can have an impact on
timing of vote decision, (Lavine 2001; Nir 2005; Nir and Druckman 2008). Certainly,
if there is an issue which a person disagrees with his party, he will be more inclined
to defect (Hillygus and Shields 2008). But, neither implies an individual with a more
general state of attitudinal conﬂict such as partisan ambivalence will take longer to
deliberate or, perhaps, defect.

To begin, imagine an individual who does not suffer from attitudinal conﬂict but is
at least somewhat interested in the campaign: his attitudes align with his partisanship
and we should thus expect him to have little trouble making up his mind and he should
therfore make a decision relatively early on in the process. An indifferent individual
will not tune into the process until later. Consequently, we should expect him to
decide later. Indifferent individuals tend to not be very interested in the campaign
and are less likely to identify with one of the two parties. As shown in Figure 5.1,
indifferent individuals are considerably less likely to identify with one of the parties,
espcially strongly identify.

Ambivalent respondents, in contrast, do tend identify with the parties, though not
as strongly as those who are strictly one-sided in their thinking. They certainly are
not uninterested. In fact, they are slightly less likely to report being uninterested than
those classiﬁed as one-sided (see Figure 5.2). Unlike the one-sided individual, however,
the ambivalent individual might be more willing to listen to the other side and to

consider new information during the duration of the campaign. If this is the case, we

95

 

. Ambivalent " - Indiffere‘r'ft' , One-sld'

 

Percent of total
N
o
I

 

 

 

 

 

 

I I I
060 3:60 ﬁegl‘o‘Q

gave 0,.»96 6 gave

I I
309
(‘59 0,39% :0 09:“009 059509 0031009
\0 \

Figure 5.1: Attitudinal states and intensity of partisan identifcation.

should expect ambivalent individuals to take a longer time to decide. However, if the
View that “for most people the tie between party identiﬁcation and voting behavior
involves subtle processes of perceptual adjustment by which the individual assembles
an image of current politics consistent with his partisan allegiance” (Stokes 1962:691)
is correct, ambivalent individuals should not consistently defect from their preferred
party. Instead, we should expect that while the decision is delayed, the strength of
one’s partisanship should overcome a general state of attitudinal conﬂict. Such a
conclusion is not inevitable, as “one may vote for a Republican candidate and yet feel

part of a Democratic team” (Green, Palmquist and Schickler 200228).

 

Ambivalent Inﬂlﬁ'erent one—sided

 

40- —
30m —
20— —
10- -
O-I . , . . . -

 

 

Percent of total

 

 

 

 

I I I I I I

9‘ cl 1‘ 9‘ cl c)“ 9‘ cl
0609:1199” \le' 0090 63‘“ \le 69" «313° \19
8‘ 9°‘° $9 9°“ \‘p 9°
Interested in following campaign?

Figure 5.2: Attitudinal states and interest in the campaign.

Two speciﬁc hypotheses can be derived from the above discussion. First, be

96

cause ambivalent individuals are interested in the campaign but more open to new

information, we should expect them to delay their vote choice.

Hypothesis 5.1: Partisan ambivalence should lead to a delay in one ’3 vote

decision.

Second, the process of perceptual adjustment should lead these individuals to vote
for their preferred party. However, indifferent individuals, while also deciding late,
should be more likely to defect as they are less tied to either party or the political

system.

Hypothesis 5.2: Indifference, not ambivalence, should predict partisan defection.

5.2 Research Design

One problem with the ﬁndings from Chapter 4 is that perhaps the methods did not
adequately account for the relationship between vote choice and ambivalence. Below, I
utilize a method that does so. I examine the role ambivalence plays when an individual
makes his ﬁnal decision. Following this, I examine if comparative ambivalence predicts

defection.

5.2.1 Measures

There are four primary variables of interest in this portion of the analysis. First, parti-
san ambivalence is measured using the measure of comparative ambivalence discussed
in Chapter 2. When examining the role ambivalence can have on voting behavior,
a measure of choice ambivalence may be appropriate. It is dichotomized so scores
greater than zero are categorized as ambivalent. Those who offer no likes or dislikes
about the parties are deemed indifferent. Third, one’s timing of vote decision is coded

so that those who knew all along are coded as one, those who decided from during

97

the primaries to the time of the convention are coded two, and those who decided
after the conventions are coded as three. This splits the proportion of individuals
into roughly thirds.1 Past research suggests hat voters can recall and report reliable
estimates of their times of decision (Fournier et al. 2001).

I include as control variables: demographic variables; interest in the campaign;
intensity of party identiﬁcation; how much the individual cares about the election;
a need for cognition; as well as measures of value conﬂict between egalitarianism
and moral traditionalism, and egalitarianism and limited government; and in 2000 a
measure of network heterogeneity. The coding is the same as the previous chapters.
As the expectations for these variables have been well documented previously, I avoid

a discussion here.

5.2.2 Modeling Ambivalence and Timing of Vote Decision

Ambivalence and timing of vote decision could be modeled as a recursive system of

equations:

yiz’ = xiil31+ 612' (5-1)

932' = x'2z'l32 + 73/1,- + 52,-, (52)

where yfz- is the latent tendency to be ambivalent, ygi is the latent tendency guiding
the time of vote decision, xllz’Bl and x’2i [32 are a set of explanatory variables and
coefﬁcients for individual i, and 'y is the inﬂuence of y1 on y; The question of
interest is what is the inﬂuence of y1 on y2, while taking into account the recursive
relationship between the two variables. This would be considered a “semi-ordered”
bivariate probit—in this case bivariate refers to the fact that there are two dependent

variables—as only one of the two variables is ordered, the other is simply dichotomous

 

1This is the case for 2004, in each year, I split the respondents into thirds.

98

(Armstrong and McVicar 2000; McVicar and McKee 2002), which is an extension of
the bivariate probit model, and a special case of the bivariate ordered probit model
(Weiss 1993). The model treats the two equations as independent, except for modeling
the underlying errors as jointly distributed (Greene and Hensher 2008:146) with a
correlation parameter p. Because the system is identiﬁed, if p equals zero, one can
estimate (5.2) alone.2 A somewhat more formal discussion of the model is presented
in the appendix.

If 'y is greater than zero it would be an indication that individuals experienc-
ing partisan ambivalence are more likely to make a decision later in the campaign.
The results, for 1980 and 2004, are in Tables 5.1 and 5.2, with standard errors in
parentheses.

The control variables behave, for the most part, as expected. For example, the
intensity with which an individual holds his partisanship is associated with a reduction
in the probability of being ambivalent and deciding early in the campaign. While the
behavior of the variables is not identical in each year (or in other years, which are
presented in the appendix), in no case are the results contrary to what one would
expect. For example, while education plays no role in 1980, in 2004 it predicts
ambivalence. The variables behave predictably and concord with previous ﬁndings
and expectations.

Of interest to the research question at hand, in 2004 ambivalence predicts voting
later in the campaign. It has no effect in 1980. Some support is found for hypothesis
1. To what extent to these ﬁndings generalize beyond 1980 and 2004? Figure 5.3
indicates that comparative ambivalence more consistently predicts delaying one’s de-
cision as compared to indifference; it has for the four most recent elections examined

predicted voting later in the campaign. Although in all years, there is a bivariate

 

2In most ears, a hypothesis that p equals zero is rejected, or nearly rejected. I
use this mo el in all years for two reasons. First is for the sake of comparability.
Second, I wish to demonstrate the relationship between vote choice and ambivalence
is robust to model speciﬁcation.

99

Table 5.1: Ambivalence and Timing of vote decision in 1980

 

 

 

 

Ambivalence Timing of vote decision
Attitudinal states
Ambivalence -0 . 578 (1 . 1 1 6)
Indifference 0.239Jr (O. 129)
Timing of vote control variables
Issues -0.061 (0.073)
Gender -0.009 (0.078)
Minority -0.027 (0.024)
Age —0.006* (0.003)
Income -0.001 (0.007)
Ambivalence control variables
Government involvement -0.018 (0.028)
Joint control variables
Education 0.005 (0.007) 0.008 (0.007)
Interest 0.208** (0.063) -0.034 (0.078)
Political knowledge 0.219** (0.033) 0.133** (0.044)
Party ID intensity -0.014 (0.044) -0.092T (0.047)
Care who wins 0.057 (0.088) -0.429** (0.107)
Intercept 0.686 (0.435)
71 -1.246** (0.346)
72 -0.302 (0.287)
N 569
Log-likelihood -1435.226
Q27) 173.088

 

Signiﬁcance levels : i : 10% * : 5% ** : 1%

100

Table 5.2: Ambivalence and Timing of vote decision in 2004

 

 

 

 

Ambivalence Timing of vote decision

Attitudinal states
Ambivalence 0.601l (0.356)
Indifference 0.255Jr (0.136)
Timing of vote control variables
Issues -0.165** (0.058)
Gender 0.027 (0.089)
Minority 0.107 (0.100)
Age -0.003 (0.003)
Income -0.009 (0.008)
Ambivalence control variables
Need for cognition 0144* (0.066)
Opinionation 0.263 (0.232)
Egalitarianism -1.358 (0.868)
Limited government -0.280 (0.464)
Moral traditionalism -1.193 (0.823)
Egalitarianism x 0.884 (0.747)

limited government
Egalitarianismx 1.992 (1.335)

moral traditionalism
Joint control variables
Education 0079* (0.031) 0.003 (0.032)
Interest 0.080 (0.076) -0.273** (0.073)
Political knowledge 0.114** (0.031) -0.023 (0.036)
Party ID intensity 0032i (0.047) -0.205** (0.053)
Care who wins -0.060 (0.149) -0.585** (0.188)
Intercept -0.908l (0.537)
r1 -2.305** (0.328)
7'2 -1.269** (0.313)
N 903
Log-likelihood -1308.784
X94) 191.562

 

Signiﬁcance levels : i : 10% * : 5% ** : 1%

101

relationship between indifference and when one makes up one’s mind, once control-
ling for other factors, the relationship is not statistically signiﬁcant, save for 1980 and

1996.

