Content uploaded by John Victor Kane
Author content
All content in this area was uploaded by John Victor Kane on Apr 23, 2019
Content may be subject to copyright.
Final draft accepted for publication at Pubic Opinion Quarterly (04/05/19).
(May undergo slight editing before official publication.)
Enemy or Ally?
Elites, Base Relations, and Partisanship in America
John V. Kane
New York University
Center for Global Affairs
15 Barclay Street
New York, NY 10007
212-992-3326
John.Kane@nyu.edu
Running Header:
“Enemy or Ally?”
Word Count: 6,495 (excluding table/figure placeholders)
Keywords: Partisan polarization; Coalitions; Cues; Trust; Compromise; Motivated Reasoning
Author Note:
JOHN V. KANE is an Assistant Professor in the Center for Global Affairs at New York University,
New York, NY, USA. The author wishes to sincerely thank Jason Barabas, Stanley Feldman,
Leonie Huddy, David Jones, Chuck Taber, Milton Lodge, Lily Mason, Marc Hetherington, Yanna
Krupnikov, John Barry Ryan, Andy Delton, Jennifer Jerit, Chris Wlezien, Yamil R. Velez,
Benjamin J. Newman and attendees of the Identity Politics Research Group meeting, Tom Hayes
and the members of the UConn Political Economy Workshop for extremely helpful comments on
presentations and earlier drafts of the manuscript. The author also wishes to thank the editor and
anonymous reviewers for their valuable insights, guidance and support. *Address correspondence
to John V. Kane, Center for Global Affairs, New York University, 15 Barclay Street, New York,
NY 10007, USA; email: jvk221@nyu.edu.
ABSTRACT
A wealth of research demonstrates that partisans dismiss information that challenges their attitudes
toward political elites, especially when citizens are aware of these elites’ party membership.
Relatively little is known, however, about the conditions under which partisans will adjust their
support for such elites. Drawing upon research on the group-foundations of partisanship, I
hypothesize that, issues and policy stances aside, partisans’ support for, and willingness to
compromise with, a given elite is contingent upon how well the elite relates to the groups associated
with his or her party (i.e., the party’s “base”). In short, partisans should be inclined to exhibit
greater political support for, and greater willingness to compromise with, the enemy of their socio-
political enemies, but less support for the enemy of their socio-political allies. Findings from survey
experiments and observational analyses involving real-world executives offer strong empirical
support for these contentions. Thus, while acknowledging the powerful effects of cues involving
elites’ party labels, this study reveals that “base relations” cues can potentially counteract the
motivated reasoning processes that arise from partisans’ attentiveness to party cues alone. More
broadly, this study advances our understanding of polarization by demonstrating an important way
in which politically-aligned social groups underpin American partisanship and public opinion.
1
“Judge me by the enemies I have made.”
—Franklin Delano Roosevelt, 1932
Trends in how partisans think, feel, and behave suggest that contemporary partisanship is unlike
anything observed in decades past. Notably, the sharp rise in affective polarization indicates that
partisans increasingly dislike members of the opposing party (Iyengar, Sood, and Lelkes 2012; Mason
2018), distrust elites from the opposing party (Hetherington and Rudolph 2015), and resist deliberating
and compromising with “the other side” (Gutmann and Thompson 2010; Levendusky 2013; Strickler
2018). These trends persist, at least in part, because information about Democratic or Republican elites
that challenges partisans’ existing views is habitually dismissed (e.g., Ramirez and Erickson 2014), and
may even serve to further polarize attitudes (Taber and Lodge 2006). Put simply, partisans “seek to
reach conclusions that reinforce existing loyalties” (Flynn, Nyhan, and Reifler 2017, 133). Indeed,
because new information is typically of little consequence for changing partisans’ attitudes toward
political elites, commentators now regularly lament that “nothing matters” (e.g., Burns, Martin, and
Flegenheimer 2018).
Given these trends, the following questions have become all the more pressing: Under what
conditions, if any, will partisans meaningfully adjust their attitudes toward political elites? Are there
circumstances in which partisans will be more supportive of an outparty elite? Are partisans ever
capable of turning against an inparty elite? Because citizens are typically supplied with party cues (i.e.,
information that attaches the Democratic/Republican party labels to elites), partisans largely rely upon
these cues to evaluate policies and elites, and generally resist adjusting these attitudes (Cohen 2003;
Bolsen, Druckman, and Cook 2014; cf. Bullock 2011). However, drawing upon literature on the group-
foundations of partisanship, I theorize that elites’ reported relations with the key social groups that
constitute the party’s coalition (or, “base”) serve as a powerful cue to partisans—a cue that is capable of
transcending the “perceptual screen” that becomes activated when party cues are supplied (Campbell et
2
al. 1960). These base relations cues, in other words, can change partisans’ attitudes toward elites, even
when the party label of the elite is known. Such information should be efficacious, I argue, precisely
because partisan loyalties are themselves shaped, at least in part, by feelings toward the types of people
perceived to associate with each party (Miller, Wlezien, and Hildreth 1991; Green, Palmquist, and
Schickler 2002; Achen and Bartels 2016; Mason 2016). In short, partisans should be inclined to look
favorably upon the enemy of their socio-political enemies, and come to resent the enemy of their socio-
political allies.
To investigate these propositions, I focus on a particular set of political elites operating in a
relatively under-studied context. Specifically, I examine situations in which executives (i.e., state
governors and the president) have alienated, rather than appeased, their party’s political base. Except
perhaps in the rare instances in which one party commands complete control over the legislative and
executive branches, executives must regularly bargain, decide which agendas to pursue or abandon, and
actively locate and exploit political leverage in an effort to enact legislation (Aberbach and Rockman
1999; Edwards 2012; Kousser and Phillips 2012; Rosenthal 2012; Redlawsk 2015). This renders
executives especially vulnerable to disappointing key groups in their political base and, thus, ideal to
examine given the propositions outlined above. With executives’ base relations as my focus, I first
report results from survey experiments that manipulated whether respondents’ governors reportedly
appeased or alienated groups in their party’s base. In a second set of analyses, I investigate whether
partisans adjust their approval of elites following real-world events in which these executives were
reported to have displeased their respective party’s base.
Across each of these studies, the results demonstrate that partisans’ political support for elites
significantly depends upon these elites’ base relations. Specifically, when an elite alienates—rather than
appeases—key groups in his or her party, he or she enjoys greater political support from members of the
3
outparty, and lower support from inparty members. Such information also significantly affects
partisans’ willingness to support political compromise.
In revealing partisans’ sensitivity to how elites relate to key groups in their party’s coalition, this
study demonstrates that partisans are capable of transcending inclinations to rely solely upon party cues
and/or disregard challenging, politically incongruent information. Further, the results build directly upon
a growing literature which contends that the social-group characteristics of the parties underpin the
powerful influence of party identification on political behavior (Achen and Bartels 2016; Kinder and
Kalmoe 2017; Ahler and Sood 2018; Mason 2018).
PARTISANSHIP & ITS DISCONTENTED
Partisans in the American public prefer members of their own party over members of the
opposing party (e.g., Lelkes and Westwood 2016), though their antipathy toward members of the
opposing party has increased sharply (Iyengar, Sood, and Lelkes 2012). Examples of this partisan
animosity abound. Iyengar and Westwood (2015), for example, find that partisans’ discrimination
against outparty members exceeded that of race-based discrimination, while Abramowitz and Webster
(2016) find that loyalty to one’s own party in the voting booth is most strongly related to antipathy
toward the opposing party. Scholars have also documented similar patterns in partisan trust, with
outparty elites increasingly distrusted relative to inparty elites (Hetherington and Rudolph 2015).
Perhaps unsurprisingly, we also observe a formidable resistance to compromise among partisans
(Gutmann and Thompson 2010; Cillizza 2013), particularly on matters in which party elites and specific
policies are involved (Harbridge and Malhotra 2011; Harbridge, Malhotra, and Harrison 2014).
Furthermore, the literatures on party cues and motivated reasoning depict citizens who are
inclined to rely largely upon party labels when evaluating a political target (Cohen 2003; Bolsen,
Druckman, and Cook 2014), actively seek out information that reinforces their partisan loyalties, and
4
argue against information that challenges these convictions (Flynn, Nyhan, and Reifler 2017). Even
when presented with factual information, partisans resist becoming more supportive of outparty elites
and/or resist turning against inparty elites (e.g., Ramirez and Erickson 2014), and may even grow more
polarized after receiving such information (e.g., Taber and Lodge 2006; Flynn, Nyhan, and Reifler
2017). In essence, such findings suggest that partisans are inclined to dismiss information that fails to
accord with what they want to believe about particular elites. Indeed, this point echoes an argument
advanced decades ago by the authors of The American Voter, which contended that partisans regularly
succumb to approaching politics with a “perceptual screen” (Campbell et al. 1960).
Are there any conditions under which partisans might, in essence, cease to act in a partisan
fashion? Though research on this question is limited, there indeed exists some scholarly skepticism that
no amount of incongruent information can succeed in changing partisans’ minds. Redlawsk, Civettini,
and Emmerson (2010), for example, find evidence for an “affective tipping point”—i.e., a discernible
point at which partisans, in light of mounting disconfirming evidence about a preferred political elite,
begin updating evaluations in a more “rational” fashion. Green et al. (2002) similarly take issue with the
notion that partisans are incapable of updating attitudes about political targets in an unbiased fashion,
and present evidence supporting their argument that partisan identification should be treated as separate
from evaluation of partisan elites.
Such findings, though few in number, reveal that partisans are in fact capable of updating their
beliefs in ways that do not simply reinforce existing loyalties. But, in light of the pervasiveness of
partisan polarization and its debilitating consequences for governance, it is crucial to determine whether
other such conditions exist and, if so, develop a stronger theoretical foundation for why such conditions
should be consequential for partisans’ beliefs. With such ends in mind, I submit that the various
literatures on American partisanship, while rife with evidence demonstrating the efficaciousness of party
5
cues, stand to benefit from an investigation of whether partisans will adjust evaluations of elites in
response to how elites relate to their respective party’s coalition, or “base.”
PARTY COALITIONS & BASE RELATIONS
Scholars have long observed that political parties in the U.S. represent a coalition of various
groups, factions and interests (Chambers 1963; Lipset 1978). Indeed, multiple groups in society (e.g.,
racial, religious, social, etc.) are, and are widely perceived to be, disproportionately sorted into one of the
two dominant political parties (Miller, Wlezien, and Hildreth 1991; Popkin 1994; Green, Palmquist, and
Schickler 2002; Levendusky 2009; Newport 2009; Chambers, Schlenker, and Collisson 2013; Mason
2016). The groups that are highly aligned with a party (e.g., in terms of voting, identification, etc.) can
be said to constitute that party’s base—i.e., its core groups that leaders of the party must often be
concerned with appeasing (e.g., Petrocik 1996, 828–29; Hillygus and Shields 2009; Newport 2009).
The notion that parties are seen by members of the public as “coalitions of groups” (Achen and
Bartels 2016, 266) has important implications for citizens’ political attitudes and behavior. Indeed, in
order to make sense of the political world, what many citizens require is a clear signal communicating
how a party or candidate relates to such groups (Converse 1964, 14). Scholars have long noted this
“group-centric” dimension of American public opinion, wherein citizens evaluate political matters with
a keen attentiveness to how various groups stand to be affected (Nelson and Kinder 1996).
Perhaps most importantly, much scholarship contends that attitudes toward various politically-
aligned groups, in part, structure citizens’ own partisan loyalties. Citizens, in other words, will gravitate
toward the party that represents groups they feel positively toward, and/or away from the party that
represents groups they feel negatively toward. As Green, Palmquist, and Schickler (2002, 109) contend,
“Partisan identities in adults typically persist because group stereotypes persist, and the location of the
self amid various social groups persists.” Thus, citizens’ affect toward politically-aligned groups can
6
help inform, structure, and crystallize their attitudes toward each party, as well as their evaluations of
each party’s elites (Miller, Wlezien, and Hildreth 1991; Miller and Wlezien 1993).
In this vein, a recent study by Campbell et al. (2011) found that a politician who identifies as an
Evangelical Christian—versus as a member of a less politically-aligned religious group (e.g.,
Catholics)—attracts more support from Republicans (and more opposition from Democrats), and that
this cue was, in some cases, capable of neutralizing the effect of a party cue (i.e., whether the politician
was labeled a Republican or Democrat) (see also McDermott 2006). Though somewhat narrow in
scope, this finding suggests a larger possibility: relations between an elite and his or her party—and,
more specifically, groups in his or her party’s base—may communicate the extent to which the elite is
beholden to the kinds of people that inform citizens’ conceptualizations and evaluations of the parties.
Such “base relations,” in other words, should cue an elite’s fidelity to the very groups that endow the
party label with its symbolic meaning. In a mass public that is sharply polarized along partisan lines,
base relations may therefore inform partisans about the extent to which an elite, Democrat or
Republican, should be regarded as an enemy or as an ally.
Importantly, relations between elites and their respective party’s base vary considerably.
Particularly during the normal course of governing (i.e., in the interims between elections (e.g.,
Chambers 1963, 46)), such relations are often not harmonious. Instances in which executives, for
example, “alienate their base”—i.e., disappoint a particular group that is aligned with the executive’s
party—abound in the media at the national, state and local level. President Obama, to take one
prominent example, was frequently criticized for alienating the base of the Democratic Party throughout
his presidency.1 Perhaps largely stemming from the fact that the U.S. has (effectively) only two parties,
the alienation of one or more key groups within each party’s coalition may be inevitable (Popkin 1994,
1 For instructive examples, see Chapman 2010; Condon 2010; Meckler and Weisman 2010; James 2011;
Halloran 2013; Wolfgang 2013; Wyler 2013; Wheaton 2015)
7
53). By way of contrast, other executives—including President Trump—have been depicted as more
closely aligned with the base of their party (Baker 2017; The Daily Beast 2017; Hohmann 2018). In
fact, Trump has explicitly invoked the wishes of his base when defending his policy stances (Phillips
and Clement 2017).
