International Journal of Communication 14(2020), 324–345 1932–8036/20200005
Copyright © 2020 (Alberto Ardèvol-Abreu and Homero Gil de Zúñiga). Licensed under the Creative
Commons Attribution Non-commercial No Derivatives (by-nc-nd). Available at http://ijoc.org.
“Obstinate Partisanship”: Political Discussion Attributes Effects
on the Development of Unconditional Party Loyalty
Universidad de La Laguna, Spain
HOMERO GIL DE ZÚÑIGA
University of Vienna, Austria
Universidad Diego Portales, Chile
Scholarly work has placed political discussion at the center of a healthier democracy.
However, this might not always be the case considering the vast amount of different
discussion attributes and their effects. This study extends existing research on the influence
of different discussion attributes (cognitive elaboration, network size, exposure to
disagreement, and online/offline discussion) on political attitude change. To do so, we
introduce the concept of obstinate partisanship and explore different discussion attributes
as its antecedents. Obstinate partisans remain loyal to their political party irrespective of its
performance, just as staunch sports fans do with their team. Results from a survey
conducted in three democracies show that discussion network size, discussion disagreement,
and offline discussion are all negative predictors of obstinate partisanship. Conversely, online
discussion fosters this negative orientation. We finally examine the moderating role of
discussion disagreement and network size on some of these relationships.
Keywords: obstinate partisanship, party identification, discussion elaboration, discussion
network size, discussion disagreement, online discussion, offline discussion
The idea that political discussion can help transcend ideological, religious, or cultural cleavages by
shaping attitudes has been the scope of great academic debate. Thus, the past decades have seen a plethora
of studies exploring the effects of political talk on a wide range of prodemocratic dispositions. These have
proven to be largely positive. For example, political talk has been found to increase discussants’ sense of
Alberto Ardèvol-Abreu: email@example.com
Homero Gil de Zúñiga: firstname.lastname@example.org
Date submitted: 2019‒06‒14
This research was supported by Grant FA2386-15-1-0003, Asian Office of Aerospace Research and
Development. Alberto Ardèvol-Abreu is funded by the Viera y Clavijo Program from the Agencia Canaria de
Investigación, Innovación y Sociedad de la Información and the Universidad de La Laguna. The authors are
grateful to James H. Liu and all participants of the Digital Influence World Project for their help with the data
collection for this study.
International Journal of Communication 14(2020) “Obstinate Partisanship” 325
political efficacy, perception of community, political knowledge, and social capital, among others (see
Burkhalter, Gastil, & Kelshaw, 2002; Wellman, Haase, Witte, & Hampton, 2001). Nonetheless, some studies
have cautioned against an overoptimistic view of political discussion. In a context of discussion with unlike-
minded people, certain individuals under certain conditions “dig in their heels” and move their views in the
direction opposite to the one advocated by their discussion partners (what is known as attitude polarization
and “repulsion effects”; Gastil, Black, & Moscovitz, 2008; Zaller, 1992).
Much less explored, however, is the influence of the different attributes of political discussion on
attitude change (Gastil et al., 2008). In this article, we extend prior research by studying how certain
interpersonal discussion attributes such as cognitive elaboration, network size, discussion disagreement,
and online/offline discussion (Gil de Zúñiga, 2017) affect individuals’ political views. To do so, we introduce
a new attitudinal outcome that we label as obstinate partisanship (henceforth referred to as OP). Building
on the group theory of politics, OP arises as a degraded outgrowth of party identification. An obstinate
attitude can be explained in terms of in-group–out-group perceptions, because it emerges from feelings of
belonging to a party-based political community in competition for power with other distinct political
communities. Connected to these feelings and perceptions, obstinate partisans develop unconditional loyalty
and commitment to their political party, which in turns leads to an increased tendency to support it—mainly
in the form of voting—regardless of policy proposals or party performance.
Using data from a demographically diverse survey collected in three countries (Spain, United
States, and New Zealand), we find that, first, OP is a viable, reliable, and valid construct. Second, OP also
performed distinctively when examining convergent and discriminant associations with the related construct
of strength of ideology. Third, results reveal that having large discussion networks, being frequently exposed
to disagreement, and discussing politics face-to-face help individuals mitigate their political attitude of
obstinacy. On the contrary, online political discussions seems to foster an obstinate attitude in the political
realm. Finally, the study also examines the moderating role of discussion disagreement and network size
on the relationships between online/offline discussion and OP.
Party Identification as a Substrate of OP
Party preference and stability of vote choice have received considerable attention from political
science and sociology scholars during the past decades (Smith & Mackie, 2007). From the United States to
Italy, from Great Britain to Spain, a substantial number of voters demonstrate a long-term support and
commitment to a political party. Behind this tendency there lies what has been called, depending of the
theoretical perspective, “party identification” or “partisanship” (Bartle & Bellucci, 2014; Butler & Stokes, 1974).
Social-psychological approaches explain these predispositions in terms of group identification.
According to expressive accounts, individuals perceive their political party as a social group to which they
belong, and through which they partially define and express their social identity (Smith & Mackie, 2007).
This “we feeling” often results in an enduring emotional attachment that explains that many partisans enjoy
the victory of their side per se, “quite apart from the uses which the party might make of power” (Butler &
Stokes, 1974, p. 37). This political identity is therefore rather stable and relatively immune to external
events, but does not imply unconditional or blind loyalty to one’s party. From a different approach,
326 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
instrumental perspectives explain partisanship as a rational choice of voters, one that is consistent with
their issue positions and reexamined in the light of external events (Fiorina, 1981; MacKuen, Erikson, &
Stimson, 1989). Although these permanently updated evaluations generate mobility in the direction and
intensity of party preference, the dynamic is limited because newly acquired information is compared with
prior knowledge, in a “summing up of considerations accumulated to date in the voter’s life” (Johnston,
2006, p. 333; see Fiorina, 1981).
Building on expressive and instrumental perspectives, this study introduces the concept of OP,
understood as a quarrelsome, factitious, and degraded outgrowth of party identification. Obstinate partisans
are a specific subgroup of partisans that remain completely indifferent to any short-term force, offering their
parties a democratically dangerous “all-weather support” (see Bartle & Bellucci, 2014). Contrary to the
democratic ideal of holding political power accountable, obstinate ones deprive their vote of any meaningful
function, giving political actors free hands to seek their self-interest.
