Do Violent Protests Affect Expressions of Party Identity? Evidence from the Capitol - OSF
←
→
Page content transcription
If your browser does not render page correctly, please read the page content below
Do Violent Protests Affect Expressions of
Party Identity? Evidence from the Capitol
Insurrection?
Gregory Eady† Frederik Hjorth§ Peter Thisted Dinesen∗
The insurrection at the United States Capitol on January 6, 2021 was the most dra-
matic contemporary manifestation of deep political polarization in the United States.
Recent research shows that violent protests can shape political behavior and attach-
ments, but several questions remain unanswered. Using day-level panel data from
millions of US social media users to track changes in the political identities expressed
in their Twitter biographies (profiles), we show that the Capitol insurrection caused
large-scale de-identification with the Republican Party and Donald Trump with no
indication of re-identification in the weeks following the insurrection. This finding sug-
gests that there are limits to party loyalty: a violent attack on democratic institutions
sets boundaries on partisanship in the US, even among avowed partisans. Further,
the finding that political violence can deflect co-partisans carries the potential positive
democratic implication that those who encourage or associate themselves with such
violence pay a political cost.
Word count: 3,998
?
This research is funded by the Independent Research Fund Denmark (grant number: 9038-00123B).
We thank participants at the Annual Conference of the NYU Center for Social Media and Politics, and
the Aarhus Research on Online Political Hostility project, and Charles Breton and Evelyne Brie for their
thoughtful comments and suggestions on early versions of this manuscript.
†
Corresponding author. Department of Political Science and the Center for Social Data Science (SODAS),
University of Copenhagen. Address: Øster Farimagsgade 5, DK-1153 Copenhagen K, Denmark. E-mail:
gregory.eady@ifs.ku.dk.
§
Department of Political Science and the Center for Social Data Science (SODAS), University of Copen-
hagen. Address: Øster Farimagsgade 5, DK-1153 Copenhagen K, Denmark. E-mail: fh@ifs.ku.dk.
∗
Department of Political Science, University of Copenhagen. Address: Øster Farimagsgade 5, DK-1153
Copenhagen K, Denmark. E-mail: ptd@ifs.ku.dk.Introduction
The insurrection at the United States Capitol on January 6, 2021 is widely considered one
of the most remarkable examples of a violent attack on democratic institutions in a mature
democracy in recent times (Bright Line Watch, 2021). Yet although many politicians and
pundits condemned the Capitol insurrection, and approval of President Trump decreased in
its aftermath (Bump, 2021), we know little about the broader effects of political violence
such as this on mass political behavior.
In this letter, we take an important first step in uncovering the consequences of the Capitol
insurrection for the strength of political affiliations by investigating changes in political self-
identification with the Republican Party and President Trump in the days immediately
following the event. More specifically, drawing on recent work that uses social media self-
descriptions as indicators of political identities (Rogers and Jones, 2021), we study changes in
identification with the Republican Party and then-President Donald Trump in the personal
‘bios’ of users on the micro-blogging platform Twitter. We use data from a sample of 3.4
million active US Twitter users to track the bio of each user each day starting approximately
7 months prior to January 6, 2021. This yields a panel dataset of nearly 1 billion user-
day observations of social media profiles. We apply a flexible difference-in-differences (event
study) model to these data to estimate the causal effect of the insurrection on self-expressions
of political identity. Our findings demonstrate that the insurrection caused an exceptionally
clear and sharp decrease in expressions of identification with the Republican Party and
‘Trumpism’, a result that is consistent across a wide series of robustness checks.
Our analysis of voters’ reactions to the Capitol insurrection contributes to a nascent liter-
ature on the consequences of violent protests for political behavior and political attachments.
