Civic engagement,social capital,and ideological extremity: Exploring online political engagement and political expression on Facebook

Abstract

Using a method incorporating both survey and trace data measures, this study presented and tested a theoretical model for understanding political expression on Facebook. The data suggested that self-reported measures of offline civic engagement, bonded social capital, and ideological extremity were predictive of a self-reported measure of general online political engagement. For its part, self-reported levels of online political engagement were positively and significantly associated with observed political expression on Facebook. These results are discussed in the context of both on and offline political connection and communication.

Keywords

Civic engagement democracy Facebook political participation social capital trace data measure

There exists increasing and consistent evidence that people use social networking platforms such as Facebook to connect with others for the purposes of expressing their political views (e.g. Gil de Zúñiga et al., 2012, 2017). This research also indicates that political exchanges on social media are linked to general patterns of civic, political, and social interaction (e.g. Dahlgren, 2000; Gil de Zúñiga et al., 2010; Vitak et al., 2011). Despite, however, these findings, there exists a relative paucity of studies that specifically investigate the individual-level factors that are associated with, and presumably stimulate, political talk (Gil de Zúñiga et al., 2016). The gap in the literature is further exacerbated in online contexts, where researchers have generally failed to investigate how user attributes and the connective potentials of digital media combine to facilitate political expression online. Those studies that do exist either focus on very specific user groups (e.g. political blog readers; Gil de Zúñiga et al., 2010), are platform non-specific (e.g. Lane et al., 2019; Yoo et al., 2017), or are limited by their exclusive use of self-report data (e.g. Gibson and Cantijoch, 2013; Gil de Zúñiga et al., 2010; Yoo et al., 2017).

In light of the foregoing, this study suggested that certain user attributes play important roles in the decision to take advantage of the Internet’s connective capabilities for the purposes of political engagement and expression. In this study, we propose (for various theoretical reasons) that the Internet’s political connective capabilities are most likely to be taken advantage of those who are civically engaged offline, those high in bridged social capital, those low in bonded social capital, and those with ideologically extreme views. General use of the Internet for political engagement, we predict, serves as a critical predictor of the use of Facebook for political expression. These contentions were tested using a relatively novel approach that combined self-reported and Facebook-based trace data.

This connection-based, general-to-specific approach to understanding online political expression adds to the literature in four potentially important ways. First, it addresses a previously identified need for additional studies assessing factors associated with and predictive of political talk (e.g. Gil de Zúñiga et al., 2016). In so doing, this study is one of the first scholarly attempts to understand how individual-level social and civic attributes may be associated with the disproportionate activation of the Internet’s connective potentials. Second, this study builds on prior research by offering further assessment of how general (i.e. non-Internet specific) aspects of the self relate to Internet-specific behavioral outcomes. By considering both online and offline aspects of political engagement and expression, this work builds on and extends prior studies that have addressed online outcomes within broader patterns of civic engagement and culture (e.g. Dahlgren, 2000; Gil de Zúñiga et al., 2010, 2012). Third, our focus on Facebook is important. According to data released by the Pew Research Center (2019), Facebook is the most widely used social media platform in the United States. Facebook has also become a central place for learning and communicating about politics. Recent studies, for instance, indicate that almost 40% of Americans learned about the 2016 election on Facebook (Pew Research Center, 2016a), that upward of 45% of Americans get the news on Facebook (Pew Research Center, 2019), and that nearly half of all Facebook-using Americans post political content (Pew Research Center, 2016b). Fourth, this study employed a novel combination of survey-based self-report data and observed social media behavior to uniquely assess the relationship between offline and online behaviors. Prior studies concerned with political communication online have, almost exclusively, relied on either self-report or trace data. Combination of the two data collection approaches allows for a comparatively rich theoretical and methodological assessments of online political expression. Such inquiry is especially important in the context of Facebook, where behavioral outcomes are often understudied due to data access limitations.