Although they are different concepts, we posit that party identification and OP are interrelated.
First, extensive empirical evidence indicates that party identification is directly and strongly related to vote
stability, which suggests partisan voters are comparatively less reactive to external events—including party
performance or policy actions (Bartle & Bellucci, 2014; Converse, 1964). Second, and related to the above,
strong partisans’ political views and attitudes tend to be stronger and more resistant to change, making
persuasion, even in politically heterogeneous environments, less likely (Taber & Lodge, 2006). Finally, the
affective component that connects a partisan to their party provides a special motivation to be consistent in
their political attitudes and behaviors (see “affective consistency” in Sniderman & Bullock, 2004). All these
forces pushing strong partisans toward opinion reinforcement give us reasons to argue for a theoretical and
empirical connection between party identity and OP:
H1: Strength of party identification is positively associated to obstinate partisanship (OP).
Exposure to (Contentious) Discussion: Moderating and Polarizing Effects
For the proponents of the deliberative theory, political talk is the most appropriate way to tackle
public disputes and reach a rational consensus with a view to the common good (Cohen, 1989; Dahl, 1989).
The basic assumption is that political talk provides opportunities to become aware of different opinions and
perspectives, making it more likely for discussants to reconsider their views and shift their attitudes toward
common ground (Cohen, 2007). Along these lines, research examining citizen-to-citizen political discussion
has generally found positive effects on a range of prodemocratic outcomes, including perceptions of
community, internal efficacy, or political participation, among many others (Burkhalter et al., 2002; Wellman
et al., 2001).
Notwithstanding the above positive picture, the academic debate on the effects of political talk on
attitudes and behaviors is far from over. As Wojcieszak and Price (2010) note, one research area that
deserves further attention and clarification is the relationship between discussion disagreement and attitude
change. Although deliberative theorists hope that political talk across lines of difference can help bring
discussants together, a growing literature about biased processing suggests that this is not always the case
International Journal of Communication 14(2020) “Obstinate Partisanship” 327
(Gastil et al., 2008; Taber & Lodge, 2006; Wojcieszak & Price, 2010). When exposed to disagreement, some
discussants under specific circumstances will “dig in their heels” so that their initial attitudes become
stronger in the direction of polarization (Festinger, Riecken, & Schachter, 1956; Taber & Lodge, 2006).
Empirical evidence on the issue provides mixed results, suggesting that both deliberative theory
and confirmation bias each explain a different part of the puzzle. It seems reasonable that different
discussion contexts (e.g., individuals’ investment in the issues under discussion, ideological distance
between discussants, attributes of the discussion) result in different attitudinal outcomes (Gastil et al., 2008;
Taber & Lodge, 2006). For example, in a study examining the effects of online discussion disagreement on
attitudes about sexual minority rights, Wojcieszak and Price (2010) found evidence of both polarization and
moderation of prediscussion political positions. On the one hand, strong opponents of sexual minority rights
increased their oppositional views after the discussions (polarization effect). On the other hand, strong
supporters of gay and lesbian rights at the outset felt less favorable toward these minorities as a result of
the exposure to other opinions (moderation effect).
Three different psychological mechanisms have been proposed to explain the sometimes-observed
relationship between exposure to disagreement and attitude polarization (Taber & Lodge, 2006; Zaller,
1992). First, and according to the prior attitude effect, congenial arguments are usually perceived as
stronger and more convincing than uncongenial or challenging information. Second, people tend to spend
significant mental resources on scrutinizing—and often denigrating—arguments with which they disagree,
while they usually accept agreeable arguments at face value. This has been called disconfirmation bias.
Finally, when people are free to choose the content they want to be exposed to, the confirmation bias will
make them prefer attitude-consistent information and avoid counter attitudinal messages—a phenomenon
also called selective exposure. Although prior attitude effect and disconfirmation bias can be triggered in
any discussion involving disagreement, face-to-face discussants frequently cannot filter out opposing
arguments coming from discussion partners in spontaneous discussions—unless they decide not to talk to
them again. In these cases, the confirmation bias will hardly work. Discussion in the online arena is different,
because Internet users can easily seek for opinion-reinforcing discussion spaces and screen out opinion-
challenging ones (“echo chamber” effect; see Iyengar & Hahn, 2009). In this highly personalized digital
environment, selective exposure can more easily come into play.
Discussion Attributes and OP
This study uses OP as a probe or pointer of political polarization. Following the above theoretical
accounts, we expect different discussion attributes to differently affect individuals’ level of OP (i.e., to move
them toward either temperance or polarization). While partisanship (and probably also OP) is a relatively
stable trait, even the more canonical expressive perspectives concede that its enduring nature “wanes
somewhat when, year after year, it presides over hard times or lacks effective leadership” (Green Palmquist,
& Schickler, 1998, p. 986). Closer to the democratic ideal, instrumental approaches view partisanship as
partially determined by short-term influences such as political and economic performance (MacKuen et al.,
1989) or personal evaluations of the candidates (Garzia, 2011). In this study, we understand OP as a
relatively stable individual attitude, subject to change as a result one’s involvement with political discussion
328 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
Discussion Cognitive Elaboration
When individuals are confronted with new information about their environment, they need to
engage in intellectual effort to integrate the gained knowledge and, if appropriate, reorganize their
existing cognitions (Perse, 1990). In the context of political talk, some participants strive to develop a
coherent discourse, make reasoned replies to their discussion partners’ considerations, and connect new
information to their prior knowledge and experiences (Hively & Eveland, 2009; Petty, Briñol, & Priester,
2009). These intellectual strategies, collectively referred to as cognitive elaboration, have been
theoretically and empirically connected to long-lasting and resistant attitude change (Petty et al., 2009).
Based on these observations, we expect discussion cognitive elaboration to affect discussants’ levels of
OP. What is more difficult to predict is the direction of the change. Following deliberative theorists, it
could be expected that cognitive efforts made during or after the discussion would moderate extreme
positions, because participants would be trying to integrate others’ point of view into their cognitive
models. However, previous work on biased perception has found that this is not always the case, as
discussants tend to allocate more intellectual resources to “denigrate or counter [contrary] arguments
and bolster their prior convictions” (Taber & Lodge, 2006, p. 762). It could therefore be the case that
these cognitive efforts would end up reinforcing prior attitudes (such as OP). Based on these conflicting
accounts, we ask our first research question:
RQ1: What is the relationship between discussion cognitive elaboration and OP?
