SHORT ARTICLE

Intraparty Polarization in American Politics

Eric Groenendyk, University of Memphis
Michael W. Sances, Temple University
Kirill Zhirkov, University of Michigan

Research shows that elite polarization and mass sorting have led to an explosion of hostility between parties, but how do
Republicans and Democrats feel toward their own respective parties? Have these trends led to more cohesion or more
division within parties? Using the American National Election Studies time series, we first show that intraparty po-
larization between ideologically extreme and ideologically moderate partisans is on the rise. Second, we demonstrate
that this division within parties has important implications for how we think about affective polarization between
parties. Specifically, the distribution of relative affect between parties has not become bimodal but merely dispersed.
Thus, while the mean partisan has become affectively polarized, the modal partisan has not. These results suggest po-
larization and sorting may be increasing the viability of third-party candidates and making realignment more likely.

R ecent studies demonstrate that the average Repub-

lican loathes the Democratic Party, and the average
Democrat loathes the Republican Party, to a degree
unprecedented in modern American politics (Abramowitz
and Webster 2016; Iyengar and Krupenkin 2018; Iyengar,
without uniting, we should only see increased dispersion,
not bimodality. This would indicate that elite polarization
and mass sorting may be opening the door for third-party
candidates and even realignment, not reinforcing our current
party division.
Sood, and Lelkes 2012; Iyengar and Westwood 2014; Mason
2018). But, how do Republicans and Democrats feel about THE EFFECT OF ELITE POLARIZATION
their own respective parties? Is partisan and ideological sorting ON THE ELECTORATE
making parties more cohesive, or are these forces actually Research shows two very different yet logically compatible
tearing party coalitions apart? We find evidence of intraparty trends in how the masses have reacted to elite polarization.
polarization—increased dispersion in partisans’ feelings to- On one hand, the public seems to be following the lead of
ward their own party—explained in part by divisions between Congress. Although the electorate may not be particularly po-
moderate partisans and their more ideologically extreme larized on policy issues, it has sorted (Fiorina, Abrams, and
counterparts. Pope 2011), bringing party identification and ideology into
How does this recognition of affective polarization within closer alignment (Levendusky 2009). This alignment has cre-
parties affect our understanding of affective polarization be- ated a stronger sense of social identity among sorted partisans
tween parties? If partisans are dividing into more cohesive (Mason 2018), leading these individuals to become more en-
groups with stronger preferences for their own party, the gaged (Abramowitz 2010) and more disdainful toward the
distribution of differences between party feeling thermom- opposition (Iyengar et al. 2012).
eters—a common indicator of affective polarization—should On the other hand, much of the public is frustrated with
appear increasingly bimodal. But, if partisans are dividing polarization. Studies show the uncivil nature of contemporary

Eric Groenendyk (grnendyk@[Link]) is associate professor of political science at the University of Memphis, Memphis, TN 38152. Michael W. Sances
(msances@[Link]), assistant professor of political science at the University of Memphis when this article was submitted, is now assistant professor of
political science at Temple University, Philadelphia, PA 19122. Kirill Zhirkov (kzhirkov@[Link]) is a PhD candidate in the Department of Political
Science at the University of Michigan, Ann Arbor, MI 48109.
Data and supporting materials necessary to reproduce the numerical results in the article are available in the JOP Dataverse ([Link]
/dataverse/jop). An online appendix with supplementary material is available at [Link]

The Journal of Politics, volume 82, number 4. Published online August 6, 2020. [Link]
q 2020 by the Southern Political Science Association. All rights reserved. 0022-3816/2020/8204-0033$10.00
 Volume 82 Number 4 October 2020 / 1617