Political participation in a digital age

In the past, scholars primarily studied political participation through offline behaviors (Milbrath and Goel, 1977). These behaviors continue to be important; however, due to the widespread proliferation of Internet-enabled communication devices, it has become apparent that citizens perform a wide variety of behaviors in digital spaces and, therein, that these behaviors may complement and shape more traditional forms of political engagement (e.g. Gibson and Cantijoch, 2013). Within communication science subfields, scholars have generally defined the construct specifically relative to behaviors concerning interaction with government representatives and entities and interaction between citizens. For instance, recent work by Gil de Zúñiga et al. (2017) operationalized online political participation as activities such as signing/sharing an online petition, signing up to volunteer for a political cause, and starting a political group on a social media site. Similarly, Dozier et al. (2016) operationalized online political engagement in terms of emailing elected officials, signing online petitions, and donating money to political organizations. Gibson and Cantijoch (2013) cast an even wider net when defining the construct, and included factors such as citizen-to-citizen discussion and joining a political party online. In a recent study exploring political engagement among young people on social media platforms, Kahne and Bower (2018) emphasized connections between citizens’ activities such as circulating content and connecting with others to discuss important politic issues.

Offline civic engagement and online political engagement

While most researchers acknowledge that no single definition of civic engagement exists (Gil de Zúñiga et al., 2012), the most common operationalization of the concept comes from the work of Verba et al. (1995) who defined it as voluntary civic activity. Accordingly, this study conceptualizes civic engagement as any activity that intersects with community issues. These activities are not necessarily political in nature, but do, in their sum, contribute to political functioning. For example, this could include anything from volunteering for a charity or political candidate, and also buying a product that supports a shared mission. By voluntary, this activity cannot result in a financial benefit.

In its whole, civic engagement represents one’s investment in her or his surrounding community. Civic engagement is believed to be the backbone of American democracy (Verba et al., 1995). According to Gil de Zúñiga et al. (2016), involvement with civic activities fosters “ties, connections and communities in such way that embeds individuals in a thriving relationship between the state of power and individuals, enhancing a healthier democracy” (p. 536). Thus, while civic engagement behaviors may, themselves, occur in a local context, they ultimately speak to an overall concern for social well-being (Schneider, 2007). This is because when people engage with others in a civic context, they obtain important information about social others, build meaningful feelings of connection, and learn to identify points of compromise. In the long term, these mechanisms of civic socialization encourage “interpersonal trust, and political knowledge among individuals,” which, in turn, result in a “citizenry that is more interested, motivated, and active politically” (Pasek et al., 2009).

Most research on the relationship between civic and political engagement has occurred in offline contexts, but an emerging body of work provides indication that that offline civic participation is also positively associated with use of the Internet for political purposes. For instance, Pasek et al. (2009) suggested that Facebook’s affordances may be especially attractive to those who are civically engaged offline and, then, that the platform may help facilitate a variety of forms of political engagement and interaction. Gil de Zúñiga et al. (2014) found that features of civic engagement were positively associated with digital and social media use. Notably, while these studies seem to collectively suggest that offline civic engagement is related to the use of digital media for political purposes, they do not directly test the contention. However, in one longitudinal study examining the intersection between online and offline political participation among young people, Kim et al. (2017) found that online participation drove offline participation in a sample of young adolescents, while offline participation drove online participation is a sample of young adults. While this study is concerned with a related, albeit distinct, construct (offline civic engagement) and a different group of users (adults), Kim et al.’s (2017) findings support the notion that there exists a meaningful linkage between offline civic engagement and online political engagement. Thus, we suggest that offline civic engagement breeds general feelings of political interest, trust, and connectedness (Gil de Zúñiga et al., 2016; Pasek et al., 2009). Because the Internet can complement, supplement, and reinforce offline political behaviors (Dahlgren, 2000; Kim et al., 2017; Vitak et al., 2011), we predict that those who engage civically offline will be comparatively likely to use the Internet for an array of political purposes:

H1: Offline civic engagement will be positively related to online political engagement.

Social capital and online political engagement

Scholars broadly define social capital as the resources, both tangible and otherwise, that people attain through direct relationships with others or through membership in both official and unofficial groups; these resources are not necessarily positive, and social capital can be directed in a malevolent manner (Putnam, 2000). However, Putnam (2000) essentially argued that social capital accumulates through social connections and that the sustainability and stability of democratic societies rely on this. In this way, social capital speaks broadly to issues of mutuality and trust that are central to civic and political society.