Therefore, while much literature suggests that partisans generally respond to new information in
ways that reinforce existing loyalties, it stands to reason that when a Democratic or Republican elite
alienates (rather than appeases) his or her party, he or she should attract greater support from outparty
members. For example, compared to when a Democratic governor appears to be strongly aligned with
groups in the Democratic Party base (e.g., African-Americans, environmentalists, etc.), a Democratic
governor who alienates these groups should be perceived as less loyal to the party and, therefore, receive
more support from Republicans, despite continued possession of the Democratic Party label. Stated
more broadly, an elite who displeases one’s “enemies” in the outparty should begin to look more like a
“friend,” and therefore enjoy greater political support. 2
H1. “Enemy of My Enemies” Hypothesis: Partisans should increase support for an outparty
elite as the elite appears less aligned with his or her party’s base.
Of course, such actions by an elite may also have consequences among partisans within the
elite’s party. Because inparty members are likely to value fidelity to the groups within the party
(Marques and Yzerbyt 1988; Roccas and Brewer 2002), we should also expect that elites who are
willing to alienate groups in their own party’s base will be perceived by inparty members as less loyal to
the party. Common political epithets such as “RINO” and “DINO” (Republican/Democrat “in name
only,” respectively) can be said to reflect such distrust of inparty elites (e.g., see Reuters 2014; Martin
2 This expectation is also consistent with psychological research demonstrating individuals’ need to achieve
psychological “balance” in response to new information about how multiple objects are related (Heider
1946). For example, a Republican may look upon Democrats negatively, and thus a Democratic governor
negatively as well. But if this governor alienates other Democrats, the Republican will likely be cognitively
motivated to view the governor more positively.
8
2015). Such a perception among inparty members should therefore predict lower political support versus
if the elite had sought to appease the party’s base.3
H2. “Enemy of My Allies” Hypothesis: Partisans should reduce support for an inparty elite
as the elite appears less aligned with his or her party’s base.
Lastly, a well-known consequence of affective political polarization is a rigid unwillingness to
compromise with members of the outparty (Gutmann and Thompson 2010; Harbridge, Malhotra, and
Harrison 2014). Might an elite’s base relations also affect partisans’ willingness to support bipartisan
policy compromise? Relying upon the reasoning outlined above, partisans should also be more
amenable to compromising with an outparty elite who is willing to display disloyalty to his or her party.
H3. “Bipartisan Compromise” Hypothesis: Partisans should be more willing to support
compromise with an outparty elite as the elite appears less aligned with his or her party’s base.
While the hypotheses apply to all partisans, it is worth noting that there exists some reason to
expect asymmetric effects between partisan groups. For example, Republicans may be
disproportionately less willing to negotiate with a Democrat than Democrats a Republican (e.g.,
Grossmann and Hopkins 2016, 50–53). In addition, Grossman and Hopkins (2016, 37) argue that
Republicans are relatively more concerned about an “abstract conflict over the proper role of
government,” and are less group-centric, compared to Democrats. Such findings imply that Republicans
may be less influenced by base relations cues. On the other hand, Mason and Wronski (2018, 257) argue
that, because the Republican Party is relatively more homogenous in terms of social identities,
“Republicans are more susceptible to identity-based politics,” which could include base relations. Thus,
while the analyses primarily aim to test the hypotheses for partisans in general, given these competing
implications, we can also investigate whether such partisan asymmetries exist.
3 Given the “negative partisanship” argument (Abramowitz and Webster 2018)—which contends that
outparty animus, more so than inparty favoritism, drives partisan behavior—it may be the case that empirical
support for H2 is stronger than support for H1. This possibility is also investigated below.
9
In sum, tests of these hypotheses stand to provide important insights into the powerful role that
politically-aligned groups continue to play in shaping American partisanship, as well as how base
relations may counteract partisans’ inclinations to rely primarily upon elites’ party labels when
processing political information and forming political attitudes.
DATA AND ANALYSES
The present section is organized as follows: I first discuss survey experiments that featured real-
world state governors, and that manipulated how these governors were reported to relate to their
respective political base. I then discuss an analysis of a real-world event involving President Obama. In
each case, the key independent variable of interest is whether the featured executive appeases or
alienates his or her party and, more specifically, prominent groups in his or her party’s base.
Survey Experiments
To test the aforementioned hypotheses, two survey experiments were conducted from 2016-
2017.4 The first study (Qualtrics) included a sample of 1,002 American adult citizens, and was collected
online by Qualtrics.5 In this study, quotas were used to ensure that the sample would be nationally
representative on age, race/ethnicity, and geographic region. The second study (MTurk) included a total
of 1,005 adult U.S. citizens collected via Amazon’s Mechnical Turk. (See Supplemental Appendix for
sample information.) Recent research finds that experimental treatment effects found in MTurk samples
are comparable to those found in studies using nationally representative samples (Mullinix et al. 2015).
4 An additional MTurk experiment, which was fielded in 2015 and featured a similar design, yielded the
same pattern of results as reported below (see Supplemental Online Appendix B).
5 Details regarding Qualtrics’s sampling procedures are located at http://success.qualtrics.com/rs/qual-
trics/images/ESOMAR%2028%202014.pdf
10
Experimental Design
Apart from the outcome measures, the designs of the Qualtrics and MTurk experiments are
identical. Both experiments feature a partisan governor who, in proposing a particular change to the
state budget, either appeases or alienates groups in his or her party’s base. A key objective was to have
citizens evaluate an executive with whom they were already familiar. To accomplish this, respondents
indicated their home state at the start of the survey. The survey was programmed such that this
information allowed for identification of each respondent’s own governor, as well as whether the
respondent lived in a state headed by a Republican or Democratic governor. Then, “blocking” on the
latter variable (see Gerber and Green 2012), respondents were randomly assigned to one of two
conditions which referenced the respondent’s own state governor by name.
In each study, the two conditions informed respondents that their governor would soon release a
budget proposal for the coming fiscal year. This budget proposal was said to “closely mirror” the budget
of the previous year, but would also “decrease funding for the state’s courts and detention systems.” In
taking this position (which did not vary across experimental conditions), the governor is reported to
have either pleased specific groups that are aligned with the governor’s party (Base Appeasement), or to
have displeased these groups (Base Alienation). Table 1 features the text from these vignettes.
[Table 1 about here]
It is worth stressing that the vignettes were designed to circumvent directly policy-based (i.e.,
spatial voting) considerations. Again, both the policy, as well as the governor’s stance on that policy,
were held constant across conditions. Thus, across the two conditions, the governor cannot be
perceived as taking a position closer to, or farther away from, the respondent’s own stance on that issue.
While policy-based considerations may play an indirect role (explored below), we can therefore rule out
the possibility that any observed differences in outcomes between the Base Appeasement and Base
11
Alienation groups are due to differences in how respondents viewed their governor’s stance on funding
courts and detention systems relative to their own.
Measures
The independent variable is a binary indicator of whether respondents were assigned to the Base
Appeasement condition (=0), or the Base Alienation condition (=1). The primary dependent variables
for testing H1 and H2 are as follows: Trust (Qualtrics study), which captures each respondent’s
perception of his/her governor’s trustworthiness on a seven-point scale6; and, Vote Governor (MTurk
study), which indicates respondents’ likelihood of voting for their state governor in the next election
(also a seven-point scale, with higher values indicating greater likelihood).
To test the “Bipartisan Compromise” hypothesis (H3), respondents in the Qualtrics study were
asked how willing they would be to “support policy compromise” between their governor and the
opposing party. Responses were measured on a five-point scale ranging from “Unwilling” (=1) to
“Willing” (=5). In the MTurk study, respondents were asked if they would support a budget measure
that would cut social welfare spending and raise taxes on the wealthy. Given Republicans’ general
opposition to tax increases (even on wealthy individuals), and Democrats’ support for social welfare
spending, this proposal reflects a policy compromise, and one that is similar to those proposed by
governors to shore up state budget shortfalls (e.g., Tims 2016). Additionally, respondents were asked,
“Suppose your representative in the state legislature were to compromise with [respondent’s governor].
How likely is it that you would vote for this representative in the next election?” This question is
designed to extend beyond the aforementioned measures of compromise by capturing willingness to
support a representative who compromises with the governor. Both of these questions were measured
6 As Hetherington (2005, 51) argues, “Other things equal, if people perceive the architect of policies as
untrustworthy, they will reject its policies; if they consider it trustworthy, they will be more inclined to embrace
them.”
12
on seven-point scales, with higher values indicating greater support. Table 2 provides an overview of
the key details in each study.
[Table 2 about here]
Experimental Results for H1 & H2
Does the manner in which a governor relates to his or her party, even on issues with little policy
substance, affect the extent to which the governor is supported by partisan citizens? For the “Enemy of
My Enemies” hypothesis (H1) to find empirical support, average perceived trustworthiness of the
outparty governor should significantly increase as we move from the Base Appeasement condition to the
Base Alienation condition.7 Indeed, as shown in Figure 1, partisans with an outparty state governor were
significantly more trusting of their governor when the latter alienated, rather than appeased, key groups
in his or her party’s coalition. This effect was on the order of 7 percentage points (20%), and attained
statistical significance (p<.001).8
[Figure 1 about here]
Figure 1 also provides strong support for the “Enemy of My Allies” hypothesis (H2). When
inparty governors alienated groups in their respective party’s base, partisans perceived these elites to be
significantly less trustworthy. In this case, the effect size was nearly identical to that found for outparty
governors—a 7 percentage-point (or, 10%) decrease (p<.001). Thus, partisans significantly adjusted
perceptions of their state governors, bestowing upon outparty governors greater trust when they
perceived the governor to be less loyal to the outparty’s base, and reducing trust in their inparty
governor upon learning that the latter is disloyal to the inparty’s base.
7 Consistent with previous research, respondents who reported leaning toward a party were coded as
partisans (see Hawkins and Nosek 2012; Petrocik 2009). The results do not substantially differ when
“leaners” are excluded.
8 Because groups of respondents will, of course, share the same governor, significance tests for all
experimental analyses were conducted both without and with standard errors clustered by state. No
substantial differences were found.
13
In the MTurk study, the experiment examined whether base relations affect partisans’ likelihood
of voting for their governor. The results of this analysis are featured in Figure 2. First, as can be seen
among respondents with an outparty governor, moving from the Base Appeasement condition to the
Base Alienation condition predicts a significant increase in the likelihood of voting for one’s outparty
governor (p<.05). And while this effect is modest in size (a .38 increase, or 15%), it is notable that any
positive effect was found given partisans’ general unwillingness to cast a vote for outparty elites (e.g.,
Abramowitz and Webster 2016). Conversely, among respondents with an inparty governor, the
likelihood of voting for one’s governor declined by .70 (or, 14%) when respondents were told that the
governor alienated (vis-à-vis appeased) his or her party’s coalition (p<.001). In contrast to the previous
analysis, the treatment effect was notably larger for respondents evaluating an inparty (vis-à-vis
outparty) governor (difference = .31, p=.01).
[Figure 2 about here]
With regard to H2, one potential concern is that, because the studies featured specific social
groups in the experimental vignettes, it is respondents’ membership in these social groups—and not
their membership in the Democratic or Republican parties—that is actually driving the results. For
example, because the Democratic governors were said to have alienated African-Americans, perhaps it
is only African-American Democrats—rather than Democrats in general—who were responsive to the
treatment. This possibility was tested by specifying a series of interactive regression models across each
experiment, each one estimating treatment effects among groups that 1) were in some way referenced in
one of the experimental vignettes, and 2) could be identified using responses to pre-treatment questions
in the surveys. The results of these analyses are featured in the Supplemental Appendix. Taken
together, the results do not indicate that the treatment effects were primarily driven by particular social
group memberships rather than party membership: co-partisans generally responded to the experimental
14
treatments in a similar fashion, regardless of their own particular social group memberships. These
findings therefore offer additional evidence for H2.
In sum, consistent support was found for the “Enemy of My Enemies” and “Enemy of My
Allies” hypotheses: Partisans are substantially more supportive of an outparty elite when the elite
displays disloyalty—rather than loyalty—to his or her party’s base, but partisans are consistently less
supportive of an inparty elite who exhibits such disloyalty.
What Did the Cues Communicate? Investigating Mechanisms
The experimental results offer consistent evidence that cues involving executives’ relations with
their party’s base affect how these executives are evaluated by partisans. But what, exactly, is
communicated by such cues? At first glance, one possibility is that the treatment effects are largely
driven not by partisan considerations, but by virtue of the governors acting in a counter-stereotypical—
rather than stereotypical—fashion. However, this suggestion cannot account for why the direction of
the treatment effects depends upon whether the governor is of a respondent’s inparty or outparty—i.e.,
why inparty members punish counter-stereotypical behavior, but outparty members reward it. Setting
this possibility aside, then, theoretically, it seems likely that the cues transmit a mixture of at least two
pieces of compelling information about the executive. First, base relations cues should communicate
that the executive is, in a general sense, willing to demonstrate disloyalty toward, and risk displeasing,
typical members of his or her party—i.e., the very kinds of people that inform citizens’
conceptualizations of the party (e.g., Green, Palmquist, and Schickler 2002). Second, base relations may
also indirectly signal an elite’s ideological position.