Discussion Network Size and Discussion Disagreement
For individual political attitudes to change, one should have the opportunity to come in contact
with different information, ideas, and opinions. Some early findings from social comparison research fit
these intuitive expectations, in the sense that deliberation tends to decrease within-group attitude
variance and guide group opinion in the direction of the majority (see Gastil et al., 2008; Wojcieszac &
Price, 2010). Under this theoretical framework, individuals’ size of their discussion network and their
frequency of exposure to disagreement should increase their familiarity with alternative, legitimate
perspectives, and thus provide them with more balanced judgments (Wojcieszac & Price, 2010).
However, as explained above, messages that question one’s political attitudes—especially those
connected to one’s social identity—are more likely to activate perceptual biases (e.g., prior attitude
effect and confirmation bias). Because of these conflicting theoretical and empirical accounts, we pose
the following research questions:
RQ2: What is the relationship between discussion network size and OP?
RQ3: What is the relationship between discussion disagreement and OP?
Offline and Online Political Discussion
Until just a few decades ago, all forms of political discussion demanded face-to-face interaction,
except in the rare cases when discussants talked about politics over the telephone. More recently,
International Journal of Communication 14(2020) “Obstinate Partisanship” 329
technological developments in communication have opened new venues for political talk, in the context of
online discussions. Interactive features of social media, news webs, blogs, and mobile messaging apps allow
individuals to engage, at minimal cost, in synchronous or asynchronous discussions about current events
and politics (Gil de Zúñiga, Ardèvol-Abreu, & Casero-Ripollés, 2019). Following the above theoretical
considerations, we expect discussants to respond—either in the direction of moderation or polarization—to
the ideological diversity of their discussion environment. In other words, what is important in our theoretical
approach is whether individuals are locked in an “echo chamber” of political discussion, or their political
conversation networks are inclusive of different perspectives. To explore these influences, and considering
the conflicting findings regarding attitudinal responses to disagreement, we test (1) the main effect of offline
and online discussion on OP, and (2) the potential interaction with our two indicators of discussion
heterogeneity (discussion disagreement and network size):
RQ4: What is the relationship between offline discussion and OP?
RQ5: What is the relationship between online discussion and OP?
RQ6: Does discussion disagreement moderate the effect of offline (RQ6a) and online (RQ6b) discussion
frequency on OP?
RQ7: Does discussion network size moderate the effect of offline (RQ7a) and online (RQ7b) discussion
frequency on OP?
The data used for this article come from a panel study conducted collaboratively between research
groups based at Massey University (New Zealand) and—at that time—at the University of Vienna (Austria).
We collected online survey data from 20 countries and two cities from Europe, the Americas, Asia, and South
Africa (see the Appendix in Gil de Zúñiga, Ardèvol-Abreu, Diehl, Patiño, & Liu, 2019, for details). To deploy
the surveys, researchers contracted Nielsen, a polling firm based in the United States that partners with
companies providing opt-in panel respondents from all over the world. Nielsen generated samples at the
country level whose characteristics mirrored those reported by official census agencies in at least three key
demographic parameters. The first wave (W1) of the data was collected from September 14 to 24, 2015.
For the second wave (W2), the same participants were asked to reanswer the questionnaire between March
22 and April 1, 2016.
The three questions of the OP scale were asked only in the second wave of the survey in three of
these 22 countries: Spain, United States, and New Zealand. We chose these specific countries because we
sought to examine the novel construct in a culturally diverse sample, which increases the external validity
of the study. At the same time, we needed to pinpoint certain social and political homogeneity, especially in
terms of democratic culture, freedom of expression and of the press, in a multiparty—or at least biparty—
330 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
political system. The three selected countries are well-established Western democracies from three different
continents, which may serve to provide a benchmark for future comparisons with other contexts.
The Democracy Index of 2015 (Economist Intelligence Unit, 2016) rates all three countries as “full
democracies,” among the first 20 positions in the full list of 165 analyzed countries. In addition, citizens of
the three countries live in “free media environments” (Dunham, 2016) and are also free to engage in open
political debate. In W1, overall cooperation rate averaged 77% across the 22-country sample (American
Association for Public Opinion Research, 2016). For this study, the total sample size was 3,337 respondents
in W1: Spain (n1 = 1,019); United States (n1 = 1,161); New Zealand (n1 = 1,157). The retention rate for W2
was 41.83%, for an overall sample size of 1,396 respondents: Spain (n2 = 302); United States (n2 = 489);
New Zealand (n2 = 605).
Variables of Interest
Our dependent variable comprises attitudes of blind passion and unconditional fidelity to a certain
political party. Because voters in democracies are free to choose the criteria by which they evaluate party
performance, our item wording does not prime any external standard (i.e., the media, approval ratings in
polls) that respondents could use as a benchmark. The statements we used allow for respondents’ own
standards to emerge: “I will always vote for the same political party, no matter what they do,” “I support
my political party, even when they make a mistake,” and “being loyal to my party is important, both when
they are doing well and not so well” (three items rated on a 7-point scale, 1 = strongly disagree to 7 =
strongly agree; W2 Cronbach’s α = .89, M = 3.18, SD = 1.58). Among all respondents, 50.7% scored 3 or
less on the OP scale (3 = partially disagree); 38% ranged >3 to ≤5 (5 = partially agree); 7.6% ranged >5
to ≤6 (6 = agree with the statements); and 3.7% scored more than 6.
This index measures individuals’ relative frequency of political talk with non-like-minded individuals
(see Gil de Zúñiga, 2017). We created a subtractive index of relative exposure to political disagreement, in
which participants’ frequency of discussion with “people whose political views are similar to [theirs]” was
subtracted from the frequency of discussion with “people whose political views are different from [theirs]”
(1 = never to 7 = all the time). We then recoded the result variable to a 7-point scale, where 1 represents
respondents with the lowest proportion of disagreement in their political discussion diet, and 7 represents
those with the highest (W1 M = 2.78, SD = 0.58).
Building on previous measures of the construct (Cho et al., 2009), we asked respondents how often
(1 = never to 7 = all the time) they talk about politics or public affairs online with “spouse or partner,”
“family, relatives, or friends,” “acquaintances,” and “strangers” (four items averaged scale; W1 Cronbach’s
α = .88, M = 2.16, SD = 1.42).