There are multiple forms of social capital. Bridged social capital describes social connections with large swaths of acquaintances or membership within large and more heterogeneous groups of likeminded people. People with this type of social capital often feel part of a community, but lack close, tight relationships. Bonded social capital accrues from strong ties within more homogeneous groups such as families, circles of friends, or work organizations. People with this type of social capital often have very strong ties with their close social circles, but sometimes display intolerance of, or a lack of interest in, others (Putnam, 2000). However, as humans, the vast majority of people strive to find the close relationships that come with bonded social capital (Adler and Kwon, 2002). Because bridged social capital relates to loosely reciprocal relationships with a large and diverse group of social others, scholars have historically associated its accumulation with civic and political outcomes related to factors such as generalized social trust, political knowledge obtainment, political deliberation, and a willingness to be politically and civically engaged (e.g. Gil de Zúñiga et al., 2012). Alternately, because bonded social capital refers to strong connections to a relatively small group of social others, scholars have found that it offers limited democratic potential (Adler and Kwon, 2002; Uslaner, 1999). Indeed, in some cases, high levels of bonded social capital have been associated with withdrawal from civic and political life (Sommerfeldt, 2013). It should be noted that the two forms of social capital are not an either/or proposition (Sajuria et al., 2015).

From research on the intersection of social capital, digital technologies and political engagement, we know that digital spaces such as social networking sites can be used to generate and maintain bridged and bonded social capital; that searching for and obtaining political information on social networking sites is associated with enhanced levels of social capital; and that online discussion networks characterized by weak ties are associated with offline community engagement (e.g. Gil de Zúñiga et al., 2012). Together, these findings suggest that broad feelings of social connection may be consequentially linked to the use of the Internet for political purposes. That said, to-date research has broadly failed to distinguish between bridged and bonded social capital when exploring the concept’s association with political engagement. As illustrated above, there is reason to believe that bridged and bonded social capital work in oppositional manners as it pertains to online political engagement. Specifically, bridged social capital is associated with loose connections with diverse others and is thought to facilitate political mobilization (Gil de Zúñiga et al., 2012). Conversely, bonded social capital is associated with strong relationships with a small number homogeneous others (Putnam, 2000), the formation of particularized trust (Uslaner, 1999), and a withdrawal from political and civic life (Sommerfeldt, 2013). Drawing on the notion that Internet-based political behaviors are linked to general patterns of social and political interaction (Kim et al., 2017; Sajuria et al., 2015; Vitak et al., 2011), we hypothesize:

H2: Bridged social capital will be positively related to online political engagement.

H3: Bonded social capital will be negatively related to online political engagement.

Ideological extremity and online political engagement

For our fourth hypothesis, we suggest that ideological extremity is positively related to online political engagement. Research on political expression has shown an association between ideological extremity and discussion-based outcomes (e.g. Binder et al., 2009). Overall, this research tends to indicate that those with politically extreme political views tend to see the Internet as an especially useful tool for political engagement. This is for two inter-linked reasons. First, people who hold ideologically extreme beliefs tend to integrate these beliefs into their overall self-identity (Frederico and Hunt, 2013). Because salient parts of one’s self-identity warrant attention and reinforcement (Coleman and Williams, 2015), those with extreme positions may make comparatively greater investments in their political self (Frederico and Hunt, 2013). Online political engagement, for its part, is subject to a reduced number of resource-based limitations, heightening the likelihood of its regular use in these scenarios. Second, as it directly pertains to social connection, research has shown that the Internet gives people the ability to connect with likeminded others (e.g. Ancu and Cozma, 2009). One important way that people connect with others is through visible performances of self-identity (Papacharissi, 2011), such as posting political comments on the Internet. Online political engagement can thus fulfill personal identity and self-articulation needs, which, subsequently, play an important role in finding, joining, and engaging with groups of likeminded others. This may be especially important for ideologically extreme people, who may not have access to likeminded others in the physical world. Thus,

H4: Ideological extremity will be positively related to online political engagement.