Delay vote decision

 

 

Ambivalence Indifference
g 1.0- + i l _ 1.0— + -
g 0.5— — 0.59 —
E 0.0— —- — ——————————— — 0_0__i’__¢._.o__é____¢____
O -0.5“ h -0.5“" '-
—1.0— i' -1.0~ —

 

 

 

 

 

 

1980 1984 1988 1992 1996 20002004 1980198419881992199620002004

Figure 5.3: Attitudinal states and time of decision.

There are a variety possible explanations for such a ﬁnding. One possibility is
that an ambivalent individual may have a fundamental disagreement with his party.
We would expect the individual to therefore have a difﬁcult time making up his
mind. Another possibility is that the individual is less reliant on his partisanship and
therefore more systematically and thoroughly examines the available information,
as suggested by Basinger and Lavine (2005). These are just two of many possible
explanations. No matter the precise cause, which likely varies across individuals,

evidence suggests ambivalent individuals take longer to make a decision.

5.2.3 Ambivalence and Detection

While it has been established that ambivalence is associated with individuals deciding
later in the election cycle, it is still an open question as to whether or not ambivalence
alters one’s vote choice. In order to examine whether or not this is the case, vote is
modeled as partisans defecting from their party (accordingly, independents are not

considered in this portion of the anlaysis). In this case, vote choice is coded as 0 if a

102

partisan votes for his party, and 1 if the individual defects. Therefore, there are two
dependent variables in a recursive system similar to equations (5.1) and (5.2), but in

this case both are dichotomous. Thus, the model

Pr(.91: 1, 92 = 1) = (192(X1i31,X2I32,/))

is estimated using bivariate probit, where 311 is ambivalence, y2 is if the respondent
defects, and (D2(-, -, p) is the cumulative distribution function of the bivariate standard
normal distribution, with correlation parameter p. The results are in Tables 5.3
and 5.4, with standard errors in parentheses. Once again, the control variables behave
as expected.

Ambivalence does not predict defection. In both years, indifference but not am-
bivalence is associated with partisans voting for the other party. Ultimately, this result
is not terribly surprising. Individuals are not ambivalent due to a lack of interest.
Indeed, the opposite is true: those variables associated with political sophistication
have long been associated with ambivalence. So, while ambivalent individuals are
aware that they may have some problems with their preferred party—or see virtues
in the other—and therefore seek more information, in the end this conﬂict is not
enough to overcome their partisanship. While speciﬁc instances of issue disagreement
may alter one’s vote choice (Hillygus and Shields 2008), a general state of attitudinal
conﬂict does not. On the other hard, indifferent individuals are more likely to defect.
These individuals tend not to identify as strongly with either party, generally have
less political information, and are less likely to care who wins the election.3 There-
fore, it is unsurprising that they are more likely to defect as they are less interested
in and know less about the political system.

The overall trend is presented in Figure 5.4. In every year besides 1980, indiffer-

 

3Evidence of indifferent individuals not caring who wins and having less informa-
tion is presented in the appendix.

103

Table 5.3: Ambivalence and defection in 1980

 

 

 

 

Ambivalence Timing of vote decision
Attitudinal states
Ambivalence -0.414 (0.395)
Indifference 0.242 (0.180)
Timing of vote control variables
Issues -0.032 (0.107)
Gender -0.049 (0.141)
Minorityaid -0.069 (0.056)
Age 0.004 (0.004)
Income 0.014 (0.013)
Ambivalence control variables
govti -0.031 (0.029)
Joint control variables
Education 0.001 (0.008) -0.029 (0.029)
Interest 0.222** (0.068) -0.013 (0.116)
Political knowledge 0.227** (0.036 -0.012 (0.076)
Party ID intensity -0.030 (0.048) -0.120 (0.098)
Care who wins 0.042 (0.095) -0.412** (0.154)
Intercept -1.570** (0.220) -0.006 (0.548)
N 569
Log-likelihood -767.947
szo) 105.052
Signiﬁcance levels : T : 10% : 5% ** : 1%

104

Table 5.4: Ambivalence and defection in 2004

 

 

 

 

Ambivalence Defection

Attitudinal states
Ambivalence 0.176 (0.339)
Indifference 0591** (0.162)
Defection control variables
Interest 0105* (0052)
Gender 0279* (0.135)
Minority -0.068 (0.143)
Age -0005 (0.004)
Income -0017 (0.012)
Ambivalence control variables
Need for cognition 0.180** (0.065)
Egalitarianism -0653 (0.827)
Limited government -0.046 (0.462)
Moral traditionalism -0.816 (0.839)
Egalitarianismx 0.532 (0.751)

limited government
Egalitarianismx 1.271 (1.345)

moral traditionalism
Joint control variables
Education 0072* (0.031) -0.054 (0.050)
Political knowledge 0.129** (0.030) -0010 (0.042)
Party ID intensity 00791 (0.046) -0.362** (0.088)
Care who wins -0.011 (0.142) 0.097 (0.198)
Intercept -1.124* (0.521) -0590 (0.433)
N 703
Log-likelihood -810.884
X01) 155.584
Signiﬁcance levels : i : 10% * : 5% ** : 1%

105

ence predicts defection. Conversely, only in 1992 does ambivalence predict defection.
One’s attachment to one of the two major parties is enough to overcome a state of
attitudinal conﬂict. However, indifferent individuals, who are characterized by low
levels of interest and attachment, are more likely to defect. Ironically, while these
individuals are perhaps the easiest to persuade, they are the hardest to target. The
evidence in the two ﬁgures offers support for the claim that partisanship, over the
course of a campaign, may alter one’s perception of events. More concretely, they are
evidence that even if an individual has questions about his choice, these often are not

enough to convince him to vote for the other side.

 

 

 

 

 

 

 

 

Detection
Ambivalence Indifference

‘5 1.0- + — 1.0- r-
0
“g 0.5— i i — 05—5:> ¢ 3 ¢ + ‘i’ é —
g 0.0— -- — ——————— ———~— 0.01 --—- ----------- -
8 -0.5‘ + — _o,5— _

-1'0-l -1.0— —

1980 1984 1988 1992 1996 2000 2004 1980 1984 1988 1992 1996 2000 2004

Figure 5.4: Attitudinal states and defection in vote choice.

5.3 Conclusion

While ambivalence is enough to give individuals pause, in the end it only delays the
choice they likely would have made anyway. The ﬁndings therefore offer indirect
support for the argument that one’s vote choice is a function of a gradual adjustment
of one’s choice becoming aligned with one’s party. More concretely, the ﬁndings
provide support for the notion that partisanship is powerful enough to overcome
mixed thoughts about the choice between parties. Ambivalent voters are, perhaps,
more open to hearing multiple points of views and to considering more information,

and consequently taking longer to decide. But, ultimately these individuals are not

106

any more likely to defect and vote for the other party. These ﬁndings coupled with
those from Chapter 4 offer fairly compelling evidence that ambivalence does not have
a consistent effect on vote choice in presidential elections, robust to measures of
ambivalence and vote choice, and to different model speciﬁcations.

Indifference, on the other hand, is capable of altering one’s choice. Indifference is
an attitudinal state characterized by low levels of attachment and interest. As a con-
sequence, these voters may be more likely to vote based on candidate characteristics,
or based on the nature of the times. This ﬁts in nicely with the evidence provided
in Chapter 2 that indifference is a distinct attitudinal state. Not only is it different
from ambivalence and one-sidedness, but it has consequences for political behavior.

Such ﬁndings are consistent with a growing body of research regarding ambiva-
lence and political behavior. Speciﬁcally, it is known that ambivalent individuals are
less reliant on partisanship (Basinger and Lavine 2005) and moderate candidate eval-
uations (Meffert, Guge and Lodge 2000). The ﬁnding is also consistent with research
suggesting partisanship is a powerful psychological attachment capable of shaping
and modifying attitudes (e.g., Bartels 2002; Campbell et al. 1960; Lewis-Beck et al.
2008). Perhaps not powerful enough to overcome disagreements over speciﬁc issues,
(Hillygus and Shields 2008), but still enough to overcome a general state of conﬂict
over the choice between parties. Phrthermore, so long as the conﬂict is not great or
speciﬁc enough, an individual’s partisan identity should prevail, and while the voting
decision is delayed, it is not altered. This and the previous chapter’s ﬁndings offer
clear evidence that ambivalence, like its forerunner cross-pressure, does not alter one’s
vote choice in presidential elections.

