Professional Documents
Culture Documents
Goode Dissertation 8 2 2016
Goode Dissertation 8 2 2016
Goode Dissertation 8 2 2016
Dissertation
By
2016
Dissertation Committee:
Kathleen McGraw, Advisor
Paul Beck, Co-Advisor
William Minozzi
Michael Neblo
c Copyright by
Jess Goode
2016
Abstract
Can political elites use general frames to set the stage for political debate?
Most current political communication research focuses on the impact of
specific issue frames - but less on whether campaigns, parties and political
discourse as a whole can be framed. This dissertation presents and tests
a theory that elites use master frames, summaries of key ideological prin-
ciples, in order to make their subsequent campaign or policy arguments
more e↵ective. First, a content analysis of President Obama’s statements
leading up to the 2012 election provides evidence that he used a “fair-
ness” master frame. Second, results of a subsequent time series analysis
show that the president’s use of the fairness master frame was associated
with his reelection support. Third, the theory is tested in a series of three
experiments, two of which show master framing e↵ects. Experimental re-
sults from a nationally-representative sample show that exposure to an
equality master frame makes individuals more likely to support taxing
the wealthy and that exposure to an individualism master frame makes
subjects less supportive, but only for subjects who were also exposed to a
liberal tax frame. Results also show that exposure to an equality master
frame makes subjects support more liberal immigration framing. The first
two experiments find that exposure to individualism master frames polar-
ize views while equality master frames have more uniform e↵ects. The
third and final experiment failed to produce any master framing e↵ects.
ii
This is dedicated to the my wife, Jayne Goode, and my mother, Sharon Maloney
Kingan
iii
Acknowledgments
This project would not have been possible without the advice, insight and support
of many. I’m deeply indebted to more people than I can possible list here, but I
would like to acknowledge and thank my fellow graduate students, my advisors and
my family.
First, this dissertation was expertly guided and greatly strengthened by each mem-
and William Minozzi for several years of thoughtful advice and encouragement. And
I’m deeply thankful to my advisor, Paul Beck, for his unparalleled advice and sug-
Second, I had the great fortune to share my doctoral studies with an exception-
ally talented group of fellow students, many of whom o↵ered important suggestions
and constructive criticism. In particular, I would like to thank Paul DeBell, Jessy
Defenderfer, Nicholas Felts, Matt Hitt, Yalidy Matos and Katy Powers.
Finally, this would not have happened without the encouragement of friends and
family. I deeply value each of you and appreciate you more than I can say. In
particular, I want to thank my mother. None of what I’ve accomplished in life would
have been possible without your love, support and many sacrifices. Finally, but
certainly not least, I want to thank – from the bottom of my heart – my wife, Jayne,
and our two boys, Séamus and Isaiah. Because of you, I not only had the daily
iv
support needed to finish this dissertation, but I also woke up every day knowing I
v
Vita
Fields of Study
vi
Contents
Page
Abstract . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . ii
Dedication . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iii
Acknowledgments . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . iv
Vita . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vi
List of Tables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . x
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Research Motivation . . . . . . . . . . . . . . . . . . . . . . 1
1.2 Background and Literature Review . . . . . . . . . . . . . . . . . . 3
1.2.1 Framing . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.2 Issue Frames . . . . . . . . . . . . . . . . . . . . . . . . . . 4
1.2.3 Master Frames . . . . . . . . . . . . . . . . . . . . . . . . . 5
1.2.4 Master Frames and Stage Setting . . . . . . . . . . . . . . . 12
1.2.5 Master Frames and Priming . . . . . . . . . . . . . . . . . . 13
1.3 Model of the Master Frame E↵ect on Attitudes . . . . . . . . . . . 14
1.3.1 Master Frame Adjusted E↵ect . . . . . . . . . . . . . . . . . 17
1.4 Equality and Individualism as Master Frames . . . . . . . . . . . . 19
1.5 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5.1 Impact of Personality Types . . . . . . . . . . . . . . . . . . 22
1.6 Following Chapters . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
vii
2. Master Frames at the Macro Level . . . . . . . . . . . . . . . . . . . . . 27
2.1 Hypotheses . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
2.2 Methods . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.1 Content Analysis . . . . . . . . . . . . . . . . . . . . . . . . 32
2.2.2 E↵ect of Measurement Error . . . . . . . . . . . . . . . . . 37
2.2.3 Variables . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
2.2.4 Missing Data . . . . . . . . . . . . . . . . . . . . . . . . . . 42
2.2.5 Estimation . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
2.2.6 Interpreting Lags . . . . . . . . . . . . . . . . . . . . . . . . 44
2.3 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
2.3.1 Additional Analysis . . . . . . . . . . . . . . . . . . . . . . 61
2.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 65
3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 69
3.2 Experiment 1: Immigration Experiment . . . . . . . . . . . . . . . 69
3.2.1 Immigration Experiment Design . . . . . . . . . . . . . . . 69
3.2.2 Immigration Experiment Results . . . . . . . . . . . . . . . 70
3.2.3 Moderators of Master Frame E↵ect . . . . . . . . . . . . . 76
3.3 Experiment 2: Tax Policy . . . . . . . . . . . . . . . . . . . . . . . 80
3.3.1 Tax Experiment Design . . . . . . . . . . . . . . . . . . . . 80
3.3.2 Tax Experiment Results . . . . . . . . . . . . . . . . . . . . 81
3.3.3 Personal Disposition as Moderator of Master Frame E↵ects
in Tax Experiment . . . . . . . . . . . . . . . . . . . . . . . 88
3.4 Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 91
4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 93
4.2 Research Question and Hypotheses . . . . . . . . . . . . . . . . . . 93
4.3 Methods and Variables . . . . . . . . . . . . . . . . . . . . . . . . . 94
4.3.1 Equality Treatments . . . . . . . . . . . . . . . . . . . . . . 95
4.3.2 Individualism Treatments . . . . . . . . . . . . . . . . . . . 96
4.3.3 Issue Frame Treatments . . . . . . . . . . . . . . . . . . . . 97
4.3.4 Dependent Variable . . . . . . . . . . . . . . . . . . . . . . 97
4.3.5 Independent Variables . . . . . . . . . . . . . . . . . . . . . 97
4.4 Results . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 98
4.5 Conclusion and Discussion . . . . . . . . . . . . . . . . . . . . . . . 112
viii
5. Conclusion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 114
Appendices
Bibliography . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 127
ix
List of Tables
Table Page
3.2 OLS Estimated E↵ect of Master Frames on Support for Raising Taxes 84
4.1 E↵ect of Master Frames on Support for Raising Taxes on the Wealthy 100
x
4.2 E↵ect of Master Frames on Support for Taxing the Wealthy, Moderated
by Openness to Experience . . . . . . . . . . . . . . . . . . . . . . . . 104
4.3 CCES Tax Experiment: E↵ect of Master Frames on Support for Taxing
the Wealthy Among People 30 and under . . . . . . . . . . . . . . . . 110
xi
List of Figures
Figure Page
3.3 Mean Support for Raising Taxes on Wealthy, by Master Frame Condition 82
xii
3.5 Master Frame E↵ect on Support for Taxing the Wealthy, by Commu-
nity vs. Individualism Orientation . . . . . . . . . . . . . . . . . . . . 90
4.1 Neither Master Frame nor Issue Frame Treatments A↵ect Support for
Raising Taxes . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 101
4.2 Master Frame Treatments Have No E↵ect, even when Combining Lib-
eral and Conservative Treatment Groups . . . . . . . . . . . . . . . . 102
4.3 Equality Master Frame Produces More Support for Taxing Wealthy
Among People Who Are Not Open to Experience . . . . . . . . . . . 105
xiii
Chapter 1: Theory and Literature Review
1.1 Introduction
Prominent public opinion research asserts that the American electorate is neither
the same time, research finds that political elites are highly ideological and increas-
ingly polarized (Poole & Rosenthal 1984, Poole & Rosenthal 2007). But, of course,
elites cannot achieve electoral goals without building majority or plurality electoral
coalitions. This poses a puzzle. How do ideological elites engage and organize a
public that is not highly ideological? This dissertation proposes a novel solution:
elites use master frames – general summaries of key ideological principles – in order
This subject is important for a few reasons. First, since the health of a democracy
is rooted in whether the government genuinely responds to the wishes of its citizens,
it’s important to explore how political elites influence public opinion change.
1
Is it possible for elite rhetoric to set the stage for future political discourse, tilting
the odds in individual campaign or policy disputes? If master frames do not have the
expected e↵ect, then some scholars may be more confident that mass opinion change
is rooted in the true desires and wishes of citizens. In contrast, if political elites
can subtly influence the pool of political considerations that citizens use to evaluate
specific policy and campaign arguments, those same scholars may be concerned that
the government will be more responsive to elite rather than mass public desires.
I advocate a more positive view of master framing. Master framing allows political
elites to communicate with the mass public while avoiding unnecessary technical
language and policy detail. In large part, my disagreement with the former view is
as analogous to a collection of issue positions (Converse 1964). But that need not be
the case for citizens to have sensible influence on the democratic process. Instead, we
could see ideological sophistication as a dimension. On one end are individuals with a
On the other end are individuals who connect those and other ideological principles
to hundreds of specific public policy positions. In between are those who connect
general ideological principles to individual issue positions that matter most to them.
Because master frames are rooted in fundamental ideological principles, they allow
those on the more “sophisticated” end of the ideological spectrum to have meaningful
political dialog with those on the other end of the ideological spectrum. In this way,
master frames act as a rhetorical bridge to help regular citizens connect their deeply
2
Second, much political rhetoric is devoid of detailed policy content. From hagiog-
raphy to character assassination, it’s important for scholars to understand how such
Finally, as mentioned earlier, recent framing research finds that some frames are
naturally more persuasive than others (Chong & Druckman 2007a). Master framing
provides an explanation for how political elites use general ideological rhetoric to
1.2.1 Framing
individual or media outlet “defines the essential problem underlying a particular social
issue.” Other prominent definitions (Druckman 2004, Chong & Druckman 2007b,
Entman 1993, Gamson 1989) similarly emphasize that frames characterize how issues
are viewed by making some aspects of issues more relevant than others. Frames
the issue’s features and, just as important, ignoring other features. As a result, as
Entman (1993) argues, frames not only identify important elements of an issue, but
also a↵ect the weight of those elements. In other words, frames not only abridge one’s
view of the world, they help one understand what’s important when considering the
world. Not only can frames influence attitudes, it’s also easy to see why they might
be attractive to citizens. Political matters are complex, and experts and novices alike
often need help narrowing down the possible ways of thinking about them.
3
Framing e↵ects are numerous in political science, communication and social psy-
chology. Experimental evidence has demonstrated that framing news stories in terms
of strategy rather than policy increases political cynicism (de Vreese 2004), that
group frames increase the e↵ect of subjects’ own group attitudes on related policy
positions (Nelson & Kinder 1996), that partisan voters are influenced by “conflict”
frames when they come from fellow partisans (Slothuus & de Vreese 2010), and even
that slight changes in language can significantly influence whether conservatives be-
Despite the profusion of framing scholarship, almost all of it focuses on how the
news media frames public policy issues. In contrast, this dissertation will explore how
elites use general frames to shape electoral campaigns as well as political discourse.
While widely used, the framing concept means slightly di↵erent things to di↵erent
scholarly communities. For instance, Druckman (2004) argues that there is an impor-
tant distinction between issue frames studied in political science and communication
framing, which demonstrated that presenting information in terms of loss or gain can
make people favor one of two mathematically equivalent options. In contrast, po-
litical scientists tend to focus on what Druckman (2004) calls “issue frames,” policy
For example, raising taxes can be framed as positive for the economy because
it could allow the government to buy goods and services in the private sector. On
4
the other hand, raising taxes might be bad for the economy because it could reduce
individual income and limit consumer spending. Clearly, these arguments are not
the wisdom of raising taxes. Since the 1980s, many influential studies have shown
that issue framing can significantly change political and policy attitudes (Druckman
One important feature of issue frames is that they emphasize the stakes of specific
policy issues, such as whether to increase taxes on the wealthy or to o↵er amnesty to
arguments for or against specific policy questions. In contrast, master frames work
by invoking values or principles that imply the “correct” position on specific issues.
This paper focuses on the general frames that elites use: master frames. Although
acknowledged that this dissertation is certainly not the first work to study general
showed that thematic frames help members of the public hold government officials
accountable.
Likewise, de Vreese, Peter & Semetko (2001) argue that there’s an important
di↵erence between issue frames and what they call “generic” frames, such as those that
emphasize political conflict. While such work is informative, these studies concentrate
on news media-driven framing, not on how politicians, party leaders and their sta↵
5
Some scholars argue that the term “frame” should only be applied to issue or
equivalency frames (Chong & Druckman 2007b). Chong & Druckman (2007b) in-
stead favor Entman’s (1993) use of the term “script” for general frames. I disagree
for the following reasons. First, general frames still may influence the weight of con-
issue. Arguments against “big government” are vague and general, yet many Amer-
icans can apply those arguments to a number of related policy issues. Second, even
a cursory review of campaign coverage reveals that candidate rhetoric is often vague
and general but aims to a↵ect political opinion regarding specific candidates and is-
sues. When President Reagan (1981) proclaimed “government is not the solution to
argument against large government programs, which was clearly intended to under-
mine support for specific policies. In fact, directly preceding this famous phrase,
Reagan briefly mentioned no fewer than six issues: inflation, unemployment, taxes,
government spending as well as both public and private debt. Finally, using the term
“script” would obscure the common-language clarity of the term “frame.” Academics
and non-academics alike can easily grasp that “framing” something is to see a par-
ticular matter or issue in a certain way, most likely because almost everyone has had
Most important, a general framing concept already exists within sociology which,
though it is primarily used to explain how social movements organize around common
themes, could also apply to political elites and their campaigns. For the master
frame concept, I build on Snow & Benford’s (1992) social movement master frame
theory, which asserts that these general frames allow social movements to organize
6
and cohere. Specifically, they argue that master frames perform three functions: (1)
identifying the causes of (and assigning blame for) social problems (2) organizing
associated concepts and ideas and, relying on the first two functions, (3) facilitating
mass mobilization.
Entman (1993) defined frames as points of view that have moral claims, causal
master frames as having the following features: (1) They are general frames for eval-
uating the political world, relying on just a few principles, values or metaphors to
structure those evaluations. (2) These evaluations lead to causal stories about how
the political-social world does and/or should work. (3) Those causal stories result in -
or influence - views that are then used to judge political issues, events and candidates.
In short, master frames are the messages that political elites use to communicate their
that organize them may be the same as those that structure ideologies, master frames
are distinct from ideologies because they are merely abbreviated versions of the values
To summarize, master frames are general frames that organize how individuals in-
terpret and evaluate the political world, determine what causes events and allow them
to judge issues, events and candidates as a result. If ideologies are collections of many
interconnected attitudes, master frames are general frames that strengthen multiple
organized around the principle of rational self-interest. This master frame asserts that
7
the U.S. – both economically and politically – would be more prosperous if individuals
and corporations were free to seek their own self-interest, and it has implications at
the individual, economic and political levels. At the individual level, it implies that
people are and should be personally responsible for their lot in life; individual success
is rooted in individual e↵ort and individual loss is the result of one’s failure to work
hard. Following this logic, the government should not provide for a robust social
safety net because to do so would be to reward certain individuals for the failure
to take responsibility for themselves. Further, this master frame implies that the
economy will grow when economic actors are free to seek their own self-enrichment.
And, finally, this master frame means that government must be as limited as possible,
that it must get out of the way of individuals and companies as they seek their own
ends.
By communicating this master frame, conservatives may have structured the po-
litical discourse in a way that sets the agenda in favor of conservative policies and
eral issue arguments (frames), increasing receptivity to their ideology in general, and
causing the public to act in quasi-ideological ways, especially at the aggregate level. If
my theory is correct, I would expect that liberal periods, (such as the New Deal Era),
were also influenced by liberal master frames that structured the political discourse
then.
alism makes America strong might make voters more responsive to arguments for
8
lowering taxes and reducing government regulation. Likewise, emphasizing the im-
portance of equality might make voters more open to arguments for increasing Social
Security and education funding. If the public receives these general master frames on
a regular basis so that the associated concepts are chronically accessible, then a host
America Great Again” meets the definition of a master frame. It’s a general frame
that invokes (nationalist and nativist) ideological principles and which is clearly meant
to set the stage for specific policy initiatives, such as building a wall between the U.S.
and Mexico.