Figure 1. Intraparty polarization hypothesis

political discourse turns many people off to politics, leading increasing heterogeneity in partisans’ feelings toward their
to distrust (Mutz and Reeves 2005) and disengagement (Klar own party will be obfuscated.1
and Krupnikov 2016; Mutz 2006). Many individuals are
embarrassed to even admit their partisan allegiances (Klar H1. Over time, partisans’ feelings toward their own
and Krupnikov 2016), and when exposed to media coverage party will become more divided based on ideological
of polarization, they report more moderate issue positions to extremity.
distance themselves from the partisan fray (Levendusky and
Malhotra 2016). In fact, despite sorting themselves into the Figure 2 illustrates what happens when we shift focus to
ideologically “correct” parties, the average partisans have differences in partisans’ feelings toward the two parties—a
actually come to like their party a bit less over time (Groe- concept commonly labeled affective polarization. An intu-
nendyk 2018; Klar, Krupnikov, and Ryan 2018). itive way to quantify affective polarization is to calculate the
Taken together, these works suggest that focusing on the polarization of the average partisan. We do this by calcu-
mean partisan may obfuscate heterogeneity in partisans’ re- lating the average absolute difference between party feeling
actions to elite polarization over time. Although partisans’ thermometer ratings, essentially “folding” the distribution
feelings toward their party appear fairly stable on average of party feeling thermometer differences at its midpoint
(Abramowitz and Webster 2016; Iyengar and Krupenkin 2018; and taking the mean.2 By this measure, affective polariza-
Iyengar et al. 2012), simple spatial logic predicts growing tion has clearly increased (see Iyengar and Krupenkin 2018;
heterogeneity underlying this average. If voters’ reactions to Iyengar et al. 2012). However, there are two observationally
elite polarization depend on their ideological self-identification equivalent explanations for this pattern. To illustrate, fig-
(Rogowski and Sutherland 2016), ideologically extreme iden- ure 2 shows three hypothetical distributions of differences
tifiers should approve of their party’s increased extremity while in affect toward parties. In each case, we show where the
moderate identifiers should disapprove. mean falls after that distribution has been “folded.” Again,
Our simple spatial model is illustrated in figure 1. First, the this is a typical way to quantify affective polarization.
polarization of party elites has provided all partisans with a Figure 2A illustrates a clear case of low polarization. In
reason to like the other party less. Regardless of personal this scenario, partisans’ feelings are clustered around zero,
ideological extremity, the other party is now farther from
every partisan’s ideal point, leading to interparty polarization
1. Readers should not interpret our use of spatial terminology as in-
(i.e., disdain for the opposition). However, when we look at
dication that feelings must be rooted in substantive policy preferences.
partisans’ feelings toward their own party, we should see Our findings are likely driven by preferences over “style” more than “sub-
increased heterogeneity or intraparty polarization. As party stance” (Lelkes 2018). Policy aside, moderate identifiers like their party less,
elites become more extreme, moderates should come to like while strong ideological identifiers like it more (see the appendix, available
their own party less compared to their more ideologically online).
2. Some researchers subtract partisans’ feelings toward the other party
extreme counterparts. If one focuses only on means (rep-
from their feelings toward their own party. This yields a measure virtually
resented by dotted lines), the increase in negative feelings indistinguishable from ours (r p :99). See the appendix for additional
toward the opposition party will be quite apparent, but the details.
1618 / Intraparty Polarization in American Politics Eric Groenendyk, Michael W. Sances, and Kirill Zhirkov

Figure 2. Dispersion hypothesis. Tall vertical lines represent the mean after the distribution has been “folded” at its midpoint. The distance of this line from
the midpoint of the distribution indicates the polarization of the mean partisan. Compared to the low-polarization panel (A), we see mean polarization in
both the bimodality panel (B) and the high-dispersion panel (C). However, it is only in the bimodality panel where the modal partisan is polarized.