Two prominent recent studies have brought this research agenda to the fore. One study finds
that temporal and geographical proximity to violent Black-led protests in the 1960s caused
an increase in endorsements of “social control” as well as support for the Republican Party
1(Wasow, 2020). Conversely, and thereby highlighting the complex nature of the effects of
violent protests, another study finds that spatial proximity to the 1992 Los Angeles riots led
to a liberal shift in policy support, and an increase in voter registration for the Democratic
Party (Enos, Kaufman and Sands, 2019). However, beyond these few (but important) stud-
ies, “we know little about the effect of these events on political behavior” (Enos, Kaufman
and Sands, 2019, 1012).
Our study furthers our knowledge about the effects of violent protests by adding to ex-
isting studies in a number of important ways. First, in contrast to their focus on historical
cases of violent political protests, we study how a contemporary episode shapes political at-
tachments. This is important because recent political developments (e.g. increased affective
polarization, Iyengar et al., 2019) and technological changes (e.g. social media and the as-
sociated ability to disseminate and engage with news in real-time) may give rise to different
reactions than those observed in the past, where these developments were not present or
were relatively muted. Second, and relatedly, we study whether reactions to violent protests
also extend beyond their immediate geographical locus. Given the salience of such events
and the increasingly nationalized nature of American political behavior (Hopkins, 2018),
more widespread effects seem plausible. Third, whereas the two studies highlighted above
center on protests associated with the political left, we examine the consequences of violent
political protests initiated by those on the right side of the political spectrum. Fourth, by
investigating immediate behavioral changes rather than subsequent (electoral) outcomes, we
gain purchase over whether responses are driven by citizens’ spontaneous reactions rather
than those prompted by longer-term elite politicization in the aftermath of the event.
Our study also has implications beyond the developing literature on the political behav-
ioral consequences of violent protests. First, it connects to the related literature on political
violence, and more specifically to the costs and benefits of violence to political actors in the
comparatively rare setting of a developed democracy (Rosenzweig, 2021). In essence, vot-
ers’ reactions to the Capitol insurrection indicate whether political violence is an attractive
2strategy for political elites to appeal to US voters. Second, our study implicitly speaks to the
limits (or lack thereof) of partisanship in the US. Often attributed to increases in social and
affective political polarization (Iyengar et al., 2019; Mason, 2018), the strength of partisan-
ship is now so socially and politically consequential in the US that, in the words of a recent
study, “it is difficult to overstate the importance of party loyalty” (Barber and Pope, 2019,
39). As such, by analyzing whether (expressed) partisans are willing to forego identifying
with their party in the face of exceptional political violence, we examine the scope conditions
of the ‘unmovable’ character of partisanship in the United States. Finally, our results speak
to the subsequent intra-party struggles within the Republican Party over how to politically
address the insurrection (Cheney, 2021).
Data and Research Design
Obtaining user biographies
We collected data each day from the Twitter bios (profiles) of 3.4 million geo-located US
users starting 7 months prior to the January 6 “Stop the Steal” Capitol insurrection to 3
months afterward (from June 1, 2020 to March 15, 2021). The sample of users were drawn
from a population defined as active US social media users who are minimally interested in
politics. We defined this as any user who followed at least one of an ideologically diverse
set of major US news media accounts, including MSNBC, Huffington Post, The New York
Times, The Washington Post, CNN, The Wall Street Journal, FOX News, and Breitbart
News.1 We first collected the profiles of followers of each media account. To identify active
users, we include only those who sent at least one tweet in the past year; sent at least 25
tweets since creating their account; and have at least 10 followers. Finally, we include only
users geo-located to the US based on geo-coordinates and text location information (Dredze
1
By virtue of displaying some political interest, we focus on a group that is likely to be relatively con-
solidated in their political attachments, and therefore, ceteris paribus, less likely to change their political
affiliations in response to the insurrection. Our test is thus a conservative one.
3et al., 2013). In total, our data constitute a 9-month 3.4 million user panel dataset with
roughly 900 million user-day observations capturing daily changes in how users self-describe
themselves in their bios.2 These panel data allow us to investigate within-user changes to
expressed political identities that each user actively made on the days that they occur.