American political history is defined by periods when one party not only holds
more power in government, but in which its programmatic and ideological agenda also
dominates the political discourse. This paper seeks to determine whether defining the
political world through their use of appealing master frames allows parties to dominate
the political discourse by setting the terms of the political debate and automatically
ordinary people described their support for candidates and parties. And he showed
that only a small percentage of the public (under ten percent) exhibited belief sys-
tems in which attitudes on one issue predicted attitudes on other issues as well as.
Similarly, Zaller (1992) reported that survey respondents tended to state opinions
based on the salience of many inconsistent considerations rather than use anything
resembling an ideology.
9
While most individual Americans are not ideologically consistent or constrained,
in di↵erent periods. Further, he and colleagues (Stimson, MacKuen & Erikson 1995)
demonstrate that these changes in mood precede partisan electoral victories for the
party associated with the dominant ideological mood. Similarly, Page & Shapiro
(1992) shows that collective opinion is generally coherent, stable and responds to
events.
One explanation for public mood swings is that individual voters are simply re-
sponding to specific life circumstances, such as job loss, and that idiosyncratic indi-
vidual opinion swings add up to changes in national political mood. Yet individuals
may also respond to elite messages. In particular, political mood changes might result
from political master frames communicated by ideological political elites. Given that
individual members of the public are rarely ideological, elites may have the opportu-
nity to influence the ideological mood of the country or a political campaign through
the master frames they use. The advantages of doing so are clear. Creating a more
friendly climate would make it easier to win elections and to pass important legisla-
tion. As Aldrich (1995) argues, structuring political parties as “long coalitions” o↵ers
politicians a way to escape collective action problems. Master frames may provide
Political elites want more than to simply win a majority of the public to their side
on particular issues, such as social security or defense or flood mitigation. They want
want supporters that will be there for their party and the party’s priorities over time
10
and across multiple issue debates. And, because time and money and organizational
capability are all limited, elites should want to build these long-term coalitions as
frames can strongly influence how individuals think about specific issues (Iyengar &
Kinder 1987, Nelson, Oxley & Clawson 1997, Iyengar 1994). (For a comprehensive
review, see Chong & Druckman (2007b)). But this e↵ect can be mitigated by exposure
2004). My theory posits that political elites are aware of both the potential power of
framing, but also its limitations. Master frames, by being grounded in fundamental
moral values or ideologies o↵er a way of strengthening issue frames that are naturally
Elites cannot expect most Americans to adopt full ideologies and related policy
constraints. Elites have neither the time, nor money, nor organizational capacity
needed to sufficiently educate each member of the voting public to act in ideologically
consistent and coherent ways. Neither do most members of the public have the time or
political interest to develop extensive ideologies, and they can act reasonably without
them. What elites can do, however, is structure their communications in a way so
that the public acts as if they were ideological on many issues. Therefore, this theory
predicts that as a particular master frame increases its proportion of the political
discourse, the public will move in the associated ideological direction and public
opinion on particular policy issues will move in the ideological direction of the master
frame, regardless of whether the master frame mentions those policy issues.
11
1.2.4 Master Frames and Stage Setting
policy arguments more persuasive by “setting the stage” of political debate. Like
a set design, master frames arrange the elements of political scenes, moving some
values and themes into the foreground while relegating others to the background (or
even o↵-stage). While similar to agenda setting, (McCombs & Shaw 1972), stage-
salient, not specific policy issues. At the individual level, following Zaller (1992), I
expect that master frames will change the proportion of general liberal or conservative
considerations, which will in turn change the weight one gives to a view regarding a
political issue.
At the mass public level, widely-embraced master frames structure political de-
ing policies, political figures or office holders. For example, if one frequently receives a
master frame arguing that equality leads to better government and a fairer economic
system, then equality will become more influential as that person evaluates policy or
campaign rhetoric.
ual issue frames, Chong & Druckman (2007a) found that “strong” frames are more
compelling and have a greater impact on attitudes than “weak” frames. In a recent
review, they argue that “Strong frames rest on symbol, endorsements and links to
partisanship and ideology. . .” (Chong & Druckman 2007b). Master frames explain
how communicators use general political or ideological language to make some issue
frames stronger and others weaker. In other words, master frames form a connective
12
tissue between issue frames and political attitudes, allowing politically unengaged
One might object that the master frame concept is not really di↵erent from prim-
ing. This is not true for several reasons. Indeed, Druckman, Kuklinski & Sigelman
(2009) argue that political scientists have not defined the priming process in way
that di↵ers from framing. Political scientists generally define priming as a process
that raises the cognitive accessibility of mental constructs (Kinder 1998, Krosnick
& Kinder 1990, Iyengar & Kinder 1987), resulting in political judgments based on
to form political opinions. But the di↵erence between accessibility, (at least as it is
used in political science), and weight is illusory. A mental construct can’t increase
in weight without also being more accessible. Indeed, in reviewing both literatures,
Chong & Druckman (2007b) conclude that the two processes are the same. A more
theoretically clear approach to priming is the one used by Lodge & Taber (2013) in
which primes are conceived as stimuli registered at the unconscious level whose e↵ects
from framing because it increases the weight of mental constructs at the unconscious
level and framing continues to define processes that include conscious political de-
way to describe how elites shape political discourse, not only because the term is
13
already widely used within sociology, but also because it identifies e↵ects that take
What is the formal process through which master frames impact political atti-
tudes? The following diagram makes the causal process in my theory explicit. To
do so it builds upon Nelson, Oxley & Clawson’s (1997) summative attitude model
of framing e↵ects, which itself derives from the expectancy value model of attitude
The summative attitude model assumes that each attitude is the weighted sum
of views regarding di↵erent aspects of an object being evaluated. The most widely-
used model of framing (Nelson, Oxley & Clawson 1997, Chong & Druckman 2007b)
proposes that frames a↵ect attitudes by increasing the weight that particular views
have in an individual’s overall attitude toward an issue. For instance, a person might
have a variety of views that are relevant to the attitude of raising taxes on the wealthy.
On the one hand, they may believe that raising taxes will harm the economy by
limiting investment. On the other hand, they may believe that raising taxes would
be fairer for the middle class. The weight or importance that one gives to these
and other relevant views will dictate their final attitude toward raising taxes on the
wealthy.
Importantly, according to this model, individuals start with the original weight
that they give to di↵erent opinions. I assume that this original weight - or “old”
Master frames influence the underlying collection of considerations that one has to
14
Figure 1.1: Model of Master Framing E↵ect
Figure 1 shows the attitude formation process as proposed. First, individuals start with
the original weights that they give to views relating to a particular attitude. I assume
these weights are generated by an individual’s mix of political considerations. Issue frames
moderate one’s original weights by increasing weight for framed views and decreasing the
weight of other views. Master frames moderate the impact of issue frames, strengthening
their impact if ideologically congruent and weakening it otherwise. This results in new
weights which were generated by encountering an issue frame in the context of a master
frame. The master frame-adjusted weights then increase the influence of impacted views
when individuals form an attitude regarding a political question.
draw from, which in turn changes the weight that individuals apply to particular
15
As mentioned earlier, Nelson, Oxley & Clawson’s (1997) model defines attitudes
as the weighted sum of an individual’s relevant opinions and how much weight is
given to each of those opinions. While framing need not always work by increasing
the weight of particular views,(for instance, frames may also add new information,
see: (Chong & Druckman 2007b), this is the most commonly used conception in
political science. In this section, I update the summative attitude model to include:
(1) the impact of issue frames and then (2) the impact of master frames. I begin
with the original summative attitude model as proposed by Nelson, Oxley & Clawson
(1997):
X
Aj = (vi )(wi ) (1.1)
candidate or political institution), vi is the view regarding one aspect of the object
being evaluated and wi is the weight or influence that each vi has in the summative
Since issue frames influence opinions by increasing the weight a particular view or
X
Aj = (vi )(winew ) (1.2)
Where A and vi are the same as in equation 1.1 and wi new is a frame-adjusted
weight variable. Again, wi new sums to 1. It is calculated by increasing the weight for
those vi which are exposed to frames while simultaneously reducing the weight of vi
that don’t receive frames, (or for which a particular frame has no impact).
16
More specifically, wi new is defined by the following:
wiold + fi
F (wiold , fi ) = P (1.3)
1 + fi
Where fi 0 and measures the impact of issue frames regarding each vi . The
denominator of equation (1.3) ensures that the resulting frame-adjusted weight sums
to 1, reducing the weights not impacted by the issue frame. For instance, if the
attitude reflects whether to raise taxes on households above $250,000 and v1 is the
positive view that doing so will spur economic activity through government stimulus
spending, an issue frame f1 persuasively making such an argument will not only
increase the weight of v1 but will also reduce the weight of other views one might
In order to capture the e↵ect of master frames, equations 1.1 and 1.3 need to be
original weight someone brings to a view– as the issue frame impacts it. In this
conception, wi represents the weight one would apply to a particular view as a result
of elite ideologies and because I propose that they are frequently repeated in normal
political debate they strengthen or weaken the impact that issue frames have on the
X
A= (vi )(wiMF ) (1.4)
17
Where A and vi are the same as in equation 1.2 and wi MF is the master frame and
wiold + fi mi
F (wiold fi mi ) = P (1.5)
1 + fi mi
Where mi 0 and is the impact of receiving a master frame, which in turn either
magnifies or reduces the e↵ect of an individual issue frame. When a master frame is
congruent with a particular issue frame, the power of that issue frame compounds,
increasing the impact of the view being advocated and moving an individual’s attitude
in the ideological direction consistent with the issue frame. At the same time, if a
master frame is ideologically incongruent with a particular issue frame, then the
the wealthy. In this example, the person has two views relevant to their attitude:
v1 is the positive (with respect to the attitude) view that raising taxes would boost
the economy through stimulus spending and v2 is the negative view that raising such
taxes would depress the economy by limiting consumer spending. If the individual
gives equal weight to both views, their attitude is: ATAX = 3(.5) + 3(.5) = 0. In this
case, the individual’s final attitude is neutral, a classical example of conflicting views
producing an ambivalent attitude. But what if this person hears an argument that
changes their reliance on one of those views? If they receive a compelling issue frame
that emphasizes the positive impact of raising taxes through more stimulus spending,
the attitude can calculated again using equation (1.2). If the issue frame has a
.5+2 .5+0
positive impact of 2, the person’s updated attitude is: ATAX = 3 1+2+0 + 3 1+2+0 =
3(.83̄)+ 3(.16̄) = 2.49 .48 = 2.01.The issue frame e↵ectively moved the individual’s
18
attitude from a neutral position (0) to a moderately supportive position (2.01). But
the individual’s support is still not as strong as it could be given their underlying
allows one to calculate the impact of an equality master frame paired with a liberal
issue frame on raising taxes. If the equality master frame has a positive impact of 10,
.5+(2·10) .5+0
the individual’s attitude is: ATAX = 3 1+(2·10)+0 + 3 1+(2·10)+0 = 3(.98) + 3(.02) =
2.94 .07 = 2.87. In this case, both the issue frame and the master frame substantively
impacted the individual’s attitude toward taxing the wealthy, moving them from
Given that both issue frames and master frames can change attitudes, why would
political elites bother to communicate master frames in addition to issue frames? Be-
reap benefit at a comparatively lower cost of time and money. And, because they are
general, master frames are not hard to incorporate into di↵erent rhetorical contexts.
By definition, issue frames are only appropriate when addressing a particular issue.
While some issue frames may directly appeal to certain values or political prin-
ciples, the master frame is distinct because it implies “correct” positions on a range
individual issues. This means that, in order to be e↵ective, master frames must con-
nect issues with “folk” ideologies. The most e↵ective master frames are likely to be
19
based on values already in the common culture. In America, that likely means either
As the world’s first liberal democracy, some argue that the contours of American
political debate have always been defined by the tension between liberty and equality,
motivated by equity. McClosky & Zaller (1984) found that these values charac-
terize part of the American ethos that defines the common culture. More recently,
Kellstedt (2000) demonstrated that news media frames using equality or individu-
alism influenced public support for federal e↵orts to combat racial discrimination.
free-market economics and because each are so commonly used in American political
1.5 Hypotheses
The foregoing specifies not only how issue frames can increase view weights, but
how those issue frame e↵ects are influenced by master frames. If master frames do
impact public opinion, both individual policy attitudes and issue frame agreement
should be a↵ected. More precisely, when master frames and issue frames are ideolog-
ically consistent they should move attitudes in their ideological direction. Assuming
that master frames are more influential because they are based on widely-understood
following hypotheses:
20
Master Frame Stage-setting Hypothesis: Experimental groups exposed to a master
frame will show stronger support for ideologically-associated issue frames and less
will be more supportive of liberal issue frames and less supportive of conservative issue
frames. Likewise, it predicts that people exposed to the individualism treatment will
be more supportive of conservative issue frames but less supportive of liberal issue
frames. If issue frame agreement is influenced by master frames, then that provides
evidence that public opinion and political discourse can be changed by appealing to
influenced by master frames, then that suggests that public opinion and political dis-
course is mostly a function of those individual issue frames, (as well as environmental
posed to a master frame and an ideologically-congruent issue frame will show stronger
support for the policy advocated by the policy frame and less support for policies op-
This hypothesis predicts that people exposed to equality master frames and liberal
policy arguments will adopt more liberal policy views and that those exposed to indi-
vidualism master frames and conservative policy frames will adopt more conservative
policy views. Similarly, this hypothesis predicts that when issue frames and master
frames are incongruous, the master frame will be more influential. In these cases,
an equality master frame paired with a conservative issue frame would result in a
more liberal attitude, (compared to those receiving only a conservative issue frame).
21
Likewise, an individualism master frame paired with liberal issue frame should result
Because communication e↵ects emanate from sources outside the receiver, they
ski & Sulloway 2003, Nyhan & Reifler 2010, Lodge & Taber 2013). Though social
for human behavior, robust support for both suggest that scholars will find lever-
age by studying the interaction between the two. Lewin (2013) famously proposed
As McGraw (2006) argues, research which aims to explain human behavior should
take into account the interaction between the individual and social context. Indeed,
early persuasion models posited that e↵ective persuasion was contingent on one’s pre-
dispositions along with other factors, such as mode, communicator credibility, and,
of course, the quality of the message itself (Hovland, Janis & Kelley 1982).
and master frames) and personal predispositions, what predispositions should be con-
sidered? Jost’s (2006) work linking Big Five Personality traits to ideology suggests
predispositions. Given his consistent finding that conservatives are low in openness
22
to experience and while liberals are more open, this trait, in particular, should im-
pact how people respond to political messages. As a result, I will test the impact of
tant for three reasons: first, master frames could make supporters more loyal to the
ideological side associated with their personality disposition. Second, master frames
might serve to inoculate mass public members against the policy arguments of the
other side. And, as a result, such a moderating impact from personality would help
will cause high openness people to be more responsive to liberal policy frames, whereas
self-directed master frames (individualism) will cause low openness people to be more
produce a backlash among low openness people, making them take more conservative
Figure 1.2 shows experimental expectations under conditions in which the mas-
ter frames have no impact and also when master frames have their predicted e↵ect
under that attitude change hypothesis. If master frames have no impact, people in
the conservative issue frame condition will show less support for the liberal policy
position, but that lower support will be the same for all master frame and control
conditions. Likewise, exposure to a liberal issue frame will increase support for liberal
views, but that higher support level will be the same for all master frame conditions.
However, if master frames have their predicted impact, the pattern in the right panel
23
will be observed. For instance, equality conditions will show the most support for
raising taxes, control (no master frame) conditions will show middle support, and
individualism conditions show the least support. The gray line represents people who
also receive a liberal tax frame, (which should also produce more support for rais-
ing taxes) and the black line shows people who also receive a conservative tax frame
The following chapters test the master frame theory in three di↵erent sets of
studies.
First, chapter two reports results of a content analysis of President Barack Obama’s
speeches and statements leading up to the 2012 election. In the chapter I measure
and test the e↵ect of the president’s fairness-based master frame using machine-coded
content analysis and time series regression. By coding and analyzing speeches and
remarks from 2011 through the 2012 election, I hope to show first that Obama used
a compelling master frame in his reelection campaign. Second, I estimate the daily
association that master frame message had with his support as well as support for
Governor Mitt Romney. Finally, I test the moderating impact of Obama’s master
frame usage on the e↵ect of Democratic Party identification on Obama and Romney
support.