Online political engagement and political expression on Facebook

Up until this point, we have treated the concept of online political engagement as the application of the Internet for the accomplishment of political ends. However, Internet applications are governed by varying affordances, which, ultimately, exert a shaping influence on the patterns of interaction that occur on-platform. Thus, as a final step in this study, we set out to explore the relationship between a general measure of political engagement and the more targeted behavior of political expression on Facebook. As social media platforms such as Facebook increasingly augment more traditional interpersonal means of community formation and involvement, it is of fundamental importance to obtain a nuanced understanding of how the overarching perception of the Internet as a political tool is specifically linked to platform and behaviorally specific outcomes. Indicators suggest that Facebook may be single most used Internet technology for political learning and discussion (Pew Research Center, 2016a, 2016b, 2019). Empirical assessment of the degree of the frequency to which political discussion occurs on the platform, and how such participation is linked to other online political behaviors, is thus a critical step forward in understanding the overall role played by Facebook in democracy. In so doing, the current analysis allows for critical evaluation of the validity of self-reported measures on online political engagement, which tend to be the default means of operationalizing the phenomena (Dozier et al., 2016).

We suggest that general levels of online political engagement will be positively associated with the specific activity of posting political content on Facebook. Drawing from prior work on media selection and use (Quan-Hasse and Young, 2010), the overarching contention here is that political expression on Facebook is one important means of engaging politically online. In this sense, we proceed from the general to the specific by theorizing that individual-level attributes pertaining to social, civic, and political connection needs guide the overall selection of the Internet for political engagement. Subsequently, and in light of Facebook’s near-total diffusion, digitally engaged citizens are increasingly likely to use of Facebook because the combined social and technical affordances of the platform conveniently and readily allow for political expression:

H5: Online political engagement will be positively related to political expression on Facebook. This relationship will exist even after controlling for the effects of offline civic engagement, bonded social capital, bridged social capital, and ideological extremity.

Finally, we wondered about the degree to which political talk on Facebook was specifically associated with other online political engagement behaviors. This inquiry is important because past research has differentially explored political expression relative to other forms of online political engagement (Gibson and Cantijoch, 2013; Gil de Zúñiga et al., 2017). Moreover, in light of H5, exploring the relationship between observed Facebook talk and an array of Internet-facilitated behaviors will help provide a fuller understanding of the needs fulfilled, specifically, by Facebook-based political communication:

RQ1: What are the associations between various forms of online political engagement and political expression on Facebook?

Method

This study combined survey self-report data with trace data, scraped from the survey respondents’ Facebook accounts. By combining data sources, the current approach offers a unique means of assessing the correspondence between estimated self-reported behavior and actual behavior on social media. We recruited a sample of US respondents using Qualtrics, a US-based data provider. Sample inclusion was predicated upon respondents being US citizens, holding accounts on both Facebook and Twitter, and having at least moderate levels of interest in US politics. We also controlled for an approximate 50/50 gender split. Before participating in the study, respondents were informed of the requirements governing participation and provided with a consent form. Regarding the collection of social media data, respondents were told that social media messages “will be collected anonymously.” Upon agreement to the terms of the study, respondents were asked to authorize a custom application that was used to harvest their social media data. The following information from each participant’s Facebook profile was retrieved using the Facebook Graph API: mobile_status_update, created_note, shared_story, created_event, wall_post, app_created_story, published_story. All identifiers were deleted from the data. After authorizing the application, respondents were piped into the survey environment. Self-report and social media data were joined using an anonymous identification code that was assigned by the web application. The authors’ institutional review board vetted all study procedures. The application used to collect the Facebook data conformed with the platform’s terms of service at the time of study execution and was approved by Facebook. Data were downloaded on 6 June 2017. All relevant Facebook user data were harvested from the date of their first post through the download date. Of those that engaged with the study materials, 13.5% provided valid data. In all, 783 valid survey responses with accompanying trace data were collected.

Measures

Political expression on Facebook

The dataset contained hundreds of thousands of Facebook content instances. Given the large number of messages, the researchers adopted a supervised machine learning (ML) approach to identify text-based political content. These efforts were focused specifically on identifying user-generated commentary concerning political issues. Accordingly, the following content instances were assessed for political content: mobile_status_update, created_note, shared_story, wall_post, published_story.