International Journal of Communication 14(2020) “Obstinate Partisanship” 331
We asked respondents the same question as for online discussion frequency, but this time referring
to discussions “face-to-face or over the phone” (four items averaged scale, W1 Cronbach’s α = .80, M =
3.17, SD = 1.32).
This variable measures the ability of political discussion to stimulate thinking at the individual level
(see Hively & Eveland, 2009). On a 7-point scale, respondents indicated their level of agreement with the
following assessments: “I often find myself thinking about my conversations with other people about politics
and public affairs after the discussion has ended,” and “I often think about how my conversations with other
people about politics and public affairs relate to other things I know” (W1 Spearman–Brown ρ = .93, M =
3.50, SD = 1.74).
Discussion Network Size
We asked respondents to report on the number of people they talked to about politics or public
affairs, on and offline, during the preceding month (two open-ended questions; see Gil de Zúñiga, Ardèvol-
Abreu, & Casero-Ripollés, 2019). The resulting averaged scale was highly skewed (W¹ M = 5.85, Mdn =
2.00, SD = 22.13, skewness = 14.79). To minimize the effect of this interference, we transformed the
variable by using the natural logarithm (W¹ M = .43, Mdn = .35, SD = .41, skewness = 1.09).
Strength of Ideological Identification
We used this aggregate measure as a proxy for strength of party identification. While both concepts
are not equivalent, highly ideologized respondents are more likely to be strong partisans, and the reversed.
We asked respondents to indicate their ideological orientation (0 = strong conservative/right-leaning to 10
= strong liberal/left-leaning) with regard to “political issues” (e.g., government size or functioning),
“economic issues” (e.g., taxation policies, trade), and “social issues” (e.g., gun control, abortion). Each item
was then folded into a 6-point scale (from 0 = no ideological identification to 6 = strong ideological
identification, either with conservatives/right-wingers or liberals/left-wingers). The final aggregate measure
was rescaled to 0–10 (W1 Cronbach’s α = .91, M = 3.81, SD = 2.90).
Our models included two measures of news media uses as control variables: Internet news use (W1
Cronbach’s α = .79, M = 4.07, SD = 1.43) and traditional news use (W1 Cronbach’s α = .57, M = 4.53, SD
= 1.32). We also expected different forms of social trust to have cross-influences with some of the key
variables of the study. For example, individuals with extreme positions on an issue (and probably obstinate
partisans as well) tend to believe that the mass media is biased against their side, and are therefore more
likely to report lower levels of media trust regarding the coverage of that issue (Gunther, 1988). Similarly,
political trust exerts a positive effect on voters’ feelings about incumbents and parties (Hetherington, 1998).
332 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
To isolate these confounding effects, our models included measures of media trust (W¹ Cronbach’s α = .72,
M = 3.28, SD = 1.06) and political trust (W¹ Cronbach’s α = .86, M = 3.11, SD = 1.28).
We also included a set of sociopolitical antecedents that have been frequently associated to various
political attitudes, orientations, and behaviors: political knowledge (W¹ Cronbach’s α = .48, M = 1.61, SD
= 0.88), and internal political efficacy (W¹ Sp earma n–Brown coefficient = .66, M = 3.60, SD = 1.50). Finally,
we also included five demographic control variables: age (M = 47.00 years, SD = 16.21), gender (55.7%
female), race (84.8% Whites), income self-perception (from 1 = people who are the least well off in society
to 10 = people who are the most well off; M = 5.87, SD = 1.83), and education (1 = elementary school to
6 = graduate school or higher; M = 4.11, Mdn = some college).
OP showed a very good internal consistency: whole sample W2 Cronbach’s α = .89; Spain = .93,
United States = .89, New Zealand = .86. However, the equivalence of the measure between countries
cannot be ascertained until invariance tests are performed. This can be achieved by evaluating the model
fit of three increasingly restrictive multigroup confirmatory factor analyses (CFAs) testing for configural,
metric, and scalar invariance (Byrne, 2012).
In the first model—referred to as baseline—the only constraint
is that the items are associated with the same latent variable across countries (configural). Building on
configural invariance, metric invariance is attained if the factor loadings of the manifest items are equivalent
across countries. Achieving metric invariance indicates that the relationships between the manifest items
and the latent factor are the same across groups (i.e., the construct has the same meaning across
countries). Finally, and building on metric invariance, scalar invariance requires the equivalence of items’
intercepts (Byrne, 2012). Achieving scalar invariance is necessary for making meaningful mean-level
comparisons, which is beyond the scope of this study.
CFAs were conducted using Mplus (Version 8.0).
International Journal of Communication 14(2020) “Obstinate Partisanship” 333
Table 1. Zero Order Correlations Among Independent (W1) and Dependent (W2) Variables in the Study.
Note. Cell entries are zero-order correlation coefficients (n = 1,313). ap < .05; bp < .01; cp < .001.
2. Gender (female)
3. Race (White)
6. Strength of ideology
7. Political knowledge
8. Internal efficacy
9. Political trust
10. Media trust
11. Traditional news use
12. Internet news use
13. Discussion network size
14. Discussion elaboration
15. Offline discussion
16. Online discussion
17. Discussion disagreement
18.Obstinate partisanship W2
334 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
Tables 2 and 3 summarize the results of these nested CFAs. Since the configural model was just
identified (27 parameters and 0 degrees of freedom), model fit could not be assessed. However, models
with zero degrees of freedom can still be compared with other nested models, considering the value of zero
for chi-square and zero for the degrees of freedom (see Muthen, 2014).
Table 2. Factor Loadings for the Configural Model Across Countries.
Note. Overall sample: n = 1,436. Spain: n = 302; United States: n = 528; New Zealand: n = 606.
Coefficients are standardized.
As reported in Table 3, metric invariance model indicates good fit to the data, showing no
differences with the fit of the configural model (chi-square difference test), χ2(4, n = 1,313) = 6.33, p =
.18. This result is suggestive of factor loadings invariance across countries. Scalar invariance, however, was
not attained, which prevents us from making mean-level comparisons between countries. As the chi-square
difference test indicates, the scalar model has a significantly poorer fit than the metric, χ2(4, n = 1,313) =
56.91, p < .001.
Table 3. Measurement Invariance Tests—Model Fit Indices.