By measuring Obama’s master frame usage and its e↵ect in the 2012 presidential
election campaign, I hope to show that not only do politicians’ regularly rely on
such generally messaging but that those general messages have real and measurable
24
Chapter three reports results of experiments that test master frame e↵ects at the
individual level. Chapter four attempts to replicate chapter three’s results as well as
25
Figure 1.2: Experimental Expectations for Attitude Formation Hypothesis
Figure 1.2 displays expected results if master frame treatments have their predicted e↵ect
under the attitude formation hypotheses. The left panel shows null findings: flat lines for
tax increase support across individualism, equality and control (no master frame) condi-
tions. This means that tax attitudes change only as a result of liberal or conservative tax
issue frames and not as result of equality or individualism master frames. The right panel
shows expectations if master frame treatments are e↵ective. In this case, equality conditions
show the most support for raising taxes, control (no master frame) conditions show middle
support, and individualism conditions show the least support for raising taxes. The gray
line represents people who also receive a liberal tax frame , (which should also produce more
support for raising taxes) and the black line shows people who also receive a conservative
tax frame, (which should produce less support for raising taxes).
26
Chapter 2: Master Frames at the Macro Level
In September 2011, the Obama White House scrambled to recover from a series
losses in the 2010 midterm elections, Obama and his budget negotiators spent most
of the spring and summer trying to get congressional agreement to raise the debt
ceiling, a budget technicality that had rarely generated partisan acrimony in the
past. By early August, Congress had agreed on a bill to expand the debt ceiling
and reduce the deficit. But in the aftermath opponents saw the president as a failed
leader (Adler 2011) and supporters saw him as perilously o↵-message (Tankersly
2011). Between May 2011 and August 2011 his disapproval rating rose from 45 to 52
percent (Huffington Post Pollster 2016a) and, despite the budget agreement, his job
According to his memoir, former senior White House advisor David Axelrod
watched these events with increasing anxiety (Axelrod 2015). In his view, the pres-
ident risked losing the voters who had elected him in 2008. After a late August dis-
cussion with the president, Axelrod sent him a strategic memo diagnosing Obama’s
electoral problems and outlining a message to get back on track (Axelrod 2015). He
argued that Obama needed to return to the message of fairness and collective action
27
Your message ultimately carried for two reasons: One is that it was rooted
in the idea that we are a better and stronger country when everyone gets a
fair shot and a fair shake; when hard work and responsibility is rewarded,
and all of us are accountable. The second is that it recognized that there
are things we must do together as a country to ensure a brighter future.
(Axelrod 2015)
Axelrod’s insight formed the basis of the president’s economic address to Congress
later that September. It was in this address that the president first used the following
cans] believed in a country where everyone gets a fair shake and does their fair share
– where if you stepped up, did your job, and were loyal to your company, that loy-
alty would be rewarded with a decent salary and good benefits; maybe a raise once
in a while” (Obama 2011a). This sentence perfectly captures what I have defined
as a master frame in earlier sections of this dissertation: a general frame that pro-
debate. Shortly after this speech, the President began using the following refinement
of the same master frame: “I believe that this country succeeds when everyone gets
a fair shot, when everyone does their fair share, when everyone plays by the same
rules” (Obama 2011b). A Google search shows that slight variations of the latter
sentence were reported more than 500,000 times since September, 2011.
Figure 2.1: Google Search Results for Obama Master Frame Sentence
28
Axelrod’s fairness master frame was not chosen carelessly. As journalist Wol↵e
(2013) recounts in his book on the 2012 Obama campaign, after leaving the White
House pressure cooker, Axelrod began refining a message that acknowledged middle
class Americans’ post-recession struggle: they were not getting a “fair shake” while
As Wol↵e writes:
The fair shake and the fair share: it was a classic Axelrod turn of phrase.
A piece of writing that encapsulated a world-view more than a policy
prescription for hard economic times. To the right, it sounded faintly
socialist with its insistence on fairness. To the left, it sounded like justice.
And to the middle-class voters who would decide the election, it sounded
like their plight at a time when wages were getting pushed down and costs
were getting pushed up. (Wol↵e 2013)
message intended to frame and strengthen the campaign’s subsequent, more specific,
policy and political arguments. Going forward, I will use the term master frame
frames, Axelrod’s master frame formed the core reelection pitch. It was intended to
boost support for the president by strengthening support for his ideological beliefs:
equality, fairness (particularly economic fairness) and collective action through gov-
ernment intervention. If successful, the fairness master frame would strengthen his
electoral support by solidifying the liberal coalition and giving it a compelling general
Obama’s campaign master frame provides an opportunity to test the master frame
theory using real-world data. Though observational studies rarely provide the causal
29
identification found in experiments, they do make it possible to test a theory’s ecologi-
cal validity. Taken together with positive experimental results, observational evidence
of master frames would lend credence to the contention that master frames not only
a↵ect certain special experimental outcomes but that they also have a measurable
Though the Obama master frame did not focus solely on one specific ideological
principle, it did fit well under the general rubric of fairness. And because of its
similarity to the master frame tested in experiments in chapters one and two, Obama’s
campaign setting.
2.1 Hypotheses
The fairness master frame contained three related but distinct principles. First,
it argued that America is strengthened when everyone is treated fairly and equally.
Second, it contended that citizens should be rewarded for their hard work and ac-
countable when they don’t work hard. Finally, the message maintained that many
three and four will test the e↵ect of master frames on subsequent policy frames
on taxes and immigration debates. This chapter will test whether master frames
30
generated by strengthening the loyalty and support from a candidate’s ideologically-
In this chapter, I test the following hypotheses, which are modified versions of the
Hypothesis One: Increased Obama fairness master frame usage will be associated
If Obama’s campaign team was right about how to appeal to disa↵ected voters, the
fairness master frame should be associated with increased support for the president’s
reelection bid.
Hypothesis Two: Increased Obama fairness master frame usage will precede higher
increasing the number of people who identify with the party ideologically-associated
with the master frame. In this case, the fairness master frame is a distillation of liberal
as a result.
Hypothesis Three: Increased Obama fairness master frame usage will be associated
with a stronger relationship between Democratic Party identification and support for
Obama.
All things being equal, one would expect that increasing percentages of Demo-
cratic identification would be associated with more support for President Obama.
Because master frames distill and communicate general ideological principles, they
31
a candidate’s master frame should increase her support by consolidating and intensi-
fying support among members of the candidate’s political party. In this case, I expect
that higher Obama fairness rhetoric will be associated with a larger positive impact
Hypothesis Four : Increased Obama fairness master frame usage will not increase
support for Romney nor will it a↵ect Democratic support for Romney.
Master frames should only increase support for ideologically-aligned policies and
candidates.
2.2 Methods
Obama fairness rhetoric as well as time series of party identification, Obama support
and Romney support. Because of the sheer volume of the President’s speeches, I used
text classification algorithms to code Obama speeches for fairness master frame sen-
tences, constructing a time series of daily Obama fairness rhetoric and then estimating
The content used in this analysis was a collection of published remarks on the
White House “Speeches and Remarks” web page (The White House 2015). Content
was web scraped with the assistance of the rvest package for R (Wickham 2015).
In addition to public speeches, the “Speeches and Remarks” web page also included
formal press conferences and presidential debates. After scraping content, text was
parsed by speaker to ensure that only speeches or statements by President Obama was
used for analysis. Finally, the collected text was separated into more than 108,000
32
individual sentences. Only content from September 2011 through November 2012 was
used for the time series analysis, a time period in which Obama introduced and used
Sentences were used as the unit of analysis because they were more likely to
analysis might have been done with a dictionary method by counting word frequencies
that approach allowed a more precise identification of the meaning of words in their
approach would not have been possible for one researcher without computer-aided
text classification.
Following is the procedure used to classify the corpus of Obama’s speeches and
statements. A random sample of sentences were first hand coded into two categories:
“Fairness” if they represented one or more of the three elements of the Obama fairness
master frame and “Non-Fairness” otherwise. The hand-coded data was then used to
The following coding criteria was used to classify each hand-coded sentence and
3. Arguments for rewarding hard work as well as holding people accountable for
33
that made explicit or implicit arguments in favor of more equitable distribution
Rather than using a dictionary method or looking for key words, I read each
sentence in the sample used for hand coding. I then classified a sentence as “fairness”
if it matched at least one of the themes identified above and “non-fairness” if did
not meet those criteria. It’s important to note that some “fairness” sentences also
contained, usually brief, mentions of policy. Though I have conceived of master frames
as policy-free, general frames, the rambling nature of campaign speeches meant that
such a clear delineation wasn’t possible in this content analysis. As a result, I will not
investigate the e↵ect the president’s fairness rhetoric had on support for issue frames
and instead will focus on the e↵ect it had on his political support.
After hand coding more than 1,800 sentences, the hand-coded data was used
to train text classification algorithms to automatically code the full sentence dataset.
The RTextTools package was used to train algorithms and classify the full dataset (Jurka,
Collingwood, Boydstun, Grossman & van Atteveldt 2012). The package implemented
several supervised machine learning text classification algorithms which “learn” from
human coded sentences and attempt to apply the same human-coded criteria to un-
classified sentences. I trained and combined classification results from the following
tion, boosting, bagging, Random Forests, and decision trees. Although each algorithm
provided a unique classification for each sentence, I followed common practice and
Wilkerson (2012) found that classification accuracy was greatest when four or more
34
algorithms agreed on a particular classification. Following their advice as well as the
recommendation of Jurka et al. (2012), the final “fairness” or “not fairness” classifi-
The specific procedure used for training and classification is as follows. A total of
1,864 sentences were hand coded. Of these, 1,264 hand-coded sentences we used to
train the algorithms, 300 hand-coded sentences were held back to test the accuracy
of the trained algorithms and an additional 300 sentences were held back to validate
the accuracy of the test data set. Because the proportion of “fairness” sentences
was lower than the non-fairness sentences in the hand-coded dataset, more fairness
sentences than average were included in the training set and were oversampled to
increase the proportion of fairness sentences used for training the algorithms.
After training and choosing the four ensemble agreement method, results were
verified using the test and validation datasets. The test data set showed 80.3 per-
cent agreement with hand-codes for the same sentences. Furthermore, agreement
between the machine coded sentences and hand-coded sentences produced a Cohen’s
Kappa score of .57 in the test data set. Cohen’s Kappa is a conservative inter-rater
reliability measure, which takes into account the probability of coding agreement by
chance (Landis & Koch 1977). There are no set rules for when Cohen’s Kappa should
what would be expected by chance. Landis & Koch (1977) suggest that values between
.40 and .60 be considered “moderate” and values between .61 and .80 be considered
“substantial.” The test data set’s Cohen’s Kappa .57 was on the very high end of
tion, I validated the text classification on a second dataset and in that dataset there
35
was 80 percent agreement between the machine codes and my manual codes. The
Cohen’s Kappa score for the validation data set was .56. The fact that the automatic
text classifications so closely matched my hand-coded data in two separate data sets
that were not used to train the algorithms and that they perform well even with
the more conservative Cohen’s Kappa provides confidence that time series calculated
from this data were an accurate measure of changes in Obama’s fairness rhetoric
percentage of machine classifications that are accurate and recall is defined as the
percentage of human coded documents that are correctly identified by the machine
In this case, precision was the percentage of fairness-classified sentences that were
the test dataset, precision was .71, meaning that 71 percent of sentences machine-
classified as fairness were actual fairness sentences. This means that 29 percent of
machine-classified fairness sentences were not actual fairness sentences. Recall was
.73, meaning that 73 percent of fairness sentences were accurately identified. However,
this also means that roughly 27 percent of fairness sentences were missed by the
machine classification. Taken together, the two results provide strong evidence that
classified sentences truly capture the president’s fairness message and that the vast
36
2.2.2 E↵ect of Measurement Error
Unfortunately, the validation process made clear that there was at least some
dependent variable measurement error does not bias OLS estimates, it’s been es-
tablished that independent variable measurement error biases its estimate toward
zero (Greene 2008a). Though not ideal, such attenuation likely provided a more
conservative test for the e↵ect of Obama’s fairness master frame on Obama vote sup-
port. If variables originating from machine-coded data were positive – even after
downward bias– one could reasonably assume those coefficients would have been even
In this case, it was possible to check potential machine-coded bias in the key
independent variable by using the 600 sentences that were hand coded but reserved
for testing and validation. In this small dataset, the hand-coded fairness proportion
of Obama daily speech and daily Obama vote support were correlated at .13 (p <
.24). Likewise, the machine-coded daily fairness proportion produced a .04 (p < .67)
correlation with the Obama vote support time series. Though such a small sample
cannot supply much statistical precision for either correlation, it is reassuring that
the predicted data shows a positive but weaker correlation compared to the hand-
coded data. This dampening e↵ect is exactly what would be expected from random
1
However, as Greene (2008a) notes, the impact of measurement error on other variables in a
model is harder to predict. Blackwell, Honaker & King (2015) suggest using multiple imputation, or
what they call “overimputation”, to correct measurement error and outline such a procedure using
the Amelia package in R. I conducted a series of tests using a multiple imputation version of the
Obama daily fairness proportion variable and the results largely mirrored those reported later in
this chapter.
37
measurement error and supports the contention that any measurement error bias
2.2.3 Variables
time series of Obama vote support and a second daily time series of Romney vote
support. To construct these time series, I aggregated daily reported data from Hu↵-
ington Post Pollster, which cataloged 2012 survey results from more than sixty public
pollsters (Huffington Post Pollster 2012). I calculated the time series by averaging the
vote support using only polls of likely voters. Because surveys are typically conducted
over several days, each poll was averaged with all other polls that shared the same
last day of fielding. So, for instance, if a particular poll was fielded from January
1, 2012 through January 4, 2012, I averaged its results with all other polls that had
the same end date: January 4, 2012. Unfortunately, data was not available for every
single day in the time series, which resulted in irregular daily time series of 169 days
more than 108,000 Obama sentences from September, 2011 through November 2012,
each day for which data was available, I then calculated both the proportion of fair-
ness message sentences and the frequency fairness sentences. As a result, I calculated
two di↵erent time series to be used as independent variables: first, a daily percentage
of Obama fairness rhetoric and, second, a daily frequency of Obama fairness rhetoric.
38
The percentage and frequency time series each had advantages and disadvantages.
The proportion variable captured how much of the president’s daily speech was com-
posed of his fairness master frame, but it did not measure the daily communication
volume. The percentage time series had the advantage of gauging relative change in
fairness rhetoric and was a purer measure of changes in the president’s master frame
usage. On the other hand, the daily master frame percentage would not have been
likely to have an e↵ect if the amount of communication was not large enough to break
through to the public. For example, it is unlikely that a 50 percent fairness sentence
day with only one speech had as much e↵ect as a 50 percent fairness day in which 30
speeches were given. On the other hand, the frequency variable measured the daily
volume of fairness sentences, but was very highly correlated with the total number
of sentences used on a particular day (r = .97). Of course, this meant that its e↵ect
was hard to separate from total volume of the president’s daily communication and
was correlated with his other messages. (But that means that total Obama speech
volume was also a proxy for the fairness message volume). Because each measure had
its own advantages, I reported regression results using each proportion and frequency
message time series. However, I focused more on the percentage time series because
Figure 2.2 displays the daily frequency of Obama’s fairness sentences from Septem-
ber, 2011 through the 2012 election. One can observe a series of high-fairness spikes
after the introduction of the message in September 2011. This is followed by a slight
lull around the holidays and the turn of the year before resuming a higher spikes in
the usage of fairness rhetoric. Starting in the summer of 2012, there is a noticeable
39
Figure 2.2: Daily Obama Fairness Sentence Frequency
up-tick in fairness rhetoric, with typical fairness frequency rising from the mid-50s to
more than 100 fairness sentences per day. The graph also suggests that the time se-
ries may not be stationary because the variance appears to rise and fall over di↵erent
time periods and the mean appears to rise as the election draws near. This change
in volume makes sense as one would expect the president to give more speeches and
Figure 2.3 displays the daily Obama fairness percentage from the fall of 2011
through early November 2012. The pattern of Obama’s daily fairness percentage is
40
Figure 2.3: Daily Obama Fairness Sentence Percentage
similar to the daily frequency variable in the beginning of the time series. There is
a rise in fairness rhetoric after the president’s September budget address and there
rhetoric through the holidays, suggesting that the volume of Obama communication
dropped during this period but that the fairness proportion of that communication
rose. The fairness percentage drops o↵ somewhat through early spring before spiking
again in April and again in the run-up to the fall campaign. In the fall, Obama’s
fairness message largely stabilizes at close to the average of 28 percent per day. This
41
period has few spikes above 40 percent and few drops below 20 percent, suggesting
that the president was repeating the same message on the stump, a message which
time series of daily national Democratic Party identification. Using Huffington Post
Pollster reported data of public polls, this time series represents the average percent
of Americans who said they identified as Democrats on a given day (Huffington Post
Pollster 2016b). (Due to a lack of data, I was not able to restrict this time series
to likely voters). As with the Obama and Romney support time series, the time
series averaged results on the last day a set of polls collected results, and data was
not available for every day in the time series. Although not needed for this chapter’s
initial hypothesis, a time series of national independent identification was also created
As mentioned, daily data did not exist for every single day in each time series.