To develop the algorithm, 100 messages were first chosen at random and annotated by two independent coders for the presence of political talk. A post was determined to be political in nature if user-posted text mentioned a political figure; discussed public policy; discussed legislation/legislative actions; discussed municipal/local political issues; mentioned high-profile social issues; mentioned an election or voting; or was related to the Supreme Court or other high-profile judiciary proceedings with political ramifications. For this study, posts were required to contain user-generated text. Of the 100 decisions, the two coders disagreed once. A random sample of 1000 additional posts were then selected and distributed to the two coders, who each annotated 500 additional messages for political talk. The annotations were used to build a ML algorithm inside of the DataRobot platform. The DataRobot tool allows researchers to estimate and comparatively evaluate a series of ML solutions. Here, a neural network ensemble model had the highest performance scores and was ultimately chosen. After the initial model was built, subsequent rounds of messages were randomly chosen, stratifying across highly scored predictions, middle predictions, and low predictions to help reinforce learning across all classes. In total, 5006 annotations were made by the two coders. Performance metrics were subject to a 10-fold cross validation, each time training on a randomly selected 64% of the data. The final ML model for political talk had F₁ and AUC scores of 0.88 and 0.98. An accuracy of 94.7%, a false positive score of 3.7% and a Matthews Correlation Coefficient of .85 all indicated that despite bias in classes, the algorithm distributed its misclassifications evenly.

Online political engagement

Online political engagement was measured using five items (where response categories were 1 = never, 7 = frequently) developed from prior measures employed by Dozier et al. (2016) and Gil de Zúñiga et al. (2017). These questions asked participants to indicate how frequently over the last 12 months they emailed political officials, signed online petitions, liked or followed a politician on social media, commented about a politician on social media, and used social media to ask a politician a question. In its sum, this measure accounted for political engagement between and among citizens, and citizens’ political engagement with various governmental actors.

Offline civic engagement

Offline civic engagement was measured using modified versions of the scales from Gil de Zúñiga and Valenzuela (2011) and McFarland and Thomas (2006). All items were placed on seven-point scales where 1 = never, 7 = frequently. Examples of the items used to measure offline civic engagement include the following: Over the past 12 months . . . about how often have you worked or volunteered for nonpolitical charitable organizations? and . . . about how often have you attended a meeting to discuss problems relevant to your community?

Bonded social capital

A measure of bonded social capital was developed from the literature (e.g. Williams, 2006). Specifically, the current measure was reflected in three items: I have strong personal relationships with my family members; I have strong personal relationships with my close friends; and I have people in life who would help me if I needed it, no matter what. All items were on seven-point scales (1 = strongly disagree, 7 = strongly agree).

Bridged social capital

Using the extant literature (e.g. Gil de Zúñiga et al., 2012) three general measures, all on seven-point scales (1 = strongly disagree, 7 = strongly agree), of bridged social capital were developed: I like to keep a large network of acquaintances; I have a large network of people with whom I am friendly with; and I feel like I am part of my community.

Ideological extremity

Ideological extremity was measured by assessing participant conservatism using three measures (e.g. Generally speaking, what is your political ideology?; Generally speaking, I tend to support candidates who are . . .; I think of myself as a . . .), all placed on seven-point scales where higher scores equaled higher levels of conservatism and lower scores equaled higher levels of liberalism. Each of these three measures were subsequently recoded into four-point scales where those who selected options at the scale poles were assigned higher numbers (i.e. a 1 or 7 was coded as a 4, a 2 or 6 was coded as 3, and so on).

Control measures

A number of adjustment factors were assessed, including the following: political interest, political party identification, conservatism, political news surveillance, Facebook usage intensity, Facebook activity duration, participant gender, and participant age in years. General political interest was assessed using four questions (I’m interested in politics, I like to learn as much as I can about politics, I follow politics closely, and I enjoy talking about politics with others), all of which were on seven-point scales where higher scores represented higher levels of interest. We included the general political talk indicator because prior research indicates that generalized motivations toward political talk are an important predictor of online political engagement and expression (Gil de Zúñiga et al., 2010). Incorporation of this measure helped us guard against potentially confounded findings. Party identification was measured by asking participants to indicate the political party (1 = Democratic, 2 = Republican, 3 = Independent, 4 = Other) that they most identified with. For subsequent statistical analyses, this variable was dummy-coded with Democratic set as the contrast category. Conservatism was measured using the three items described in the above section on construction of the political extremity measure. Political news surveillance was assessed by asking respondent to indicate how often they read the newspaper, watch the news on network television, watch the news on cable television, purposefully seek out political information on social media, and read news blogs. All items were on scales where 1 = never and 7 = frequently. These five items were collapsed into a single composite index. Facebook usage intensity was measured by asking respondents to indicate how often they use Facebook (1 = never, 7 = frequently). Finally, using the harvested trace data, we calculated the length of each respondent’s platform activity, which was computed by measuring the number of years between the respondent’s first and last post.