indicating similar feelings toward the two parties. Figure 2B The results displayed in figures 3A and 3B support our
illustrates a clear case of high polarization. Compared to fig- prediction of divergence within both parties. Liberal Demo-
ure 2A, partisans have divided into two distinct groups, cre- crats now report more positive feelings toward their party, and
ating a bimodal distribution. Here the mean partisan and the moderate Democrats now report less positive feelings toward
modal partisan are both farther from the midpoint of the their party. Conservative Republicans’ feelings conform to
distribution. Notably, bimodality is the standard Fiorina et al. expectations by rising through the 1980s but then decline
(2011) use to evaluate policy-based polarization, concluding back to approximately their previous level. Nonetheless, we
that the electorate is closely divided (i.e., symmetrically dis- see divergence because of the steady decline in moderate
tributed) but not deeply divided over policy (i.e., not bimod- Republicans’ feelings toward their party.
ally distributed). In figure 2C, feelings toward the parties have Figure 3C reinforces this pattern. Here we plot the or-
polarized on average (as indicated by the higher average ab- dinary least squares regression coefficient associated with
solute difference in thermometer ratings compared to fig. 2A), moderation in each year of the ANES time series:
but the modal partisan remains near the midpoint of the dis-
ownpartyi p a 1 b # moderationi 1 ei ;
tribution. This is the type of distribution that would result
if strong liberals and conservatives were polarizing while where ideology is measured on the standard seven-point
moderate partisans were rating both parties more negatively scale, moderation is operationalized as 2jideologyi 2 4j, and
over time, leaving them relatively unpolarized. ownparty represents one’s own party rating on a 100-point
feeling thermometer.
H2. The distribution of relative affect toward the Re- Figure 3C shows downward trends in the coefficient on
publican and Democratic parties will become more moderation for both Republicans’ feelings toward the Re-
dispersed, rather than bimodal, over time. publican Party and Democrats’ feelings toward the Demo-
cratic Party. This indicates that moderates increasingly dislike
EVIDENCE OF INTRAPARTY POLARIZATION their party relative to those who identify as more ideologically
We have hypothesized that, while ideological extremists extreme.4
should approve of the polarization of their party’s elites, self- Looking across figures 3A–3C, the most notable exceptions
identified moderates should be frustrated and have nowhere appear to “prove the rule.” In 1990, George H. W. Bush agreed
to sort. If this is true, over time we should see divergence to raise taxes to reduce the budget deficit, working across the
between the feelings that extreme ideological identifiers and aisle with Democrats. At this time point, we see a momentary
moderates report toward their party. We test this prediction drop in conservative Republicans’ feelings toward their party.
using the seven-point measure of ideological identification We see a similar pattern among liberal Democrats in 2016,
and the party feeling thermometers available in the American
National Election Studies (ANES) time series.3
within-respondent effect during years when both versions were included
and adding the mean difference to the corresponding feeling thermometer
3. Analyses include partisan “leaners.” However, results appear quite before the change. We analyze Republicans and Democrats separately, to
robust to the exclusion of leaners and inclusion of all independents (see account for possible partisan asymmetry (Grossman and Hopkins 2016).
the appendix). All analyses exclude online respondents. Tests account for 4. See the appendix for discussion of differences between Republicans
change in wording of party feeling thermometers, by calculating the and Democrats in fig. 3.
 Volume 82 Number 4 October 2020 / 1619

Figure 3. Intraparty polarization explained by ideological extremity: A, Democrats; B, Republicans; C, coefficients

when the party nominated a moderate, Hillary Clinton, over a ploy Esteban and Ray’s (1994) axiomatically derived measure
socialist, Bernie Sanders. Previous primary battles did not of polarization (bimodality), which captures changes in the
produce this sort of ideological fracturing within the party, distribution of affect between parties (see the appendix).
which speaks to the increased influence of ideology, not just Although we remain agnostic about the semantics, Esteban
on feelings between parties but also within parties.5 and Ray argue that, by definition, polarization requires bi-
In attempt to gain causal leverage, we ran cross-lagged modality.7
regression models using each of the ANES panel data sets that Figure 4 shows the trends in mean difference (fig. 4A)
included the necessary variables (see the appendix). In two of and bimodality (fig. 4B) from 1964 to 2016. Trends were es-
the three data sets, moderate ideology is associated with large timated using a simple polynomial regression model, where t
decreases in partisans’ feelings toward their own party from signifies year:
time 1 to time 2. We see weak and inconsistent evidence of
polarizationt p a 1 b1 # t 1 b2 # t 2 1 et :
feedback, with partisans’ feelings toward their party only
affecting their ideology in one of the three periods.6 Consistent with the polarization literature, figure 4A
shows that the average absolute value of the difference in
partisans’ feelings toward the two parties has indeed grown
HOW POLARIZATION WITHIN PARTIES AFFECTS over time. However, in figure 4B, where we shift focus to the
POLARIZATION BETWEEN PARTIES distribution of differences in party feeling thermometer
Have Americans divided into more cohesive groups, or have ratings (see theoretical distinction explained in fig. 2), we see
they just divided? Using a common method of estimating a very different pattern. Rather than seeing an upward trend
affective polarization, we first calculate the average absolute in bimodality, as one might expect in a polarizing electorate,
difference between party feeling thermometers for each year the bimodality of the distribution appears to have decreased
(mean difference). However, this method has the potential to from 1964 to 2000 and then risen back up to only about half
obfuscate the heterogeneity observed above, so we also em- of its prior level. Thus, while the average distance between
Republicans’ and Democrats’ feelings toward one another
has been growing (mean difference), there is little evidence to
suggest that partisans’ feelings are consolidating into more
5. Coefficients in fig. 3 are from separate regressions fit to each year of
the ANES data. To test the significance of this trend, we modeled the in-
distinctive groups (bimodality). Some partisans are becom-
teraction between moderation and time with year dummies included sep- ing more affectively polarized, driving up mean affective
arately. The negative interaction terms for both Republicans (p ! :01) and polarization as the distribution disperses. But the modal
Democrats (p ! :01) support our theory that time is making the effect of partisan remains near the midpoint of the scale, reporting
moderation more negative (see table A1; tables A1–A3 are available online).
similar feelings toward both parties. This suggests that the
We have also replicated the patterns shown in fig. 3 with specifications
including relevant controls (see the appendix). existing literature is right to point out the growing mean
6. Because of space constraints, results are presented in the appendix.
All models included controls for party identification strength and feelings
toward the other party. Effects noted are statistically significant at p ! :05.
While cross-lagged analyses aid causal inference, it should be noted that 7. Distributions of own party, other party, and difference between
they cannot establish definitive causality. parties are plotted for each year in the appendix.
1620 / Intraparty Polarization in American Politics Eric Groenendyk, Michael W. Sances, and Kirill Zhirkov