The Obama speech variables covered the most days because there few days in which
the president did not give a speech or release a statement. In all there were 334 days
in the Obama fairness master frame time series. There were more missing days in
the vote support and party identification time series because public polling was not
conducted on every single day between September, 2011 and November, 2012. In
total, there were 169 days in the two vote support time series and 140 days each of
the party identification time series. Since OLS commonly uses list-wise deletion when
faced with missing data, models could only be estimated using days in which each
42
time series had data. This meant that N-sizes for models were lower than the 400 or so
days that would be available if there was full data in each time series. This issue was
exacerbated somewhat by adding lags and moderators. However, because there was
no reason to believe that the missing data pattern in these variables was correlated
with observed data, I assumed the data was Missing Completely at Random (MCAR)
Sidani & Figueredo 2007). The missing data did reduce N-size and statistical power,
but, once again, this limitation likely meant more stringent tests for most of my
hypotheses.
2.2.5 Estimation
I estimated results using OLS regression with lagged dependent variables as needed
of the president’s fairness master frame. Though there are abundant options for
modeling multivariate time series data, Keele & Kelly (2006) demonstrated that OLS
with lagged dependent variables performs better than other options when the data
reasonable to assume that past values of Obama vote support or party identification
influence present values. Using lagged dependent variables also had the advantage
of controlling for all other expected e↵ects on the 2012 election, including campaign
An important requirement for estimating OLS with time series variables is that
each variable must be stationary, i.e. have constant means and variances across
time. Regression with non-stationary variables can lead to spurious regression results,
43
correlations that appear strong and significant only because of an underlying shared
time trend. Because of this threat to validity, time series variables must be tested
di↵erence from the previous time period – 1st di↵erencing- is a common time series
As suggested by Greene (2008b) and others (Cowpertwait & Metcalfe 2009), sta-
tionarity was tested using the augmented Dickey-Fuller and the Kwiatkowski-Phillips-
unit roots, or “long-memory” trends that violate stationarity and its null hypothesis
stationarity as its null hypothesis. Each tests’ null is the reverse of the other, so a
time series would need a significant ADF result but a non-significant KPSS result for
both tests to suggestion stationarity. Because of the danger that spurious regression
poses to causal inference, I di↵erenced time series that showed mixed results. For
robustness, I presented results using di↵erenced and non-di↵erenced time series for
hypothesis one, but after I focused only on di↵erenced time series models.
As the results in table 2.1 report, every time series except Obama’s daily fair-
columns of the table show results for both ADF and KPSS tests before first di↵er-
encing and the right 3 columns show results for the same tests after first di↵erencing.
the models below. Because both the Obama vote support and the Democratic Party
44
Table 2.1: Time Series Stationarity Tests
Time Series ADF KPSS Results ADF Di↵. KPSS Di↵. Results After Di↵.
Obama Vote Support 3.52⇤ 1.87⇤⇤ Mixed 7.34⇤⇤ 0.01 Stationary
Romney Vote Support 3.38 1.81⇤⇤ Not Stationary 7.31⇤⇤ 0.01 Stationary
Obama Fairness Freq. 6.57⇤⇤ 1.54⇤ Mixed 7.13⇤⇤ 0.10 Stationary
Obama Fairness Sent. Prop. 5.94⇤⇤ 0.31 Stationary n/a n/a Di↵erencing not necessary
National Democratic Party ID 4.08⇤⇤ 0.53⇤ Mixed 7.91⇤⇤ 0.01 Stationary
National Independent ID 4.66⇤⇤ 0.49⇤ Mixed 7.67⇤⇤ 0.01 Stationary
Identification time series were aggregated by the last day of each polls field period, any
independent variable lag of one to three days will generally not be before the polling
results were collected, but while they were being collected. If Obama’s average for a
particular day was 49 percent, the data for that day was most likely collected on that
day and the preceding two or three days, meaning that a speech frequency or speech
percentage lag of three days from the polls last fielding date most likely occurred
during the first day of data collection for a particular set of polls.
Since the dependent variable is not a complete daily time series, dependent vari-
able lags are not in days but in previous periods of available data. This means the
dependent variable lag coefficients should not be interpreted as the e↵ect of the previ-
ous days polling results but the previous period’s polling results, whenever they may
Because there was not a theoretical prediction for how long it would take for
Obama’s messaging to have the most impact, I estimated results with independent
variables lagged one through four days and kept those lags that were statistically
significant. Dependent variable lags were chosen by visually inspecting the variables’
45
2.3 Results
Table 2.2 reports the estimated e↵ect of Obama’s frequency of daily fairness sen-
tences on his reelection support. Model one reports results for un-di↵erenced time
series variables and Model two reports results after the time series were di↵erenced
lation and to increase statistical precision for the independent variables. The high
Breusch-Godfrey p-values reported at the bottom of the table mean that a null of
no autocorrelation could not be rejected, suggesting there were no issues with serial
correlation.
Results for both models are quite similar. In Model one, each fairness sentence
used two days before the polling period ended was associated with a .006 rise in the
percent of Americans who said they would vote for Obama for president. Of course,
the e↵ect of each individual sentence is quite small. But another way to interpret
results would be to calculate the e↵ect of days in which Obama used a minimum num-
ber of fairness sentences (1) compared to days in which he used the maximum number
of sentences (319). Moving from the minimum to the maximum number of fairness
predicted 1.9 percent increase in Obama’s polling support. In model one, the e↵ect
of fairness sentences used on the last day of survey collection also had a significant
and positive estimated e↵ect on Obama’s presidential support. Specifically, each sen-
tence was expected to garner an additional .009 of a percent for Obama’s presidential
support. For this “same day” variable, the increase from minimum fairness sentence
to the maximum fairness sentences would garner a 2.82 percent rise in Obama’s sup-
port. The combined increase from minimum sentences used to maximum sentences
46
Table 2.2: Estimated E↵ect of President’s Daily Frequency Of Fairness Rhetoric on
2012 Obama Support
Daily Obama Vote Support Percentage Model 1 Model 2
(Intercept) 24.901⇤⇤⇤ 0.085
(5.539) (0.183)
Obama Vote Support - 1 Period Lag 0.070
(0.104)
Obama Vote Support - 2 Period Lag 0.118
(0.096)
Obama Vote Support - 3 Period Lag 0.039
(0.096)
Obama Vote Support - 4 Period Lag 0.203⇤
(0.094)
Obama Vote Support - 5 Period Lag 0.014
(0.088)
Obama Fairness Rhetoric (Daily Freq.) 0.009⇤⇤⇤
(0.003)
Obama Fairness Rhetoric (Daily Freq.)- 2 DL 0.006⇤
(0.003)
Di↵. Obama Vote Support - 1 Period Lag 0.774⇤⇤⇤
(0.097)
Di↵. Obama Vote Support - 2 Period Lag 0.587⇤⇤⇤
(0.126)
Di↵. Obama Vote Support - 3 Period Lag 0.489⇤⇤⇤
(0.132)
Di↵. Obama Vote Support - 4 Period Lag 0.214⇤
(0.097)
Di↵. Obama Fairness Rhetoric (Daily Freq.) - 3 DL 0.005⇤
(0.002)
2
R 0.375 0.470
2
Adj. R 0.321 0.436
Breusch-Godfrey 0.612 0.340
Num. obs. 89 84
RMSE 1.688 1.620
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
47
used is a substantial 4.72 increase in Obama’s polling support. Of course, days with
the maximum number of sentences were very rare. To measure the impact of average
days, the average number of sentences (70.51) would be expected to garner .42 of a
percent increase for the two-day lag and .63 of a percent increase for the same day
variable. Combined, that means the e↵ect of any average “fairness sentence” day
Model two reports results for the estimated impact of the frequency of fairness
sentences after taking first di↵erences of both the daily Obama election support vari-
able and the daily frequency of fairness sentence variable. After di↵erencing, the
immediate e↵ect of sentence frequency was no longer significant below the traditional
.05 level and was dropped from the model. Instead, a three-day lag of the frequency
variable was positive and significant, although the coefficient reports little more than
half the total e↵ect as the e↵ect of the immediate frequency and 2 day lagged vari-
able combined2 . Specifically, the three day lagged Obama fairness sentence e↵ect is
estimated to be .005 increase in Obama support per each sentence. This means that
on days when Obama used 70 fairness sentences, his presidential support percentage
would be expected to rise .35 percent, all things being equal. The maximum e↵ect
of the three-day lag would be 318*.005, or a 1.59 percent increase in support for
president Obama.
As mentioned earlier, the Obama fairness frequency variable had the advantage
of measuring the e↵ect of master frame volume, but, because it is highly correlated
with Obama’s overall speech volume, it’s not the best test for the pure e↵ect of
2
A two day lag of the frequency variable was negative and borderline significant below the p < .10
level, but it lost significance and became positive when added to a model with a positive and
significant three day lag of frequency and therefore was dropped from the final model.
48
Obama’s fairness master frame. To more precisely estimate that e↵ect, I used the
daily percentage of fairness sentences as an independent variable and report the results
Table 2.3 reports results for the e↵ect Obama’s daily percentage of master frame
fairness had on his subsequent electoral support. Model 1 shows the e↵ect of Obama’s
daily percentage of the fairness master frame on his support for president. Model 1
reports results for an un-di↵erenced dependent variable and shows that a three-day lag
of Obama’s percentage of fairness rhetoric has a positive and but not quite significant
e↵ect (p < .06) on his subsequent vote support. (This e↵ect was only significant
when including period dependent variable lags). Specifically, model one shows that
for each percent increase in fairness rhetoric three days before a poll’s final day,
Obama’s vote support is expected to rise by .04. At the average 28 percent fairness
rhetoric, Obama’s vote support would be expected to rise 1.12 percent (compared to
vote support would be expected to rise more than two percentage points.
Model two reports results with the dependent variable di↵erenced to ensure sta-
tionarity. The independent variable was not di↵erenced because it was deemed sta-
tionary by both the ADF and KPSS test without transformation. In this model, a
three-day lag of Obama fairness percentage has a positive and statistically significant
association with his subsequent vote support. The coefficient is slightly stronger than
model one and more significant. (The same coefficient is borderline significant below
the .10 level with only three dependent variable lags). In this model, an average
day of Obama fairness rhetoric at 28 percent would be expected to increase his vote
support rate by 1.34 percentage points.Though a little under 1 and a half percent
49
Table 2.3: Estimated E↵ect of President’s Daily Percentage Of Fairness Rhetoric on
2012 Obama Support
Daily Obama Vote Support Percentage Model 1 Model 2
(Intercept) 10.204⇤ 1.350⇤
(4.928) (0.636)
Obama Vote Support - 1 Period Lag 0.166
(0.086)
Obama Vote Support - 2 Period Lag 0.160
(0.083)
Obama Vote Support - 3 Period Lag 0.164
(0.087)
Obama Vote Support - 4 Period Lag 0.188⇤
(0.085)
Obama Vote Support - 5 Period Lag 0.078
(0.089)
Obama Fairness Rhetoric (Daily Percentage) - 3 Day Lag 0.040+ 0.048⇤
(0.021) (0.021)
Di↵erenced Obama Vote Support - 1 Period Lag 0.775⇤⇤⇤
(0.084)
Di↵erenced Obama Vote Support - 2 Period Lag 0.588⇤⇤⇤
(0.108)
Di↵erenced Obama Vote Support - 3 Period Lag 0.383⇤⇤⇤
(0.108)
Di↵erenced Obama Vote Support - 1 Period Lag 0.146
(0.086)
R2 0.337 0.447
Adj. R2 0.302 0.423
Breusch-Godfrey 0.525 0.231
Num. obs. 120 120
RMSE 1.710 1.744
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05, +p < 0.10
50
may not initially seem like a large increase, it is quite large considering that Obama’s
final winning margin was just under four percent (Leip 2012). Likewise, a day with
the maximum observed fairness rhetoric –54 percent –would be expected to increase
the rate of his vote support by 2.6 percentage points – a substantial amount in such
Figure 2.4 plots the predicted e↵ect of model two on the di↵erenced Obama vote
support variable, which is the rate of change in his vote support. Overall, it shows that
higher fairness speech percentages increase Obama’s vote support rate. In particular,
low fairness days pull his vote support down below average and high fairness days
push his support above his average rate of support. The most significant negative
e↵ects are at days in which fairness percentages were below 20 percent and the most
Tests using both the Obama fairness frequency and daily percentage variables
and using di↵erenced and un-di↵erenced variables all show a consistent positive e↵ect
from Obama’s fairness master frame. The results are robust to di↵erent ways of mea-
suring Obama’s fairness master frame and in models that make di↵erent stationarity
variable lags to remove autocorrelation and to act as controls. Taken together, these
results present strong evidence in support of my first hypothesis: all things being
equal, Obama’s fairness master worked just as he an his campaign intended. It set
the terms of campaign debate and had a substantial and positive e↵ect on his electoral
support.
the positive relationship between Obama’s fairness master frame and his electoral
51
Figure 2.4: E↵ect of Fairness Percentage on Obama Support
Figure 2.4 shows the estimated e↵ect of increasing percentage of Obama’s master frame on
his electoral support. The x-axis is three-day lag of Obama’s daily percentage of fairness
rhetoric. The y-axis is the first di↵erence of Obama’s daily electoral support. The shaded
area is the regression line’s 95 percent confidence interval.
support is due to some other variable. What might that be? Though it is not possible
to verify with this data, it is plausible that increases in Obama’s fairness rhetoric also
52
Table 2.4: Estimated E↵ect of Obama Fairness Rhetoric on Romney 2012 Support
Daily Romney Vote Support Percentage Model 1
(Intercept) 0.14
(0.54)
Di↵erenced Romney Vote Support – 1 Period Lag 0.71⇤⇤⇤
(0.08)
Di↵erenced Romney Vote Support – 2 Period Lag 0.50⇤⇤⇤
(0.09)
Di↵erenced Romney Vote Support – 3 Period Lag 0.07
(0.08)
Obama Fairness Rhetoric (Daily Percentage) – 2 Day Lag 0.01
(0.02)
2
R 0.47
Adj. R2 0.45
Num. obs. 121
Breusch-Godfrey 0.486
RMSE 1.65
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
coincided with campaign visits and with TV ad spending. In other words, it could
be that the associated increases with Obama’s fairness rhetoric are really the result
of e↵ective campaign stops and/or powerful TV ads. Indeed, Sides & Vavreck (2013)
vote share, but only when dominating Romney’s advertising volume by more than 3
ads per viewer. On the other hand, it is likely that Obama’s advertisements mirrored
his messaging on the stump, so its possible they were e↵ective because of the master
frame content they contained. Clearly, an additional content analysis of Obama’s ads
53
Table 2.5: Estimated E↵ect of Speech Proportion on Party ID
Di↵. Daily Nat. Dem Party ID Model 1
(Intercept) 2.25
(1.37)
Di↵erenced National Democratic Party ID – 1 Period Lag 0.55⇤⇤⇤
(0.08)
Obama Fairness Rhetoric (Daily Percentage) – 1 Day Lag 0.09⇤
(0.04)
R2 0.34
2
Adj. R 0.33
Num. obs. 107
Breusch-Godfrey 0.106
RMSE 4.25
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
Table 2.4 reports the estimated e↵ect of Obama’s daily fairness percentage on
Romney’s vote support. Because the Romney support time series failed both sta-
tionary tests, only a di↵erenced model was reported. In this model, none of the
lags of Obama’s daily fairness percentage were significant, but the two-day lag was
Obama’s daily percentage of fairness rhetoric had no e↵ect on Romney’s support. The
coefficient is slightly positive but not statistically di↵erent from zero. This null e↵ect
provides partial support for hypothesis four’s expectation that the fairness master
frame would not increase support for Romney. Paired with the strong evidence that
Obama’s fairness master frame increased support for president, this suggests that the
54
In order to test hypothesis two, I regressed daily national democratic party iden-
tification on the daily fairness percentage time series. Table 2.5 reports those results
and, once again, provides support for hypothesis two. In this case, a 1-day lag of
Obama’s daily fairness percentage is estimated to have a positive and significant e↵ect
on the percentage of voters who subsequently identify with the Democratic Party3 .