Measuring expressed partisanship
We measure partisanship by identifying the terms in users’ Twitter profiles that expressly
indicate partisanship. Partisan keywords are identified using a keyword expansion algo-
rithm, which is shown to be superior to ad hoc selection for social media data (King, Lam
and Roberts, 2017). The keyword expansion technique begins with a minimal set of seed
words (“Democrat” and “Republican”), and applies a supervised learning model to generate
candidate keywords for potential inclusion. The algorithm is based on the principle that
researchers are poor at generating relevant terms from memory, but adept at recognizing if
a given keyword captures the relevant concept when prompted. This allows us to identify
relevant terms that users would include in their profiles to explicitly indicate their partisan-
ship (or remove to de-identify from it). Details of the procedure, and the terms indicating
Democratic and Republican partisanship are presented in online Appendix A.
Difference-in-differences model
To estimate the effect of the Capitol insurrection on Republican partisan de-identification,
we apply a flexible difference-in-differences (event study) model to data collected within a 10-
day window around the event from users whose profiles include a Republican or Democratic
keyword on at least one day within this time period. This allows us to capture the dynamics
of the effect of the insurrection on Republican partisan identification relative to Democrats,
the natural counterfactual group. Event study estimates from the pre-insurrection period
2
On a small number of days (June 5, July 11, August 19, and September 14-16), the data are incomplete
for technical reasons. However, because days with missing data are well before the insurrection, this does
not affect our results.
4also allow us to visually assess the parallel trends assumption that is necessary for causal
identification. Our model is defined as follows:
X
yit = αi + λt + βt Republicani + it , (1)
t6=0
where the outcome variable yit is a binary variable indicating whether user i’s profile contains
a keyword representing their partisan identity on day t, and Republicani is a binary variable
indicating whether user i’s partisan identity as measured by keyword use during the period is
Republican (Republicani = 1) or Democratic (Republicani = 0). The parameters of interest,
βt , capture the estimated difference in differences between partisan-identifying Democrats
and Republicans on a given day t relative to a pre-treatment baseline (set as t = 0). User
and time fixed effects are denoted by αi and λt respectively. In online Appendices B and C
we define an alternative outcome yit as the count of Republican and Democratic terms in
users’ profiles (with equivalent results). Finally, given our event study setup, it is possible
that the observed effect is driven by an increase in Democratic identity. As we show in online
Appendix B, however, we observe no major discontinuity in Democratic identification in the
period of interest. In effect, the observed effects presented in the Results section appear
wholly driven by changes in Republican (de-)identification.
Results
To examine the effect of the Capitol insurrection on expressions of Republican identification,
we begin by presenting a descriptive tally of aggregate changes in identification. Figure 1
shows daily net changes in the number of users who indicate a Republican identification
across the entire data collection period. The figure shows that in the days after the Capitol
insurrection, there is a dramatic net decrease in the number of users identifying with the
Republican Party and President Trump: each day sees a net de-identification of several hun-
dred Republican-identifying Twitter users in our sample. Moreover, although the magnitude
50
Net change in Republican Party ID
−200
−400
−600
Capitol insurrection −
−800
Election day −
Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar
Figure 1: Change in Republican Party identification over time. This figure presents
daily net changes among Twitter users who include a term indicating a Republican Party
identity from June 1, 2020, to March 15, 2021. Values below zero indicate a net decrease in
users with Republican identity terms compared to the previous day. Loess regression lines
included for visual reference.
of de-identification diminishes, we see continual net de-identification throughout the study
period. In the three weeks immediately following the insurrection, a substantial 1 in 14 users
(7.1%) remove Republican-identifying terms. This compares with 1 in 108 users (0.9%) with
Democratic terms (see online Appendix B for parallel analyses for Democrats). For com-
parison, in the three weeks before the Capitol insurrection, identification loss of users with
Republican and Democratic terms was equivalent (∼0.5%). The post-insurrection drop is
also far more pronounced than that in the aftermath of the presidential election on November
3, 2020.