As in the proceeding chapter, it should be noted that these ﬁndings are only
for presidential elections, and may or may not generalize to other elections. It is
easy to conceive that ambivalence will have less impact on presidential vote choice

as it is perhaps more centrally related to one’s partisan identiﬁcation and therefore

107

more immune to attitudinal conﬂict. Therefore, one might expect to see individuals
to defect in Congressional elections, and therefore we would expect the effects of
ambivalence to manifest itself as split-ticket voting. Or, maybe only under certain
circumstances, as suggested by Lavine, Johnston, and Steenbergen (2009), should
we see ambivalence inﬂuence vote choice. For now at least, there is no compelling
evidence that a general state of attitudinal conﬂict is enough to make partisans defect

in presidential elections.

108

Appendix to Chapter 5

Recursive System of Equations

In order to model equations (5.1) and (5.2) I use semi-orederd bivariate probit As
is traditionally the case when estimating models with binary and ordered dependent

variables,
1ifr0=—ooSy*-<rl
1 if yfz. > 0 2‘

0 otherwise
3 ifr1=r2§y§i<oo

The goal is to then estimate,

P1“(91 =1, 92 =J'IX1131,X2B2,T) =
Pr(91 = 1, 92 = 1) = <I>2[x1I31,(r1— 791— X232),p]
Pr(91 = 1, 92 = 2) = (P2IX1I31: (T2 - 991 - X252lml , (5-3)
-‘I’2IX1B1,(7'1 - 791 - X2B2Ml
Pr(91=1, 92 = 3) =1— (P20151472 - 791- X292)n0l

where <I>2[-, -, p] is the cumulative distribution function of the bivariate standard nor-
mal distribution, with a correlation coefﬁcient between the two random elements, 61
and 62, p,

6i]. ~N 0 l ,0

fig 0 p 1

Because those terms in equation (5.3) are what would go into the usual likeli-
hood function, unlike in linear simultaneous equations models, if the two dependent

variables are jointly determined, the dependent variable can just be put each on the

109

right-hand side of the other equation and one can the estimate the model as if were
no simultaneity problem existed (Greene 1998:295; Greene and Hensher 2008:146).
Doing so treats the two equations as independent aside from modeling the underlying
errors as jointly normally distributed. The models were estimated using the “cmp”

routine in Stata (Roodman 2007).

110

Model Results: 1984-2000 for Timing

Table 5.5: Ambivalence and Timing of vote decision in 1984

 

 

 

Ambivalence Timing of vote decision
Attitudinal states
Ambivalence 0.165 (0.313)
Indifference 0.112 (0.091)
Timing of vote control variables
Issues -0102* (0.050)
Gender -0043 (0.070)
Minority -0.010 (0.024)
Age -0.006** (0.002)
Income -0.015* (0.006)
Ambivalence control variables
Egalitarianism -0.145 (0.129)
Limited government -0.048 (0.092)
Egalitarianism 0.048 (0.032)

x limited government

Joint control variables

 

Education 0.125M (0.024) 0.011 (0.028)
Interest 0.027 (0.030)

Political knowledge 0112’” (0.019) -0.023 (0.024)
Party ID intensity 0.066 (0.037) -0.173** (0.039)
Care who wins 0.020 (0.029) -0.007 (0.032)
Intercept -1.235** (0.392)

71 -1.821** (0.227)
T1 -0.700** (0.223)
N 1311

Log-likelihood -2033.367

£27) 186.751

 

Signiﬁcance levels : T : 10% * : 5% ** : 1%

111

Table 5.6: Ambivalence and Timing of vote decision in 1988

 

 

 

 

 

Ambivalence Timing of vote decision

Attitudinal states
Ambivalence 0.510 (0.477)
Indifference 0.027 (0.030)
Timing of vote control variables
Issues ~01 05* (0.046)
Gender -0.146l (0.075)
Minority -0.006 (0.112)
Age 0002 (0.002)
Income 0016* (0.007)
Ambivalence control variables
Egalitarianism 0.248l (0.150)
limited 0.091 (0.075)
Egalitarianism -0.036 (0.033)

x limited government
Joint control variables
Education 0140" (0.023) -0042 (0.033)
Interest -0150** (0.055)
Political knowledge 0.186** (0.029) -0.026 (0.042)
Party ID intensity —0.035 (0.039) -0.242** (0.042)
Care who wins —0.037 (0.081) -0.380** (0.090)
Intercept -2.127** (0.351)
71 -1.321** (0.247)
72 -0.513* (0.245)
N 1055
Log-likelihood -1905.091
X07) 265.076
Signiﬁcance levels : T : 10% * : 5% ** :

112

Table 5.7: Ambivalence and Timing of vote decision in 1992

 

 

Ambivalence

Timing of vote decision

 

Attitudinal states

 

 

Ambivalence 0.840** (0.302))
Indifference 0.231i (0.127)
Timing of vote control variables
Issues -0169** (0.061)
Gender 0.086 (0.078)
Minority 0.017 (0.025)
Age -0004 (0.002)
Income -0006 (0.006)
Ambivalence control variables
Egalitarianism -0.322* (0.142)
Limited government -1.320** (0.311)
Moral traditionalism -0.067 (0.130)
Egalitarianism 0369*”< (0.115)

x limited government
Egalitarianism 0.001 (0.052)

x moral traditionalism
Joint control variables
Education 0099’" (0.027) -0027 (0.028)
Interest -0014 (0.033) 0080* (0.033)
Political knowledge 0.102** (0.023) 0.001 (0.026)
Party ID intensity -0.173** (0.042) -0.216** (0.053)
Care who wins 0.018 (0.113) -0.192 (0.122)
Intercept 0.292 (0.422)
T1 -0.731* (0.291)
72 0.282 (0.283)
N 827
Log-likelihood -1524.132
X97) 227.376
Signiﬁcance levels : f : 10% * : 5% ** : 1%

113

Table 5.8: Ambivalence and Timing of vote decision in 1996

 

 

 

 

Ambivalence Timing of vote decision

Attitudinal states
Ambivalence 1.201“‘ (0.235)
Indifference 0.611** (0.164)
Timing of vote control variables
Issues 0.012 (0.071)
Gender 0.053 (0.097)
Minority -0.020 (0.036)
Age 0.000 (0.003)
Income 0.015Jr (0.008)
Ambivalence control variables
Egalitarianism -0.599* (0.264)
Limited government -0497 (0.500)
Moral traditionalism -0.135 (0.205)
Egalitarianism 0.204 (0.150)

x limited government
Egalitarianism 0.085 (0.062)

x moral traditionalism
Joint control variables
Education 0099’” (0.027) -0.027 (0.028)
Interest -0014 (0.033) 0080* (0.033)
Political knowledge 0.102** (0.023) 0.001 (0.026)
Party ID intensity -0.173** (0.042) -0.216** (0.053)
Care who wins 0.018 (0.113) -0192 (0.122)
Intercept 0.042 (0.921)
T1 -0.731* (0.291)
r2 0.282 (0.283)
N 477
Log-likelihood -860.533
9523271 201.795

 

Signiﬁcance levels : T : 10% * : 5% ** : 1%

114

Table 5.9: Ambivalence and Timing of vote decision in 2000

 

 

 

 

 

Ambivalence Timing of vote decision

Attitudinal states
Ambivalence 0825** (0.311)
Indifference 0.023 (0.124)
Timing of vote control variables
Gender 0.027 (0.077)
Minority -0.265** (0.101)
Age -0005* (0.002)
Income -0.006 (0.010)
Ambivalence control variables
Need for cognition 0250* (0.111)
Egalitarianism -1.182* (0.516)
Limited government -0015 (0.309)
Moral traditionalism -0760 (0.536)
Egalitarianismx 0.365 (0.485)

limited government
Egalitarianism x 1698* (0.799)

moral traditionalism
Joint control variables
Education 0.115** (0.026) -0.021 (0.031)
Interest 0.005 (0.026) ~0070* (0.034)
Political knowledge 0097** (0.025) -0063* (0.028)
Party ID intensity -0.139** (0.037) -0.293** (0.057)
Care who wins 0282’” (0.100) -0363** (0.124)
# Bush supporters 0.036 (0.041) -0149** (0.041)
# Gore supporters -0.002 (0.050) -0.037 (0.047)
# Gore supportersx 0.052 (0.047) 0.072 (0.045)

# Bush supporters
Intercept -0854* (0.372)
1'1 -2.526** (0.289)
7'2 -1.190** (0.249)
N 514
Log-likelihood -1709.097
€27) 294.982
Signiﬁcance levels : i : 10% * : 5% ** : 1%

115

Model Results: 1984—2000 for Detection

Table 5.10: Ambivalence and defection in 1984

 

 

 

 