Indeed, this e↵ect is even stronger than the master frame e↵ect on Obama’s presiden-
tial support. For each percentage point Obama’s fairness rhetoric increases, national
national Democratic Party ID by 2.5 percentage points. Even more, the maximum
table 2.5 provides strong evidence that Obama’s fairness master frame strengthened
and solidified the Democratic coalition. By increasing the number of voters who
identified as Democrats, one might expect that this is the source of Obama’s master
Unfortunately for my third hypothesis, that did not appear to be the case. Indeed,
as table 2.6 and figure 2.3 report, Obama’s use of the fairness master frame appears to
have no e↵ect on the impact national Democratic party identification had on support
for President Obama. One would expect that as the percentage of voters who identify
with the Democratic party rose, so would support for president Obama. If the master
55
Table 2.6: E↵ect of Dem. Party ID Moderated by Obama Fairness Rhetoric on
Obama Vote support
Di↵. Daily Obama Vote Support Model 1
(Intercept) 0.67
(1.24)
Di↵erenced National Democratic Party ID 0.05
(0.24)
Obama Fairness Rhetoric (Daily Percentage) – 1DL 0.02
(0.04)
Di↵. National Dem. Party ID X Obama Fairness Rhetoric Perc. 1DL 0.00
(0.01)
R2 0.03
2
Adj. R -0.02
Num. obs. 59
Breusch-Godfrey 0.753
RMSE 2.34
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
56
coalition, the positive e↵ect of national democratic party ID should grow as the
percentage of Obama’s fairness rhetoric increases. But as figure three makes clear,
the marginal e↵ect of Democratic Party ID is not significantly di↵erent from zero
at any Obama fairness percentage. In other words, regardless of how much fairness
rhetoric Obama used, the percent of national Democratic Party ID had no e↵ect on
Why might this be the case? It is possible that the highly polarized political
liberal coalition. It could be that with such high levels of partisanship, not many more
people could have been persuaded to join the cause who were not already members of
the cause. On the other hand, it may be possible that Democratic party members and
“leaners” have already heard and been persuaded by the fairness message because its
distillation of fundamental liberal values in the United States. In other words, the
master frame may not have had an e↵ect because voters open to its argument had
already been persuaded by similar appeals to liberal values and have already joined
The evidence suggests otherwise, however. Table 2.7 and figure 2.6 demonstrates
that Obama’s master frame had the opposite than expected e↵ect on support for Rom-
and significant e↵ect on support for Romney. Although for roughly eighty percent of
the days, Obama’s fairness master frame had no statistically significant e↵ect on sup-
port for Romney, days with more than 35 percent fairness rhetoric had a statistically
57
Figure 2.5: Moderating E↵ect of Fairness Rhetoric by Dem. ID on Obama Support
support. In other words, days with a very high fairness rhetoric were predicted to in-
crease Democratic support for Romney. At high values the marginal e↵ect is roughly
.4, meaning that each percentage point in Democratic ID would be translated into a
58
Table 2.7: Democratic Party ID Interacted with Fairness Percentage e↵ect on Romney
Support
half a percentage point in support for Romney. This finding provides strong evidence
against part of hypothesis four and argues against hypothesis two as well. This result
suggests that the positive master frame e↵ect must have been carried by a mechanism
What might cause this counterintuitive result? Data is not available to investigate
which Democratic voters might be shifting support for Romney after high fairness
days. It is possible that as the Democratic coalition has grown more professional and
less reliant on the working class that there is a certain subset of Democratic voters
who found the fairness master frame alienating and that caused a rise in support
59
Figure 2.6: Moderating E↵ect of Fairness Rhetoric by Dem. ID on Romney Support
for Romney. It is also possible that those high fairness days coincided with other
campaign events or rhetoric and they caused a small increase in support for Romney.
60
Table 2.8: Fairness Rhetoric Moderating E↵ect of Independent Identification on Rom-
ney Support
Di↵. Daily Romney Vote Support Model 1
(Intercept) 0.36
(1.15)
Di↵erenced National Independent ID 0.24
(0.13)
Obama Fairness Rhetoric (Daily Percentage) – 2DL 0.01
(0.04)
Di↵erenced National Ind. ID X Obama Fairness Rhetoric Perc. – 2DL 0.01⇤
(0.00)
2
R 0.16
2
Adj. R 0.11
Num. obs. 64
Breusch-Godfrey 0.961
RMSE 2.21
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
Given these last set of results, what would help explain the increase in support
driven by Obama’s fairness master frame? One obvious place to look would be for
any e↵ect the master frame had among independents. By their very nature, we might
expect an increasing percentage of independents to have no e↵ect one way or the other
on support for president Obama. Although this was not included in the hypotheses
for this chapter, I estimated models to explore the relationship between increasing
the fairness master frame and support for Obama and Romney among independents.
Table 2.8 and figure 2.7 show that Obama’s fairness master frame produces a
61
Figure 2.7: Moderating E↵ect of Fairness Rhetoric on Independent Support for Rom-
ney
starting at just above the median for Obama’s fairness master frame percentage, the
marginal e↵ect from the national independent identification time series has a sig-
nificant and negative association with support for Romney. At the maximum of 55
62
percent fairness a two-day lag would be expected to produce nearly a -.3 marginal
e↵ect for each percentage point independents increased in the electorate. Based on
the estimates of this model, an increase of independent identification from the bot-
fairness rhetoric period – would be expected to reduce Romney’s vote support by 4.5
percentage points.
Just as important, table 2.9 and figure 2.8 report results for the interaction of
a 2 day lag of Obama’s fairness master frame percentage and the e↵ect of the per-
model make clear that increasing Obama fairness rhetoric produced a positive e↵ect
is not statistically significant until the two-day lag of Obama’s fairness percentage
ness rhetoric in turn increases the positive e↵ect that independent identification has
followed two days later by a .3 increase in Obama’s vote support for each percentage
Since the data sets used for this analysis do not report leaners, it’s likely that
Though it is not possible with the available data to uncover which independents were
most responsive to the Obama fairness message, it’s plausible that leaning Democrats
63
Table 2.9: Fairness Rhetoric Moderating E↵ect on Independent Identification support
for Obama
Daily Obama Vote Support Percentage Model 1
(Intercept) 0.15
(1.25)
Di↵erenced National Independent ID 0.27
(0.14)
Obama Fairness Rhetoric (Daily Percentage) – 2DL 0.01
(0.04)
Di↵erenced National Ind. ID X Obama Fairness Rhetoric Perc. – 2DL 0.01⇤
(0.00)
2
R 0.09
2
Adj. R 0.04
Num. obs. 64
Breusch-Godfrey 0.412
RMSE 2.40
⇤⇤⇤ p < 0.001, ⇤⇤ p < 0.01, ⇤ p < 0.05
64
were most a↵ected by the fairness message and more likely to support president
Obama and less likely to support Romney as a result. The fact that these particular
polls used the more expansive definition of independent suggests that many of the
measured “independents” likely had some latent political affiliation and were therefore
persuade because they are the most knowledgeable and therefore most fixed voters
on the issues, it follows that individuals who were truly independent might be most
responsive to a well-crafted master frame. The fact that the Obama fairness master
frame generated support for Obama and opposition to Romney among independents
suggests that this message was e↵ective among the most persuadable voters in the
electorate.
Given the accounts of the message development process that Axelrod (2015)
and Wol↵e (2013) both recounted, the fact that the Obama fairness master frame
was most e↵ective among independents fits perfectly with the campaign’s goal of
building a winning coalition in a very closely divided electorate. They sought a mes-
sage that would both motivate the base and persuade independent voters to reelect
the president. Taken together, these results suggest that the message persuaded in-
dependents, but it may not have worked as intended among some Democratic base
voters.
2.4 Conclusion
This chapter set out to test the master frame theory in the real-world context
65
Figure 2.8: Moderating E↵ect of Fairness Rhetoric on Independent Support for
Obama
ness rhetoric. Using time series analysis, this chapter provided strong evidence that
66
changes in Obama’s fairness master frame were associated by fairly rapid movements
Not only do the results in this chapter provide evidence that Obama used a con-
sistent and measurable master frame, but also that his master frame usage had a
positive and substantial association with his support for reelection. Of course, the
Obama campaign employed several strategies to persuade voters and rally support-
ers, including extensive field activities. But this chapter’s findings underscore that
In general, the results show strong support for the master frame theory at the
macro level. Obama’s fairness master frame increased his electoral support, boosted
independents.
Just as the master frame theory predicts, frames that define the campaign can have
fairness master frame were not associated with support for Romney. Changes in
Obama’s master frame usage also had a strong and positive e↵ect on Democratic
partisanship. Put simply, when Obama used more fairness rhetoric, Democratic par-
However, the mechanism through which Obama’s master frame had its e↵ect
on his vote support was unclear. The master frame did not intensify the impact
of partisan support for Obama, instead it had no e↵ect on partisans support for
Obama. The master frame appeared to have the opposite e↵ect, boosting Romney
when fairness rhetoric made up a high percentage of Obama’s daily speeches. This
result suggests that, while master frames have the predicted e↵ect overall - boosting
67
support for ideologically aligned candidates and policies, it’s less certain how that
e↵ect is actually made real. It may be in some cases that master frames work by
intensifying partisan support. But it may also be that master frames work by more
68
Chapter 3: Experimental Evidence for Master Frame E↵ects
3.1 Introduction
In the first chapter I proposed a theory that master frames set the stage for politi-
cal debate and that, as such, they should increase the e↵ectiveness of individual issue
and/or campaign frames. To test this theory, I conducted two sets of experiments:
sity. The goal of these experiments was to test whether master frames indeed increase
and taxes (study two). I also explored the moderating influence personality disposi-
To test the four hypotheses in chapter one at the individual level, I conducted the
Experiment 1 was a 3x2 design in which subjects were first randomly assigned
trol). After the master frame assignment, subjects were randomly assigned to receive
69
either a liberal immigration frame or a conservative immigration frame. (Full word-
ing, including issue frame treatments, is in the appendix). The dependent variable
immigration frame to which they were assigned. (Subjects were assigned only to one
Subjects who received the equality master frame treatment were asked to read the
following:
Subjects who received the individualism master frame treatment were asked to
Subjects
Experimental subjects were drawn from a sample of 277 students from a Mid-
western university. Students were given extra credit in return for participation. The
in the experiment at a time of their own choosing and on a computer of their own
For this analysis, the dependent variable is coded so that liberal positions take
positive values and conservative positions take negative values. In the liberal im-
migration frame conditions, positive values represent agreement with the frame, 0
70
is neutral and negative values represent opposition to the frame. For conservative
neutral and negative values represent agreement with the conservative frame.
71
Figure 3.1: Immigration Issue Frame E↵ects Stronger than Master Frames
Figure 3.1 shows means for each experimental condition. Bars are 95 percent confidence
intervals. Means are plotted first by whether subjects received a conservative (left panel)
or liberal (right panel) immigration frame. Within each panel, means of subjects who re-
ceived an individualism, equality or no (control) master frame are plotted. The y-axis is
the level of support for the conservative or liberal immigration frame. For ease of compar-
ison, the dependent variable is coded so that negative values represent either support for
the conservative frame or opposition to the liberal frame, zero is neutral and positive values
represent either support for the liberal frame or opposition to the conservative frame. The
plotted means show a strong e↵ect from issue framing, but a modest master frame e↵ect.
However, subjects who received an equality master frame and a liberal immigration frame
showed the strongest support for the liberal immigration frame. (The master frame e↵ect is
more notable when a personality moderator is taken into account - see figure 3.2).
72
As is clear in Figure 3.1, the experiment appeared to resulted in a large issue
framing e↵ect. However, it is important to note that this dependent variable di↵ers
from the dependent variable used in the next experiment. The present dependent
variable measures the level of support for a immigration frame. In contrast, the
CCES experiment later in this chapter uses support for tax policy as its dependent
variable. In short, the first experiment measures support for an issue frame while
the second experiment measures attitudes regarding a policy issue. It is possible that
these issue framing results rose from a tendency among subjects to simply agree with
59 percent of participants who read the conservative immigration frame and were
in the control (no master frame) condition supported the conservative immigration
frame. In contrast, 70 percent of participants who read the liberal immigration frame
and were in the control (no master frame) condition supported the liberal position.
and that it’s important to follow the law, participants were more likely to support re-
stricting citizenship. At the same time, emphasizing that the people in question were
“undocumented” immigrations, (rather than “illegal”), and were taken to the United
Though these issue frame e↵ects are relevant to understanding how policy argu-
ments shape immigration opinions, this experiment was conducted to test whether
master frames increase the e↵ectiveness of issue frames. In particular it was conducted
to test the stage-setting hypothesis, which predicts that exposure to the equality mas-
ter frame will increase support for the liberal immigration frame and that exposure to
73
the individualism master frame will increase support for the conservative immigration
frame.
As the means plotted in figure 3.1 show, the results provide conflicting evidence
for the stage-setting hypothesis. First, people who received both the equality master
frame and the liberal immigration argument were more supportive of the liberal frame.
The mean support for this group was 1.02, or “somewhat support.”
Participants who read an individualism message and the liberal immigration frame
also reported slightly higher, but statistically insignificant support. Compared to the
control group, people exposed to the individualism master frame and the conservative
immigration frame were not more likely to support the conservative view than people
Table 1 reports OLS results testing the e↵ect of master frame treatments. For each
model the dependent variable was whether subjects supported either the conservative
or liberal immigration frames. Models 1 and 2 estimate the e↵ect of issue frames and
Individualism are coded “1” for those who received the treatment and “0” for those
who were in the control (no master frame condition). This allows a comparison
frame treatment. The issue frame variable is coded “1” for those who received the
liberal immigration frame and “0” for those who received the conservative immigration
frame. Consistent with Figure 3.1, the positive interaction between receiving an
equality master frame and an liberal issue frame means that subjects in that condition
reported stronger support for a liberal immigration frame, compared to the e↵ect of
the equality master frame in the conservative tax frame condition. This interaction
74
Table 3.1: OLS Estimates of Master Frame E↵ect on Immigration Frame Support
Immigration Frame Support
Model Model Model Model
(1) (2) (3) (4)
p < .05 level(model 2, one-tailed) when pretreatment covariates were added. Contrary
also supported the liberal immigration master frame at a slightly higher level than
the control group, but the di↵erence was not statistically significant.
new experience. Specifically, the individualism master frame did have a substantive
and significant e↵ect on support for both the liberal and conservative immigration
new experiences - was derived from the relevant questions on the Ten Item Person-
ality Inventory (TIPI), a shortened measure of the Big Five Personality Inventory
Models 3 and 4 show the interaction between the individualism master frame and
subject’s personality. Model 3 tests for a two-way interaction between personality and
exposure to master frame treatments. Model 4 tests for a three-way interaction be-
tween personality, master frame treatment and issue frame treatment. In both models
the two-way interaction is positive and statistically significant. The three way inter-
action is not significant in model 4, though the plotted values in figure 4 show notable,
In both model 3 and 4, the individualism and the control variables are coded “1” if
subjects were in those respective conditions and “0” if subjects were in the equal-
76
frame and an equality frame treatment as well as a comparison between exposure to
the equality master frame treatment and the control (no master frame) condition.
evidence for the backlash and personality intensification hypotheses. Consistent with
the backlash hypothesis, figure 3.2 shows that receiving the individualism master
frame caused high openness subject to take more liberal immigration positions, not
less. Likewise and consistent with the intensification hypothesis, the same interaction
shows that exposure to the individualism master frame caused low-openness subjects
master frame did not have backlash and intensification e↵ects. Instead, the e↵ect
of equality was relatively uniform across subjects’ level of openness. The fact that
equality does have significant tax-frame moderated e↵ects in models 1 and 2 while
individualism shows only openness-moderated e↵ects suggests that the e↵ect of some
master frames – such as equality – is consistent with the stage-setting and attitude
change hypotheses, while the e↵ect of other master frames – such as individualism
–is polarizing.