Figure 4. Comparing indicators of polarization between parties: A, mean difference; B, bimodality

distance between partisans’ feelings toward the two parties, Groenendyk, Eric. 2018. “Competing Motives in a Polarized Electorate:
Political Responsiveness, Identity Defensives, and the Rise of Partisan
but this pattern is clearly not representative of all partisans.
Antipathy.” Political Psychology 39 (S1): 159–17.
Grossman, Matt, and David A. Hopkins. 2016. Asymmetric Politics. New
York: Oxford University Press.
DISCUSSION Iyengar, Shanto, and Masha Krupenkin. 2018. “Partisanship as Social Identity:
We believe these results have vital implications for under- Implications for the Study of Party Polarization.” Forum 16 (1): 23–46.
standing our political climate and the stability of the current Iyengar, Shanto, Gaurav Sood, and Yphtach Lelkes. 2012. “Affect, not Ide-
party system. One might conclude from the extant literature ology: A Social Identity Perspective on Polarization.” Public Opinion
Quarterly 76 (3): 405–31.
that the current partisan and ideological cleavage is solidifying
Iyengar, Shanto, and Sean J. Westwood. 2014. “Fear and Loathing across
and growing deeper. Our findings suggest the opposite: the Party Lines: New Evidence on Group Polarization.” American Journal
partisan divide may be growing more fluid. Despite partisan of Political Science 59 (3): 690–707.
and ideological sorting, parties are becoming internally di- Klar, Samara, and Yanna Krupnikov. 2016. Independent Politics. New York:
Cambridge University Press.
vided. Stronger ideological identifiers continue to express very
Klar, Samara, Yanna Krupnikov, and John Ryan. 2018. “Affective Polari-
positive feelings toward their party, but ideologically moderate zation or Partisan Disdain? Untangling a Dislike for the Opposing Party
partisans like their party less, and they have nowhere to sort. from a Dislike of Partisanship.” Public Opinion Quarterly 82 (2): 379–90.
This lack of positive attachment leaves these partisans vul- Lelkes, Yphtach. 2018. “Affective Polarization and Ideological Sorting: A
nerable to third parties and antiestablishment candidates, Reciprocal, Albeit Weak Relationship.” Forum 16 (1): 67–79.
Levendusky, Matthew. 2009. The Partisan Sort. Chicago: University of Chi-
increasing the likelihood of electoral realignment. cago Press.
Levendusky, Matthew, and Neil Malhotra. 2016. “Does Media Coverage of
Partisan Polarization Affect Political Attitudes?” Political Communi-
REFERENCES cation 33 (2): 283–301.
Abramowitz, Alan I. 2010. The Disappearing Center. New Haven, CT: Yale Mason, Lilliana. 2018. Uncivil Agreement. Chicago: University of Chicago
University Press. Press.
Abramowitz, Alan I., and Steven Webster. 2016. “The Rise of Negative Mutz, Diana C. 2006. Hearing the Other Side. New York: Cambridge Uni-
Partisanship and the Nationalization of U.S. Elections in the 21st Cen- versity Press.
tury.” Electoral Studies 41 (2): 12–22. Mutz, Diana C., and Bryon Reeves. 2005. “The New Videomalaise: Effects
Esteban, Joan-Maria, and Debraj Ray. 1994. “On the Measurement of of Televised Incivility on Political Trust.” American Political Science
Polarization.” Econometrica 62 (4): 819–51. Review 99 (1): 1–15.
Fiorina, Morris P., Samuel J. Abrams, and Jeremy C. Pope. 2011. Culture Rogowski, Jon C., and Joseph Sutherland. 2016. “How Ideology Fuels Af-
War? The Myth of a Polarized America. 3rd ed. New York: Pearson. fective Polarization.” Political Behavior 38 (2): 485–508.