Ambivalence Defection

Attitudinal states
Ambivalence 0.330 (0.349)
Indifference 0586’” (0.129)
Defection control variables
Issues -0031 (0.077)
Gender 0.061 (0.106)
Minority 0.071i (0.038)
Age -0001 (0.004)
Income 0.003 (0.010)
Ambivalence control variables
Egalitarianism -0.186 (0.129)
limited government -0069 (0.091)
Egalitarianism 0.059T (0.032)

x limited government
Joint control variables
Education 0.116** (0.024) -0.069l (0.042)
Interest 0.232** (0.056) -0.107 (0.085)
Political knowledge 0091** (0.019) -0.051 (0.034)
Party ID intensity 0095* (0.038) -0073 (0.070)
Care who wins 0.003 (0.031) -0006 (0.046)
Intercept -1.521** (0.398) -0845* (0.348)
N 1043
Log-likelihood -1222.955
X320) 173.531
Signiﬁcance levels : I : 10% * : 5% ** :

116

Table 5.11: Ambivalence and defectign in 1988

 

 

 

Ambivalence Defection
Attitudinal states
Ambivalence -0.368 (0.653)
Indifference 0293* (0.125)
Defection control variables
ssuesi 0.031 (0.061)
Gender -0199* (0.100)
Minority -0.114 (0.136)
Age 0.006l (0.003)
Income -0002 (0.009)
Ambivalence control variables
Egalitarianism 0.288l (0.149)
Limited government 0.102 (0.075)
Egalitarianism -0.043 (0.033)
x limited government
Joint control variables
Education 0136** (0.023) -0.016 (0.053)
Interest -0.068* (0.029) 0.000 (0.042)
Political knowledge 0.166” (0.030) -0.011 (0.063)
Party ID intensity -0.039 (0.039) -0.239** (0.065)
Care who wins -0.077 (0.082) -0313** (0.104)
Intercept -1.919** (0.360) 0.465 (0.474)
N 1250
Log-likelihood -1232.638
2 210.838

X(21)
Signiﬁcance levels: 1‘ : 10% * : 5% ** : 1%

 

117

Table 5.12: Ambivﬂence and defection in 1992

 

 

 

Ambivalence Defection
Attitudinal states
Ambivalence 1.177** (0.429)
Indifference 0591** (0.162)
Defection control variables
Issues -0145 (0.102)
Gender -0062 (0.124)
Minority 0.080 (0.044)
Age 0007* (0.004)
Income -0.006 (0.010)
Ambivalence control variables
Egalitarianism -0378* (0.147)
Limited government -1.301** (0.315)
Moral traditionalism -0.116 (0.135)
Egalitarianism 0.348** (0.117)
x limited govermnent
Egalitarianism 0.030 (0.055)
x moral traditionalism
Joint control variables
Education 0.101 ** (0.027) -0061 (0.043)
Interest -0.018 (0.033) 0034 (0.056)
Political knowledge 0.098** (0.023) 0.003 (0.041)
Party ID intensity -0173** (0.042) -0320** (0.097)
Care who wins 0.026 (0.113) 03011‘ (0.177)
Intercept -0.616l (0.350) 0301 (0.460)
N 761
Log-likelihood -909.206
190.4

X2
(21)
Signiﬁcance levels : 'l' : 10% at : 5% ** : 1%

 

118

Table 5.13: Ambivalence and defection decision in 1996

 

 

 

 

Ambivalence Defection

Attitudinal states
Ambivalence 0.382 (0.311)
Indifference 0662’” (0.156)
Defection control variables
Issues -0058 (0.134)
Gender 0.047 (0.175)
Minority 0.078 (0.064)
Age 0.008 (0.005)
Income 0.016 (0.016)
Ambivalence control variables
Egalitarianism 0277 (0.326)
Limited government -0069 (0.659)
Moral traditionalism -0.095 (0.231)
Egalitarianism 0.056 (0.197)

x limited government
Egalitarianism 0.064 (0.070)

x moral traditionalism
Joint control variables
Education 0129** (0.034) 0.012 (0.069)
Political knowledge 0.185** (0.042) 0014 (0.071)
Interest 0152i (0.084) 0060 (0.139)
Party ID intensity -0.157** (0.058) -0.332** (0.118)
Care who wins 0.156 (0.147) 0538* (0.229)
Intercept -1.032 (0.761) 0.343 (0.474)
N 503
Log-likelihood -546.734
57%”) 122.808
Signiﬁcance levels : I : 10% * : 5% *4: : 1%

119

Table 5.14: Ambivalence and defection decision in 2000

 

 

 

 

Ambivalence Defection

Attitudinal states
Ambivalence 0.382 (0.311)
Indifference 0662** (0.156)
Defection control variables
Gender -0250* (0.126)
Minority 0.141 (0.149)
Age 0002 (0.004)
Income -0010 (0.018)
Ambivalence control variables
Need for cognition 0285* (0.111)
Egalitarianism 0870i (0.523)
Limited government 0.100 (0.317)
Moral traditionalism -0.541 (0.547)
Egalitarianism x 0.193 (0.498)

limited government
Egalitarianismx 1.314 (0.815)

moral traditionalism
Joint control variables
Education 0.115** (0.026) -0040 (0.045)
Interest -0001 (0.026) -0143* (0.057)
Political knowledge 0092** (0.025) -0009 (0.043)
Party ID intensity -0143** (0.037) -0337** (0.078)
Care who wins 0.275** (0.100) 0315* (0.154)
# Bush supporters 0.038 (0.041) -0092 (0.074)
# Gore supporters -0003 (0.050) -0159l (0.087)
# Gore supportersx 0.050 (0.047) 0.050 (0.088)

# Bush supporters
Intercept -1.052** (0.375) 0.157 (0.352
N 579
Log-likelihood -1125.751
X07) 252.894
Signiﬁcance levels : I : 10% * : 5% ** : 1%

120

Chapter 6

Conclusions

This study sought to achieve three primary objectives. Before examining substantive
issues of partisan ambivalence, I ﬁrst wanted to examine common measures. The next
objective was to consolidate and elaborate on an informational theory of ambivalence.
The ﬁnal objective was to examine how and if partisan ambivalence inﬂuences voting
behavior in presidential elections. In this chapter, I summarize and synthesize the
ﬁndings from proceeding chapters. I will then document my perceived shortcomings

in an attempt to shed light on future possible studies.

6.1 Findings

As there is a growing consensus around the appropriateness of objective measures,
it is appropriate the properties of the measures be scrutinized to rigorous empirical
examination. Speciﬁcally, I examined the level of measurement with the results quite
clearly indicating that the measures are not continuous, or even ordinal. As was
shown, substantive conclusions are altered if the variable is treated as interval rather
than nominal. These ﬁndings result primarily from those individuals who, when given
the opportunity to do, offer no thoughts about the two parties. Objective measures

traditionally treat zero as a state in between one-sidedness and ambivalence; the

121

problem is that the vast majority of zeroes are made up individuals who offer no
thoughts about the parties. These individuals appear to be qualitatively different from
other respondents. While others have coded these individuals as indifferent (Basinger
and Lavine 2005), this represents the ﬁrst systematic empirical study of whether or not
such a practice is appropriate. The measures are capable of categorizing individuals,
but appear unsuited to gauge the intensity of conﬂicted thoughts.

Also examined was the appropriateness of a measure of comparative ambivalence.
Speciﬁcally, the question of whether or not negative responses about one party can be
combined with positive about the other. Just as with the proper level of measurement,
this assumption can be empirically examined. As the measure tries to capture the
intensity and similarity of conﬂicted thoughts, it explicitly assumes negative and
positive thoughts are related. While several tests were inconclusive, the results of
a factor analysis quite clearly indicate that a measure of comparative ambivalence
is appropriate. Furthermore, scales created of positive thoughts of one party and
negative thoughts were shown to be reliable, whereas expressed thoughts about the
same party appear to be largely unrelated. With a clear idea of how to measure the
concept I moved on to substantive issues.

The ﬁrst substantive goal was to elaborate and offer further evidence for the in-
formational theory of ambivalence. The theory suggests ambivalence is caused by
exposure to multiple points of view. This suggests an ambivalent individual has ac-
cepted both positive and negative messages. Consequently, I modeled the probability
an individual could recall at least one positive and one negative thought about the
two parties (implied by recall are both the reception and acceptance of a message).
The two probabilities were found to be independent. The independence of the prob-
abilities allowed me to compare the product of the two with the probability of being
classiﬁed as ambivalent toward that party. The two were shown to be very highly

correlated with one another, providing compelling evidence that the acceptance of

122

messages about the two parties are causes of ambivalence and that the theoretical
framework of Chapter 3 is appropriate. This dovetails well with the ﬁndings from
Chapter 2, which showed it was knowledge that best predicted ambivalence and listing
more than one statement about the political parties.

To further test the theory, I looked to see if potential messages in a given campaign
were sources of ambivalence. I selected messages which were prominent in the cam-
paigns, meaning they were most likely to be heard by many respondents. Although
the substantive effects of the campaign messages selected are not large, this is not
terribly surprising, as there is much error involved when measuring issue attitudes
(Ansolabehere, Rodden and Jr. 2008). What was directly provided was evidence
that these inﬂuence ambivalence. That the ﬁnding was robust to different ways of
conceptualizing and modeling partisan ambivalence adds to the strength.