To further validate our measure of OP, we assessed the scale in terms of convergent and
discriminant validity. We first estimated the relationship between the measure of OP and the measure
of the separate but related concept of strength of ideology (see Campbell & Fiske, 1959). Table 4 shows
the multitrait–multimethod correlation matrix of OP (items H1–H3) and strength of ideology measures
(items I4–I6). Figures show a consistent pattern of lower values for discriminant coefficients (roman
type) than for convergent coefficients (bold type), which can be interpreted as supportive evidence of
construct validity (Campbell & Fiske, 1959; Raykov, 2011). Second, we confirmed the statistical
significance of these differences between discriminant and convergent correlation point estimates with
the use of latent variable modeling methods (see Raykov, 2011, Model 4). With this factor analytical
approach, we calculated a confident interval for mean difference in convergent and discriminant validity
coefficients, M = .636, 95% CI [.598, .674]. This means that in the population, the true difference
between convergent and discriminant validity coefficients (in favor of the former) ranges between .598
Always vote for the same party
Always support my party
Being loyal my party is important
International Journal of Communication 14(2020) “Obstinate Partisanship” 335
Table 4. Multitrait–Multimethod Matrix Obstinate Partisanship/Strength of Ideology.
H1. Always vote for . . .
H2. Always support my party . . .
H3. “Being loyal to ‘my’ . . . .”
I4. “On political issues . . .”
I5. “On economic issues . . .”
I6. “On social issues . . .”
Note. Items H1–H3 comprise the measure of obstinacy, whereas I4–I6 comprise the measure of strength
of ideology. Convergent validity coefficients are in boldface, and discriminant coefficients are in roman
type. Cell entries are zero-order correlation coefficients (n = 1,291). cp < .001.
H1 predicted a positive effect of strength of party identification on OP. We found empirical support
for H1. The first lagged regression model
in Table 5 (first column) shows a direct, positive influence of strength
of ideology (used as a proxy measure) on OP (β = .148, p < .001). Within this first regression analysis, the
sociopolitical antecedents block explained the most part of variance in obstinacy (ΔR2 = .071, p < .001),
followed by discussion attributes I (ΔR2 = .033, p < .001), demographics (ΔR2 = .014, p < .01), and discussion
attributes II (ΔR2 = .007, p < .05). The third block, containing news use variables, did not reach statistical
significance (ΔR2 = .001, p = .89). Within the block of sociopolitical controls, both trust variables (political
trust, β = .157, p < .001; and media trust, β = .120, p < .001) were positively and strongly related to
obstinacy. Within the first block of demographic variables, age was a positive predictor of obstinacy (β = .090,
p < .01), whereas education level was negatively related to the dependent variable (β = .062, p < .05). That
is, older individuals and those with lower education tend to show higher levels of OP.
Regression analyses were performed using SPSS Statistics 21.0, with the support of the PROCESS macro
(Hayes, 2013) to test moderation effects.
336 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
Table 5. Lagged Regression Models Predicting Obstinate Partisanship
Obstinate Partisanship W²
Block 1: Demographics W¹
Gender (1 = female)
Race (1 = White)
Block 2: Sociopolitical antecedents W¹
Strength of ideology
Internal political efficacy
Block 3: News media uses W¹
Traditional news media use
Internet use for news
Block 4: Discussion attributes I W¹
Discussion network size
Block 5: Discussion attributes II W¹
Offline political discussion
Online political discussion
Block 6: Moderation effects W1
Offline Discussion × Disagreement
Online Discussion × Disagreement
Offline Discussion × Network Size
Online Discussion × Network Size
Note. Cell entries are beta coefficients from a lagged ordinary least squares regression. To maximize
statistical power, missing values on control variables (except gender and race) have been replaced with
the mean. Sample size = 1,110. *p < .05; **p < .10; ***p < .001 (two-tailed). W1 = Wave 1, W2 =
International Journal of Communication 14(2020) “Obstinate Partisanship” 337
These overall outcomes become more meaningful when we account for moderating effects, as
addressed in RQ6 and RQ7 (second and third columns in Table 5, plots in Figures 1 and 2). Research
Question 6 asked about the possible moderating influence of discussion disagreement on the relationships
between offline (RQ6a) and online discussion (RQ6b; predictors) and OP (outcome variable). We found that
exposure to disagreement does not moderate the effect of face-to-face political discussions on obstinacy
(second model in Table 5; β = −.015, p = .93). Thus, offline political discussions, regardless of their level
of disagreement, seem to have a negative effect on the development of OP (i.e., frequent offline discussants
tend to reduce their unconditional loyalty to their party).
Looking at the interaction in the online arena (RQ6b), we did find a moderating role of discussion
disagreement on the relationship between Internet discussion and OP (β = .284, p < .05). However, the
effect is not as straightforward as might be expected. As Figure 1 shows, the positive, direct effect of online
discussion on obstinacy is rather small for those with a lower exposure to disagreement. That is, those who
primarily discuss politics with people they agree with will show higher base levels of obstinacy, and
increasing their frequency of online discussion will barely raise their already high obstinate attitude.
However, for those who encounter high levels of disagreement in their discussions, online political talk has
a markedly positive effect: the more often they discuss online, the more obstinate they become.
Figure 1. Interaction between online political discussion and exposure to disagreement (W1) on
obstinate partisanship (W2). The R2 increase due to interaction is statistically significant: F(1,
1091) = 4.81, p < .05. n = 1,110. Interaction estimated from model in Table 3. Values for the
moderator are the mean and plus/minus one standard deviation from the mean.
Complementary to RQ6, RQ7 aimed to explore the moderating effect of discussion network size on
the relationships offline discussion—obstinacy (RQ7a), and online discussion—obstinacy (RQ7b). As in the
previous set of research questions, we found a moderating effect of network size only for online (third model
in Table 5; β = −.245, p < .05), but not for offline (β = .060, p = .63) discussions. Thus, similarly as above,
offline discussions contribute to reducing polarization (i.e., to toning down obstinate attitudes), irrespective
of the size of the discussion network. In online environments, however, the effect of political discussion on
338 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
obstinacy is conditioned by discussion network size (RQ7b; see Figure 2). For those respondents with a
smaller discussion network, talking politics on the Internet has a strong, direct effect on obstinacy: The
more they discuss, the higher their levels of OP. For this group of respondents, this finding may be of
particular concern, as they already show higher base levels of obstinacy (Figure 2). For the other group of
respondents (those with a larger discussion network), the relationship between online discussion and OP is
also positive (the more they discuss online, the higher they score on obstinacy in time), although the
strength of the effect is much smaller.