15
A series of additional analyses, including a follow-up MTurk experiment featuring two
subjective manipulation checks9 regarding the governor’s perceived relations with his/her party, were
conducted in an effort to better identify the mechanism(s) underlying these experimental results. Due to
spatial constraints, the details of these analyses appear in the Supplemental Appendices. Overall,
however, I find that, while the experimental cues likely communicated a mixture of important
information about the executive, the weight of the evidence suggests that considerations about the
executive’s ideology likely played a relatively minor role in influencing changes in support compared to
the more symbolic considerations involving elites’ base relations.10
Support for Bipartisan Compromise
Beginning with the Qualtrics data, we can also test whether executives’ base relations are
consequential for partisans’ support for compromise with an outparty executive (H3). Figure 3 displays
the difference in mean support for policy compromise (recoded to range from 0 to 1) when moving from
Base Appeasement to Base Alienation. Similar to the previous analyses, we observe greater willingness
to support bipartisan compromise in the Base Alienation condition (in which the governor is perceived
as less associated with groups in his or her party’s coalition) than in the Base Appeasement condition.
This effect was on the order of 6.4 percentage points (11%), and attained statistical significance (p<.05).
Thus, the governor’s willingness to alienate groups in his or her party’s base appears to have garnered
greater support for bipartisan policy compromise from members of the outparty, providing initial
support for H3.
[Figure 3 about here]
9 See Kane and Barabas (2019) for an overview of different types of manipulation checks.
10 For example, only in the case of Democratic governors did respondents perceive a significant, though
substantively small, change in their governor’s position on the 7-point ideology scale across the two
experimental conditions.
16
As noted above, the MTurk study featured two alternative measures of support for compromise:
(1) likelihood of voting for a representative in the state legislature who compromised with the
respondent’s governor, and (2) support for the governor’s budget proposal compromise (i.e., cutting
social spending and increasing taxes on the wealthy). The results of this study are displayed in Figure 4.
The left panel of the figure indicates that moving from the Base Appeasement condition to the Base
Alienation condition again predicts a significant increase (approximately 11%) in willingness to vote for
a legislator who compromises with the respondent’s governor (p<.01). Next, the right panel of Figure 4
demonstrates that willingness to support the governor’s policy compromise also significantly increased
(by approximately 17%) when the governor was portrayed as less aligned with his or her party’s base.
This effect attained statistical significance (p<.001) and, like the left panel of Figure 4, provides further
support for H3.
[Figure 4 about here]
Taken together, the Qualtrics and MTurk studies offer substantial support for the “Bipartisan
Compromise Hypothesis” (H3). Several different measures of support for compromise were employed
across the two studies, yet the results consistently demonstrate that partisans are more willing to
negotiate with an outparty elite when that elite is perceived as less aligned with his or her party’s base.
Lastly, each of the aforementioned analyses was redone in an effort to investigate whether base
relations cues were more powerful for Democrats compared to Republicans (or vice versa). I find no
systematic pattern suggestive of asymmetric effects. That is, overall, Democrats and Republicans
tended to respond to the experimental manipulation in a relatively similar fashion (see Supplemental
Appendix for details).
17
The Budget Compromise of 2010 & Approval of President Obama: A Case Study
We can further test H1 and H2 by examining public opinion surrounding President Obama’s
budget compromise with congressional Republicans in 2010. The event was notable in that it
signaled Obama’s willingness to disappoint key portions of his political base—a fact not lost on
major media outlets. Obama’s compromises reportedly “infuriated” many of those in his party’s
coalition (Herszenhorn and Calmes 2010). Moreover, the outcry from many Democrats reportedly
prompted President Obama to famously refer to these critics as “sanctimonious” (Meckler and
Weisman 2010), thus further driving a wedge between the president and the base of the Democratic
Party. Media outlets of all ideological persuasions took notice of this intra-party strife, with a
headline in the conservative-leaning Wall Street Journal proclaiming, “Obama Lashes Out at Critics
in His Base” (Meckler and Weisman 2010).11 As such, this event serves as an ideal test case for
both H1 and H2.
Data & Measurement
H1 and H2 were tested by merging data from six separate national surveys (administered by
Gallup, Pew, NBC/WSJ, and ABC/Washington Post). The budget compromise first appeared in The
New York Times on Monday, December 6th, 2010 (Herszenhorn and Calmes 2010). Three of the
surveys were administered immediately prior to the budget compromise event (between 12/1/10 and
12/6/10), with the remaining three administered immediately after the event (between 12/9/10 and
12/13/10).12
The independent variable of interest, Base Alienation, is a binary indicator equaling “0” for
respondents surveyed prior to the date of the event (i.e., before Obama alienated members of his
11 The prominence of this event was also confirmed by a content analysis of major newspaper articles
appearing immediately after the compromise (see Supplemental Online Appendix F).
12 See Supplemental Online Appendix E for additional details regarding these surveys.
18
party over the budget deal), and equaling “1” for respondents surveyed immediately after the event.
The dependent variable of interest is presidential approval (Approval), coded “0” (“1) for
respondents who did not (did) approve of Obama’s performance as president. The variable was
constructed using survey items that asked respondents whether they approve or disapprove of the
“job Obama is doing.” Lastly, the variable Republican is coded such that “Strong
Democrat/Democrat/Lean toward Democrat” equal “0” (39.29%) and “Strong
Republican/Republican/Lean toward Republican” equal “1” (38.21%).
Results
The results from several logistic regression analyses are reported in Table 3.13 The first two
models test H1 and H2, respectively. Model 1 demonstrates that, consistent with H1, Republican
approval of Obama significantly increased between the pre- and post-event periods. Controlling for
other variables in the model, the change in Republicans’ predicted probability of approving of
Obama’s job performance is on the order of (positive) 7 percentage points (p<.001). No significant
change is found among Democrats once covariates are taken into account (see Model 2). However,
the Base Alienation X Republican interaction term in Model 3 confirms that there was indeed a
significant difference between how partisans responded to the base-alienating event, while Model 4
confirms that this significant interaction holds even after accounting for the interaction between
Base Alienation and respondents’ ideology. Notably, this latter interaction is non-significant, which
again suggests that policy-based concerns may have played a relatively minor role in accounting for
changes in partisans’ approval of Obama.
[Table 3 about here]
13 A simplified analysis of changes in mean approval among partisan groups appears in the Supplemental
Online Appendix E.
19
Therefore, while some citizens may have shifted evaluations of Obama based upon policy-
relevant (i.e., ideological) considerations, a simpler—though not mutually exclusive—explanation
is that evaluations of Obama changed because many citizens merely learned of Obama’s
willingness to exhibit disloyalty toward the people in his party. These results provide additional,
real-world evidence for H1, though no significant evidence for H2. However, in a second set of
analyses featuring an event in which former New Jersey Governor, Chris Christie, alienated his
political base, I do find support for both H1 and H2.14 Because of spatial constraints, these latter
analyses are featured in the Supplemental Appendix.
Finally, it should be noted that, while these effects may at first appear relatively modest in
size, there are manifold reasons (beyond partisan motivated reasoning) why we might have
expected to find no significant change in approval of Obama, including citizen inattentiveness to the
event as well as the broad wording of the outcome measure. Due to spatial constraints, however, a
more detailed discussion of these considerations appears in the Supplemental Appendix.
CONCLUSION
This study finds that a heretofore under-explored type of political cue—base relations—can
counteract citizens’ well-documented inclination to maintain existing attitudes toward partisan elites. In
all, the findings offer an important contribution to the growing literatures on partisanship and motivated
reasoning, which repeatedly demonstrate that partisans discount information that challenges their
existing attitudes and loyalties (Flynn, Nyhan, and Reifler 2017). Moreover, because the study finds that
both Democratic and Republican citizens were similarly susceptible to base relations cues, the study
contributes more broadly to recent research investigating the social group-bases of Democratic and
14 Regarding these observational studies, finding support for H1 and H2 among Republicans, but only
support for H1 among Democrats, is suggestive of a partisan asymmetry (as discussed above).
20
Republican citizens’ political behavior (Ahler and Sood 2018; Grossmann and Hopkins 2016; Mason
2016; Mason and Wronski 2018).15
Several finer points are worth emphasizing. Crucially, if the party label were all-powerful,
and/or if citizens did not view the parties as coalitions of societal groups, the consistent pattern of results
found in this study should not have been observed. Second, the experimental treatment effects were
found using relatively obscure policy issues, suggesting that partisans can change attitudes as a result of
executives’ symbolic actions toward groups in their party’s base. That is, though base relations over hot-
button policy issues may be sufficient for finding similar patterns of effects, it was not necessary that
such issues were at stake. Relatedly, the experimental treatments found significant changes in partisans’
attitudes even though the policies at stake—as well as elites’ stances on those policies—remained
constant across conditions. Thus, consistent with recent research (e.g., Mason 2015), this study provides
further evidence that the intensity of affective polarization need not stem primarily from policy-based
considerations.
The results of this study also come with a variety of notable implications. In particular,
executives may be able to gain political support from outparty members at opportune moments by
conspicuously distancing themselves from key groups in their respective party’s base (e.g., by openly
criticizing these groups, refusing to meet with them, etc.). For example, ceteris paribus, in attempting to
gain political support for, or defend, a given policy agenda, an executive would likely obtain greater
support from outparty members by making it known that groups in his/her base are displeased with the
policy. This is especially likely given recent research which finds that news stories involving an elite
who alienates (versus appeases) her/his political base are more likely to attract attention from members
15 The results provide, at best, mixed support for a pertinent implication of the negative partisanship
argument (i.e., that evaluations of inparty (versus outparty) elites are more malleable). Results from the
MTurk analyses support this contention, but the opposite pattern is found in the Obama study, as well as the
Christie and supplemental MTurk studies (see Supplemental Online Appendices G and B, respectively),
while no pattern was found in the Qualtrics study.
21
of the outparty (Kane 2019). Thus, beyond the study’s contribution to the growing literatures on
partisanship, political cues, and motivated reasoning, the results have notable implications for the
strategies that elites can employ to garner greater support from members of the opposing party.
Certainly, the results of this study raise a number of questions that merit exploration in future
research. For example, which groups matter most to partisans when deciding the extent to which an
executive is loyal to the party, and might the answer vary depending upon the issue at stake? Second,
when would it be most advantageous, and/or least costly, for elites to strategically alienate members of
their own party in the interest of attracting greater outparty support? Investigation of these questions
stands to provide further insight into how partisans use base relations cues to evaluate political elites.
Finally, such implications raise normative questions regarding what leaders should do. Is
alienating one’s political base, for example, ever the right course of action in a representative
democracy? Important as they are, such questions fall outside the scope of the present study.
Nevertheless, it is worth emphasizing that the experimental results consistently demonstrate that
alienating the groups in one’s base is predictive of lower support among inparty members, which may
well be problematic in the context of a competitive primary or general election. However, during the
normal course of governing, wherein garnering some modicum of support from “the other side” can
often mean the difference between key legislation passing or failing, distancing oneself from the groups
known to associate with one’s party could potentially be strategically wise, even if normatively dubious.
Supplementary Data
Supplementary data are freely available at Public Opinion Quarterly online.
22
REFERENCES
Aberbach, Joel D., and Bert A. Rockman. 1999. “Hard Times for Presidential Leadership? (And
How Would We Know?).” Presidential Studies Quarterly 29:757–78.
Abramowitz, Alan I., and Steven Webster. 2016. “The Rise of Negative Partisanship and the
Nationalization of U.S. Elections in the 21st Century.” Electoral Studies 4:12–22.
Abramowitz, Alan I., and Steven W. Webster. 2018. “Negative Partisanship: Why Americans
Dislike Parties But Behave Like Rabid Partisans.” Political Psychology 39:119–35.
Achen, Christopher H., and Larry M. Bartels. 2016. Democracy for Realists: Why Elections Do Not
Produce Responsive Government. Princeton: Princeton University Press.
Ahler, Douglas J., and Gaurav Sood. 2018. “The Parties in Our Heads: Misperceptions about Party
Composition and Their Consequences.” The Journal of Politics 80:964–81.
Baker, Peter. 2017. “In Rejecting Popular Paris Accord, Trump Bets on His Base.” The New York
Times, June 1, 2017, sec. Climate. https://www.nytimes.com/2017/06/01/climate/paris-
accord-trump-conservative-base.html
Bolsen, Toby, James N. Druckman, and Fay Lomax Cook. 2014. “The Influence of Partisan
Motivated Reasoning on Public Opinion.” Political Behavior 36:235–62.
Bullock, John G. 2011. “Elite Influence on Public Opinion in an Informed Electorate.” American
Political Science Review 105:496–515.
Burns, Alexander, Jonathan Martin, and Matt Flegenheimer. 2018. “4 Takeaways From Tuesday’s
Primary Elections.” The New York Times, June 28, 2018, sec. U.S.
https://www.nytimes.com/2018/06/27/us/politics/primary-results-highlights.html
23
Campbell, Angus, Philip E. Converse, Warren E. Miller, and Donald E. Stokes. 1960. The American
Voter. Chicago: University of Chicago Press.
Campbell, David E., John C. Green, and Geoffrey C. Layman. 2011. “The Party Faithful: Partisan
Images, Candidate Religion, and the Electoral Impact of Party Identification.” American
Journal of Political Science 55:42–58.
Chambers, John R., Barry R. Schlenker, and Brian Collisson. 2013. “Ideology and Prejudice: The
Role of Value Conflicts.” Psychological Science 24:140–49.
Chambers, William Nisbet. 1963. Political Parties in a New Nation: The American Experience,
1776-1809. New York: Oxford University Press.
Chapman, Steve. 2010. “Obama and Liberals: Why Obama Disappoints Liberals.” Chicago
Tribune. December 12, 2010. http://articles.chicagotribune.com/2010-12-12/news/ct-oped-
1212-chapman-20101212_1_barack-obama-house-democrats-tax-cut-deal
Cillizza, Chris. 2013. “Compromise IS a Dirty Word.” The Washington Post, September 4, 2013.
http://www.washingtonpost.com/blogs/the-fix/wp/2013/09/04/compromise-is-a-dirty-word/
Cohen, Geoffrey L. 2003. “Party Over Policy: The Dominating Impact of Group Influence on
Political Beliefs.” Journal of Personality and Social Psychology 85:808–22.