The next goal was to examine the extent to which partisan ambivalence might
inﬂuence voting behvaior. I began ﬁrst by using measures of party speciﬁc ambiva-
lence. Indirect evidence was provided suggesting that as parties polarize, individuals
become more ambivalent. The exact nature of this relationship needs to be addressed
more fully, examining individual level results for all election years available from the
inception of the NES. However, the indirect evidence is suggestive. More concretely,
the evidence in Chapter 4 offers limited support for the hypothesis that as parties
polarize, individuals experiencing ambivalence will actually be more likely to stay
with their preferred party. The most interesting ﬁnding is that for the most part
ambivalence plays little role in electoral decision-making.

This ﬁnding was replicated when running a different model as above this time
using comparative ambivalence. While not inﬂuencing the ﬁnal decision, I hoped
to see how ambivalence might inﬂuence when one makes up one’s mind during a
campaign. The argument was that one should expect ambivalent individuals to delay

but not alter their ﬁnal decision. The ﬁndings were consistent with those from the

123

previous chapter. Speciﬁcally, from 1980 to 2004 ambivalent plays little role in the
ﬁnal decision of voters. Consistent with those who suggest a need for cognition and
information cause ambivalence and people that say ambivalent individuals seek out
and use information differently, it does delay the decision.

Indifference, on the other hand, is related to vote choice. Individuals who do not
care are more likely to change parties. Decades of previous research has provided evi-
dence there is a class of voters who are unaware about, uninvolved with, uninterested
in, and could generally be described as lacking attachment to politics. A measure of
indifference seems to be able to categorize these individuals quite well. Thus, it is
not surprising these individuals, who for the most part do not feel connected to the
system or the parties, are more likely to vote inconsistently with their party identiﬁ-
cation. More speciﬁcally, the measure of indifference appears to be a good proxy for
these related concepts which try to capture those individuals who are inattentive yet
open to persuasion.

A few general themes begin to emerge, which are largely consistent with previous
research on ambivalence and electoral behavior more generally. There is a growing
base of evidence that ambivalence regarding the political parties is largely related to
information. Although ambivalence can be generated from emotional responses, this
does not seem to be the case here. Over the last decade or so a growing body of
literature is beginning to suggest that value conﬂict may not play as large of a role in
determining political attitudes as originally suggested. More speciﬁcally, in the case
of partisanship, it seems quite clear that value conﬂict plays at most a complimentary
role to informational sources.

A second theme is that ambivalence does not inﬂuence voting behavior. Robust
to different conceptualizations and measurements, and consistent with decades of
research on political behavior and partisanship, one’s attachment to one of the two

parties is enough to overcome sources of paritsan conﬂict. As noted earlier, in many

124

ways, ambivalence is a more nuanced version of the concepts of cross-presssure and
attitudinal conﬂict. Yet, with this new measure and more appropriate modeling
techniques, it still appears the relationship between one’s partisan identity and vote
choice is for the most part unaffected by conﬂicted attitudes.

Finally, from the earliest studies of voting it was recognized that there is a certain
class of voters who weakly identify with the parties, and politics more generally.
Berelson et al. (1954) noted that although these individuals are most receptive to
campaigns, they are least able to make use of the informational environment of a
campaign. Interestingly, while the measures of ambivalence did not originally intend
to categorize individuals as indifferent, they do just as well at this task as they do at

categorizing individuals as ambivalent.

6.2 Future Directions

The ﬁnding that most begs further research is the lack of a relationship between
ambivalence and voting behavior. As mentioned, presidential elections in the US.
might be the least likely scenario in which one would expect ambivalence to inﬂuence
behavior. For many, presidential voting is the primary political act, and likely the
most related to their partisan identity. That no effects were found here does not mean
the ﬁnding will be replicated in other situations. Work, therefore, should proceed to
look at other scenarios where ambivalence could, perhaps more likely, inﬂuence the
electoral behavior of citizens.

This means moving in a variety of directions. First, one could stay in the US. con-
text and examine elections with a strong third party candidate. Ambivalent Southern
Democrats in 1968, for example, might have opted for a third party candidate. Am-
bivalence might also be able to mediate the role of partisanship and vote choice in

sub-presidential elections, and therefore we would expect to see split-ticket voting as

125

an empirical manifestation of ambivalence. There also exists the possibility of moving
to a comparative context. Systems where there are more than two alternatives could
give ambivalent voters a viable alternative. Especially those systems where there are
two dominant parties and one fairly strong third party, such as the British political
system. Certainly, the results presented in chapters 5 and 6 do not indicate that
ambivalence will never alter behavior suggesting further research about the impact
ambivalence has on voting is necessary.

Relatedly, as indifference turns out to be a. key indicator of voting behavior, in-
vestigating these individuals in more detail would be useful. In this case, the primary
objective would most likely be to see how well they correspond to those voters already
identiﬁed by other scholars as unattached to the political system (e. g., Granberg and
Holmberg 1990; Jacoby 2009; MacKuen 1984). Are there systematic patterns to how
they defect, or is it largely random, meaning defection cancels out?

And ﬁnally, and somewhat unrelatedly, is to further investigate the process of
listing likes and dislikes. The last section of Chapter 2 is just a starting point. For
example, as it was shown that PD and NR are related, it could be possible this is a
form of Democratic accessibility. If true, then scholars would have at their disposal a
measure that does not rely on response latencies (e.g., Fazio 1990; Grant, Mockabee
and Monson 2009; Huckfeldt, Sprague and Levine 2000) dating back to the inception
of the National Election Studies.

To summarize, future work should move in three primary directions. Find and
examine scenarios where ambivalence is more likely to inﬂuence vote choice. Second,
to see if there are systematic implications of indifference. And, to see what other

information can be extracted from statements listed by respondents.

126

Bibliography

Abelson, Robert P., Donald R. Kinder, Mark D. Peters and Susan T. Fiske. 1982.
“Affective and Semantic Components in Political Person Perception.” Journal of
Personality and Social Psychology 42(4):619—630

Achen, Christopher. 1992. “Social Psychology, Demographic Variables, and Linear
Regression: Breaking the Iron Triangle in Voting Research.” Political Behavior
14(3):195—211.

Achen, Christopher H. and Larry M. Bartels. 2006. “It Feels Like We’re Thinking:
The Rationalizing Voter and Electoral Democracy.” Presented at the 2006 Annual
Meeting of the American Political Science Association.

Adams, Greg D. 1997. “Abortion: Evidence of an Issue Evolution.” American Journal
of Political Science 41(3):718—737.

Ai, Churong and Edward C. Norton. 2003. “Interaction Terms in Logit and Probit.”
Economic Letters 80(1):123—129.

Allport, Gordon W. 1935. Attitudes. In Handbook of Social Psychology, ed. Carl A.
Murchison. New York: Russell and Russell.

Alvarez, R. Michael. 1990. “The Puzzle of Party Identiﬁcation: Dimensionality of an
Important Concept.” American Politics Quarterly 18:476—491.

Alvarez, R. Michael and John Brehm. 1995. “American Ambivalence Towards Abor-
tion Policy: Development of a Heteroskedastic Probit Model of Competing Values.”
American Journal of Political Science 39(4):1055-1082.

Alvarez, R. Michael and John Brehm. 1997. “Are Americans Ambivalent Towards
Racial Policies?” American Journal of Political Science 41(2):345-374.

Alvarez, R. Michael and John Brehm. 2002. Hard Choices, Easy Answers: Values,
Information, and American Public Opinion. Princeton, NJ: Princeton University
Press.

Alvarez, R. Michael and Jonathan Nagler. 1995. “Economics, Issues, and the Perot
Candidacy: Voter Choice in the 1992 Presidential Election.” American Journal of
Political Science 39(3):714—744.

127

Ansolabehere, Stephen, Jonathan Rodden and James M. Snyder Jr. 2008. “The
Strength of Issues: Using Multiple Measures to Gauge Preference Stability, Ideolog-
ical Constraint, and Issue Voting.” American Political Science Review 102(2):215—
232.

Armitage, Christopher J. and Mark Conner. 2005. Attitudinal Ambivalence and
Political Opinion: Avenues for Further Research. In Ambivalence, Politics, and
Public Policy, ed. Stephen C. Craig and Michael D. Martinez. New York: Palgrave
Macmillan pp. 144—166.

Armstrong, David and Duncan McVicar. 2000. “Value Added in Further Education
and Vocational Training in Northern Ireland.” Applied Economics 32(13):1727—
1736.

Bafumi, Joseph. 2004. “The Stubborn American Voter.” Presented at the 2004 Annual
Meeting of the American Political Science Association, Chicago, Illinois.

Bafumi, Joseph and Michael C. Herron. 2007. “Preference Aggregation, Representa-
tion, and Elected American Political Institutions.” Presented at the 2007 Annual
Meeting of the American Political Science Association.

Baldassarri, Delia and Andrew Gelman. 2008. “Partisans Without Constraint: Po-
larization and Trends in American Public Opinion.” American Journal of Sociology
114(2):Forthcoming.