Because Figure 1 considers only Republicans, the rapid change in identification may also
reflect cross-partisan alienation from politics rather than a reaction specifically among Re-
publicans. To account for this possibility, we compare changes in expressions of party iden-
tification among Republican-identifying versus Democratic-identifying Twitter users. For
simplicity, we exclude a small subset of users who include both Democratic and Republican
60
−.01
Republican Party ID
−.02
−.03
Capitol insurrection −
−.04
Dec 28 Jan 1 Jan 5 Jan 9 Jan 13 Jan 17
Figure 2: Event study estimates. Each point (with 95% CIs) represents a difference-
in-differences estimate of (de)identification between Democratic and Republican identifiers
relative to January 6, 2021. Data were collected each morning, and thus profiles on January
6 itself were collected before the insurrection.
terms. Based on daily observations for these two partisan groups, we fit a difference-in-
differences model as described above, comparing changes in each group’s party identification
relative to the day immediately prior to the insurrection.
Figure 2 presents estimates from the model, using data collected within a 10 day window
around the insurrection. Each point in Figure 2 represents the difference between Repub-
lican and Democratic users in the predicted probability of party identification relative to
a pre-insurrection baseline (the day prior to the insurrection). Negative values imply that
Republican users de-identify more (i.e. drop all party-related terms) compared to Democrats.
Figure 2 clearly shows that prior to the insurrection, Republican and Democratic users
changed their expressed identification similarly. This implies that there is no differential de-
identification in trends among Republicans prior to the insurrection, and provides evidence
consistent with the parallel trends assumption necessary for causal identification. After the
insurrection, the change is dramatic. Within a few days, Republican users were on average 2
percentage points less likely to express a party identity relative to Democrats than they were
7before the insurrection. Over the 10-day post-insurrection window shown in Figure 2, this
relative difference increases further to around 4 percentage points. Hence, these estimates,
based on a narrower time frame than shown in Figure 1, imply that within only a week and
a half after the insurrection, roughly 1 in 25 Republican identifiers had removed markers of
partisan identification from their biographies. This result is substantively equivalent when
we examine the count of partisan terms in users’ bios (online Appendices B and C), and
robust to alternative choices of partisan keywords (online Appendices D, E, and F).
Partisan terms indicating identification with Trumpism (e.g. ‘Trump’, ‘#MAGA’) are
more frequent than those referencing the Republican Party itself, which raises the question
of whether the effect is wholly driven by de-identification with the President rather than the
party (for details on keyword usage across users, see online Appendix E). Unsurprisingly,
users overlap in their use of these terms. However, by considering only the terms ‘Repub-
lican’ and ‘Democrat’, we can gauge whether the effect is driven solely by de-identification
with ‘Trumpism’. As we report in online Appendix E, the effect is similar, although di-
minished in magnitude, when using only party labels. Considering that party identity is
often characterized as being as stable as racial and religious identities (Green, Palmquist
and Schickler, 2004), this result is important. It suggests that, if dramatic enough, violent
events can weaken partisanship among avowed partisans to the extent that they actively
remove expressions of partisan identity from their self-descriptions.
A natural follow-up question to this set of findings is whether those who de-identified
as a result of the insurrection re-identified shortly afterward. Was the effect long-lasting
or short-lived? To answer this, we use the fact that our data are panel data to track users
who removed Republican partisan terms from their bios within the first week after the
insurrection. For each user who de-identified in the first week after the insurrection, we
examine each day afterward to investigate whether they re-included Republican partisan
terms. In short, this behavior is rare. To demonstrate this, Figure 3 presents the proportion
of previously Republican-identifying users who re-included Republican partisan terms in
8100
90
80
Percent re−identifying 70
60
50
40
30
20 1 week 2 weeks 3 weeks 1 month
3% 4% 5% 5%
10
0
0 10 20 30
Number of days since de−identifying
Figure 3: Percentage of Republican de-identifiers who re-identified within 30
days. This figure presents the percentage of previously Republican-identifying users who
re-identify within 30 days of de-identification. Data include any Republican-identifying user
who de-identified within a week after the Capitol insurrection. Vertical markers indicate the
percentage of users who re-identified at specific points in time.
their bios within 30 days of having removed them in the immediate aftermath of the event.