Condon, Stephanie. 2010. “Obama Still Opposes Same-Sex Marriage.” August 5, 2010.
http://www.cbsnews.com/news/obama-still-opposes-same-sex-marriage/
Converse, Philip E. 1964. “The Nature of Belief Systems in Mass Publics (1964).” Critical Review
18:1–74.
Edwards, George C. 2012. The Strategic President: Persuasion and Opportunity in Presidential
Leadership. Reprint edition. Princeton: Princeton University Press.
24
Flynn, D.J., Brendan Nyhan, and Jason Reifler. 2017. “The Nature and Origins of Misperceptions:
Understanding False and Unsupported Beliefs About Politics.” Political Psychology
38:127–50.
Gerber, Alan S., and Donald P. Green. 2012. Field Experiments: Design, Analysis, and
Interpretation. New York: W. W. Norton & Company.
Green, Donald P., Bradley Palmquist, and Eric Schickler. 2002. Partisan Hearts and Minds:
Political Parties and the Social Identities of Voters. New Haven: Yale University Press.
Grossmann, Matthew, and David A. Hopkins. 2016. Asymmetric Politics: Ideological Republicans
and Group Interest Democrats. New York: Oxford University Press.
Gutmann, Amy, and Dennis Thompson. 2010. “The Mindsets of Political Compromise.”
Perspectives on Politics 8:1125–1143.
Halloran, Liz. 2013. “Obama’s Year Of Disappointing The Liberal Base.” NPR.Org. December 17,
2013. http://www.npr.org/blogs/itsallpolitics/2013/12/17/251983136/the-year-of-
disappointing-the-liberal-base-obama-2013
Hanmer, Michael J., and Kerem Ozan Kalkan. 2013. “Behind the Curve: Clarifying the Best
Approach to Calculating Predicted Probabilities and Marginal Effects from Limited
Dependent Variable Models.” American Journal of Political Science 57:263–77.
Harbridge, Laurel, and Neil Malhotra. 2011. “Electoral Incentives and Partisan Conflict in
Congress: Evidence from Survey Experiments.” American Journal of Political Science
55:494–510.
Harbridge, Laurel, Neil Malhotra, and Brian F. Harrison. 2014. “Public Preferences for
Bipartisanship in the Policymaking Process.” Legislative Studies Quarterly 39:327–55.
25
Hawkins, Carlee Beth, and Brian A Nosek. 2012. “Motivated Independence? Implicit Party Identity
Predicts Political Judgments among Self-Proclaimed Independents.” Personality & Social
Psychology Bulletin 38:1437–52.
Heider, Fritz. 1946. “Attitudes and Cognitive Organization.” The Journal of Psychology 21:107–12.
Herszenhorn, David M., and Jackie Calmes. 2010. “Obama and G.O.P. Agree on Tax Cuts.” The
New York Times, December 6, 2010, sec. U.S. / Politics.
http://www.nytimes.com/2010/12/07/us/politics/07cong.html
Hetherington, Marc J. 2005. Why Trust Matters: Declining Political Trust and the Demise of
American Liberalism. Princeton, NJ: Princeton University Press.
Hetherington, Marc J., and Thomas J. Rudolph. 2015. Why Washington Won’t Work: Polarization,
Political Trust, and the Governing Crisis. Chicago: University Of Chicago Press.
Hillygus, D. Sunshine, and Todd G. Shields. 2009. The Persuadable Voter: Wedge Issues in
Presidential Campaigns. Princeton: Princeton University Press.
Hohmann, James. 2018. “Two in Three Republicans Would Rather See a Shutdown than Trump
Compromise on the Border Wall.” The Washington Post’s “The Daily 202” (blog).
December 11, 2018. https://www.washingtonpost.com/news/powerpost/paloma/daily-
202/2018/12/11/daily-202-two-in-three-republicans-would-rather-see-a-shutdown-than-
trump-compromise-on-the-border-
wall/5c0f399c1b326b2d6629d42b/?utm_term=.1eb2ffabc3c9
Iyengar, Shanto, Gaurav Sood, and Yphtach Lelkes. 2012. “Affect, Not Ideology A Social Identity
Perspective on Polarization.” Public Opinion Quarterly 76:405–31.
Iyengar, Shanto, and Sean J. Westwood. 2015. “Fear and Loathing across Party Lines: New
Evidence on Group Polarization.” American Journal of Political Science 59 (3):690–707.
26
James, Frank. 2011. “Obama’ ‘Stop Complaining’ Order To Black Caucus Causes Stir.” NPR.Org.
September 26, 2011.
http://www.npr.org/sections/itsallpolitics/2011/09/26/140802831/obama-stop-complaining-
order-to-cbc-fires-up-some-folks
Kane, John V. 2019. “Fight Clubs: Media Coverage of Party (Dis)Unity and Citizens’ Selective
Exposure to It:” Forthcoming at Political Research Quarterly.
Kane, John V., and Jason Barabas. 2019. “No Harm in Checking: Using Factual Manipulation
Checks to Assess Attentiveness in Experiments.” American Journal of Political Science
63:234–49.
Kinder, Donald R., and Nathan P. Kalmoe. 2017. Neither Liberal nor Conservative: Ideological
Innocence in the American Public. Chicago: University of Chicago Press.
King, Gary, Jason Wittenberg, and Michael Tomz. 2003. “CLARIFY: Software for Interpreting and
Presenting Statistical Results.” Journal of Statistical Software 8 (1).
Kousser, Thad, and Justin H. Phillips. 2012. The Power of American Governors: Winning on
Budgets and Losing on Policy. New York: Cambridge University Press.
Lelkes, Yphtach, and Sean J. Westwood. 2016. “The Limits of Partisan Prejudice.” The Journal of
Politics 79:485–501.
Levendusky, Matthew. 2009. The Partisan Sort: How Liberals Became Democrats and
Conservatives Became Republicans. Chicago: University of Chicago Press.
———. 2013. How Partisan Media Polarize America. Chicago, IL: University of Chicago Press.
Lipset, Seymour Martin, ed. 1978. Emerging Coalitions in American Politics. San Francisco:
Transaction Publishers.
27
Marques, José M., and Vincent Y. Yzerbyt. 1988. “The Black Sheep Effect: Judgmental Extremity
towards Ingroup Members in Inter-and Intra-Group Situations.” European Journal of Social
Psychology 18:287–292.
Martin, Jonathan. 2015. “First in the Nation Forum: Jeb Bush.” The New York Times - First Draft
(blog). April 17, 2015. http://www.nytimes.com/politics/first-draft/2015/04/17/first-in-the-
nation-forum-jeb-bush/
Mason, Lilliana. 2015. “‘I Disrespectfully Agree’: The Differential Effects of Partisan Sorting on
Social and Issue Polarization.” American Journal of Political Science 59:128–45.
———. 2016. “A Cross-Cutting Calm How Social Sorting Drives Affective Polarization.” Public
Opinion Quarterly 80:351–77.
———. 2018. Uncivil Agreement: How Politics Became Our Identity. Chicago: University of
Chicago Press.
Mason, Lilliana, and Julie Wronski. 2018. “One Tribe to Bind Them All: How Our Social Group
Attachments Strengthen Partisanship.” Political Psychology 39:257–77.
McDermott, Monika L. 2006. “Not for Members Only: Group Endorsements as Electoral
Information Cues.” Political Research Quarterly 59:249–57.
Meckler, Laura, and Jonathan Weisman. 2010. “Obama Lashes Out at Critics in His Base.” The
Wall Street Journal. December 7, 2010.
http://www.wsj.com/articles/SB10001424052748703963704576005914263946634
Miller, Arthur H., and Christopher Wlezien. 1993. “The Social Group Dynamics of Partisan
Evaluations.” Electoral Studies 12:5–22.
28
Miller, Arthur H., Christopher Wlezien, and Anne Hildreth. 1991. “A Reference Group Theory of
Partisan Coalitions.” The Journal of Politics 53:1134–1149.
Mullinix, Kevin J., Thomas J. Leeper, James N. Druckman, and Jeremy Freese. 2015. “The
Generalizability of Survey Experiments.” Journal of Experimental Political Science 2:109–
138.
Nelson, Thomas E., and Donald R. Kinder. 1996. “Issue Frames and Group-Centrism in American
Public Opinion.” The Journal of Politics 58:1055–1078.
Newport, Frank. 2009. “Republican Base Heavily White, Conservative, Religious.” Gallup.Com.
June 1, 2009. http://www.gallup.com/poll/118937/Republican-Base-Heavily-White-
Conservative-Religious.aspx
Petrocik, John. 2009. “Measuring Party Support: Leaners Are Not Independents.” Electoral Studies
28:562–72.
Petrocik, John R. 1996. “Issue Ownership in Presidential Elections, with a 1980 Case Study.”
American Journal of Political Science 40:825–50.
Phillips, Kristine, and Scott Clement. 2017. “Trump Says His Base ‘Really Wants’ a Border Wall.
Polls Show Most Americans Don’t.” Washington Post. April 24, 2017.
https://www.washingtonpost.com/news/the-fix/wp/2017/04/24/trump-says-his-base-really-
wants-a-border-wall-polls-show-most-americans-dont/
Popkin, Samuel L. 1994. The Reasoning Voter: Communication and Persuasion in Presidential
Campaigns. Chicago: University of Chicago Press.
Ramirez, Mark D., and Nathan Erickson. 2014. “Partisan Bias and Information Discounting in
Economic Judgments.” Political Psychology 35:401–15.
29
Redlawsk, David P., ed. 2015. The American Governor: Power, Constraint, and Leadership in The
States. New York: Palgrave Macmillan.
Redlawsk, David P., Andrew J. W. Civettini, and Karen M. Emmerson. 2010. “The Affective
Tipping Point: Do Motivated Reasoners Ever ‘Get It’?” Political Psychology 31:563–593.
Reuters. 2014. “All Eyes on Governor’s Race in Florida Primary, Turnout Is Key.” The New York
Times, August 25, 2014. http://www.nytimes.com/reuters/2014/08/25/us/politics/25reuters-
usa-florida-primary.html
Roccas, Sonia, and Marilynn B. Brewer. 2002. “Social Identity Complexity.” Personality and
Social Psychology Review 6:88–106.
Rosenthal, Alan. 2012. The Best Job in Politics: Exploring How Governors Succeed as Policy
Leaders. Los Angeles: CQ Press.
Strickler, Ryan. 2018. “Deliberate with the Enemy? Polarization, Social Identity, and Attitudes
toward Disagreement.” Political Research Quarterly 71:3–8.
Taber, Charles S., and Milton Lodge. 2006. “Motivated Skepticism in the Evaluation of Political
Beliefs.” American Journal of Political Science 50:755–769.
The Daily Beast. 2017. “Trump’s Reported DACA Deal Greeted With Scorn by His Base.” The
Daily Beast. September 14, 2017. http://www.thedailybeast.com/trumps-reported-daca-deal-
greeted-with-scorn-by-his-base
Tims, Dana. 2016. “Gov. Brown Proposes Cuts, New Taxes to Close $1.7 Billion Budget Shortfall.”
OregonLive.Com. December 1, 2016.
http://www.oregonlive.com/politics/index.ssf/2016/12/gov_brown_proposes_cuts_new_ta.ht
ml
30
Wheaton, Sarah. 2015. “Obama Admonishes His Base on Trade.” POLITICO. April 23, 2015.
http://www.politico.com/story/2015/04/barack-obama-elizabeth-warren-trade-117301.html.
Wolfgang, Ben. 2013. “James Carville: Obama ‘likes’ Upsetting His Liberal Base.” Washinton
Times. Washington Times: Inside Politics (blog). April 12, 2013.
http://www.washingtontimes.com/blog/inside-politics/2013/apr/12/james-carville-obama-
likes-upsetting-his-liberal-b/
Wyler, Grace. 2013. “Outraged Liberals Say Obama Is About to Screw Over the Very People Who
Got Him Elected.” Business Insider. April 9, 2013. http://www.businessinsider.com/liberals-
threaten-obama-democrats-social-security-cuts-2013-4
31
TABLES AND FIGURES
TABLE 1. Experimental Vignette Example
Text Shown to Respondents with a Republican Governor
“Details have recently been released regarding Governor [respondent’s governor’s last name]'s
budget proposal for the next fiscal year.
The Republican Governor’s budget proposal will likely be presented to the state legislature in the
coming months. The proposal closely mirrors the budget of the previous year. However, the proposal
also includes a significant decrease in funding for the state’s courts and detention systems. The
Governor’s decision to decrease this funding has [pleased / upset] a variety of politically active
groups, particularly Evangelical Christian, business, and gun rights activist groups.
Said one representative of a prominent Evangelical Christian group, “The Governor’s decision comes
as a [welcomed surprise / terrible shock]. In doing this, the Governor has truly [pleased / upset]
many of the people in the organization I represent.”
Note: Bolded text indicates language that was manipulated between the Base Appeasement and Base Alienation
conditions, respectively.
32
TABLE 2. Overview of Survey Experiment Designs & Measures
Notes: In the Qualtrics and MTurk studies, respondents were first identified as residing in states with a Republican
governor or states with a Democratic governor. Then, blocking on this variable, respondents were randomly assigned to
read a (fabricated) story in which their state governor (who was referenced by name) either 1) appeased, or 2) alienated
groups in the governor’s party’s coalition. In both studies, the featured policy, as well as the governor’s stance on it,
remained constant across conditions. See Supplemental Online Appendix A for question/response wordings.
29
TABLE 2. Overview of Survey Experiment Designs & Measures
Table X. Key Components of MTurk and Qualtrics Survey Experiments
Qualtrics Study & MTurk Study
Variable
Manipulated
Elite target either appeases or alienates
a group in his or her party’s coalition.