77
Figure 3.2: Individualism Master Frame and Personality Create Strong Polarization
E↵ect
Figure 3.2 shows the interaction between exposure to master frames and the subjects’ level
of openness to new experience, controlling for issue frame e↵ect. Predictions for the in-
dividualism master frame condition include a 95% confidence interval. On the y-axis is
support for liberal or conservative immigration frames. The x-axis is subjects’ level of open-
ness. It is not balanced at zero because more subjects had high open rather than low open
dispositions. Positive values represent support for the liberal frame or opposition to the
conservative frame. Negative values represent support for the conservative frame or oppo-
sition to the liberal frame. The x-axis shows subject’s level of openness to new experiences,
measured using the TIPI. Negative values represent less open dispositions and higher values
more open dispositions. The slope of the black line shows that as people are more more
open, they are more likely to backlash against the individualism master frame and voice
more liberal positions. Likewise, low-openness subjects are made even more conservative by
exposure to the individualism master frame. In short, the individualism master frame has a
polarizing e↵ect depending on one’s personality type. In contrast, the equality master frame
e↵ect is relatively stable across all personality types. Though the equality master frame slope
is clearly di↵erent from the individualism slope, the equality condition isn’t significantly dif-
ferent from control condition. Thus, the moderating e↵ect of personality is largely a function
of how people react to the individualism master frame.
78
Figure 3.2 makes the 3-way interaction between master frames, issue frames and
openness more clear. Though the coefficient for the 3-way interaction is not sig-
nificant, it is well established that interaction coefficients often do not provide full
information regarding moderating e↵ects (Brambor, Clark & Golder 2006). Plotting
sonality predispositions. This strong polarizing e↵ect was observed both for subjects
were also received a liberal immigration frame and those who received a conserva-
tive immigration frame. As expected, exposure to the liberal immigration frame (left
panel) produced higher overall liberal positions and exposure to the conservative im-
migration frame (right panel) produced more conservative positions. Among subjects
in the liberal immigration frame condition, low-openness subjects who received the
the same conditions voiced solid support for the liberal immigration frame. Like-
wise, low-openness people in the conservative immigration frame condition and who
received the individualism master frame, voiced support for the conservative frame
while high-openness people expressed slight opposition. The figure shows that the
predicted e↵ect of individualism is di↵erent from the equality and control conditions
for most values of openness, but particularly for the 74 percent of subjects who were
at the high or low end of the spectrum. Figure 3.2 also shows that, interacted with
openness, the e↵ect of the equality master frame was the same as control for people
in the conservative immigration condition, but di↵erent for those who also received a
79
liberal immigration frame. Specifically, low-openness subjects who received an equal-
ity master frame on top of a liberal immigration frame took more liberal positions
than those in the control (no master frame) condition. Overall, these results suggest
that equality and individualism master frames produce di↵erent e↵ects. Equality
master frames either overwhelm personality dispositions or are not a↵ected by them.
simply, equality has more universal appeal while individualism polarizes. This same
divergent pattern between equality and individualism is seen again in the next study,
Study several days after the November 2012 election. Though the complete study
of 806 respondents. The analysis was conducted using the provided sample weights
to ensure representativeness.
Experiment #2 was a 3x2 design in which subjects were first randomly assigned to
an equality master frame, individualism master frame, or a control (no master frame)
treatment. After the master frame assignment, subjects were randomly assigned to
receive either a liberal tax frame or a conservative tax frame. The dependent variable
was the extent to which respondents supported raising taxes on households that earn
more than $250,000 per year. Subjects were assigned only to one master frame (or
control) and one liberal or conservative issue frame. (Full wording, including issue
80
frame treatments, is provided in the appendix).This experiment was designed to test
the attitude change hypothesis. Unfortunately, the survey did not include the TIPI,
so was I unable to replicate the moderating e↵ect of openness tested in experiment #1.
People assigned to the individualism treatment were asked to read the following:
People assigned to the equality treatment were asked to read this statement:
First, the experiment showed no issue frame e↵ects. This is particularly important
because a typical framing study would make this comparison and perhaps falsely
conclude that it is not possible to change subjects tax attitudes. Subjects exposed
only to liberal or conservative tax frames showed roughly the same level of support for
raising taxes on households with incomes above $250,000. Among subjects who didn’t
get a master frame, there was .58 mean support for raising taxes in the liberal tax
frame group and .51 mean support in the conservative tax frame condition. In other
words, both groups voiced moderate support for raising taxes. However, among those
who received the liberal tax frame, as predicted by the attitude change hypothesis,
81
the equality master frame increased support for raising taxes on the wealthy and the
Figure 3.3: Mean Support for Raising Taxes on Wealthy, by Master Frame Condition
Figure 3.3 shows mean support for taxing the wealthy by experimental condition. Bars are
95 percent confidence intervals. The left panel reports mean support for subjects who also
received a liberal tax frame and the right panel reports mean support for subjects who also
received a conservative tax frame. The left panel perfectly matches the attitude change hy-
pothesis. People who received the equality master frame voiced the most support for raising
taxes, control (no master frame) subjects voiced middle support and people in the individ-
ualism master frame conditions were the least supportive of raising taxes. Means in the
conservative tax frame condition were not statistically di↵erent from each other.
82
Figure 3.3 shows the means of support for taxing the wealthy by condition. Op-
position to taxing the wealthy was coded with negative values and support for taxing
the wealthy was coded with positive values, with neutral coded as zero. In the liberal
tax frame condition, as predicted, subjects exposed to the individualism master frame
were less supportive of raising taxes while people exposed to equality were more sup-
portive. Mean support for taxing the wealthy among those exposed to the equality
master frame was .87 while mean support among those exposed to the individualism
master frame was .28, the di↵erence of which is a treatment e↵ect of .59 and in a t-test
is statistically significant below the standard .05 level (p< .003, one-tailed). In the
conservative tax frame conditions, exposure to both the equality and individualism
for raising taxes. Mean support for increasing taxes in the conservative tax frame
conditions among those exposed to the equality master frame was .72, a treatment
e↵ect of .21. Likewise, the treatment e↵ect for subjects exposed to the individualism
Table 3.2 reports results from four OLS models that test the e↵ect of master
frame treatments on support for taxing the wealthy. In both models Equality is
coded “1” if subjects received an equality master frame and “0” if they received an
individualism master frame. Control is coded “1” of subjects received no master frame
and “0” if they received an individualism master frame. Equality is interacted with
a dummy variable which is coded “1” for people who received the liberal tax frame
and “0” for those received the conservative tax frame. In model 3, the master frame
83
Table 3.2: OLS Estimated E↵ect of Master Frames on Support for Raising Taxes
Tax Increase Support Model Model Model Model
(1) (2) (3) (4)
84
vs. community orientation. Model 4 includes a three-way interaction between master
As in figure 3.3, model 1 and 2 show that the interaction between receiving the
equality master frame and receiving a liberal tax frame is positive and significant,
was associated with more support for taxing the wealthy when subjects also read a
liberal tax frame. In model 2, marital status, gender, partisan identification, ideology
and race were added as covariates, which yielded a more precise estimate of the e↵ect
tax frame they were more likely to support the liberal tax position. And when people
received the individualism master frame in addition to the liberal tax frame they were
less influenced by the liberal tax frame and voiced less support for raising taxes as a
result.
85
Figure 3.4: Master Frame Impact on Support for Taxing the Wealthy
Figure 3.4 shows the e↵ect of master frames on tax views for comparison with figure 1.2.
When compared to experimental expectations, its clear that the liberal tax frame conditions
closely match figure 1.2 while the conservative tax frame conditions does for all but one
condition.
These results provide compelling evidence that simply exposing people to a short
86
they will support or oppose and import policy issue such as tax policy. As the theory
predicts, these results show that political elites have much to gain by emphasizing
ideological principles – they make later arguments more e↵ective. Comparing figure
3.4 to expectations in figure 1.2, the overall pattern in the CCES tax experiment’s
issue frame conditions reported higher support for taxing the wealthy. The support
was greater among people who also received an equality master frame and was weaker
among people who received an individualism message. Likewise and also consistent
with experimental expectations, people in the conservative tax frame conditions were
generally less supportive of raising taxes. However, the pattern of results in the
conservative tax frame conditions was not entirely consistent with predictions.
To better understand the practical e↵ect of these results, it’s helpful to explore the
to the liberal tax frame. The starkest di↵erence is among people citing strong support
for taxing the wealthy. Among the equality group, 45 percent of subjects reported
strong support for increasing taxes on households above $250,000. In contrast, only
31 percent of the individualism group and 35 percent of the control group expressed
strong support. The percentage of subjects expressing strong opposition was higher
among those exposed to the individualism master frame. 20 percent of the individ-
ualism group was strongly opposed, while only 8 percent of the equality group was
strongly opposed. Overall, 66 percent of the equality group supported raising taxes on
the wealthy, compared to 60 percent of the control group and only 48 percent of the
individualism group. The practical e↵ect is that master frames moved support from
87
3.3.3 Personal Disposition as Moderator of Master Frame
E↵ects in Tax Experiment
or reverse the e↵ect of master frames, depending on whether the master frame and
did not include a measure for openness to new experience, so it was not possible to
test for whether it had the same e↵ect as in the immigration experiment. However,
communitarian orientation, which is, of course, the same value dimension the master
frame treatments are intended to magnify. For this reason, I estimated the interac-
tion between master frame exposure and subjects’ individualism vs. communitarian
orientation as another way of capturing the moderating e↵ect of personality. The in-
dividualism vs. communitarian variable is coded so that lower values represent more
The key coefficient in model (3) is the interaction between master frame exposure
tive. Therefore, the equality and individualism master frame slopes were di↵erent
across values of self-other orientation. The three-way interaction in model (4) is not
significant, but, as with study #1, there are subtle di↵erences between the liberal
and conservative tax frame conditions. Most notably, the interaction between master
frames and disposition is stronger in the liberal tax frame condition. For this reason,
marginal predictions for all conditions are plotted in Figure 3.5. As the graphs re-
polarizing e↵ect. In the liberal tax frame condition, the individualism master frame
88
made individualistic people less supportive of taxing the wealthy while it made com-
munitarian people more supportive. The same pattern for individualism is observed
for subjects in the conservative tax frame condition, but the interaction is weaker.
Also similar to study #1, receiving a liberal tax frame and an equality master frame
support raising taxes. However, individualistic subjects who got a conservative tax
frame and an equality master frame showed roughly the same opposition as people
89
Figure 3.5: Master Frame E↵ect on Support for Taxing the Wealthy, by Community
vs. Individualism Orientation
Figure 3.5 shows the moderating e↵ect of subjects’ individualism vs. communitarian ori-
entation on master frame treatments. The interaction is shown for subjects exposed to a
liberal tax frame (left panel) as well as subjects exposed to a conservative tax frame (right
panel). The interaction is strongest in the liberal tax condition, but in each individualism
tends to polarize participants making their predispositions stronger. In the left panel, mod-
erate to high individualists take even stronger conservative positions when exposed to an
individualism master frame. At the same time, high communitarians take more liberal posi-
tions after receiving an individualism master frame, but those positions are not statistically
di↵erent from control and equality treatments. Also among people in the liberal tax fame
conditions, high individualist subjects became more liberal after receiving an equality master
frame, overwhelming their predispositions. The same general pattern can be seen in subjects
exposed to the conservative tax frame, but the di↵erences between conditions do not reach
statistical significance at the standard p<.05 level.
90
3.4 Conclusion
Though not conclusive, results in this chapter suggest that master frames can
have powerful e↵ects on both support for subsequent policy arguments and on policy
attitudes themselves. However, those e↵ects are not uniform nor are they universal.
are strongly moderated by one’s personality type. People who are less open to new
ter frames and people who are more open appear to be more liberal. At the same
time, both the immigration and CCES tax experiments showed that equality master
frames can cause people to adopt more liberal positions regardless of - or despite of
- personality type.
Furthermore, the e↵ect of master frames in the CCES tax experiment is conse-
quential for policy outcomes because it shows that master frames can either generate
majorities for or against raising taxes. Since the experiment was conducted on a rep-
resentative sample of Americans, these results show that elites can build or destroy
majority policy coalitions simply by mixing certain issue frames with master frames
that summarize key ideological principles. This demonstrates that elites can swing
support for or against important policy arguments – not through issue arguments
alone – but also through separate stage-setting appeals to key ideological values.
Even with these results, some might wonder if opinion change was due to a
“double-framing” e↵ect rather than a master framing e↵ect. However, it’s not likely
that this is the case. A double framing e↵ect would need to mention the specific
policy question in both frames. For instance, two separate frames favoring increased
taxes on the wealthy could each influence tax views, but it’s not clear why a general
91
frame that makes no mention of taxes would have any e↵ect at all. Such a general
frame does not directly frame the issue because it does not mention the issue. On
the other hand, such a general master frame could have an e↵ect if, as hypothesized,
Future research should extend these findings by testing for the moderating e↵ect
future experiments should test whether master frames might a↵ect policy attitudes,
regardless of whether individuals receive a specific policy argument. Finally, the next
chapter will investigate whether repeated master frames increase their e↵ect and
might explain how certain issue frames seem to be naturally stronger than others.
92
Chapter 4: Experimental Test of Repeated Master Frame
Exposure
4.1 Introduction
Experimental evidence from the last chapter suggests that master frames signif-
icantly a↵ect policy views on taxes and immigration, particularly when paired with
extension of those experimental designs. First it replicates the spirit of the CCES
experiment by testing master frame e↵ects on tax policy views using a di↵erent set
of participants. Second, it again tests whether master frame e↵ects are moderated
There were two primary research questions investigated in this chapter: first, just
as with the preceding chapter, I sought to answer whether master frames can a↵ect
policy attitudes by increasing or decreasing the e↵ect of issue frames. Second, since
93
a master frame increases its e↵ect. In addition, multiple scholars have shown that
repeated messages have a persuasive impact (Hasher, Goldstein & Toppino 1977,
an ideologically-congruent master frame will show more support for the position ad-
ments will increase master frame e↵ects relative to only receiving one master frame
treatment as in chapter 3.
Finally, I will also test the two personality moderation hypotheses in chapter three.
will make openness-oriented people more a↵ected by liberal issue frames, whereas
master frames that emphasize the individual will make avoidant personalities more
make less open personalities more likely to support conservative positions and master
frames that emphasize individualism will make openness-oriented people more likely
science classes, and they were o↵ered extra credit for participation. Extra credit was
94
given regardless of whether participants completed the experiment. Participants took
and pretreatment items, randomized treatments and then a measure of support for
After pre-treatment items, participants were first exposed to either repeated equal-
ity, individualism or a control (no master frame) condition. After the master frame
servative tax frame treatment. Participants who received the equality master frame
treatment were asked to read the following statements in succession, each in its each
own individual screen. After each statement, participants were asked to describe
what they just read in their own words to encourage them to pay attention to the
statement.
• America is strongest when we recognize that we are greater together than when
• We are all Americans and it’s our responsibility to help our fellow Americans
vote.
• Our country is at its best when we treat each fellow American the way we would
95
• America can only grow and prosper if all of its people have a fair and equal
• Our country is strongest when everyone gets a fair shot, when everyone does
their fair share, and when everyone plays by the same rules.
read the following six statements in succession and were asked to summarize each
• Our country is strongest when individual entrepreneurs are freed to unleash the
• America is strongest when people don’t get handouts, but instead profit from
• All Americans should accept responsibility for themselves and not expect help if
• American democracy depends on the leadership and guidance of the most tal-
• Being an American means having the freedom to follow your dreams and build
• America can only grow and prosper when individuals and businesses have the
96
4.3.3 Issue Frame Treatments
Finally, those in the control (or, no master frame) condition did not receive either
participants were then randomly assigned to either a liberal tax frame treatment or
Some argue that we should raise taxes on people making more than $250,000
per year because we need the money to pay for essential investments in ed-
ucation, health care and economic development.
And this was the language used in the conservative tax frame:
Some argue that we should not raise taxes on people making $250,000
per year because doing so would be punishing success and that instead we
should encourage the entrepreneurship and risk-taking that drive economic
growth.
The dependent variable for the experiment was whether participants supported
were asked whether they would favor raising taxes on households with incomes above
$250,000. Responses were coded so that opposition to raising taxes on the wealthy
took negative values and support for raising taxes took positive values. Neutral
was exposed to the repeated equality master frame treatment, it was coded “1” if a
97
participant received the equality treatment and “0” if one received the individualism
treatment.