As Figure 3 shows, the vast majority of users who de-identified in the week immediately
following the insurrection did not re-identify within the following month. Within a week
after first de-identifying, only 3% re-included Republican terms in their social media bios;
within a month, only 5%. In other words, the observed Republican de-identification was not
merely an ephemeral shift in response to the insurrection, but appears relatively long-lasting.
The online Appendix presents a number of additional robustness tests and auxiliary anal-
yses. First, while we interpret the decline in identification as a reflection of Republican users’
de-identification in response to political violence committed by co-partisans, an alternative
explanation is that Republican users changed their bios in the wake of the insurrection out
of fear of legal prosecution (including but not limited to actual rioters). Finding a significant
drop when considering only the partisan label speaks prima facie against fear of prosecution
as an alternative explanation. However, we can address this possibility more directly by
9excluding users who scrubbed their timelines of potentially incriminating tweets. As a proxy
for this behavior, we identify users who removed tweets (as captured in their daily profile
data) on the same day that they dropped Republican partisan terms from their bio. In
online Appendix G we show that the findings are robust to excluding users who deleted one
or more tweets on the day that they removed partisan terms.
Second, there is the potential concern that the results are driven by the fact that Twitter
deleted QAnon-related accounts—a loosely knit group of political conspiracy theorists, some
of whose profiles may overlap with the set of Republican-identifying users—in the weeks after
the insurrection (Singh, 2021). Deleted users do not directly affect the event study results due
to the inclusion of user-level fixed effects, and because deleted users are not coded as having a
de-identified profile. QAnon supporters may nevertheless have pre-emptively scrubbed their
timelines and profiles to potentially prevent detection. In online Appendix H, we show that
our findings are substantively unaffected when excluding users whose accounts were deleted,
suspended, or made private following the insurrection.
Third, as noted in the introduction, we interpret the effect as a national-level shock
rather than one driven by users geographically close to the Capitol riots. In online Appendix
I, we substantiate this interpretation by demonstrating that the effect is unchanged when
excluding any user geo-located to Washington D.C. and neighboring states.
Fourth and finally, we consider the possibility that the findings are driven primarily by
social desirability bias rather than ‘true’ de-identification. In online Appendix J, we evaluate
this by comparing event study models among users whose user names match and do not
match a first name in US Social Security Administration records. If social desirability were
a factor, one would expect that users who use a real name would be more likely to de-identify.
We do not find this to be the case.
10Conclusion
Studying the effect of the US Capitol insurrection on expressed partisanship, we find that
the insurrection caused a substantial number of Republican partisans to actively remove
expressions of identification with the Republican Party and Donald Trump. Our findings
add to our knowledge about the effects of violent protests on mass political behavior in several
ways. Complementing studies of historical cases of left-wing protest, we provide evidence of
the effects of violent protests in a contemporary setting and on the political right. Second,
we show that mass de-identification in response to the insurrection is nationalized, i.e. it
extends beyond its immediate geographical focus, and occurs immediately, with most of the
effect manifested within days. Further, we document that this immediate effect persists,
with only a small minority of de-identifiers re-identifying in the following month.
More broadly, our findings suggest that extreme events, such as those that violate demo-
cratic norms, can drive even some avowed partisans to distance themselves from their party.
In the context of the ongoing debate about the negative consequences of polarization in the
United States (Iyengar et al., 2019; Finkel et al., 2020), this finding is encouraging as it
indicates that there appear to be limits to partisan loyalty. Our results thus complement
recent work finding that exposure to incivility in same-party media leads partisans to dis-
tance themselves from their party (Druckman et al., 2019). Furthermore, this carries the
positive democratic implication that political violence potentially deflects and de-mobilizes
co-partisans, raising the political cost of using such tactics.