Figure 2. Interaction between online political discussion and size of the discussion network (W1)
on obstinate partisanship (W2). The R2 increase due to interaction is statistically significant: F(1,
1091) = 6.88, p < .01. n = 1,110. Interaction estimated from model in Table 3. Values for the
moderator are the mean and plus/minus one standard deviation from the mean.
Research Question 1 asked about the effect of discussion elaboration on OP. Within the first block
of discussion attributes variables (see Table 5), discussion cognitive elaboration showed no meaningful
effects on the dependent variable (β = −.006, p = .87). Differently, we did find a direct, negative influence
of discussion network size on OP (RQ2; β = −.158, p < .001). Also, in Table 5, a similar negative effect on
obstinacy was found for discussion disagreement (RQ3; β = −.133, p < .001). That is, respondents
discussing politics with a larger number of others, and more frequently exposed to disagreement during
their conversations (in W1) tend to show lower levels OP (in W2).
With respect to RQ4 and RQ5, we found the effects of offline and online discussion on OP to be
opposite in sign (see Table 5, fifth block): Offline political discussion has a negative influence on obstinacy
(β = −.100, p < .01), and the effect of online discussion on obstinacy is positive (β = .074, p < .05). In
other words, respondents who more frequently talk about politics face-to-face tend to moderate the extreme
attitudes that characterize OP, while those who regularly discuss public affairs via online media are more
likely to “dig in their heels” and become more extreme in their positions.
International Journal of Communication 14(2020) “Obstinate Partisanship” 339
Dialogue between people representing different interests, views, and opinions is credited by many as
intrinsic to democracy. Proponents of deliberative theories argue that political talk should be the main tool to
resolve deep social conflicts along ideological, cultural, or religious cleavages, and thereby prevent political
polarization (Cohen, 1989; Dahl, 1989). As research on biased perception indicates, however, not all is good
about political discussion. Exposure to disagreement—for certain individuals and under certain conditions—can
lead to attitude change in the opposite direction than expected. Put differently, discussants can “dig in their
heels” and shift their attitudes in the direction of strengthening prior beliefs, resulting in polarization (Gastil et
al., 2008; Zaller, 1992).
In this study, we introduce the concept of OP as an indicator of polarization. Building on expressive
and instrumental approaches to partisanship, we conceptualize OP as a negative drift of the feelings of party
identification. Our results indicate that the construct has the same meaning to the respondents across countries
as different as Spain, the United States, and New Zealand. According to our theoretical perspective, obstinate
partisans tend to simply disregard their party performance as a benchmark to guide their vote. The issue is no
longer that obstinate individuals use partisan heuristics to evaluate party candidates, or that they distort new
political information to avoid any threat to their partisan identity (Taber & Lodge, 2006). Thus, even though
political obstinates may consider that their party is performing badly, making mistakes, or not doing well, they
will remain loyal and not change their vote.
Another contribution of the study is the description of the distinct but complementary nature of the
concepts of partisanship and OP. As convergent and discriminant validity test confirm, these two concepts are
theoretically and empirically related, but not equivalent. Obstinate partisans are necessarily strong partisans,
but not all those showing high levels of partisanship are automatically obstinate partisans. Thus, OP may serve
as a predictor of how different persons (including strong partisans) will respond to external events related to
political parties and leaders. Partisans with low levels of OP will be more likely to reevaluate and adjust their
party preferences according to external events (e.g., party and candidate issue positions, economic and political
performance), as instrumental approaches to partisanship suggest. Conversely, and more in line with expressive
perspectives, partisans that also score high on OP will be more likely to be unconditionally loyal an essentially
immune to any external event.
Our results also shed some light on the important question of which factors determine an obstinate
attitude, providing us with some hints on how to counter their negative effects in the interest of democracy. We
found that different discussion attributes affect OP differently, which makes us believe that both deliberative
theorists and researchers on confirmation bias provide different pieces of the puzzle of the effects of exposure
to disagreement. First, we found that discussion cognitive elaboration has no effect on obstinacy. Because prior
research indicates that cognitive elaboration is an antecedent of attitude change (Petty et al., 2009; Taber &
Lodge, 2006), we suspect that this absence of aggregate effects of cognitive elaboration is concealing individual
attitude change in both directions. In other words, depending on individual and contextual factors, discussion
cognitive elaboration could lead to either increases or decreases in obstinacy. Previous, more general findings
on the effects of cognitive elaboration on attitude change offer some support for this interpretation. For example,
according to Petty and Cacioppo’s (1986) elaboration likelihood model, extensive issue or argument cognitive
340 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
processing can be conducive to either persuasion (if the received messages evoke positive thoughts) or
resistance (if the messages elicit counter-arguing; see also Petty et al., 2009).
Secondly, and consistent with the expectations of deliberative theory, we found an overall negative
effect of both discussion network size and exposure to disagreement on OP. This means that those with larger
and ideologically diverse discussion networks are more likely to critically examine the performance of political
parties and use their vote as a means of rewarding or punishing them at the ballot box. On the contrary, people
with smaller and like-minded political discussion networks tend to show a more intense obstinate attitude,
remaining loyal to their political parties. These findings, however, do not preclude the possibility that biased
processing occurs in large and ideologically diverse discussion networks—at least for some people or under
certain circumstances. For example, the emotional investment of the discussants in the topic, the perceived
democratic quality of the discussion, or the strength of discussion network ties could moderate these
relationships and determine different outcomes for different groups or in different situations. Further research
should explore possible indirect and conditional effects of network size and disagreement on obstinacy.
Finally, the direct effects of offline and online discussion on OP are opposite in sign. Thus, offline
discussion seems to have an overall beneficial influence on obstinacy, meaning that those who discuss politics
face-to-face on a frequent basis tend to score lower on OP. The same cannot be said for online talk. Our findings
suggest that those who frequently talk about politics on the Internet will ultimately increase their OP levels. A
number of reasons could explain these opposing effects of offline versus online discussion on obstinacy, but we
shall attempt a theoretical justification. Some scholars have raised concerns about the (frequently low) quality
and (not always) democratic nature of political talk in the online arena (Graham & Wright, 2014; Noveck, 2000).