Control – This variable records whether a participant was exposed to the no-
treatment condition. It was coded “1” for those in the control condition and “0” for
Liberal Tax Frame – This dummy variable measures whether a participant was
exposed either to the liberal tax frame, which was coded “1”, or the conservative tax
in the chapter three CCES tax experiment was 30 or under years of age, coded “1”,
for 30 or under, and coded as “0” otherwise. The variable was used to reexamine the
4.4 Results
The first hypothesis was tested by observing the impact experimental conditions
had on participant’s support for taxing the wealthy after receiving either a liberal
In keeping with master frame attitude change and intensity hypotheses, I expected
that exposure to a particular master frame would cause tax attitudes to move in the
ideological direction of that master frame and that the master frame would intensify
the e↵ect of an ideologically-consistent tax frame. For example, the master frame
attitude change hypothesis predicts that exposure to the equality master frame treat-
ment would increase support for taxing the wealthy and the master frame intensity
hypothesis predicts that receiving an equality master frame will increase support even
98
more than simply receiving one master frame treatment and a liberal tax frame by
itself. Experimental e↵ects were estimated using OLS regression using support for
taxing the wealthy as the dependent variable. Table 4.1 reports results of two models,
first the e↵ect of master frame treatments interacted with the tax frame treatment,
and, second, the direct e↵ect of master frame exposure on participants’ support for
demonstrates that exposure to master frames when paired with an ideological tax
frame failed to change participants’ tax views. The interaction between exposure to
the equality master frame treatment and exposure to the liberal tax frame treatment
is not significantly di↵erent from zero, suggesting that neither the individualism nor
the equality master frame treatments produced meaningful changes in tax attitudes
when paired with the liberal tax frame. The equality master frame coefficient is
positive both in model one and in model two, suggesting a positive relationship be-
tween support for raising taxes and receiving the equality master frame treatments
both in the conservative tax frame condition and in combined issue frame conditions.
meaning that tax views among people who received the equality master frame treat-
ment were statistically indistinguishable from those who received the individualism
Figure 4.1 displays means and 95 percent confidence intervals for each experimen-
tal condition. As is readily apparent, none of the master frame conditions produces
tax views that are meaningfully di↵erent from the others. Comparing the control
conditions for the two ideological tax frames, it’s equally clear that there were no
99
Table 4.1: E↵ect of Master Frames on Support for Raising Taxes on the Wealthy
Support for Raising Taxes (1) (2)
Model Model
100
tax frame e↵ects. The equality treatment in the conservative tax frame condition
did appear to produce slightly higher support for taxing the wealthy compared to
participants who received the equality master frame and the liberal tax frame, but
the .44 mean di↵erence was not statistically significant (se = .31, p < .15).
Figure 4.1: Neither Master Frame nor Issue Frame Treatments A↵ect Support for
Raising Taxes
Figure 4.1 displays means and 95 percent confidence intervals for each experimental condi-
tion. Neither issue frames nor master frames appear to have significantly a↵ected support
for taxing the wealthy.
101
Figure 4.2 compares support for taxing the wealthy in di↵erent master frame
(i.e., the “Equality” condition reports the mean support for taxing the wealthy among
people were received the equality master frame including both liberal and conserva-
tive tax frame participants). As in figure 4.1, support for taxing the wealthy was
Figure 4.2: Master Frame Treatments Have No E↵ect, even when Combining Liberal
and Conservative Treatment Groups
Figure 4.2 displays means and 95 percent confidence intervals for combined master frame
conditions.
102
Taken together, these results fail to support either of the first two hypotheses.
Master frame treatments did not change participant’s tax views, much less move
them in the same ideological direction as the master frame treatment. Similarly,
experiment failed to o↵er any support for the master frame theory or its primary
hypotheses.
In table 4.2, I report results testing the personality intensification and the person-
ality backlash hypotheses. Model one estimates the e↵ect of receiving a master frame
treatment compared to the control condition, moderated by the Big Five personality
item openness to experience. (The same item was used to test personality modera-
tion in chapter three). Neither the interaction with the individualism nor the equality
master frame treatment produced significant moderating e↵ects. In short, the e↵ects
experience predisposition.
However, model two does reveal a significant three-way interaction between the
equality master frame treatment, liberal tax frame and openness to experience. In the
liberal tax frame condition, openness to experience does moderate the e↵ect of the
equality master frame treatment, but in the opposite direction predicted by the per-
sonality intensification and backlash hypotheses. If the hypotheses were correct, one
would expect the coefficient for the three-way interaction to be positive, meaning that
when openness to experience increases, the equality master frame treatment drives
higher support for raising taxes. In other words, we should expect that people who
103
Table 4.2: E↵ect of Master Frames on Support for Taxing the Wealthy, Moderated
by Openness to Experience
Support for Raising Taxes (1) (2)
Model Model
Individualism -0.121 -1.128
(1.085) (1.424)
104
public policy that benefits the greater good rather than the individual. Likewise, the
hypotheses predict that people low in openness would backlash against arguments
that emphasize the common good and downplay the role of the individual. This
Figure 4.3: Equality Master Frame Produces More Support for Taxing Wealthy
Among People Who Are Not Open to Experience
Figure 4.3 predicted e↵ects of master frames by subjects’ openness to experience. The shaded
area is the 95 percent confidence interval for the equality master frame treatment. The
pattern in this graph is the opposite of what the personality hypotheses predicted.
105
Figure 4.3 more fully demonstrates the relationship between the master frame
treatments and openness to experience in the liberal tax frame condition. Both the
individualism and control (no master frame) conditions follow the same general pat-
tern. Subjects low in openness oppose raising taxes on the wealthy and those high in
openness to experience support taxing the wealthy. The individualism treatment and
the control treatment groups are not significantly di↵erent from each other, meaning
there was no master frame e↵ect for the individualism group. However, subjects in
the equality master frame reported starkly di↵erent views on raising taxes from the
control group. Specifically, equality master frame subjects who were low in openness
show much stronger support for taxing the wealthy than those in the control condition.
Indeed, the e↵ect of the equality master frame is to nearly reverse support for raising
taxes compared to the individualism and control conditions. Control subjects with
the lowest level of openness strongly opposed raising taxes, but equality treatment
subjects with the lowest level of openness supported raising taxes on the wealthy.
The di↵erences appear starkest among subjects who were low in openness, but only
2.96 percent of subjects scored a two or lower on the openness measure. Nearly ten
percent of subjects scored a 4.5 or higher on the openness measure and therefore –
though the di↵erences are smaller between the control and equality treatments – the
interaction had the greatest practical e↵ect among high openness participants.
The bottom line conclusion from figure three is that the personality intensification
and backlash hypothesis could not be confirmed as they were in chapter three. Indeed,
at least among participants who also received a liberal tax frame, the results directly
106
Taken as a whole, results from this experiment provide no support for the master
frame theory or for the personality moderation predictions. If these were the only
experimental results, one might be comfortable dismissing the theory altogether. Yet,
the CCES tax experiment and the immigration experiment reported in chapter three
imply strong support for the theory. This raises the reasonable question of why the
Three explanations are most compelling. First, it’s certainly possible that the
CCES and immigration experiments’ results were simply statistical artifacts that co-
incided with expectations. After all, even in those experiments not every treatment
produced statistically significant results. On the other hand, the CCES experiment
used nearly three times the number of participants and had more statistical power as
sample, it certainly had greater external validity than either the immigration experi-
ment or the experiment reported in the present chapter. A second explanation is that
the present experiment was not actually a strict replication of the CCES experiment,
but rather a conceptual replication instead. It may have failed to strictly replicate the
or individualism.
Though counterintuitive, the experiment’s master frame repetition may not have
validly reflected how messages are repeated in political campaigns in which the rep-
etition takes places over days rather than a span of minutes. This could be the case
for two reasons. First, by asking participants to read a series of statements, one
right after the other, the experiment may have alerted participants to the purpose
107
of the experiment and encouraged them to consciously stick to their previously held
opinion on tax policy. Similarly, repetition in such a short period of time may have
caused participants to think more critically about the treatment information, making
its influence weaker as a result. In contrast, political advertising that repeats over the
course of a few days might garner advantage from repeated exposure without risking
backlash. A final explanation is that di↵erences between the CCES and the present
Though random assignment ensures that control and treatment groups are roughly
similar on average, it may also be the case that separate experimental groups can dif-
might make them more or less likely to respond to treatment. The most striking dif-
ference between this experiment and the CCES experiment was age. In this chapter’s
experiment, which was conducted among college undergraduate students, more than
the CCES subjects were thirty years old or younger. Because young people are still in
the process of developing a set of political beliefs and, in some cases, systematic ideo-
logical thinking, it is reasonable to believe that master frames would a↵ect them less
because they were less able to comprehend and apply a master frame to subsequent
issue frames. If this is true, master frame e↵ects observed in the CCES experiment
should be attenuated when tested among subjects who were 30 years old or younger.
Table 4.3 reports results of the e↵ect of master frame treatments in the CCES
experiment interacted with a dummy variable that identifies subjects were 30 years
old or below. The interaction e↵ects themselves are not significant and neither were
simple e↵ects. Figure four plots the means and 95 percent confidence intervals for
108
the support for taxing the wealthy by each experimental conditions. As is readily
apparent by visual inspection, almost all the experimental conditions produced the
same support for taxing the wealthy among people 30 or below. The one exception
is the individualism condition among people who also received the liberal tax frame.
Though support for taxing the wealthy is not significantly di↵erent from people in
the equality master frame/liberal tax frame condition, it is statistically di↵erent from
people who only received a liberal tax frame (p < .03, se = .33). This result notwith-
standing, master frame e↵ects among people 30 years and younger were smaller and
less robust compared to the whole sample of experimental participants. Though ex-
ploratory, this suggests that chapter four’s null results were due to the experiment’s
109
Table 4.3: CCES Tax Experiment: E↵ect of Master Frames on Support for Taxing
the Wealthy Among People 30 and under
(1)
Support for Raising Taxes
Control -0.225
(0.266)
Equality 0.0738
(0.235)
Under30 0.296
(0.365)
Constant 0.624⇤⇤⇤
(0.168)
Observations 803
Adjusted R2
OLS Regression. Standard errors in parentheses
⇤
p < 0.05, ⇤⇤ p < 0.01, ⇤⇤⇤ p < 0.001
110
Figure 4.4: When CCES Tax Experiment Was Restricted to Participants 30 and
Below, Master Frames Have Small E↵ect
Figure 4.4 means of CCES tax experiment conditions for just subjects under 30.
Beyond this, it is likely that there were other di↵erences between the CCES pop-
ulation and the experimental group. The experiment was conducted among political
science students at a major public university. It is possible that students were resis-
tant to treatment for two reasons. First, they may have been more aware of ideological
arguments that are similar to the master framing treatments used and may have dis-
counted them because they knew they were intended to have a persuasive e↵ect. It is
also possible that, because the students were taking the experiment for extra credit,
111
students skimmed treatments but didn’t consider them with as much care as survey
respondents did in the CCES. Unfortunately, because of a lack of data, I was not able
The foregoing chapter had three goals: first to provide a conceptual replication
of the tax experiment using CCES data in chapter three, second to test the e↵ect of
repeated master frame treatments and third to test the impact of personality when in
the context of repeated master frame messaging. The results in this chapter did not
replicate the core findings of chapter three, namely that master frames can increase or
decrease subsequent issue frames and that those e↵ects can themselves be enhanced or
mitigated by one’s openness to experience. Neither did this chapter’s results support
the hypothesis that repeated master frame messages would increase the e↵ect of
master frames. But, as already explored, these results are at odds with the first set of
experimental results and with the macro-level master frame e↵ects reported in chapter
two. Which set of results provide the most credible test of the master frame theory
and which should guide our conclusions? I believe the CCES experiment provides
more information. First, the sample size was larger, providing more statistical power.
Second, the CCES experiment was conducted using a representative sample, providing
external validity and overcoming the potential limitation of muted e↵ects among
in the CCES results, the failure to replicate in the present experiment means that
support for the master frame theory is weaker than it first appeared. In the following
112
chapter, I will o↵er concluding thoughts in light of these conflicting results and how
113
Chapter 5: Conclusion
This dissertation has attempted to explain how political elites use general ideo-
logical frames – master frames – to strengthen their policy arguments and win elec-
tions. The theory argues that political elites use these general ideological frames in
order to pull the political discourse in their ideological direction, strengthening con-
gruent policy messages and political candidates as a result. Specifically, the theory
predicts that using ideologically-aligned master frames will strengthen the e↵ect of
ideologically-aligned policy frames on policy views. Further, the theory predicts that
master frames will increase the relevance of ideologically-friendly concepts that set
the stage for political debate, giving advantages to allied candidates. The theory
predicts that political elites use master frames because doing so is an efficient use
of their time and resources. Put simply, it is much cheaper and takes less campaign
time to communicate one general message than multiple specific policy messages.
114
5.2 Review of Results
At the macro level, results in chapter two provide evidence that President Obama’s
fairness master frame increased his vote share, often by more than a percentage
point. Not only did his master frame message fail to e↵ect Romney’s vote share,
campaigns, such vote share swings can mean the di↵erence between winning and
losing an election.
causal e↵ects. To provide causal evidence, chapters three and four both tested master
frames in experimental settings. In general, chapter three also found strong evidence
for master frame e↵ects, particularly when paired with a liberal immigration or tax
frame. In additional, the experiments in chapter three also confirmed that person-
influenced how master frames a↵ected subjects. Equality master frames tended to
sonal orientation. On the other hand, individualism master frames tended to polarize,
causing people to more strongly support the ideological position more aligned with
their disposition.
Unfortunately for the theory, the experiment in chapter four did not replicate the
results found in the first set of experiments. The experiment was not an exact repli-
cation of the CCES tax experiment, so it was possible that null e↵ects were due to the
in the final experiment made them resistant to master frames. Exploratory analysis
115
suggests this may be the case because CCES experiment master frame e↵ects were
Master frame theory is useful to the discipline for three reasons. First, it explains
behavior, and eventually polarization, can arise among a mass public that not as
politically engaged as politicians, advocates and lobbyists. Second, the theory allows
single issue. Campaigns don’t have enough money, candidates don’t have enough
time and the public doesn’t have enough attention to give. Master frames allow
campaigns and activists to efficiently move the discourse in their ideological direction
and strengthen their subsequent arguments as a result. Finally, the theory advocates
studying general frames in addition to issue frames. Though political science has
made great progress by documenting the e↵ect of issue frames on public opinion,
general frames that politicians and their supporters use. This, of course, implies that
in those general frames exist hierarchically above individual issue frames, influencing
116
5.4 Implications for Democracy
What does this mean for democracy? Some could argue that any communication
politically-sophisticated as politicians, lobbyists and other elites, then are they more
easily swayed from what’s in their best interest by compelling - but inherently simple
– messages. The danger of such an outcome is that citizens are removed from steering
the ship of state and are instead only proxies for power.
That need not be the case. All in all, I think the optimistic view is more warranted.
Master frames rooted in ideological values may actually help citizens by making elec-
toral choices clearer and more relevant. In national elections, citizens likely benefit
when campaigns place their arguments in terms of their general vision for the country.
Because the United States lacks the institutional coherence of a parliamentary sys-
tem, consensus must be forged between factions and branches of government. Master
and the presidency, citizens presently have the opportunity to participate in political
parties that are ideologically distinct from each other. While this has undoubtedly
increased partisan warfare, it has also made government action responsive to popular
the public simple guideposts to decide, based on the principles they value, which side
they want to stand on. The normative challenge going forward is to craft avenues
117
for compromise between two coalitions that are increasingly estranged. Nevertheless,
master frames could give everyday citizens a way to act as rational actors without
having to invest the time political elites put into learning the issues.