Elite Featured
in Vignette
Governor of respondent’s own state
Budget Issue
Highlighted
Governor proposed to “decrease funding for the
state’s courts and detention systems.”
Groups
Appeased or
Alienated by
Governor’s
Stance on
Budget Issue
If Republican State Governor:
Evangelical Christians, Business, Gun Rights Groups
If Democratic State Governor:
Labor Unions, Environmentalists, African- Americans
Outcome
Measure(s)
Qualtrics Study
1) Perceived trustworthiness of one’s governor
2) Support for policy compromise between one’s governor and the
opposing party
MTurk Study
1) Likelihood of voting for one’s governor
2) Support for a specific compromise policy
3) Likelihood of voting for a representative who compromises with
the governor
Notes: In the Qualtrics and MTurk studies, respondents were first identified as residing in states with a Republican
governor or states with a Democratic governor. Then, blocking on this variable, respondents were randomly
assigned to read a (fabricated) story in which their state governor (who was referenced by name) either 1)
appeased, or 2) alienated groups in the governor’s party’s coalition. In both studies, the featured policy, as well as
the governor’s stance on it, remained constant across conditions. See appendix for question/response wordings.
33
TABLE 3. Regression Analyses of Obama Approval Pre- and Post-Budget Compromise
1
2
3
4
Republicans
Democrats
All Partisans
All Partisans
Base Alienation
0.73
(0.20)***
0.03
(0.15)
0.09
(0.15)
0.30
(0.21)
Republican
--
--
-3.41
(0.17)***
-3.48
(0.18)***
Base Alienation X Republican
--
--
0.59
(0.22)**
0.71
(0.23)***
Ideology
-1.46
(0.26)***
-0.36
(0.20)†
-0.96
(0.16)***
-0.72
(0.23)**
Base Alienation X Ideology
--
--
--
-0.44
(0.31)
Female
0.07
(0.17)
-0.07
(0.14)
-0.00
(0.11)
-0.00
(0.11)
Income
-0.13
(0.23)
0.31
(0.22)
0.05
(0.15)
0.06
(0.15)
Nonwhite
0.74
(0.21)***
0.69
(0.16)***
0.71
(0.13)***
0.70
(0.13)***
Age
-0.23
(0.05)***
-0.05
(0.04)
-0.14
(0.03)***
-0.14
(0.03)***
Education
-0.66
(0.34)*
1.03
(0.26)***
0.29
(0.20)
0.29
(0.20)
Constant
0.14
(0.38)
1.09
(0.31)***
2.21
(0.26)***
2.10
(0.27)***
Pseudo-R2
.09
.04
.41
.41
N
1,486
1,483
2,969
2,969
∆ pr(Approve of Obama)
+.07
.00
--
--
Notes: Dependent variable is a dichotomous measure, with 0=Disapprove of the job Obama is doing as president, and 1=Approve of the job Obama is doing as
president. Model is binomial logistic regression. Standard errors appear in parentheses. Base Alienation equals 0 for individuals surveyed before the base
alienation event, and equals 1 for individuals surveyed after the event. Republican =0 for Democrats and =1 for Republicans. Ideology (liberal to conservative),
Income and Education were recoded to range from 0 to 1; Age was divided by 10. Changes in the predicted probability that y=1 were estimated using observed
values of the independent variables (Hanmer and Kalkan 2013). † significant at p<.10; * significant at p<.05; **significant at p<.01; *** significant at p<.001
(one-tailed hypothesis tests given directional hypotheses).
34
FIGURE 1. Base Relations & Partisans’ Trust in Their State Governors
Notes: Qualtrics Study. Dependent variable is perceived trustworthiness of respondent’s state governor, recoded here to
range between 0 (lowest trust) and 1 (highest trust) for interpretive ease. The “Base Appeasement” condition involves
the respondent’s state governor pleasing his/her party’s base; the “Base Alienation” condition involves the respondent’s
state governor alienating his/her party’s base. “Outparty Governor” indicates that a respondent’s governor is of the
opposing party; “Inparty Governor” indicates that the respondent’s governor is of the same party. Thus, the figure
indicates that, when a respondent’s outparty governor alienates (vs. appeases) the party’s base, the respondent’s trust in
the governor significantly increases. However, when a respondent’s inparty governor alienates (vs. appeases) the
party’s base, the respondent’s trust significantly declines. ** significant at p<.01 level (one-tailed tests given directional
hypothesis).
.36
.43**
.60
.53**
0.2 .4 .6 .8
Mean Trust in State Governor
Outparty Governor Inparty Governor
Base Appeasement Base Alienation Base Appeasement Base Alienation
35
FIGURE 2. Base Relations & Partisans’ Willingness to Vote for Their State Governors
Notes: MTurk Study. Dependent variable is likelihood of voting for one’s governor in next election, measured on a 7-
point scale (higher values indicate greater likelihood). * significant at p<.05; *** significant at p<.001 (one-tailed tests
given directional hypothesis).
2.50
2.88*
4.82
4.12***
11.5 22.5 33.5 44.5 5
Likelihood of Voting For Governor
Outparty Governor Inparty Governor
Appeasement Alienation Appeasement Alienation
36
FIGURE 3. Base Relations & Partisans’ Support for Compromise (Qualtrics)
Notes: Qualtrics Study. Dependent variable is willingness to support policy compromise between the respondent’s
outparty governor and the respondent’s inparty, recoded to range between 0 and 1 for interpretive ease. “Base
Appeasement” condition involves state governor pleasing his/her party’s base; “Base Alienation” condition involves
state governor alienating his/her party’s base. Partisans appear more willing to support political compromise between
their party and the outparty governor when the governor is not perceived as aligned with his or her party’s base.
*=p<.05 (one-tailed hypothesis tests).
0.56
0.63*
0.1 .2 .3 .4 .5 .6 .7
Support For Policy Compromise With Outparty Governor
Base Appeasement Base Alienation
37
FIGURE 4. Base Relations & Partisans’ Support for Compromise (MTurk)
Notes: MTurk data. Left portion: Dependent variable is likelihood of voting for a representative in state legislature who
compromised with governor, measured on a 7-point scale (higher values indicating greater likelihood). Right portion:
Dependent variable is support for governor’s proposal to cut social spending and increase taxes, measured on a 7-point
scale (higher values indicating greater support). “Appeasement” condition involves state governor pleasing his/her
party’s base; “Alienation” condition involves state governor alienating his/her party’s base. Partisans appear more
willing to support political compromise when the outparty governor is not perceived as aligned with his or her party’s
base. *** significant at p<.001 (one-tailed tests given directional hypothesis).
3.34
3.73**
11.5 22.5 33.5 44.5
Likelihood of Voting For Representative Who Compromised
Appeasement Alienation
3.21
3.76***
11.5 22.5 33.5 44.5
Support For Governor's Policy Initiative
Appeasement Alienation
1
ONLINE SUPPLEMENTAL APPENDICES
Enemy or Ally?
Elites, Base Relations, and Partisanship in America
John V. Kane
New York University
TABLE OF CONTENTS
A) Experiments’ Sample Characteristics, Question/Treatment Wordings & Design
Features
B) Additional MTurk Study (Spring 2015)
C) Robustness Tests and Manipulation Checks (Qualtrics and MTurk Studies)
D) Asymmetric Effects Across Partisan Groups?
E) Details Regarding the Obama Budget Compromise Analyses
F) Content Analysis of Major Media Coverage of 2010 Budget Agreement
G) Base Relations & Partisan Approval of New Jersey State Governor Christie: Case
Study 2
H) Debriefing Language for Experimental Studies
2
Supplemental Appendix A
Experiments’ Sample Characteristics, Question/Treatment Wordings & Design Features
*Note: Detailed information regarding Qualtrics’s sampling procedures can be found at:
http://success.qualtrics.com/rs/qualtrics/images/ESOMAR%2028%202014.pdf
The Qualtrics sample was obtained via online sampling by Qualtrics from January 1st – January
26th, 2016. Though not a probability sample, the sample was selected to ensure that it would be
nationally representative on three key variables according to U.S. Census data: Age,
Race/Ethnicity, and Region. The specific targets for these variables were as follows:
Ø Age
o 18-24: 11.3%
o 25-34: 17.8%
o 35-44: 17.3%
o 45-54: 19.1%
o 55-64: 16.5%
o 65+ : 18%
Ø Race / Ethnicity
o Caucasian: 63%
o African-American: 13%
o Asian-American: 5%
o Hispanic: 17%
o Other: 2%
Ø Geographic Region
o Northeast: 17.9%
o Midwest: 21.7%
o South: 37.1%
o West: 23.3%
These quotas enabled the survey to be considerably more representative of the U.S. adult
population than popular online convenience samples. Table A1 displays demographic
characteristics of the sample.
Respondents were required to be at least 18 years of age, citizens of the United States, and
currently reside in one of the fifty the United States (excluding Alaska). The survey was
programmed by the author and administered in English. The survey was fielded (and data were
collected) by Qualtrics. No external funding was obtained for this study.
3
TABLE A1. Sample Characteristics For Each Study
QUESTION WORDINGS
Race
Would you describe yourself as:
m American Indian / Native American (1)
m Asian (2)
m Black / African American (3)
m Hispanic / Latino (4)
m White / Caucasian (5)
m Pacific Islander (6)
Table X: Sample Characteristics of MTurk and Qualtrics Studies
2015 MTurk
Sample
2016 Qualtrics
Sample
2017 MTurk
Sample
2016 ANES
Sample
Median Income
$30k-39k
$40k-49k
$40k-49k
$55-59k
Mean Age (SD)
35.71 (12.43)
46.28 (16.41)
36.83 (12.47)
49.58 (17.58)
Female
44.96%
59.28%
48.24%
52.77%
White
78.06%
63.77%
76.23%
71.68%
Black
7.21%
12.08%
7.55%
9.39%
Hispanic
5.53%
16.97%
5.64%
10.62%
Democrat
54.35%
46.51%
52.82%
34.17%
Independent
21.15%
22.55%
17.20%
36.84%
Republican
24.51%
30.94%
18.81%
28.99%
Liberal
53.56%
32.63%
--
24.26%
Moderate
22.33%
34.33%
--
43.36%
Conservative
24.12%
33.04%
--
32.37%
Note: The MTurk Study was fielded in the Spring of 2015, with a total N of 1012 adult U.S. citizens. The
Qualtrics study was fielded in January of 2016, and had surveyed a total of 1002 adult U.S. citizens. The
Qualtrics sample was selected to be mirror U.S. Census data on Age (18-24; 25-34; 35-44; 45-54; 55-64;
65+), Race (White; Black; Hispanic; Asian; Other), and Geographic Region (West; Midwest; Northeast;
South). The Qualtrics Study contains respondents from 47 U.S. states. The second MTurk study surveyed
a total of 1005 adult U.S. citizens. Ideological self-placement was not measured in this survey. The 2016
ANES sample is featured for the purposes of comparison.
4
State of Residence (Qualtrics Study Only)
Which U.S. state do you currently live in?
m Alabama (1)
m Alaska (2)
m Arizona (3)
m Arkansas (4)
m California (5)
m Colorado (6)
m Connecticut (7)
m Delaware (8)
m Florida (9)
m Georgia (10)
m Hawaii (11)
m Idaho (12)
m Illinois (13)
m Indiana (14)
m Iowa (15)
m Kansas (16)
m Kentucky (17)
m Louisiana (18)
m Maine (19)
m Maryland (20)
m Massachusetts (21)
m Michigan (22)
m Minnesota (23)
m Mississippi (24)
m Missouri (25)
m Montana (26)
m Nebraska (27)
m Nevada (28)
m New Hampshire (29)
m New Jersey (30)
m New Mexico (31)
m New York (32)
m North Carolina (33)
m North Dakota (34)
m Ohio (35)
m Oklahoma (36)
m Oregon (37)
m Pennsylvania (38)
m Rhode Island (39)
m South Carolina (40)
m South Dakota (41)
m Tennessee (42)
m Texas (43)
m Utah (44)
5
m Vermont (45)
m Virginia (46)
m Washington (47)
m West Virginia (48)
m Wisconsin (49)
m Wyoming (50)
m Washington D.C. (District of Columbia) (57)
m Other (58)
Party Identification
Generally speaking, do you consider yourself to be a(n):
m Strong Democrat (1)
m Democrat (2)
m Independent, But Leaning Democrat (3)
m Independent (4)
m Independent, But Leaning Republican (5)
m Republican (6)
m Strong Republican (7)
Religion
What, if any, is your religious affiliation?
m Protestant (1)
m Catholic (2)
m LDS / Mormon (3)
m Jewish (4)
m Other (5)
m No Preference / No Religious Affiliation (6)
m Prefer Not To Say (7)
Ideological Self-Placement of Self and Governor
Below is a 7-point scale on which the political views that people might hold are arranged from
extremely liberal to extremely conservative. Where would you place [yourself / (Respondent’s
governor)] on this scale?
m Extremely Liberal (1)
m Liberal (2)
m Slightly Liberal (3)
m Moderate / Middle Of The Road (4)
m Slightly Conservative (5)
m Conservative (6)
m Extremely Conservative (7)
6
Income
What is your annual income range?
m Below $20,000 (1)
m $20,000 - $29,999 (2)
m $30,000 - $39,999 (3)
m $40,000 - $49,999 (4)
m $50,000 - $59,999 (5)
m $60,000 - $69,999 (6)
m $70,000 - $79,999 (7)
m $80,000 - $89,999 (8)
m $90,000 Or More (9)
Trust
How trustworthy do you find your state's current governor, Governor {last name inserted}, to
be?
m Highly Untrustworthy (1)
m Untrustworthy (2)
m Slightly Untrustworthy (3)
m Neutral (4)
m Slightly Trustworthy (5)
m Trustworthy (6)
m Highly Trustworthy (7)
Compromise Outcomes (Qualtrics)
Support for Compromise (if Respondent has a Democratic Governor)
How willing would you be to support policy compromise between Governor {last name
inserted} and the Republican Party?
m Unwilling (1)
m Slightly Unwilling (2)
m Neutral (3)
m Slightly Willing (4)
m Willing (5)
7
Support for Compromise (if Respondent has a Republican Governor)
How willing would you be to support policy compromise between Governor {last name
inserted} and the Democratic Party?
m Unwilling (1)
m Slightly Unwilling (2)
m Neutral (3)
m Slightly Willing (4)
m Willing (5)
Compromise Outcomes (MTurk)
Your state governor, Governor {last name inserted}, has also outlined a long-term plan for
ensuring the state's fiscal health in the future. In particular, the governor wishes to implement
targeted cuts to specific social welfare programs, while also raising taxes that would only affect
the state's wealthiest individuals.