Informal political talk in online forums, blogs, and social media is frequently driven by emotions, and it is not
uncommon that it uses low levels of argumentation and underdeveloped opinions (Stromer-Galley, 2014). Not
only that, Internet-based discussions often fail to adhere to rules of etiquette, such as showing respect for
different opinions, listening to others, or refraining from ad hominem attacks (Stromer-Galley, 2014; see also
Papacharissi, 2002). When discussions do not meet the principles of democracy and mutual respect, persuasion
is less likely to occur, and discussants tend to buttress their prediscussion attitudes (i.e., they polarize their
views; Gastil et al., 2008). Differently, the actual physical presence in face-to-face discussions may reduce
psychological distances and increase the possibilities that discussion develops in more cordial and democratic
ways, thus contributing to a persuasion-friendly context.
These negative findings for online discussion must, however, be nuanced in light of the interaction
effects of discussion disagreement and network size. We should first note that in both interaction models,
exposure to different ideas and large discussion networks is more beneficial—in terms of subsequent levels of
OP—than the confinement in small, ideologically homogeneous discussion groups. In other words, for individuals
who use the Internet to discuss politics, no matter how frequently, the more they are exposed to disagreement
and the larger their discussion networks, the less intense their obstinate attitude will tend to be in time.
Moreover, it should be pointed out that individual levels of OP tend to increase with online discussion frequency,
even in the presence of large discussion networks, or—even more intensely—high disagreement. These findings
suggest that when it comes to preventing the development of OP, it would be advisable to discuss politics online
only in small doses, and in politically heterogeneous and large networks. In line with previous findings, these
International Journal of Communication 14(2020) “Obstinate Partisanship” 341
results indicate that frequently discussing politics online within “echo chambers” of ideological homogeneity
hampers the development of a critically engaged, cohesive society.
Political discussion attributes seem to be important, but not exclusive, antecedents of OP. An important
question considering this finding is how OP attitudes develop. We believe that expressive partisanship is the
foundation on which OP is built, and our results provide support for this assumption. But the process through
which a political partisan degenerates into an obstinate partisan may be complicated and prolonged. We
encourage future multilevel studies to examine social and cultural influences (e.g., media and educational
systems, type of governance and institutional frameworks, religious and socioeconomic cleavages) that may
hinder or favor the development of OP. At the individual level, future research should also explore personality
(e.g., agreeableness, openness) and cognitive traits (e.g., perspective taking, cognitive ability) that may be
associated with OP.
Another potentially fruitful avenue for future research could focus on the associations between OP and
trust in institutional actors (see Gil de Zúñiga, Ardèvol-Abreu, Diehl, et al., 2019). Following previous research,
we believe obstinate partisans who identify with “mainstream” parties will be more likely to develop positive
attitudes toward the political system overall, although this may be different for obstinate supporters of
“antisystem” and “outsider” parties (Söderlund & Kestilä-Kekkonen, 2009). Similarly, further research could
examine whether the positive relationship found in this study between media trust and OP relates to selective
exposure mechanisms. Obstinate partisans may select ideologically consistent news sources and show increased
trust not in “the media,” but in “their” (biased) media (Iyengar & Hahn, 2009).
The findings of this study should be interpreted in the light of its limitations. The three survey questions
for our measure of OP were only included in the second wave of the study. This forced us to test the effects of
political discussion in time base using lagged regression models, which is more appropriate than a cross-sectional
approach. However, the lack of a measure of previous levels of obstinacy (W1) prevents us from taking full
advantage of raw change score analysis, baseline-adjusted change score analysis, or autoregressive model
analysis, which are better suited to address causal relationships. In a related vein, political discussion attributes
may be considered as either antecedents or outcomes of individual levels of OP. Considering OP as a “structural
characteristic” of the individual or a more dynamic “outcome attitude” has empirical consequences concerning
the inclusion of OP as an independent or dependent variable (Cho et al., 2009; McLeod, Kosicki, & McLeod,
2009). The relationship between political discussion and OP is likely to be reciprocally influential, OP being both
a cause and effect of discussion attributes. This remains an open question for future research.
American Association for Public Opinion Research. (2016). Standard definitions: Final dispositions of case
codes and outcome rates for surveys. Retrieved from http://www.aapor.org/AAPOR_Main/media/
342 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
Bartle, J., & Bellucci, P. (2014). Introduction. In J. Bartle & P. Bellucci (Eds.), Political parties and
partisanship: Social identity and individual attitudes (pp. 1–25). New York, NY: Routledge/ECPR
Studies in European Political Science.
Burkhalter, S., Gastil, J., & Kelshaw, T. (2002). A conceptual definition and theoretical model of public
deliberation in small face-to-face groups. Communication Theory, 12(4), 398–422.
Butler, D., & Stokes, D. E. (1974). Electoral change in Britain. London, UK: Macmillan.
Byrne, B. M. (2012). Structural equation modeling with Mplus: Basic concepts, applications and
programming. New York, NY: Taylor & Francis.
Campbell, D. T., & Fiske, D. W. (1959). Convergent and discriminant validation by the multitrait–
multimethod matrix. Psychological Bulletin, 56(2), 81–105. doi:10.1037/h0046016
Cho, J., Shah, D. V., McLeod, J. M., McLeod, D. M., Scholl, R. M., & Gotlieb, M. R. (2009). Campaigns,
reflection, and deliberation: Advancing an O-S-R-O-R model of communication effects.
Communication Theory, 19(1), 66–88. doi:10.1111/j.1468-2885.2008.01333.x
Cohen, J. (1989). Deliberation and democratic legitimacy. In P. Pettit & A. Hamlin (Eds.), The good polity
(pp. 17–34). New York, NY: Basil Blackwell.
Cohen J. (2007). Deliberative democracy. In S. W. Rosenberg (Ed.), Deliberation, participation and
democracy (pp. 219–236). London, UK: Palgrave Macmillan.
Converse, P. E. (1964). The nature of belief systems in mass publics. In D. E. Apter (Ed.), Ideology and
discontent (pp. 206–261). New York, NY: The Free Press.
Dahl, R. A. (1989). Democracy and its critics. New Haven, CT: Yale University Press.