Descriptive statistics and reliability coefficients for the sample are provided in Table 1. Figure 1 shows a histogram for the political expression variable.

Table 1.

Descriptive statistics for study variables.

	M	SD	α	%
Facebook political posts	12.80	71.64
Political interest	5.30	1.42
Party ID: Democrat				40.2
Party ID: Republican				23.6
Party ID: Independent				32.5
Party ID: Other				3.7
Conservatism	3.72	1.80	.97
Political information surveillance	4.01	1.58	.83
Facebook intensity	6.46	1.07
Facebook activity duration (years)	7.25	2.33
Male				41.6
Female				58.4
Age (years)	39.35	12.98
Online political engagement	3.67	1.66	.86
Offline civic engagement	2.96	1.46	.84
Bridged social capital	4.76	1.42	.82
Bonded social capital	5.77	1.24	.77
Ideological extremity	2.47	1.10	.94

SD: standard deviation.

Reported statistics are for those that provided complete data; total number of political posts generated by analytic sample was 9227; 56.9% of the sample generated at least one political post on Facebook; among those with post counts ⩾ 1, the mean value was 22.50 posts (SD = 93.90).

Figure 1.

Histogram showing Facebook political post creation frequency.

Missing data analysis

Little’s Missing Completely at Random test was used to examine the degree to which missing data were systematic in nature. The results indicated that the data were missing completely at random, χ² = 206.40, df = 245, p > .05. Listwise deletion was employed, resulting in an analytic sample n of 721.

Data analysis

Predicting online political engagement

H1–H4 were concerned with prediction of the general measure of online political engagement. To test these hypotheses, a series of ordinary least squares (OLS) regression models were estimated. The first model contained the covariates of interest, while the second model added the independent variables of interest.

Predicting political expression on Facebook

As can be seen in Figure 1, the political expression measure was a count variable that contained a large number of zero counts. A negative binomial logit hurdle (NBLH) model thus used to test H5. Hurdle models are two-part models that comprised a binary model that addresses zero and positive counts and a zero-truncated count model that addresses all positive counts. In this particular instance, positive counts were assessed using a zero-truncated negative binomial (NB) model. NB models account for violation of the Poisson distribution’s assumption of equidispersion. Because respondents were active on Facebook for varying periods of time, the count component of the NBLH model used in this study employed platform activity duration as an offset term, resulting in coefficients that can be interpreted as representing the rate of occurrence relative to a given time interval (here, instances of political expression on Facebook per year of platform activity). Thus, taken in its whole, the modeling approach employed in this study allowed us to assess the degree to which the variables of interest were associated with (1) generating at least one instance of political expression on Facebook and (2) among those who created at least one political content item on Facebook, the rate at which content items were posted per year of site activity. For the binary component of the model, we provide both the logged odds (b) and odds ratios (OR). For the count component of the model, we provide the logged counts (b) and incidence rate ratios (IRR). To address potential confounds, the NBLH model included the independent variables identified in H1–H4 in addition to the specified control factors. Finally, for RQ1, we estimated five NBLH models that explored the bivariate associations between the individual items comprising the online political engagement scale and the observed levels of political talk on Facebook. As in the case of H5, these models employed Facebook activity duration as an offset factor in the count component of the model.

Results

Hypotheses 1–4

As shown in Table 2 (Model 1), the control variables accounted for approximately 47% of the variance in online political engagement, F(10, 710) = 62.47, p < .001. The model accounting for the independent variables of primary interest (Figure 1, Model 2) accounted for an additional 16% of the variance in the criterion variable. In all, this model accounted for approximately 63% of the variance in online political engagement, F(14, 706) = 85.30, p < .001.

Table 2.

Series of ordinary least squares (OLS) regression models with online political engagement set as the criterion variable.