To what extent would you support such an initiative?
m Strongly Oppose (1)
m Oppose (2)
m Slightly Oppose (3)
m Neutral (4)
m Slightly Support (5)
m Support (6)
m Strongly Support (7)
Suppose your representative in the state legislature were to compromise with Governor {last
name inserted}.
How likely is it that you would vote for this representative in the next election?
m Extremely Unlikely (1)
m Unlikely (2)
m Slightly Unlikely (3)
m Neutral (4)
m Slightly Likely (5)
m Likely (6)
m Extremely Likely (7)
8
TREATMENT WORDINGS
MTurk Study 1 (featured in this appendix)
Instructions: Now we'd like to ask for your opinion about different elected officials in federal
and state government. We have a database of news stories on hundreds of elected officials, and
will randomly select an excerpt from one of these news stories for you to view. We ask that you
carefully read over the information provided in the excerpt, and then answer a few questions
about your opinions. Please read the news article excerpt closely as you will be asked questions
about it afterward.
Excerpt from an article appearing in THE ASSOCIATED PRESS (AP) earlier this year:
“[Governor Branstad/Governor Dayton] Outlines Budget Proposal; Includes Funding for
Weather Detection System, [Pleasing/Upsetting] Members of His Party.”
Speaking to members of the Iowa state legislature late yesterday afternoon, [Republican
Governor Branstad / Democratic Governor Dayton] outlined his budget proposal for the next
fiscal year. Among other proposals, the Governor’s budget calls for funding for the development
of a state-of-the-art weather detection system. The Governor has previously argued that the
measure is essential for the people and state of Iowa. [However, while] satisfied with most of the
budget, the [conservative/liberal] base of the Governor’s party in the state legislature has
expressed [high praise for/deep disappointment with] the Governor’s decision to include funding
for the weather detection system. “In doing this, he has [pleased many of us in his party/let so
many of us in his party down],” said one Republican state senator. The proposal has been
forwarded to the state legislature for review and is scheduled to be voted upon within the next
month.
The [Republican/Democratic] Governor, in office since 2011, also hinted at the possibility of a
run for the presidency in 2016. Having [pleased/upset] the base of his [Republican/Democratic]
Party with his recent push for the weather detection system, it remains to be seen whether
[conservatives/liberals] in the state legislature will endorse him should he decide to run.
Qualtrics Study & MTurk Study (featured in this manuscript)
Instructions: Now we'd like to ask for your opinions on a different topic. We have a large
database of recent news stories covering elected officials in your state government, and will
randomly select an excerpt from one of these news stories for you to view. We ask that you
carefully read over the information provided in the excerpt, and then answer a few questions
about your opinions. Please read the news article excerpt closely as you will be asked questions
about it afterward.
Excerpt from recent Associated Press article covering political developments in your state:
Details have recently been released regarding Governor {respondent’s governor’s
name}'s budget proposal for the next fiscal year. The [governor’s party] Governor’s budget
9
proposal will likely be presented to the state legislature in the coming months. The proposal
closely mirrors the budget of the previous year. However, the proposal also includes a
significant decrease in funding for the state’s courts and detention systems. The Governor’s
decision to decrease this funding has [pleased/upset] a variety of politically active groups,
particularly [Evangelical Christian, business, and gun rights activist groups / labor unions,
environmentalists, and African American activist groups]. Said one representative of a
prominent [Evangelical Christian group / labor union], “The Governor’s decision comes as a
[welcomed surprise /terrible shock]. In doing this, he has truly [pleased/upset] many of the
people in the organization I represent.”
Qualtrics Sampling
Additional information regarding Qualtrics’s sampling procedures can be found at:
http://success.qualtrics.com/rs/qualtrics/images/ESOMAR%2028%202014.pdf
Design Features
Several features of the experiments are worth highlighting. First, the decision regarding
which groups to include in the experiment’s vignettes was directly informed by previous research on
the groups that citizens associate with the parties (see Campbell et al. 2010). Second, the studies did
not explicitly state that the featured groups are Democratic- or Republican-leaning, which makes for
particularly conservative tests of the hypotheses. Third, the decision to embed the governor’s
proposal within the context of the budget was informed by existing research on governors’ common
use of the state budget to effect preferred policy goals (Kousser and Phillips 2012). Fourth, direct
references to respondents’ own state governors in both the experimental vignette and outcome
measures served not only to enhance the realism of the study, but also to subject the hypotheses to a
far more conservative test insofar as changing attitudes toward known (vis-à-vis unknown) partisan
targets should be more difficult.
10
Supplemental Appendix A References
Campbell, David E., John C. Green, and Geoffrey C. Layman. 2011. “The Party Faithful:
Partisan Images, Candidate Religion, and the Electoral Impact of Party Identification.”
American Journal of Political Science 55 (1): 42–58.
Kousser, Thad, and Justin H. Phillips. 2012. The Power of American Governors: Winning on
Budgets and Losing on Policy. Cambridge England ; New York: Cambridge University
Press.
11
Supplemental Appendix B
Additional MTurk Study (MTurk Study 1)
In addition to the studies featured in the manuscript, an additional study was fielded on
MTurk in the Spring of 2015. As MTurk operates via posting the survey online as a “Human
Intelligence Task (HIT)” that all users (“workers”) can potentially observe and choose to
complete, rather than contacting workers directly, it is not possible to calculate a response rate.
This experiment had a total sample of 1,012 adult U.S. citizens. Similar to the other
experiments, in this MTurk study, respondents were randomly assigned to read a (fabricated)
news story wherein either Republican Governor Terry Branstad (R-Iowa) or Governor Mark
Dayton (D-Minnesota) appeased or alienated a prominent group in his party’s coalition. This
purportedly occurred because of the governor’s proposal to “include funding for a state-of-the-art
weather detection system”—a stance which did not vary across conditions.
Table B1. Design Features of MTurk Study 1 (2015)
Table X. Key Components of MTurk and Qualtrics Survey Experiments
MTurk Study (2015)
Variable
Manipulated
Governor either appeases or alienates a
group in his party’s coalition.
Governor
Featured in
Vignette
Terry Branstad (R-Iowa)
or
Mark Dayton (D-Minnesota)
Budget Issue
Highlighted
Governor proposed to include “funding
for a state-of-the-art weather detection
system.”
Groups
Appeased or
Alienated by
Governor’s
Stance on
Budget Issue
If Gov. Branstad (R):
Conservatives in the State Legislature
If Gov. Dayton (D):
Liberals in the State Legislature
Outcome
Measure(s)
Perceived Favorability of the Governor
Notes: In this MTurk study, respondents were randomly assigned to read a fabricated news story in which either
Governor Branstad or Governor Dayton 1) appeased, or 2) alienated a group in the governor’s party’s coalition.
As with the studies featured in the manuscript, the budget issue highlighted in the news story, as well as the
governor’s stance on it, remained constant across conditions.
Table 4. Qualtrics Sample Characteristics
2016 Qualtrics
Sample
2012 ANES
Sample
Median Income $40k-49k $40-45K
12
The dependent variable in this study is Favorability, which captures each respondent’s
perceived favorability of the featured governor and is measured on a seven-point scale with
higher values indicating greater favorability. Table B1 lists the key details of this study.
For the “Enemy of my Enemies” and “Enemy of my Allies” hypotheses (H1 and H2,
respectively) to find support, it should be the case that, relative to the Base Appeasement condition,
partisans assigned to the Base Alienation condition exhibit significantly higher favorability for the
governor in the opposing party (H1), and significantly lower favorability for the governor in their
own party (H2). This is indeed precisely the pattern we observe. Figure B1 displays mean
favorability ratings for the outparty and inparty governor, respectively. Consistent with H1,
respondents rated an outparty governor 10 percentage points (24%) more favorably in the Base
Alienation condition than in the Base Appeasement condition (p<.001). Again, for H2 to find
support, we should observe a decrease in perceived favorability as we move from Base Appeasement
to Base Alienation. Though the absolute effect size (-4 percentage points, or -7%) is smaller than in
the previous case, this is again the pattern we observe, and the effect again attains statistical
significance (p<.05). This study, therefore, provides additional support for both H1 and H2,
suggesting that the manner in which an elite relates to his or her party’s base affects how the elite is
evaluated by partisan citizens.
13
Figure B1. Base Relations & Partisans’ Evaluations of Governors
FIGURE 1. Coalition Politics & Partisans’ Evaluations of State Governors
Notes: MTurk Study 1 data. Dependent variable is perceived favorability of either Republican Governor
Branstad (R-Iowa) or Democratic Governor Dayton (D-Minnesota). Democratic respondents who read about
Republican Governor Branstad, and Republican respondents who read about Democratic Governor Dayton are
featured in the “Outparty Governor” result on the left. Republican respondents who read about Republican
Governor Branstad, and Democratic respondents who read about Democratic Governor Dayton are featured in the
“Inparty Governor” result on the right. “Base Appeasement” condition indicates that the governor pleased
members of his party, while “Base Appeasement” condition indicates that the governor displeased members of his
party. *=p<.05; ***=p<.001 (one-tailed hypothesis tests given directional hypotheses).
Figure X. Base Relations & Partisan Polarization in Trust of State Governors
Difference = .10***
(Total n=386)
Difference = -.04*
(Total n=412)
0.2 .4 .6
Mean Favorability of Governor
Outparty Governor Inparty Governor
Base Appeasement Base Alienation Base Appeasement Base Alienation
14
Supplemental Appendix C
Robustness Tests and Manipulation Checks (Qualtrics and MTurk Studies)
Treatment Effects, Subgroup Membership, and Ideological Self-Placement
Because specific subgroups were mentioned in the experimental vignettes, one possible
explanation for the results is that respondents were primarily concerned with loyalty/disloyalty to
their own subgroup rather than loyalty/disloyalty to the party’s base more broadly. To
investigate this possibility, additional analyses were done for each of the three experimental
studies. Specifically, interactions between the treatment and (when available) a measure of
subgroup membership (e.g., religious, racial, and economic groups) were specified to examine
whether the treatment effect occurred for non-group members of the party.
Overall, the evidence indicates that even non-group members of the party were affected
by the experimental manipulation and adjusted their political support for state governors. In ten
separate regression models (see Tables C1-C3), the coefficient on “Treatment” is negative
(indicating, as hypothesized, reduced political support for the elite among non-group members)
in nine cases (and at least marginally statistically significant in six cases). Importantly, in cases
where this coefficient is not statistically significant, the interaction term was not statistically
significant either. In only one case (Christians in MTurk Study 1) was the treatment effect
among non-group members positive and the interaction term statistically significant (indicating
that the treatment effect was significantly stronger for Christian Republicans than for non-
Christian Republicans). This is notable, but it is also important to stress that, with the small
sample (n=122), and the majority of Republicans in the MTurk sample being Christian (61%),
the treatment effect estimate among non-Christians contains a substantial amount of uncertainty.
Lastly, table C4 uses the Qualtrics data to test whether treatment effects depended upon
respondents’ ideological self-placement. In only one instance (Democratic respondents with
15
Republican governors) do we find evidence of the treatment effect being conditional upon
ideological self-placement (in this case, the treatment effect being significantly stronger for more
liberal respondents).
Table C1. Interactions Between Treatment & Subgroup Membership (MTurk Study 1)
Table A1. Treatment Effects Among Sub-Groups (MTurk Study)
Republican
Governor &
Respondents
Democratic
Governor &
Respondents
Group=
Christians
Group=
High
Income
Group=
African-
Americans
Treatment
.05
-.08
-.03
(.05)
(.06)
(.02)
Group
.17***
-.09
-.05
(.05)
(.07)
(.07)
Treatment X
Group
-.18**
(.07)
.07
(.11)
-.08
(.08)
Constant
.49***
.62***
.58***
(.04)
.04
(.02)
Adj. R2
.09
.00
.02
N
122
122
290
Notes: MTurk Study 1. Dependent variable is perceived favorability of a co-partisan governor (1=Highly
Unfavorable; 7=Highly Favorable). “Treatment” represents going from the respondent’s state governor
appeasing his/her party’s political base (0) to alienating this base (1). “Christian” and “African American”
groups are dummy variables for Republicans and Democrats, respectively, where self-reported membership
in the group=1 (otherwise=0). “Income” is a nine-point measure of Republican respondents’ individual
annual income ranging from <20k to >90k, recoded to range between 0 and 1. All covariates were
measured pre-treatment. †=p<.10; *=p<.05; **=p<.01; ***p<.001 (two-tailed hypothesis tests).