Additionally, master frames may give citizens avenues to join the political dis-
course with fellow citizens. While this dissertation has focused on how master frame
a↵ect opinion change, one can imagine that master frames help people deliberate with
one another. If they share agreement on individual master frame values, then those
master frames likely help citizens branch out from common agreement on principles
to discussions of specific policy. If they don’t share such agreement, master frames
Although this dissertation dealt with master frames that simplify ideological prin-
ciples, there are certainly many other types of general frames. It is possible that
master frames could have similar e↵ects without necessarily appealing to ideological
principles. Following are several other master frame types which are promising for
study.
personal stories that are meant to exemplify characteristics that the public finds
attractive. Many of those traits, such as strength and integrity, don’t have ideological
connotations. Though it’s not likely that these kind of master frames would influence
118
policy debates, they may very well have an impact on elections by framing the debate
Metaphors: Lako↵ & Johnson (2008) argue that most of our conceptual thinking
(1997) argues that Americans see politics through the metaphor of family, and that
parent” metaphor. To my knowledge, this theory has not been tested empirically and
experiments could test the e↵ect of such master frames on both issue frame strength
ideologies so the e↵ect on policy frames is harder to predict. However, there are
common political metaphors that seem to have implications that could be tested. For
instance, it’s often asserted that “government should be run like a business.” While
it does not appeal directly to ideology, this metaphor implies that government should
do “business-like” things such as make decisions based on the bottom line, efficiency
and frugality. At the same time, adopting such a view precludes other “non-business”
goals such as promoting the common good, (even if it costs more), debt for public
while the primary cleavage in American politics is between liberals and conservatives,
Donald Trump’s presidential campaign raises the possibility of other organizing ide-
ologies, such as right-wing populism or nativism, which more closely mirrors persistent
European right-wing movements. Testing such a master frame is not as clear cut as it
119
is not immediately apparent which issues should naturally align with emerging Amer-
ican populism. But, at the very least, one could imagine testing the e↵ect of Trump’s
“Make America Great Again” and “America First” rhetoric on strengthening specific
Trump’s extraordinary rise and the chaos he’s riven in the Republican party makes
such a study justified. If Trump wins election, he will have a powerful position to
pull the Republican party into a more nativist direction. If Trump does not win the
election, he may still leave behind a potent nativist faction in the Republican Party.
Whether such a movement would split the party and the conservative movement is
unclear. In any case, Trump’s success at distilling his core principles into a compelling
master frame may well determine his movement’s unity and permanence.
Though master frames are expected to increase the e↵ectiveness of policy frames,
nothing in the theory would prevent master frames from having direct e↵ects on
policy views. Just as master frames are predicted to a↵ect the extent to which issue
frames influence one’s previous views, master frames could a↵ect one’s original set of
views or considerations regarding a policy question. This suggests that master frames
would be most e↵ective among people who had previous views regarding a particular
issue, so one would expect that political knowledge would moderate any master frame
e↵ects. The set of experiments presented in this dissertation were not designed to test
120
participants are exposed to master frames and then asked opinions on issues such as
Chapter two explored estimated master frame e↵ects over the course of a year, but
the theory suggests that sustained ideological master frame usage could help explain
even longer term political changes. Though realignment literature is now in disrepute
in some quarters (see Mayhew (2002)) it is still clear that there are di↵erent ideological
eras in American political history. The New Deal Era and the Reagan Revolution are
partly distinct because they seem to have been dominated by di↵erent master frames
regarding the proper use of government power. To investigate such long-term master
frame e↵ects would be a massive undertaking, however. Rather than analyzing one
public official’s speech, one would need to collect a sizable portion of news coverage
through the twentieth century, hand-code for competing master frames, (perhaps
equality vs. individualism), and train machine learning models to code the entire,
However given time and resources, it would help answer whether shifts in general
important to study more deeply how they arise from the attitude formation process.
Several theoretical possibilities present themselves. First, rather than impacting the
weights of views which are then individually stored in memory, it could be that that
121
to a summary attitudes produced in the online process proposed by McGraw, Lodge
& Stroh (1990). If so, e↵ects should be more durable than one might suspect. One
promising way to test this possibility would be to expose subjects to master frames
and measure any e↵ect on opinion change over several weeks. If the master frames
influence long-term attitudes, one would expect that those e↵ects would be sustained
over a several week period compared to a control group. Secondly, one would expect
that those e↵ects would persist even while subjects remember little about the master
Finally, the null results in chapter four underscore the need to test and replicate
the positive results found earlier in the dissertation. Scientific confidence should be
based on how well results hold up over time. While promising, the master frame
findings here should encourage studies to confirm these results as well subsequent
indeed confirmed.
5.6 Conclusion
Democratic politics relies on the competition of ideas, but not all political ideas
deal with specific policy issues. Indeed, even those who know the most about policy
are often more concerned with which fundamental principles frame political debate.
Is America stronger if everyone is treated equally and everyone is “given a fair shot,”
or is it stronger if individuals are expected to make their own success? These views
may be easy to communicate, but they can have profound implications on which
122
The most memorable part of the Declaration of Independence is not a specific
complaint levied against the king. It’s an assertion of “self-evident” truth that leads
to fundamental rights: “We hold these truths to be self-evident, that all men are
created equal, that they are endowed by their Creator with certain unalienable Rights,
that among these are Life, Liberty and the pursuit of Happiness.” (Congress 1776)
master frame about what a proper government should protect and promote. It prob-
around equality and liberty. Both of these values are embedded in the nation’s found-
ing and, though they are not necessarily in conflict, they can lead to implications that
While framing researchers have made great strides in explaining how issue debates
are influenced by news coverage, we can extend and expand that research by studying
the general frames that political elites use to shape discourse. As noted several times
already, there are advantages for political elites to use arguments that are both simple
and appeal to broadly shared principles. Master framing explains how elites can
build coalitions of regular people while influencing policy attitudes and support for
candidates.
While master frames certainly are not the only arrow in a campaign’s quiver,
they are essential for organizing both internal strategy and supporters and allies. As
the results in this dissertation suggest, master frames can have meaningful impacts
on public opinion and candidate support. Master frames are by no means the only
way to influence the public, but I hope this dissertation demonstrates the promise of
123
Appendix A: Immigration and Tax Experimental Treatments
First, subjects are randomly assigned to one of two groups: Individualism master
[If Group B, read the following:] 2. “Many Americans believe our country is strongest
[Liberal Immigration Policy Argument] [If AL, BL, read the following:]
5. “Some argue that the U.S. should provide a path for citizenship for
undocumented immigrants who came here as teenagers or who have not
broken any additional laws and who will pay fines and back taxes. Overall,
do you support or oppose this approach? ”
<1> Strongly Support <2> Slightly Support <3>Slightly Oppose <4>Strongly Op-
124
<1> Strongly Support <2> Slightly Support <3>Slightly Oppose <4>Strongly Op-
Subjects were randomly assigned first to one of two groups: Individualism master
[If Group A read the following:] 1. “Many Americans believe our country
would be strongest if every individual was fully free to pursue their own
self-interest and profit from their own talent and hard work while accepting
responsibility if they don’t work hard enough to be successful. Do you agree
with this point of view?”
[If Group B read the following:] 2. “Many Americans believe our country
is strongest when everybody has a fair shot and plays by the same rules
and that we need to work together as a community so that everyone can
reach their full potential. Do you agree with this point of view?”
5. “Some argue that we should raise taxes on people making more than
$250,000 per year because we need the money to pay for essential in-
vestments in education, health care and economic development. Overall,
do you support or oppose raising taxes on people who make more than
$250,000 per year?”
<1> Strongly Support <2> Slightly Support <3>Slightly Oppose <4>Strongly Op-
6. “Some argue that we should not raise taxes on people making $250,000
per year because doing so would be punishing success and that instead we
should encourage the entrepreneurship and risk-taking that drive economic
growth. Overall, do you support or oppose raising taxes on people who
make more than $250,000 per year?”
125
<1> Strongly Support <2> Slightly Support <3>Slightly Oppose <4>Strongly Op-
126
Bibliography
Adler, Ben. 2011. “GOP Causes S&P Downgrade, but Republican Candidates Blame
Obama.” The Nation .
URL: http://www.thenation.com/article/gop-causes-sp-downgrade-republican-
candidates-blame-obama/
Ajzen, Icek & Martin Fishbein. 1980. Understanding attitudes and predicting social
behaviour. Prentice-Hall.
Aldrich, John H. 1995. Why Parties? The Origin and Transformation of Political
Parties in America. Chicago: The University of Chicago Press.
Axelrod, David. 2015. Believer: My Forty Years in Politics. Penguin Press p. 439.
Begg, Ian Maynard, Ann Anas & Suzanne Farinacci. 1992. “Dissociation of processes
in belief: Source recollection, statement familiarity, and the illusion of truth.”
Journal of Experimental Psychology: General 121(4):446.
Blackwell, Matthew, James Honaker & Gary King. 2015. “A Unified Approach to
Measurement Error and Missing Data Overview and Applications.” Sociological
Methods & Research pp. 1–39.
Brambor, Thomas, William Roberts Clark & Matt Golder. 2006. “Understanding
interaction models: Improving empirical analyses.” Political analysis 14(1):63–
82.
Chong, Dennis & James N. Druckman. 2007a. “Framing Public Opinion in Compet-
itive Democracies.” American Political Science Review 101:637–654.
Chong, Dennis & James N. Druckman. 2007b. “Framing Theory.” Annual Review
Political Science (10):103–126.
Collingwood, Loren & John Wilkerson. 2012. “Tradeo↵s in accuracy and efficiency
in supervised learning methods.” Journal of Information Technology & Politics
9(3):298–318.
127
Congress, US. 1776. “Declaration of independence.” Available in: http://memory.
loc. gov/cgi-bin/ampage .
Converse, Philip E. 1964. The Nature of Belief Systems in Mass Publics. In Ideology
and Discontent, ed. David E. Apter. Ann Arbor: University of Michigan Press.
Cowpertwait, Paul SP & Andrew V Metcalfe. 2009. Introductory time series with R.
Springer Science & Business Media chapter Multivariate Models, pp. 214–216.
de Vreese, Claes. 2004. “The E↵ects of Strategic News on Political Cynicism, Issue
Evaluations, and Policy Support: A Two-Wave Experiment.” Mass Communi-
cation and Society 7(2):191–214.
de Vreese, Claes H., Jochen Peter & Holli A. Semetko. 2001. “Framing politics at the
launch of the Euro: A cross-national comparative study of frames in the news.”
Political communication 18(2):107–122.
Druckman, James N., James H. Kuklinski & Lee Sigelman. 2009. “The unmet po-
tential of interdisciplinary research: Political psychological approaches to voting
and public opinion.” Political Behavior 31(4):485–510.
Druckman, James N. & Kjersten R. Nelson. 2003. “Framing and Deliberation: How
Citizens’ Conversations Limit Elite Influence.” American Journal of Political
Science 47:729–745.
Fishbein, Martin & Icek Ajzen. 1975. Belief, attitude, intention and behavior: An
introduction to theory and research.
Gosling, Samuel D, Peter J Rentfrow & William B Swann. 2003. “A very brief
measure of the Big-Five personality domains.” Journal of Research in personality
37(6):504–528.
128
Greene, William H. 2008b. Econometric Analysis. Pearson chapter Time Series and
Macroeconomics, pp. 751–754.
Hasher, Lynn, David Goldstein & Thomas Toppino. 1977. “Frequency and the con-
ference of referential validity.” Journal of Verbal Learning and Verbal Behavior
16(1):107–112.
Hovland, Carl I, Irving L Janis & Harold H Kelley. 1982. Communication and Per-
suasion: Psychological Studies of Opinion Change. Greenwood Press Reprint.
Huffington Post Pollster. 2012. “2012 General Election: Romney vs. Obama.” The
Huffington Post .
URL: http://elections.huffingtonpost.com/pollster/2012-general-election-
romney-vs-obama
Huffington Post Pollster. 2016a. “Obama Job Approval.” The Huffington Post .
URL: http://elections.huffingtonpost.com/pollster/obama-job-approval
Iyengar, Shanto & Donald R. Kinder. 1987. News thatMatters: Television and Amer-
ican Opinion. Chicago: University ofChicago Press.
Jost, John. 2006. “The End of the End of Ideology.” American Psychologist pp. 651–
670.
Jost, John T, Jack Glaser, Arie W Kruglanski & Frank J Sulloway. 2003. “Political
conservatism as motivated social cognition.” Psychological bulletin 129(3):339.
Jurka, Timothy P., Loren Collingwood, Amber E. Boydstun, Emiliano Grossman &
Wouter van Atteveldt. 2012. “RTextTools: Automatic Text Classification via
Supervised Learning. R package version 1.3.9.”.
URL: http://CRAN.R-project.org/package=RTextTools
Keele, Luke & Nathan J Kelly. 2006. “Dynamic models for dynamic theories: The
ins and outs of lagged dependent variables.” Political analysis 14(2):186–205.
Kellstedt, Paul M. 2000. “Media framing and the dynamics of racial policy prefer-
ences.” American Journal of Political Science pp. 245–260.
Kinder, Donald R. 1998. Opinion and Action in the Realm of Politics. In Handbook of
Social Psychology, ed. D. T. Gilbert, S. T. Fiske & G. Lindzey. 4th ed. London:
Oxford University Press.
129
Krosnick, Jon A & Donald R Kinder. 1990. “Altering the foundations of support for
the president through priming.” American Political Science Review 84(02):497–
512.
Lako↵, George. 1997. “Moral politics: What conservatives know that liberals don’t.”.
Lako↵, George & Mark Johnson. 2008. Metaphors we live by. University of Chicago
press.
Landis, J Richard & Gary G Koch. 1977. “The measurement of observer agreement
for categorical data.” biometrics pp. 159–174.
Leip, David. 2012. “2012 Presidential General Election Results.” David Leip’s Election
Atlas .
URL: http://uselectionatlas.org/RESULTS/national.php
Lodge, Milton & Charles S. Taber. 2013. The rationalizing voter. Cambridge Univer-
sity Press.
McClosky, Herbert & John R. Zaller. 1984. The American Ethos: Public Attitudes
Toward Capitalism and Democracy. Cambridge, MA: Harvard University Press.
McCombs, M.E. & D. Shaw. 1972. “The Agenda-Setting Function of the Mass Media.”
Public Opinion Quarterly 36:172–187.
McGraw, Kathleen M. 2006. How and Why Psychology Matters. Oxford University
Press chapter 7, pp. 131–151.
McGraw, Kathleen M, Milton Lodge & Patrick Stroh. 1990. “On-line processing in
candidate evaluation: The e↵ects of issue order, issue importance, and sophisti-
cation.” Political Behavior 12(1):41–58.
Monogan, James E. 2015. Political Analysis Using R. Springer chapter Time Series
Analysis, p. 159.
Nelson, T.E., Z.M. Oxley & R. Clawson. 1997. “Toward a Theory of Framing E↵ects.”
Political Behavior 19:221–246.
130
Nelson, Thomas E. & Donald R. Kinder. 1996. “Issue frames and group-centrism in
American public opinion.” Journal of Politics 58:1055–1078.
Nyhan, Brendan & Jason Reifler. 2010. “When corrections fail: The persistence of
political misperceptions.” Political Behavior 32(2):303–330.
Page, Benjamin I. & Robert Y. Shapiro. 1992. The Rational Public: Fifty Years of
Trends in Americans’ Policy Preferences. Chicago: University of Chicago Press.
Poole, Keith T. & H. Rosenthal. 2007. Ideology and Congress. Transaction Publishers,
New Brunswick, New Jersey.
Poole, Keith T. & Howard Rosenthal. 1984. “The Polarization of American Politics.”
Journal of Politics 46:1061–1079.
Reagan, Ronald. 1981. “Reagan’s First Inaugural: “Government is not the solution
to our problem; government is the problem.”.”.
URL: http://www.heritage.org/initiatives/first-principles/primary-
sources/reagans-first-inaugural-government-is-not-the-solution-to-our-problem-
government-is-the-problem
Schuldt, Jonathon P., Sara H. Konrath & Norbert Schwarz. 2011. ““Global warm-
ing” or “climate change”? Whether the planet is warming depends on question
wording.” Public Opinion Quarterly .
Sides, John & Lynn Vavreck. 2013. The Gamble: Choice and Chance in the 2012
Presidential Election. Princ chapter Chapter 7: The Winning Hand.
Slothuus, Rune & Claes H. de Vreese. 2010. “Political Parties, Motivated Reasoning,
and Issue Framing E↵ects.” The Journal of Politics 72(3):630–645.
Snow, D.A. & R.D. Benford. 1992. Fronteirs in Social Movement Theory. Yale
University Press, New Haven chapter Master Frames and Cycles of Protest,
pp. 133–155.
Stimson, James A. 1999. Public Opinion in America: Moods, Cycles, and Swings.
2nd ed. Boulder, CO: Westview Press.
131
Stimson, James A., Michael B. MacKuen & Robert S. Erikson. 1995. “Dynamic
Representation.” American Political Science Review 89:543–565.
Tversky, A & D Kahneman. 1981. “The framing of decisions and the psychology of
choice.” Science (New York, NY) 211(4481):453.
Zaller, John R. 1992. The Nature and Origins of Mass Opinion. New York: Cambridge
University Press.
132