	Model 1				Model 2
	b	SE	b [95% CI]	β	b	SE	b [95% CI]	β
Political interest	0.33***	0.05	[0.23, 0.42]	0.28	0.24***	0.04	[0.16, 0.32]	0.21
Democrat (0)–Republican (1) Contrast	−0.08	0.15	[−0.37, 0.21]	−0.02	−0.05	0.12	[−0.29, 0.19]	−0.01
Democrat (0)–Independent (1) Contrast	0.10	0.11	[−0.12, 0.32]	0.03	0.28**	0.10	[0.08, 0.47]	0.08
Democrat (0)–Other (1) Contrast	0.11	0.25	[−0.38, 0.59]	0.01	0.18	0.21	[−0.23, 0.59]	0.02
Conservatism	−0.04	0.03	[−0.11, 0.02]	−0.05	−0.01	0.03	[−0.07, 0.04]	−0.01
Political news surveillance	0.46***	0.04	[0.38, 0.55]	0.44	0.24***	0.04	[0.16, 0.31]	0.23
Facebook intensity	0.02	0.04	[−0.06, 0.11]	0.01	0.05	0.04	[−0.03, 0.12]	0.03
Facebook activity duration	−0.03	0.02	[−0.07, 0.01]	−0.04	−0.02	0.02	[−0.05, 0.01]	−0.03
Biological sex (1 = Female)	−0.12	0.10	[−0.31, 0.07]	−0.04	0.02	0.08	[−0.14, 0.18]	0.01
Age	−0.01	0.00	[−0.01, 0.00]	−0.04	−0.00	0.00	[−0.01, 0.01]	−0.00
Offline civic engagement					0.55***	0.03	[0.48, 0.61]	0.48
Bridged social capital					−0.03	0.03	[−0.10, 0.03]	−0.03
Bonded social capital					−0.12**	0.04	[−0.19, −0.05]	−0.09
Ideological extremity					0.13***	0.04	[0.05, 0.21]	0.09
F	10, 710 = 62.47***			14, 706 = 85.30***
R ²	.47			.63
ΔR²	.16***

CI: confidence interval; SE: standard error.

All variance inflation factors below 2.51.

p < .01; ***p < .001.

Examination of the coefficients shown in Model 2 of Table 2 suggested that offline civic engagement was positively associated with online political engagement, b = 0.55, p < .001; β = 0.48 (H1 supported). Contrary to expectations, we did not observe a significant relationship between bridged social capital and online political engagement, b = −0.03, p > .05; β = −0.03 (H2 not supported). We, did, however, observe a negative and significant association between bonded social capital and online political engagement, b = −0.12, p < .01; β = −0.09, supporting H3. Finally, we observed a significant and positive relationship between political extremity and online political engagement, b = 0.13, p < .001; β = 0.09 (H4 supported).

Hypothesis 5

Model assessment

Before interpreting the NBLH model coefficients, we assessed model performance using a hanging rootogram plot. Good models should show minimal signs of systematic over or under-prediction. As shown in Figure 2, the NBLH model accounted for the observed distribution adequately. As a function of its design, the NBLH properly accounted for the number of zero counts. As it pertained to the positive count frequencies, we observed minimal levels of systematic misfit, particularly for count bins ranging from 1 to 20, which cover the majority of the cases that posted at least one political post on Facebook.

Figure 2.

Hanging rootogram plot for the negative binomial logit hurdle (NBLH) model.

Model results

The logistic component of the model suggested that a one-unit increase in the online political engagement scale was associated with a 68% increase in the odds of making at least one political post on Facebook (b = 0.52, p < .001). Turning next to the count component of the model, we found that a one-unit increase in the online political engagement scale was associated with a 60% increase in the yearly rate of posting political commentary on Facebook (b = 0.47, p < .001). Accordingly, H5 was supported.

Among the independent variables of interest in H1–H4, we observed significant coefficients for offline civic engagement (b = −0.31, p < .001; OR = 0.73), bridging social capital (b = 0.15, p < .05; OR = 1.17), bonding social capital (b = −0.23, p < .05; OR = 0.80), and ideological extremity (b = 0.19, p < .05; OR = 1.21) in the binary component of the NBLH model. In the count component, only ideological extremity was significantly related to posting rate, b = 0.26, p < .05; IRR = 1.30. A full report of these findings is provided in Table 3.^1–3

Table 3.

Logit and count components of a negative binomial hurdle (NBLH) model with political expression on Facebook set as the criterion variable.