16
Table C2. Interactions Between Treatment & Subgroup Membership (Qualtrics Study)
Table A1. Treatment Effects Among Sub-Groups (Qualtrics Study)
Republican
Governor &
Respondents
Democratic
Governor &
Respondents
Group=
Christians
Group=
High
Income
Group=
African-
Americans
Treatment
-0.16
-0.72†
-0.50*
(0.38)
(0.41)
(0.23)
Group
0.68*
-0.43
0.49
(0.34)
(0.48)
(0.48)
Treatment X
Group
0.04
(0.47)
1.19†
(0.66)
-1.31*
(0.60)
Constant
3.97***
4.64***
4.69***
(0.27)
(0.30)
(0.16)
Adj. R2
.05
.00
.07
N
207
207
193
Notes: Qualtrics Study. Dependent variable is perceived trustworthiness of respondent’s state governor
(1=Highly Untrustworthy; 7=Highly Trustworthy). “Treatment” represents going from the respondent’s
state governor appeasing his/her party’s political base (0) to alienating this base (1). “Christian” and
“African American” groups are dummy variables for Republicans and Democrats, respectively, where self-
reported membership in the group=1 (otherwise=0). “Income” is a nine-point measure of Republican
respondents’ individual annual income ranging from <20k to >90k, recoded to range between 0 and 1. All
covariates were measured pre-treatment. †=p<.10; *=p<.05; **=p<.01; ***p<.001 (two-tailed hypothesis
tests).
17
Table C3. Interactions Between Treatment & Subgroup Membership (MTurk Study 2)
Table A1. Treatment Effects Among Sub-Groups (MTurk Study 2)
Republican
Governor &
Respondents
Democratic
Governor &
Respondents
Democratic
Governor &
Respondents
Group=
Christians
Group=
High
Income
Group=
African-
Americans
Group=
Union
Household
Treatment
-.63
-.94†
-.65**
-.68**
(.61)
(.56)
(.21)
(.22)
Group
-.16
-.44
-.22
-.01
(.52)
(.74)
(.39)
(.48)
Treatment X
Group
-.25
(.73)
.21
(1.01)
-.15
(.69)
.18
(.63)
Constant
5.13***
5.23***
4.77***
4.74***
(.43)
(.42)
(.15)
(.14)
Adj. R2
(.04)
(.04)
.03
.03
N
108
108
235
235
Notes: MTurk Study 2. Dependent variable is reported likelihood of voting for respondent’s co-partisan
state governor in the next election (1=Highly Unlikely; 7=Highly Likely). “Treatment” represents going
from the respondent’s state governor appeasing his/her party’s political base (0) to alienating this base (1).
“Christian” and “African American” groups are dummy variables for Republicans and Democrats,
respectively, where self-reported membership in the group=1 (otherwise=0). “Income” is a nine-point
measure of Republican respondents’ individual annual income ranging from <20k to >90k, recoded to
range between 0 and 1. All covariates were measured pre-treatment. †=p<.10; *=p<.05; **=p<.01;
***p<.001 (two-tailed hypothesis tests).
18
Table C4. Interactions Between Treatment & Ideological Self-Placement (Qualtrics Study)
Further Investigating the Mechanism
As briefly noted in the manuscript, there exist two likely reasons for the treatment effects we
observe. First, given the theory discussed above, the Base Alienation cue (vis-à-vis the Base
Appeasement cue) should signal a purely symbolic detachment from politically-aligned groups. That
is, base relations cues should communicate that the executive is, in a general sense, willing to
demonstrate disloyalty toward, and risk displeasing, typical members of his or her party—i.e., the
very kinds of people that inform citizens’ conceptualizations of the party (e.g., Green, Palmquist, and
Schickler 2002)
This proposition was tested directly by re-launching the experimental design featured in the
Qualtrics and second MTurk study on a separate MTurk sample. A total of 503 respondents
completed the survey, 413 of which identified as a partisan (or “leaning toward” one of the two
parties). Following the manipulation involving his or her own state governor, each respondent was
Table A2. Interactions Between Treatment & Ideological Self-Placement
Republican Governor
Democratic Governor
Democrats
Republicans
Democrats
Republicans
Treatment
1.08**
-0.35
-0.39
.23
(0.34)
(0.82)
(0.41)
(1.02)
Ideology
0.96†
1.97
-0.97
-1.38
(0.55)
(0.80)
(0.70)
(1.07)
Treatment X
Ideology
-1.75*
0.33
-1.00
0.32
(0.79)
(1.08)
(0.99)
(1.39)
Constant
2.84***
2.98***
5.09***
3.9***
(0.24)
(0.60)
(0.29)
(0.78)
Adj. R2
.03
.06
.09
.02
N
273
207
193
103
Notes: Qualtrics data. Dependent variable is perceived trustworthiness of respondent’s state governor.
“Treatment” represents going from the respondent’s state governor appeasing his/her party’s political base
(0) to alienating this base (1). Ideology was measured pre-treatment and recoded to range between zero and
one. Each model tests the possibility that treatment effects were driven partly by ideological
considerations. In only one model do we observe the interaction attaining significance. †=p<.10; *=p<.05;
**=p<.01; ***p<.001 (two-tailed hypothesis tests).
19
asked, “Given the information you just read, how loyal or disloyal would you say your governor is to
the people in his or her party?” Response options ranged from “Very Disloyal” (1) to “Very Loyal”
(4). In addition, depending upon whether the respondent’s governor was a Democrat or Republican,
each respondent was asked, “Given the information you just read, do you think typical
[Democrats/Republicans] living in your state are “Very Pleased,” (4) “Somewhat Pleased,” (3)
“Somewhat Displeased,” (2) or “Very Displeased” (1) with your governor’s actions?”
Figure C1 displays the results of these analyses. In every instance, and regardless of whether
respondents with Democratic or Republican governors were analyzed, moving from Base
Appeasement to Base Alienation resulted in sizable and statistically significant decreases in both of
these measures. These effects ranged from -.57 to -.92 on each four-point scale, and always achieved
significance at the p<.001 level. This manipulation check provides strong evidence that the vignettes
altered partisans’ perceptions of relations between executives and their respective party’s base.
In addition, and as noted in the manuscript, executives’ relations with particular groups may
also indirectly signal their ideological position. While the specific policy items featured in the
experiments were relatively ambiguous in terms of ideological content, and did not vary across
conditions, citizens may have nevertheless inferred the executive’s ideological position based on his
or her relation to the groups mentioned in the experimental vignettes (e.g., see Feldman and Conover
1983). This possibility is especially plausible given that citizens harbor perceptions of many social
groups’ ideological leanings (e.g., see Chambers, Schlenker, and Collisson 2013).
Additional analyses revealed that survey respondents did significantly adjust their
perceptions of Democratic, but not Republican, governors’ placement on the ideological scale in
response to the experimental manipulation. Specifically, both Republican and Democratic
respondents perceived Democratic governors to be more conservative in the Base Alienation
20
FIGURE C1. Manipulation Check Results
condition than in the Base Appeasement condition (p<=.06 in both cases). However, these effects
were substantively quite small (.42 and .44, respectively, on a seven-point scale). Nevertheless,
because the Qualtrics survey included both a pre- and post-treatment measure of each respondent’s
perceived ideological location of their governor, it was possible to create an individual-level variable
capturing the change in perceived ideology of one’s governor (going from pre- to post-treatment).
Even after accounting for this measure in the regression analyses, however, the experimental
treatment effects remain substantively identical.1 Lastly, as reported in Table C4 above, a series of
1 Admittedly, because change in perceived ideology is, in part, a function of the treatment, including this
post-treatment measure as a control variable runs the risk of biasing the estimated treatment effect
(Montgomery, Nyhan, and Torres 2018).
FIGURE 4. Manipulation Checks
Note: MTurk data. Top portion: Dependent variable is governor’s perceived loyalty to his or her party (listed on
the y-axis) and ranged from “Very Disloyal” (1) to “Very Loyal” (4). All differences between “Base Appeasement”
and “Base Alienation” conditions significant at the p<.001 level. Bottom portion: Dependent variable is perception
of how pleased members of governor’s party are with him/her (listed on the y-axis), and ranged from “Very
Displeased” (1) to “Very Pleased” (4). All differences between “Base Appeasement” and “Base Alienation”
conditions significant at the p<.001 level. *** significant at p<.001 (one-tailed tests given directional hypothesis).
3.18
2.61***
3.14
2.53***
Democratic Governors Republican Governors
(n = 184) (n = 318)
1 2 3 4
Governor's Perceived Loyalty
Base Appeasement Base Alienation Base Appeasement Base Alienation
3.06
2.14***
2.94
2.13***
Democratic Governors Republican Governors
(n = 184) (n = 318)
1 2 3 4
Party Members Pleased
Base Appeasement Base Alienation Base Appeasement Base Alienation
21
regression models tested for interactions between the experimental manipulation and respondents’
ideological self-placement—in only one instance (Democrats with Republican governors) did this
interaction attain statistical significance. This further suggests that ideological considerations played
a relatively minor role in generating the observed treatment effects.
In sum, while the experimental cues likely communicated a mixture of important information
about the executive, the weight of the evidence suggests that considerations about the executive’s
ideology likely played a relatively minor role in influencing changes in support compared to the
more symbolic considerations involving elites’ base relations.
Supplemental Appendix C References
Chambers, John R., Barry R. Schlenker, and Brian Collisson. 2013. “Ideology and Prejudice:
The Role of Value Conflicts.” Psychological Science 24 (2): 140–49.
Feldman, Stanley, and Pamela Johnston Conover. 1983. “Candidates, Issues and Voters: The
Role of Inference in Political Perception.” The Journal of Politics 45 (04): 810–39.
Green, Donald P., Bradley Palmquist, and Eric Schickler. 2002. Partisan Hearts and Minds:
Political Parties and the Social Identities of Voters. Yale University Press.
Montgomery, Jacob M., Brendan Nyhan, and Michelle Torres. 2018. “How Conditioning on
Post-Treatment Variables Can Ruin Your Experiment and What to Do about It.”
American Journal of Political Science (forthcoming).
22
Supplemental Appendix D
Asymmetric Effects Across Partisan Groups?
As noted in the manuscript, it is possible that treatment effects differed across partisan
groups. To investigate this possibility, I re-ran each analysis featured in the manuscript, but
included an interaction with a dummy variable indicating each respondent’s party identification.
Given the patterns identified for partisans in general, if Republicans, for example, are less
susceptible to cues involving base relations, we should observe a negative, statistically
significant interaction term in analyses involving an outparty governor, and a positive,
statistically significant interaction term in analyses involving an inparty governor. Such a pattern
would indicate that, when evaluating an outparty elite, base alienation has a less positive effect
on Republicans compared to Democrats; and, when evaluating an inparty elite, base alienation
has a less negative effect on Republicans compared to Democrats.
Table D1 displays the results of these analyses. In no case did the interaction term attain
statistical significance at the conventional (p<.05) level. Moreover, the direction of the
interaction term offers no clear pattern of asymmetric treatment effects: in the “outparty
governor” analyses (Models 1, 3, 4, 5 and 7), the interaction is positive three times and negative
twice; in the “inparty governor” analyses (Models 2 and 6), the interaction term is positive once
and negative once. In only one instance (Model 6) did the interaction term approach statistical
significance (p=.06), suggesting a tendency for Republicans to have been less influenced by base
relations in evaluating the trustworthiness of their Republican governor than were Democrats in
evaluating the trustworthiness of their Democratic governor. Nevertheless, aside from this one
result, there is no indication that one partisan group is systematically less susceptible to base
relations cues than the other partisan group.
23
TABLE D1. Treatment Effects Generally Symmetrical Across Democrats and Republicans
MTurk Study
Qualtrics Study
(1)
(2)
(3)
(4)
(5)
(6)
(7)
Outparty
Inparty
Outparty
Outparty
Outparty
Inparty
Outparty
Vote (Gov)
Vote (Gov)
Vote (Rep)
Support
Trust
Trust
Compromise
Base Alienation
0.24
-0.65**
0.39*
0.50*
0.44*
-0.73**
0.05
(0.20)
(0.21)
(0.19)
(0.19)
(0.18)
(0.23)
(0.04)
Republican
0.71*
0.28
1.02***
0.83**
-0.26
-0.34
0.03
(0.32)
(0.27)
(0.31)
(0.32)
(0.26)
(0.23)
(0.05)
Base Alienation X Republican
0.26
-0.19
-0.35
-0.12
0.00
0.60†
0.03
(0.43)
(0.37)
(0.41)
(0.43)
(0.35)
(0.32)
(0.07)
Constant
2.38***
4.74***
3.17***
3.07***
3.20***
4.75***
0.56***
(0.13)
(0.14)
(0.12)
(0.13)
(0.13)
(0.16)
(0.02)
R2
0.05
0.05
0.06
0.06
0.03
0.03
0.02
N
367
343
367
367
376
400
375
Notes: All models estimated using OLS regression (SEs in parentheses). Base Alienation = 0 for “Base Appeasement” condition, and 1 for “Base Alienation”
condition. Republican = 0 for Democrats, and 1 for Republicans. “Inparty” indicates that respondent’s governor is of the same party; “Outparty” indicates that
respondent’s governor is of the opposing party. Dependent variables are those featured in the manuscript. “Vote (Gov)” is willingness to vote for one’s
governor; “Vote (Rep)” is willingness to vote for a state legislator who compromises with the (outparty) governor; “Support” is willingness to support the
governor’s proposed fiscal policy initiative, which includes both tax increases and spending cutes; “Trust” is perceived trustworthiness of the governor;
“Compromise” is support for policy compromise between one’s own party and the outparty governor. † significant at .10; * significant at .05; ** significant at
.01; *** significant at .001 (two-tailed).