Barker, David C. and Susan B. Hansen. 2005. “All Things Considered: System-
atic Cognitive Processing and Electoral Decision-making.” Journal of Politics
67(2):319—344.

Bartels, Larry M. 1993. “Messages Received: The Political Impact of Media Expo-
sure.” American Political Science Review 87(2):267—285.

Bartels, Larry M. 2000. “Partisanship and Voting Behavior, 1952-1996.” American
Journal of Political Science 44(1):35-50

Bartels, Larry M. 2002. “Beyond the Running Tally: Partisan Bias in Political
Perceptions.” Political Behavior 24(2):117—150

Basinger, Scott J. and Howard Lavine. 2005. “Ambivalence, Information, and Elec-
toral Choice.” American Political Science Review 99(2):169—184.

Bassili, John N. 1996. “Meta-Judgmental Versus Operative Indexes of Psychological
Attributes: The Case of Measures of Attitude Strength.” Journal of Personal and
Social Psychology 71 (4):637—653.

Beck, Paul A. 2002. “Encouraging Political Defection: The Role of Personal Discus-
sion Networks in Partisan Desertions to the Opposition Party and Perot Votes in
1992.” Political Behavior 24(4):309—337.

128

Beck, Paul Allen, Russell J. Dalton, Steven Greene and Robert Huckfeldt. 2002. “The
Social Calculus of Voting: Interpersonal, Media, and Organizational Inﬂuences on
Presidential Choices.” American Political Science Review 96(1):57—73.

Berelson, Bernard R., Paul F. Lazarsfeld and William N. McPhee. 1954. Voting: A
Study of Opinino Formation in a Presidential Campaign. Chicago: University of
Chicago Press.

Bishop, George F., Alfred J. Tuchfarber and Robert .W. Oldendick. 1986. “Opinions
on Fictitious Issues: The Pressure to Answer Survey Questions.” Public Opinion
Quarterly 50(2):240—250

Brambor, Thomas, William Roberts Clark and Matt Golder. 2006. “Understanding
Interaction Models: Improving Empirical Analyses.” Political Analysis 14(1):63—82.

Breckler, Steven J. 1994. “A Comparison of Numerial Indexes for Measuring Attitude
Ambivalence.” Educational and Psychological Measurement 54(2):350—365.

Brown, Judson S. and LE. Farber. 1951. “Emotions Conceptualized as Intervening
Variables—With Suggestions toward a Theory of Frustration.” Psychological Bul-
letin 48(6):465-495.

Cacioppo, John T. and G. G. Bernston. 1994. “Relationship Between Attitudes and
Evaluative Space: A Critical Review of the Separability of Positive and NEgative
Substrates.” Psychological Bulletin 115(401—423).

Campbell, Angus, Philip E. Converse, Warren E. Miller and Donald E. Stokes. 1960.
The American Voter. Chicago: University of Chicago Press.

Cantril, Albert H. and Susan Davis Cantril. 1999. Reading Mixed Signals: Ambiva-
lence in American Public Opinion about Government. Washington, DC: Woodrow
Wilson Center Press.

Carmines, Edward G. and James A. Stimson. 1989. Issue Evolution: Race and the
Transformation of American Politics. Princeton University Press.

Carsey, Thomas M. and Geoffrey C. Layman. 2006. “Changing Sides or Changing
Minds? Party Identiﬁcation and Policy Preferences in the American Electorate.”
American Journal of Political Science 50(2):464—477.

Chaffee, Steven H. and Rajiv Nath Rimal. 1996. Time of Vote Decsion and Openess
to Persuasion. In Political Persuasion and Attitude Change, ed. Diana C. Mutz,
Paul M. Sniderman and Richard A. Brody. Ann Arbor, MI: Univ of Michigan
Press.

Chaffee, Steven H. and Sun Yuel Choe. 1980. “Time of Decision and Media Use
During the Ford-Carter Campaign.” Public Opinion Quarterly 44(1):53—69.

129

Cleveland, William S. 1979. “Robust Locally Weighted Regression and Smooth Scat-
terplots.” Journal of the American Statistical Association 74(368):829—836.

Cliff, Norman. 1993. What Is and Isn’t Measurement. In A Handbook or Data Analysis
in the Behavioral Sciences, ed. Gideon Keren and Gordon Lewis. Hillsdale, NJ:
Lawrence Erlbaum.

Converse, Philip E. 1964. “The Nature of Belief Systems in Mass Publics.” Ideology
and Discontent .

Converse, Philip E. 1966. Religion and Politics: The 1960 Election. In Elections and
the Political Order, ed. Angus Campbell, Philip E. Converse, Warren E. Miller and
Donald E. Stokes. New York: John Wiley 8!. Sons, Inc.

Converse, Philip E. 1970. Attitude and Non-Attitudes: Continuation of a Dialouge.
In The Quantitative Analysis of Social Problems, ed. Edward R. Tufte. Reading,
MA: Addison-Wesley.

Coombs, Clyde H. 1964. A Theory of Data. New York: Wiley.

Craig, Stepehn C. 1985. “Partisanship, Independence, and No Preference: Another
Look at the Measurement of Party Identiﬁcation.” American Journal of Political
Science 29(2):274—290

Craig, Stephen C., James G. Kane and Michael D. Martinez. 2002. “Sometimes You
Feel Like a Nut, Sometimes You Don’t: Citizens’ Ambivalence About Abortion.”
Political Psychology 23(2) :285-301.

Craig, Stephen C., Michael D. Martinez and James G. Kane. 1999. “The Structure of
Political Competition: Dimensions of Candidate and Group Evaluation Revisited.”
Political Behavior 21:289-304.

Craig, Stephen C., Michael D. Martinez and James G. Kane. 2005. “Core Values,
Value Conﬂict, and Citizens’ Ambivalence about Gay Rights.” Political Research
Quarterly 58(1):5—17.

Davis, James A. 1982. “Achievement Variables and Class Cultures.” American Soci-
ological Review 47(5):569—586.

Dobrynska, Agnieszka and André Blais. 2008. “Testing Zaller’s Reception and Accep-
tance Model in an Intense Election Campaign.” Political Behavior 30(2):259—276.

Dow, Jay K. and James W. Endersby. 2004. “Multinomial Probit and Multinomial
Logit: A comparison of Choice Models for Voting Research.” Electoral Studies
23(1):107—112.

Eagly, Alice H. and Shelly Chaiken. 1993. The Psychology of Attitudes. Fort Worth:
Harcourt Brace Jovanovich College Publishers.

130

Enelow, James M. and Melvin J. Hinich. 1984. The Spatial Theory of Voting: An
Introduction. New York: Cambridge University Press.

Fazio, Russell H. 1990. A Practical Guide to the Use of Response Latency in Social
Psychological Reseach. In Research Methods in Personality and Social Psychology,
ed. Clyde Hendrick and Margret S. Clark. N ewbury Park, CA: Sage Publications.

Feldman, Stanley. 1988. “Structure and Consistency in Public Opinion: the Role of
Core Beliefs and Values.” American Journal of Political Science 32(2):416—440

Feldman, Stanley. 2003. Values, Ideology, and the Structure of Political Attitudes.
In Oxford Handbook of Political Psychology, ed. David O. Sears, Leonie Huddy and
Robert Jervis. New York: Oxford University Press pp. 47 7—510

Feldman, Stanley and John Zaller. 1992. “The Political Culture of Ambivalence:
Ideological Responses to the Welfare State.” American Journal of Political Science
36(2):268—307.

Finkel, Steven E. 1993. “Reexarnining the ‘Minimal Effects’ Model in Recent Presi-
dential Campaigns.” Journal of Politics 55(1):1—21.

Fiorina, Morris P. 1976. “The Voting Decision: Instrumental and Expressive As-
pects.” Journal of Politics 38(2):390—413.

Fiorina, Morris P. 1981. Retrospective Voting in American National Elections. New
Haven: Yale University Press.

Fiorina, Morris P., Samuel J. Abrams and Jeremy Pope. 2005. Culture War?: The
Myth of a Polarized America. Pearson Longman.

Fournier, Patrick, Richard Nadeau, André Blais, Elisabeth Gidengil and Neil Nevitte.
2001. “Validation of Time-of-Voting—Decision Recall.” Public Opinion Quarterly
65(1):95—-107.

Fournier, Patrick, Richard N adeau, André Blais, Elisabeth Gidengil and Neil Nevitte.
2004. “Time—of-Voting Decision and Susceptibility to Campaign Effects.” Electoral
Studies 23(4):661—681.

Franklin, Charles. 1991. “Eschewing Obfuscation? Campaigns and the Perception
of U. S. Senate IncumbentsPolitical Science Review.” American Political Science
Review 85(4):1193—1214.

Franklin, Charles and John Jackson. 1983. “The Dynamics of Party Identiﬁcation.”
American Political Science Review 77 (4):957—973.

Free, Lloyd A. and Hadley Cantril. 1967. “The Political Beliefs of Americans: A
Study of Public Opinion”.

Giﬁ, Albert. 1990. Nonlinear Multivariate Analysis. Wiley.

131

Gopoian, J. David and Sissie Hadjiharalambous. 1994. “Late-Deciding Voters in
Presidential Elections.” Political Behavior 16(1):55—78.