These potentially positive conclusions should also not be overstated. Expressed de-
identification is an indicator of distancing from the party and its leader, but does not imply
that partisan identities are no longer salient or consequential. Republican politicians’ re-
sponses to the insurrection, for example, resulted in a struggle over the meaning of the
party’s identity, rather than its abandonment. A minority of radical partisans, as with the
Capitol insurrection, may also use violence for their own ends despite its potential costs
11to a political party. As such, our results should be seen as contributing to a still evolving
understanding of the conditions under which partisanship may be curbed or amplified in an
age of polarization (Iyengar et al., 2019; Finkel et al., 2020; Druckman et al., 2019).
References
Barber, Michael and Jeremy C Pope. 2019. “Does party trump ideology? Disentangling
party and ideology in America.” The American Political Science Review 113(1):38–54.
Bright Line Watch. 2021. “American Democracy at the Start of the Biden Presidency.”
Bright Line Watch Report, January-February 2021 Surveys.
Bump, Philip. 2021. “Disapproval of Trump Soars in Wake of Insurrection Attempt.” Wash-
ington Post, January 15.
Cheney, Liz. 2021. “The GOP is at a Turning Point. History is Watching Us.” Washington
Post, May 5.
Dredze, Mark, Michael J. Paul, Shane Bergsma and Hieu Tran. 2013. “Carmen: A Twitter
Geolocation System with Applications to Public Health.” Association for the Advance-
ment of Artificial Intelligence (AAAI) Workshop on Expanding the Boundaries of Health
Informatics Using AI (HIAI).
Druckman, James N, SR Gubitz, Ashley M Lloyd and Matthew S Levendusky. 2019. “How in-
civility on partisan media (de) polarizes the electorate.” The Journal of Politics 81(1):291–
295.
Enos, Ryan D, Aaron R Kaufman and Melissa L Sands. 2019. “Can violent protest change
local policy support? Evidence from the aftermath of the 1992 Los Angeles riot.” American
Political Science Review 113(4):1012–1028.
12Finkel, Eli J, Christopher A Bail, Mina Cikara, Peter H Ditto, Shanto Iyengar, Samara Klar,
Lilliana Mason, Mary C McGrath, Brendan Nyhan, David G Rand et al. 2020. “Political
sectarianism in America.” Science 370(6516):533–536.
Green, Donald P, Bradley Palmquist and Eric Schickler. 2004. Partisan hearts and minds:
Political parties and the social identities of voters. Yale University Press.
Hopkins, Daniel J. 2018. The increasingly United States: How and why American political
behavior nationalized. University of Chicago Press.
Iyengar, Shanto, Yphtach Lelkes, Matthew Levendusky, Neil Malhotra and Sean J Westwood.
2019. “The origins and consequences of affective polarization in the United States.” Annual
Review of Political Science 22:129–146.
King, Gary, Patrick Lam and Margaret E Roberts. 2017. “Computer-Assisted Keyword and
Document Set Discovery from Unstructured Text.” American Journal of Political Science
61(4):971–988.
Mason, Lilliana. 2018. Uncivil Agreement: How Politics Became Our Identity. Chicago, IL:
University of Chicago Press.
Rogers, Nick and Jason J Jones. 2021. “Using Twitter Bios to Measure Changes in Self-
Identity: Are Americans Defining Themselves More Politically Over Time?” Journal of
Social Computing 2(1):1–13.
Rosenzweig, Steven C. 2021. “Dangerous disconnect: Voter backlash, elite misperception,
and the costs of violence as an electoral tactic.” Political Behavior pp. 1–24.
Singh, Kanishka. 2021. “Twitter suspends tens of thousands of accounts dedicated to sharing
QAnon content.” Reuters, January 12.
Wasow, Omar. 2020. “Agenda seeding: How 1960s black protests moved elites, public opinion
and voting.” American Political Science Review 114(3):638–659.
13You can also read