24
A final set of analyses, made possible because of the pre-treatment measures featured in the
Qualtrics data, aimed to investigate whether Democrats differed from Republicans in the extent to
which objective group membership considerations—versus ideological considerations—moderated
treatment effects. Table D2 shows the results of these analyses. The “Group Member” variable
indicates membership in a party-aligned group (for Democrats, this equals 1 for African-Americans,
and 0 otherwise; for Republicans, this equals 1 for Christians, and 0 otherwise). The “Ideological
Distance” variable equals the respondent’s self-placement on the 7-point ideological scale (ranging
from extremely liberal to extremely conservative) minus the respondent’s (pre-treatment) placement
of his or her governor on this scale (thus, higher values indicate that the respondent is more
conservative relative to the governor (recoded to range from 0 to 1)). The relatively small sample
sizes and severe inflation of standard errors (the mean variance inflation factor (vif) exceeded 9 in
both models) arising from multiple interactions with the treatment should lead us to exercise caution
in interpreting these results. Nevertheless, it is notable that, though a significant interaction exists
between the treatment and Group Member among Democrats, no significant interaction (either for
Group Member or Ideological Distance) among Republicans. Overall, then, while it remains
possible that the treatment was effective for Democrats for fundamentally different reasons than it
was effective among Republicans (e.g., Democrats were more concerned with group interests,
Republicans with ideological purity as signaled by base relations), the evidence for this contention in
Table D2 is quite limited.
25
TABLE D2. Group-Based and Ideology-Based Mechanisms Among Partisans
Table C5.
Perceived Trustworthiness of
Inparty Governor
Republicans
Democrats
Base Alienation
-1.21
0.50
(1.02)
(0.91)
Group Member
0.58†
0.49
(0.34)
(0.48)
Base Alienation X Group Member
0.15
-1.28*
(0.47)
(0.60)
Ideological Distance
-2.35†
0.09
(1.25)
(1.25)
Base Alienation X Ideological Distance
1.74
-2.05
(1.80)
(1.83)
Constant
5.32***
4.65***
(0.73)
(0.61)
R2
0.06
0.10
N
206
193
Notes: Qualtrics Study. Dependent variable is perceived trustworthiness of respondent’s state governor, measured on
a 7-point scale (higher values = greater perceived trustworthiness). Treatment indicates going from the “Base
Appeasement” condition (which involves the respondent’s state governor pleasing his/her party’s base) to the “Base
Alienation” condition (which involves the respondent’s state governor alienating his/her party’s base). “Inparty
Governor” indicates that the respondent’s governor is of the same party. †=p<.10; *=p<.05; **=p<.001; ***=p<.001
level (two-tailed hypothesis tests).
26
Supplemental Appendix E
Details Regarding the Obama Budget Compromise Analyses
This section provides additional details regarding the analysis of public opinion
surrounding President Obama before and after the 2010 budget compromise with congressional
Republicans in which Obama had alienated his political base. Supplemental Appendix F (see
below) describes the content analysis that was performed on media coverage of this event.
Surveys
The surveys dates of administration, and reported survey details are as follows:
Pre-event:
• Pew (12/1/2010-12/05/2010): Nationally representative sample of 1500 adults
living in continental United States, contacted via landline or cellular phone via
RDD. Survey conducted by Princeton Survey Research Associates International.
AAPOR RR3 = 15.5% for landline, 10.2% for cell.
• Pew (12/2/2010-12/5/2010): Nationally representative sample of 1003 adults
living in the continental United States, contacted via landline or cellular phone via
RDD. Survey conducted by Princeton Survey Research Associates International.
Response rate not reported.
• Gallup (12/3/2010-12/6/2010): Random sample of 1003 national adults,
contacted via landline or cellular phone. Response rate not reported.
Post-event:
• ABC/Washington Post (12/9/10-12/12/10): Nationally representative sample of
1001 adults living in the continental United States, contacted via landline or
cellular phone via RDD. Response rate not reported.
• Gallup/USA (12/10/10-12/12/10): Nationally representative sample of 1019
national adults, contacted via landline or cellular phone via RDD (share in each
category not reported). Response rate not reported.
• NBC/WSJ (12/9/10-12/13/10): Nationally representative sample of 1000 national
adults, contacted via landline or cellular phone via RDD by Hart and McInturff
Research Companies. Response rate not reported.
The Gallup survey was conducted over four days, and includes respondents interviewed
on the exact day of the event (26.82% of the Gallup sample), which potentially increases the
probability of a Type II error by labeling these respondents as untreated. However, additional
27
analyses suggested that differences in mean support for Obama among Republicans and
Democrats did not substantively differ when the analysis omitted respondents who were
interviewed on 12/6/10.
The Pew Research Center survey, conducted between 12/2/10 and 12/5/10, did not
contain the standard “job approval” question for President Obama. Instead, the survey asked
respondents to rate their favorability of Obama on a 4-point scale ranging from “Very
unfavorable” to “Very favorable.” Thus, respondents who indicated either “Very Unfavorable”
or “Mostly unfavorable” were coded as “0,” while respondents who indicated “Very/Mostly
unfavorable” were coded as “1.” Results do not differ in any substantive way when this survey
is excluded from the analysis. Also, the Pew Research Center survey conducted between 12/2/10
and 12/5/10 did not contain a measure of respondent ideology.
Additional analyses tested whether the different survey houses mattered for mean
approval of Obama among Republicans. The results reveal that the Pew (12/02/10-12/05/10)
survey was significantly different than the other two pre-treatment surveys (perhaps because
mean approval was measured slightly differently for this survey (see below)). However,
Republican approval of Obama is significantly higher (roughly six percentage points) in this
survey than in the other two pre-treatment surveys, and thus should make it more difficult to find
a treatment effect, all else equal. Similarly, in the post-treatment period, the significant outlying
survey (Gallup/USA) provides a significantly lower mean approval among Republicans. Again,
this effectively renders it more difficult to find a significant difference between the pre-and post-
treatment periods, all else equal.
28
Covariates
Respondents identifying as female (Female; 47% of the sample) were coded as “1,” with
males coded as “0.” Respondents’ age (Age) was recorded (mean=52.17; SD=17.38), and
subsequently divided by 10 for interpretive ease. Respondents were also asked about their
respective racial identification. Because of heterogeneity in response options across the six
surveys, a simple binary variable, Nonwhite, was created wherein respondents who identified as
any race other than white/Caucasian were coded as “1” (31.15% of the sample), while whites
were coded as “0.” The variable Income is an ordinal measure capturing respondents’ reported
household income. Households reportedly earning less than $50k per year were coded as “0”
(50.87%); between $50k and $100k were coded as “.5” (31.40%); and over $100k were coded as
“1” (17.73%). Lastly, respondents’ highest level of educational attainment was recorded.1
Again, because of heterogeneity in question wording and response options, a five-category
variable (Education) was created, wherein “Less than High School” equals “0” (6.43%); “High
School” equals “.25” (24.64%); “Some College”/ “Vocational School”/ “Don’t Know”/
“Refused” responses equal “.50” (31.36%); “College” equals “.75” (22.49%); and “Graduate
degree” equals “1.0” (15.08%).
Differences in Means
The set of analyses reported here simply examines changes in mean approval rates of
President Obama among Democrats and Republicans. Specifically, a series of difference-in-
means analyses were conducted to determine whether, both within and between these groups,
approval rates of Obama significantly changed from pre-treatment to post-treatment. These
results are displayed in Table E1. First, the upper rows indicate that partisan groups significantly
29
differed in their approval of Obama between the pre-treatment and post-treatment periods.
Specifically, approval among Democrats fell approximately 4 percentage points (p=.01), while
Republicans increased their approval of Obama by nearly 3.5 percentage points (p=.02).2 3
In addition, Table E1 lists the results of a difference-in-differences (DID) analysis. When
the difference between Democratic and Republican approval of Obama in the pre-treatment
period is compared to this difference in the post-treatment period, we observe that the groups
became less polarized, as indicated by the negatively signed figure in the bottom row of the
penultimate column. Indeed, polarization in approval of Obama among Democrats and
Republicans contracted by over 7 percentage points (p<.001).
TABLE E1. Mean Approval of President Obama Pre- and Post-Budget Compromise
2 The change in approval of Obama among Independents did not approach statistical significance (p=.38).
3 The total sample size is noticeably larger than the models featured in the manuscript. This is because
the Pew (12/2/10-12/5/2010) survey did not include a measure for Ideology and the NBC/WSJ (12/9/10-
12/13/10) survey did not include a measure for Education. These two surveys were therefore dropped,
leaving two pre-treatment surveys and two post-treatment surveys to be included in Table 2 of the
manuscript.
TABLE X. Mean Approval Ratings of President Obama among Partisans and
Independents
Notes: Dependent variable is a dichotomous measure, with 0=Disapprove of the job Barack Obama is doing as
president, and 1=Approve of the job Barack Obama is doing as president. Entries are results of t-test analyses
with unequal variances specified. Standard errors appear in parentheses. Table demonstrates that Obama’s
mean approval among partisan groups changed significantly between pre-treatment and post-treatment, with
Republicans (Democrats) increasing (decreasing) in their approval of Obama’s job performance. (Independents
did not significantly change in their approval of Obama.) The table also indicates that, moving from the pre-
treatment period to the post-treatment period, the extent of polarization between Democrats and Republicans
regarding Obama’s job performance was reduced by over 7 percentage points. The last column indicates p-
values generated for values in the “Difference PreèPost” column (two-tailed hypothesis tests).
Pre-
treatment
Post-
treatment
Difference
PreèPost
p-value
Partisan Groups (n)
Republicans (2367)
.12 (.01)
.15 (.01)
+.03 (.01)
.02
Democrats (2416)
.85 (.01)
.81 (.01)
-.04 (.02)
.01
Difference-in-means
Democrats - Republicans
+.73 (.01)
+.66 (.01)
-.07 (.01)
.00
30
Importantly, as shown in the manuscript, the significant change in approval among
Democrats is not robust to inclusion of covariates. In investigating potential reasons for this,
additional analyses revealed that, among Democrats, both Nonwhite and Female are negatively
(and significantly) predictive of Base Alienation, which indicates that non-whites and women
disproportionately appeared in the pre-event (vis-à-vis post-event) period. These imbalances
likely explain why, without controlling for other covariates, Democrats’ approval of Obama
declines between the pre- and post-event periods.
Finally, as noted in the manuscript, while these results may at first appear modest, there
are numerous reasons why we might have expected to find null results. First, many partisans
likely did not “observe the treatment,” either directly or via social networks; thus, we should not
expect any changes in evaluations of Obama for those who were unaware that he disappointed
the Democratic base. Second, the post-treatment surveys were not immediately administered;
thus, treatment effects may have dissipated due to time itself or other intervening events. Third,
the dependent variable—presidential approval—is a broad measure, and may have been more
resistant to change vis-à-vis more nuanced outcome measures (e.g., evaluation of Obama’s
handling of the budget deal specifically). Lastly, Obama had alienated the Democratic base prior
to the budget agreement4—thus, the impact of this particular event may have been weak relative
to previous events. Given these considerations, therefore, any observed change in presidential
approval among partisans is noteworthy.
4 See, for example, the President’s 2009 decision to increase troops in Afghanistan, which liberals had
openly opposed (“Obama Afghanistan Strategy: More Troops in Quickly, Drawdown in 2011 -
CNN.Com” 2009), as well as the 2009 decision to ask Congress for a somewhat smaller stimulus package
in order to appease Congressional Republicans (Pfiffner 2011, 254). Thus, to the extent citizens had
already been exposed to some degree of “treatment,” this case study represents a fairly conservative test
of H1 and H2.
31
Supplemental Appendix E References
“Obama Afghanistan Strategy: More Troops in Quickly, Drawdown in 2011 - CNN.Com.” 2009.
CNN Politics. December 1, 2009.
http://www.cnn.com/2009/POLITICS/12/01/obama.afghanistan/.
Pfiffner, James P. 2011. “Decision Making in the Obama White House.” Presidential Studies
Quarterly 41 (2): 244–62.
32
Supplemental Appendix F
Content Analysis of Major Media Coverage of 2010 Budget Agreement
Two coders independently reviewed articles appearing in The New York Times, The
Washington Post, USA Today, and The Wall Street Journal that mentioned “Obama” during the
week of 12/06/10 to 12/13/10. A total of 75 unique articles mentioned “Obama,” and the coders
analyzed whether these articles 1) mentioned the budget agreement, and, if so, 2) allude to the
fact that Obama alienated groups aligned with the Democratic Party, and, if so, 3) allude to this
fact within the article headline itself (% agreement between coders = 94.51%; kappa=.8816,
p<.001). This last category is to acknowledge the disproportionate influence that headlines have
been found to have on consumption and interpretation of printed news (Dor 2003; Emig 1928).
Second, it should be noted that this collection of newspapers is used not because they represent
the entirety of “the treatment,” but rather because these news sources influence the news that is
carried by other newspapers, televised news, and/or online news (e.g., Atkinson, Deam, and
Uscinski 2014).
Table F1 lists the results of this analysis, which make clear that a substantial amount of
coverage noted that Obama had upset his Democratic base.
33
TABLE F1. Major Media Coverage of the 2010 Budget Agreement
Supplemental Appendix F References
Atkinson, Matthew D., Maria Deam, and Joseph E. Uscinski. 2014. “What’s a Dog Story
Worth?” PS: Political Science & Politics 47 (4): 819.
Dor, Daniel. 2003. “On Newspaper Headlines as Relevance Optimizers.” Journal of Pragmatics
35 (5): 695–721.
Emig, Elmer. 1928. “The Connotation of Newspaper Headlines.” Journalism and Mass
Communication Quarterly 4 (4): 53.
TABLE X. Major Media Coverage of the 2010 Budget Agreement Between President Obama
and Congress
NYT
WaPo
USAT
WSJ
Total
Content Analysis Categories
Number of Unique Articles Mentioning
President Obama
N=13
N=28
N=10
N=24
N=75
% of Previous Row that Mention
Budget Agreement/Tax Deal
62%
(N=8)
68%
(N=19)
67%
(N=6)
75%
(N=18)
68%
(N=51)
% of Previous Row that Allude to Base
Alienation
88%
(N=7)
68%
(N=13)
83%
(N=5)