Dunham, J. (2016). Freedom of the press 2016: The battle for the dominant message. Retrieved from
Economist Intelligence Unit. (2016). Democracy Index 2015: Democracy in an age of anxiety. Retrieved
Festinger, L., Riecken, H., & Schachter, S. (1956). When prophecy fails. New York, NY: Harper and Row.
Fiorina, M. P. (1981). Retrospective voting in American national elections. New Haven, CT: Yale
International Journal of Communication 14(2020) “Obstinate Partisanship” 343
Garzia, D. (2011). The personalisation of partisan attachments in the Second Italian Republic. Bulletin
of Italian Politics, 3(1), 59–78.
Gastil, J., Black, L., & Moscovitz, K. (2008). Ideology, attitude change, and deliberation in small face-to-
face groups. Political Communication, 25(1), 23–46. doi:10.1080/10584600701807836
Gil de Zúñiga, H. (2017). Attributes of interpersonal political discussion as antecedents of cognitive
elaboration. Revista Española de Investigaciones Sociológicas, 157(1) 65–84.
Gil de Zúñiga, H., Ardèvol-Abreu, A., & Casero-Ripollés, A. (2019). WhatsApp political discussion,
conventional participation and activism: Exploring direct, indirect and generational effects.
Information, Communication & Society, 1–18. doi:10.1080/1369118X.2019.1642933
Gil de Zúñiga, H., Ardèvol-Abreu, A., Diehl, T., Patiño, M. G., & Liu, J. H. (2019). Trust in institutional
actors across 22 countries: Examining political, science, and media trust around the world.
Revista Latina de Comunicación Social, 74, 237–262. doi:10.4185/RLCS-2019-1329en
Graham, T., & Wright, S. (2014). Discursive equality and everyday talk online: The impact of
“superparticipants.” Journal of Computer-Mediated Communication, 19(3), 625–642.
Green, D., Palmquist, B., & Schickler, E. (1998). Macropartisanship: A replication and critique. American
Political Science Review, 92(4), 883–899. doi:10.2307/2586310
Gunther, A. C. (1988). Attitude extremity and trust in media. Journalism Quarterly, 65(2), 279–287.
Hayes, A. F. (2013). Introduction to mediation, moderation, and conditional process analysis. New York,
Hetherington, M. J. (1998). The political relevance of political trust. American Political Science Review,
92(4), 791–808. doi:10.2307/2586304
Hively, M. H., & Eveland, W. P. (2009). Contextual antecedents and political consequences of adolescent
political discussion, discussion elaboration, and network diversity. Political Communication, 26(1),
Iyengar, S., & Hahn, K. S. (2009). Red media, blue media: Evidence of ideological selectivity in media
use. Journal of Communication, 59(1), 19–39. doi:10.1111/j.1460-2466.2008.01402.x
Johnston, R. (2006). Party identification: Unmoved mover or sum of preferences? Annual Review of
Political Science, 9(1), 329–351. doi:10.1146/annurev.polisci.9.062404.170523
344 A. Ardèvol-Abreu and H. Gil de Zúñiga International Journal of Communication 14(2020)
MacKuen, M. B., Erikson, R. S., & Stimson, J. A. (1989). Macropartisanship. American Political Science
Review, 83(4), 1125–1142. doi:10.2307/1961661
McLeod, D. M., Kosicki, G. M., & McLeod, J. M. (2009). Political communication effects. In J. Brian &
M. B. Oliver (Eds.), Media effects: Advances in theory and research (pp. 229–251). New York,
Muthen, L. K. (2014, September 18). Just-identified model [Online forum post]. Mplus.
Noveck, B. S. (2000). Paradoxical partners: Electronic communication and electronic democracy. In
P. Ferdinand (Ed.), The Internet, democracy, and democratization (pp. 18–35). London, UK:
Papacharissi, Z. (2002). The virtual sphere: The Internet as a public sphere. New Media and Society, 4(1),
Perse, E. M. (1990). Media involvement and local news effects. Journal of Broadcasting and Electronic
Media, 34(1), 17–36. doi:10.1080/08838159009386723
Petty, R. E., Briñol, P., & Priester (2009). Mass media attitude change: Implications of the elaboration
likelihood model of persuasion. In J. Bryant & M. B. Oliver (Eds.), Media effects: Advances in
theory and research (pp. 125–164). Hillsdale, NJ: Erlbaum.
Petty, R. E., & Cacioppo, J. T. (1986). Communication and persuasion: Central and peripheral routes to
attitude change. New York, NY: Springer-Verlag.
Raykov, T. (2011). Evaluation of convergent and discriminant validity with multitrait–multimethod
correlations. British Journal of Mathematical and Statistical Psychology, 64, 38–52.
Smith, D. R., & Mackie, D. M. (2007). Social psychology. Hove, UK: Psychology Press.
Sniderman, P. M., & Bullock, J. (2004). A consistency theory of public opinion and political choice: The
hypothesis of menu dependence. In W. E. Saris & P. M Sniderman (Eds.), Studies in public
opinion: Attitudes, nonattitudes, measurement error, and change (pp. 337–354). Princeton, NJ:
Princeton University Press.
Söderlund, P., & Kestilä-Kekkonen, E. (2009). Dark side of party identification? An empirical study of
political trust among radical right-wing voters. Journal of Elections, Public Opinion and Parties,
19(2), 159–181. doi:10.1080/17457280902799014
International Journal of Communication 14(2020) “Obstinate Partisanship” 345
Stromer-Galley, J. (2014). Political discussion and deliberation online. In K. Kenski & K. H. Jamieson
(Eds.), The Oxford handbook of political communication. New York, NY: Oxford University Press.
Taber, C. S., & Lodge, M. (2006). Motivated skepticism in the evaluation of political beliefs. American
Journal of Political Science, 50(3), 755–769. doi:10.1111/j.1540-5907.2006.00214.x
Wellman, B., Haase, A. Q., Witte, J., & Hampton, K. (2001). Does the Internet increase, decrease, or
supplement social capital? Social networks, participation, and community commitment. American
Behavioral Scientist, 45(3), 436–455. doi:10.1177/00027640121957286
Wojcieszak, M., & Price, V. (2010). Bridging the divide or intensifying the conflict? How disagreement
affects strong predilections about sexual minorities. Political Psychology, 31(3), 315–339.
Zaller, J. R. (1992). The nature and origins of mass opinion. Cambridge, UK: Cambridge University Press.