Granberg, Donald and Soren Holmberg. 1990. “The Berelson Paradox Reconsid-
ered: Intention-behavior Changes in US. and Swedish Election Campaigns.” Public
Opinion Quarterly 54(4):530—550

Grant, J. Tobin, Stephen T. Mockabee and J. Quin Monson. 2009. “Campaign
Effects on the Accessibility of Party Identiﬁcation.” Political Research Quarterly
doi:10.11 77 /1065912909336269.

Grant, J. Tobin and Thomas J. Rudolph. 2003. “Value Conﬂict, Group Affect, and the
Issue of Campaign Finance.” American Journal of Political Science 47(3):453—469.

Green, Donald, Bradkey Palmquist and Eric Schickler. 2002. Patisan Hearts 6? Minds:
Political Parties and the Social Identities of Voters. New Haven: Yale University
Press.

Green, Donald P. 1988. “On the Dimensionality of Public Sentiment toward Partisan
and Ideological Groups.” American Journal of Political Science 32(3):758—780

Greene, Steven. 2005. “The Structure of Partisan Attitudes: Reexamining Partisan
Dimensionality and Ambivalence.” Political Psychology 26(5):809—822.

Greene, William H. 1998. “Gender Economics Courses in Liberal Arts Colleges:
Further Results.” Research in Economic Education 19(4):291—300

Greene, William H. and David A. Hensher. 2008. “Modeling Ordered Choices:
A Primer and Recent Developments (June 15, 2008).” Available at SSRN:
http: //ssrn.com/abstract= 1213093.

Gurmu, S. 1991. “Tests for Detection Overdispersion in the Positive Poisson Regres-
sion Model.” Journal of Business 89’ Economic Statistics 9(2):215—222.

Guttmann, Louis. 1977. “What is Not What in Statistics.” The Statistician 26(2):81—
107.

Hausman, Jerry and Daniel McFadden. 1984. “Speciﬁcation Tests for the Multinomial
Logit Model.” Econometrica 52(5):1219—1240

Hetherington, Marc J. 2001. “Resurgent Mass Partisanship: The Role of Elite Polar-
ization.” American Political Science Review 95(3):619—631.

Hill, Jennifer L. and Hanspeter Kriesl. 2001. “An Extension and Test of Converse’s
“Black-and—White” Model of Response Stability.” American Political Science Re-
view 95(02):397—413.

Hillygus, D. Sunshine. 2005. “Campaign Effects and the Dynamics of Turnout Inten-
tion in Election 2000.” Journal of Politics 67(1):50—68.

132

Hillygus, D. Sunshine and Simon D. Jackman. 2003. “Voter Decision Making in
Election 2000: Campaign Effects, Partisan Activation, and the Clinton Legacy.”
American Journal of Political Science 47(4):589—596.

Hillygus, D. Sunshine and Todd G. Shields. 2008. The Persuadable Voter: Wedge
Issues in Presidential Campaigns. Princeton, N.J.: Princeton University Press.

Hochschild, J.L. 1981. What’s Fair?: American Beliefs about Distributive Justice.
Cambridege, MA: Harvard University Press.

Horan, Patrick M. 1971. “Social Positions and Political Cross-Pressures: A re—
Examination.” American Sociological Review 4:650—660

Huckfeldt, Robert, Jeanette Morehouse Mendez and Tracy Osborn. 2004. “Disagree-
ment, Ambivalence, and Engagement: The Political Consequences of Heteroge—
neous Networks.” Political Psychology 25(1):65—95.

Huckfeldt, Robert, John Sprague and Jeffrey Levine. 2000. “The Dynamics of Collec-
tive Deliberation in the 1996 Election: Campaign Effects on Accessibility, Certainty,
and Accuracy.” American Political Science Review 95(3):641—651.

Huckfeldt, Robert R. 1979. “Political Participation and the Neighborhood Social
Context.” American Journal of Political Science 23(3):579—592.

Jackson, Elton F. and Richard F. Curtis. 1972. “Effects of Vertical Mobility and
Status Inconsistency: A Body 0 Negative Evidence.” American Sociological Review
37(6):?01—713.

Jacoby, William G. 1988. “The Impact of Party Identiﬁcation on Issue Attitudes.”
American Journal of Political Science 32(3):643—661.

Jacoby, William G. 1991. Data Theory and Dimensional Analysis. Newbury Park:
SAGE Publications.

Jacoby, William G. 1999. “Levels of Measurement and Political Research: An Opti-
mistic View.” American Journal of Political Science 43(1):271—301.

Jacoby, William G. 2002. Core Values and Political Attitudes. In Understanding
Public Opinion, ed. Barbara Noreandar and Clyde Wilcox. 2nd ed. Washington,
DC: CQ Press pp. 177—201.

Jacoby, William G. 2006. “Value Choices and American Public Opinion.” American
Journal of Political Science 50(3):706—723.

Jacoby, William G. 2009. “Public opinion during a presidential campaign: Distin-

guishing the effects of environmental evolution and attitude change.” Electoral Stud-
ies doi:10.1016/j.electstud.2009.05.011.

133

Johnston, Christopher, Howard Lavine, David Perkens and Marco R. Steenbergen.
2008. “Elite Polarization, Partisan Ambivalence, and a Preference for Divided Gov-
ernment.” Presented at the 2008 Midwest Political Science Association Conference.

Judd, Charles M. and Cynthia M. Lusk. 1984. “Knowledge Structures and Evaluative
Judgments: Effects of Structural Variables on Judgmental Extremity.” Journal of
Personality and Social Psychology 46(6):1193—1207.

Kam, Cindy K. 2006. “Political Campaigns and Open—Minded Thinking.” Journal
of Politics 68(4):931—945.

Kaplan, K.J. 1972. “On the Ambivalence-Indifference Problem in Attitude Theory
and Measurement: A Suggested Modiﬁcation of the Semantic Differential Tech-
nique.” Psychological Bulletin 77(5):361-72.

Keele, Luke and David K. Park. 2007. “Ambivalent about Ambivalence: A Re-
examination of Heteroskedastic Probit Models.” Working paper.

Keele, Luke and Jennifer Wolak. 2006. “Value Conﬂict and Volatility in Party Iden-
tiﬁcation.” British Journal of Political Science 36:671—690.

Keele, Luke and Jennifer Wolak. 2008. “Contextual Sources of Ambivalence.” Political
Psychology 29(5):653—573.

Kelley, Stanley, Jr. 1983. Interpreting Elections. Princeton: Princeton University
Press.

Krosnick, Jon A. 1988. “Attitude Importance and Attitude Change.” Journal of
Experimental Social Psychology 24(3):240—255.

Krosnick, Jon A. and Robert P. Abelson. 1992. The Case for Measuring Attitude
Strength in Surveys. In Questions About Questions: Inquiries into the Cognitive
Bases of Surveys, ed. Judith M. Tanur. New York: Russel Sage Foundation.

Krosnick, Jon A., Savid S. Boninger, Yac C. Chuang, Mathew K. Berent and Cather-
ine G. Garnot. 1993. “Attitude Strength: One Construct or Many Related Con-
structs.” Journal of Personality and Social Psychology 65(6):1132—1151.

Kuhn, Thomas S. 1977. The Essential Tension: Selected Studies in Scientiﬁc Trudi-
tion and Change. Chicago: University of Chicago Press.

Lacy, D. and BC. Burden. 1999. “The Vote-Stealing and Turnout Effects of Ross
Perot in the 1992 US Presidential Election.” American Journal of Political Science
43(1):233—255.

Lavine, Howard. 2001. “The Electoral Consequences of Ambivalence toward Presi-
dential Candidates.” American Journal of Political Science 45(4):915—929.

134

Lavine, Howard, Christopher Johnston and Marco R. Steenbergen. 2009. “Ambiva-
lence, Information Processing, and Behavior.” Presented at the 2009 Annual Meet-
ing of the Midwest Political Science Association.

Lavine, Howard, Joseph W. Huff, Stephen H. Wagner and Donna Sweeney. 1998. “The
Moderating Inﬂuence of Attitude Strength on the Susceptibility to Context Effects
in Attitude Surveys.” Journal of Personality and Social Psychology 75(2):359—379.

Layman, Geoffrey C. and Thomas M. Carsey. 2002a. “Party Polarization and “Con-
ﬂict Extension” in the American Electorate.” American Journal of Political Science
46(4):786—802.

Layman, Geoffrey C. and Thomas M. Carsey. 2002b. “Party Polarization and Party
Structuring of Policy Attitudes: A Comparison of three NES Panel Studies.” Po-
litical Behavior 24(2) : 199—236.

Lazarsfeld, Paul F., Bernard R. Berelson and Hazel Gaudet. 1944. The People’s
Choice: How the Voter Makes Up His Mind in a Presidential Campaign. New
York: Duell, Sloan and Pearce.

Lewis-Beck, Michael .8, William G. Jacoby, Helmut Norpoth and Herbert F. Weis-
berg. 2008. The American Voter Revisited. Ann Arbor, MI: Univ of Michigan
Press.

Lin, Lawrence I-Kuei. 1989. “A Concordance Correlation Coefﬁcient to Evaluate
Reproducibility.” Biometrics 45(1):255—268.

Linville, Patricia W. 1982. “The Complexity-Extremity Effect and Age—Based Stereo-
typing.” Journal of Personality and Social Psychology 42(2):193—211.

Lodge, Milton, Marco R. Steenbergen and Shawn Brau. 1995. “The Responsive Voter:
Campaign Information and the Dynamics of Candidate Evaluation.” American
Political Science Review 89(2):309—326.

Long, J. Scott and Jeremy Freese. 2005. Regression Models for Categorical Outcomes
Using Stata. Second ed. College Station, TX: Stata Press.

Lord, FM. 1953. “On the Statistical Tieatment of Football Numbers.” American
Psychologist 8(3):750—751.

MacKuen, Michael B. 1984. “Exposure to Information, Belief Integration and In-
dividual Responsiveness to Agenda Change.” American Political Science Review
78(2):372—391.

Martinez, Michael D., Stephen C. Craig and James G. Kane. 2005. Pros and Cons:
Ambivalence and Public Opinion. In Ambivalence and the Structure of Public Opin-
ion, ed. Michael D. Martinez and Stephen C. Craig. New York: Palgrave Macmillan.

135

McClosky, Herbert. 1964. “Consensus and Ideology in American Politics.” American
Political Science Review 58(2):361—382.

McClosky, Herbert and John Zaller. 1984. The American Ethos: Public Attitudes
Toward Capitalism and Democracy. Cambridege, MA: Harvard University Press.

McGraw, Kathleen M. and Brandon Bartels. 2005. Ambivalence Toward American
Political Institutions: Sources and Consequences. In Ambivalence and the Struc-
ture of Public Opinion, ed. Stephen C. Craig and Michael D. Martinez. Palgrave
Macmillan.

McGraw, Kathleen M., Edward Hasecke and Kimberly Conger. 2003. “Ambiva-
lence, Uncertainty, and Processes of Candidate Evaluation.” Political Psychology
24(3):421—448.

McVicar, Duncan and Brian McKee. 2002. “Part-time Work During Post-compulsory
Education and Examination Performance: Help or Hindrance?” Scottish Journal
of Political Economy 49:393—406.

Meffert, Michael F., Michale Guge and Milton Lodge. 2000. Good, Bad, Indifferent
and Ambivalent: The Consequence of Multidimensional Political Attitudes. In The
Issue of Belief: Essays in the Intersection of Non-Attitudes and Attitude Change,
ed. Willem E. Saris and Paul M. Sniderman. Amsterdam: The Amsterdam School
of Communication Research, Universiteit van Amsterdam, The Netherlands pp. 60—
100

Miller, Joanne M. and David A.M. Peterson. 2004. “Theoretical and Empirical Im-
plications of Attitude Strength.” Journal of Politics 66(3):847—867.

Mowrer, Orval Hobart. 1960. Learning Theory and Behavior. New York: Wiley.

Mutz, Diana C. 2002. “The Consequences of Cross-Cutting Networks for Political
Participation.” American Journal of Political Science 46(4):838—855.

Newbay-Clark, Ian R., Ian McGregor and Mark P. Zanna. 2002. “Thinking and
Caring About Cognitive Inconsistency: When and for Whom Woes Attitudinal
Ambivalence Feel Uncomfortable?” Journal of Personality and Social Psychology
82(2):157—166.

Nie, Norman H., Sidney Verba and John R. Petrocik. 1979. The Changing American
Voter. Cambridge, MA: Harvard University Press Cambridge, Mass.

N iemi, Richard G. and Herbert F. Weisberg. 1976. Are Parties Becoming Irrelevant. In
Controversies in American Voting Behavior, ed. Richard G. Niemi and Herbert F.
Weisberg. San Francisco: W. H. Freeman and Company.

Nir, Lilach. 2005. “Ambivalent Social Networks and Their Consequences for Partici-
pation.” International Journal of Public Opinion Research 17(4):422—442.

136

Nir, Lilach and James N. Druckman. 2008. “Campaign Mixed-Message Flows
and Timing of Vote Decision.” International Journal of Public Opinion Research
20(3):326—346.

Parker-Stephen, Evan. 2007. “When Perception Polarize: How Motives and Informa-
tion Shape Partisan Inference.” Presented at the 2007 Midwest Political Science
Association Conference.

Pool, Ithiel de Sola, Robert P. Abelson and Samuel Popkin. 1965. Candidates, Issues
and Strageﬁes. Cambridge, MA: MIT Press.

Priester, Joseph and Richard E. Petty. 1996. “The Gradual Threshold Model of
Ambivalence: Relating the Positive and Negative Bases of Attitudes to Subjective
Ambivalence.” Journal of Personality and Social Psychology 71(3):431-449.

Ranney, Austin. 1954. The Doctrine of Responsible Party Government: Its Origins
and Present State. Urbana The University of Illinois Press.

Rokeach, Milton. 1973. The Nature of Human Values. New York: Free Press.

Roodman, David. 2007. cmp: Stata Module to Implement Conditional (Recursive)
Mixed Process Estimator. Technical report.

Rudolph, Thomas J. 2005. “Group Attachment and the Reduction of Value—Driven
Ambivalence.” Political Psychology 26(6).

Rudolph, Thomas J. and Elizabeth Popp. 2007. “An Information Processing Theory
of Ambivalence.” Political Psychology 28(5):563—586.

Schuman, Howard and Stanley Presser. 1981. Questions and Answers in Attitude
Surveys. New York: Academic Press.

Scott, William A. 1969. “Structure of Natural Cognitions.” Journal of Personality
and Social Psychology 12(4).

Small, Kenneth A. Hsiao, Cheng. 1985. “Multinomial Logit Speciﬁcation Tests.”
International Economic Review 26(3):619—267.

Sniderman, Paul, Richard Brody and Phillip Tetlock. 1991. Reasoning and Choice.
New York: Cambridge University Press.

Steenbergen, Marco R. and Paul R. Brewer. 2004. The Not—So Ambivalent Public:
Policy Attitudes in the Political Culture of Ambivalence. In Studies in Public Opin-
ion: Attitudes, Nonattitudes, Measurement Error, and Change, ed. Willem Saris
and Paul Snideran. Princeton, NJ: Princeton University Press.

Stevens, Stanley Smith. 1951. Mathematics, Measurement and Psychophysics. In
Handbook of Experimental Psychology, ed. Stanley Smith Stevens. New York: John
Wiley & Sons, Inc.

137

Stimson, James A. 2004. Tides of Consent: How Public Opinion Shapes American
Politics. Cambridge: Cambridge University Press.

Stokes, Donald E. 1962. “Party Loyalty and the Likelihood of Deviating Elections.”
Journal of Politics 24(4):689—702.

Thompson, M.M., M.P. Zanna and D.W. Griffin. 1995. Let’s Not Be Indifferent about
(Attitudinal) Ambivalence. In Attitude strength: Antecedents and consequences, ed.
Jon Krosnick. Erlbaum pp. 361—386.

Thurstone, Louis L. 1928. “Attitudes Can Be Measured.” American Journal of Soci-
ology 33(4):529—554.

Tourangeau, R., Kenneth Rasinki, Norman Bradburn and Roy D’Andrade. 1989.
“Belief Accessibility and Context Effects in Attitude Measurement.” Journal of
Experimental Social Psychology 25(5):401—421.

Vellerman, Paul F. and Leleand Wilkinson. 1993. “Nominal, Ordinal, Interval, and
Ratio Typologies Are Misleading.” The American Statistician 47(1):65—72.

Verba, Sidney and Norman H. Nie. 1972. Participation in America. Chicago: Uni-
versity of Chicago Press.

Wattenberg, Martin P. 1998. The Decline of American Political Parties, 1.9524996.
Cambridge: Harvard University Press.

Weisberg, Herbert F. 1980. “A multidimensional conceptualization of party identiﬁ-
cation.” Political Behavior 2(1):33—60

Weiss, Andrew A. 1993. “A Bivariate Ordered Probit Model with Truncation: Helmet
Use and Motorcycle Injuries.” Applied Statistics 42(3):487—499.

Wittenberg, Jason. 2008. “How Similar Are They? Rethinking Electoral Congru-
ence.” Presented at the 25th Annual Meeting of the Society for Political Method-

ology.

Young, Forest W. 1981. “Quantitative Analysis of Qualitative Data.” Psychometrika
46(4):357—388.

Young, Forest W. 1987. Theory. In Multidimensional Scaling, ed. Forest W. Young
and Robert M. Hamer. Hillsdale, NJ: Lawrence Erlbaum.

Young, Forest W., Jan De Leeuw and Toshio Takane. 1976. “Regression with Qual-
itative and Quantitative Variables: An Alternating Least Squares Method with
Optimal Scaling Features.” Psychometrika 41(4):505—529.

Zaller, John. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge
University Press.

138

Zaller, John and Stanley Feldman. 1992. “A Simple Theory of the Survey Response:
Answering Questions versus Revealing Preferences.” American Journal of Political
Science 36(3):579—616.

139