	Logit component				Count component
	b	SE	b [95% CI]	OR	b	SE	b [95% CI]	IRR
Political interest	0.06	0.09	[−0.12, 0.24]	1.06	0.34**	0.13	[0.09, 0.59]	1.40
Democrat (0)–Republican (1) Contrast	0.08	0.26	[−0.43, 0.59]	1.08	0.09	0.47	[−0.84, 1.02]	1.09
Democrat (0)–Independent (1) Contrast	0.11	0.21	[−0.31, 0.52]	1.11	0.07	0.34	[−0.59, 0.73]	1.08
Democrat (0)–Other (1) Contrast	0.99*	0.50	[0.00, 1.98]	2.70	−0.49	0.58	[−1.64, 0.65]	0.61
Conservatism	−0.10	0.06	[−0.21, 0.02]	0.91	−0.26**	0.10	[−0.46, −0.07]	0.77
Political news surveillance	−0.19*	0.09	[−0.36, −0.02]	0.83	0.02	0.11	[−0.20, 0.24]	1.02
Facebook intensity	0.16*	0.08	[0.00, 0.32]	1.17	0.04	0.14	[−0.24, 0.31]	1.04
Facebook activity duration	0.18***	0.04	[0.10, 0.25]	1.19	−0.15**	0.05	[−0.25, −0.06]	0.86
Biological sex (1 = Female)	−0.26	0.18	[−0.61, 0.09]	0.77	−1.28***	0.25	[−1.76, −0.80]	0.28
Age	0.03***	0.01	[0.02, 0.05]	1.03	0.07***	0.01	[0.05, 0.09]	1.07
Online political engagement	0.52***	0.09	[0.35, 0.69]	1.68	0.47***	0.13	[0.23, 0.72]	1.60
Offline civic engagement	−0.31***	0.09	[−0.49, −0.14]	0.73	0.07	0.12	[−0.16, 0.30]	1.07
Bridged social capital	0.15*	0.07	[0.01, 0.30]	1.17	−0.10	0.10	[−0.30, 0.10]	0.90
Bonded social capital	−0.23*	0.08	[−0.39, −0.07]	0.80	−0.08	0.12	[−0.31, 0.15]	0.92
Ideological extremity	0.19*	0.09	[0.02, 0.36]	1.21	0.26*	0.13	[0.00, 0.52]	1.30
Null loglikelihood	−1824.37
Residual loglikelihood	−1648.95
Likelihood ratio	χ² = 350.85, df = 30***

CI: confidence interval; IRR: incidence rate ratios; OR: odds ratios; SE: standard error.

All variance inflation factors below 2.70.

p < .05; **p < .01; ***p < .001.

Research question 1

RQ1 was concerned with exploring the relationship between observed Facebook expression frequency and the individual items that together comprised the self-reported online political engagement scale. As seen in Table 4, we observed generally similar relationships (positive and statistically significant) between the individual components of the general online political engagement measure and political expression on Facebook. The strongest overall association was found in the case of the item pertaining to commenting about a political figure on social media. The weakest overall association appeared in the case of the item describing the frequency with which respondents use social media to ask political figures questions.

Table 4.

Bivariate relationships between online political engagement items and political expression on Facebook.

	Logit components		Count components
	b	OR	b	IRR
Sent email to a politician	0.16***	1.17	0.38***	1.46
Signed an online petition	0.15***	1.16	0.37***	1.45
Followed or liked political figure on social media	0.15***	1.16	0.14*	1.15
Commented about a political figure on social media	0.23***	1.26	0.39***	1.47
Asked a political figure a question on social media	0.12**	1.13	0.10	1.11

OR: odds ratios; IRR: incidence rate ratios; results derived from five (negative binomial logit hurdle) NBLH models.

p < .01; ***p < .001.

Discussion

This study builds upon and extends prior research seeking to understand how individual-level factors are associated with political engagement and political expression online. This study took a general-to-specific approach and sought to understand how features of one’s overall social and political life are associated with use of the Internet for political engagement. Therein, we explored how the general perception that the Internet is viable means of engaging politically is associated specifically with the use of Facebook for political expression. While other studies (e.g. Gil de Zúñiga et al., 2010, 2012; Kim et al., 2017) have collectively explored the variables of primary interest to this study, our work is the first, to our knowledge, to simultaneously assess the ways in which offline civic engagement, social capital, and perceptions related to ideological extremity are associated with online political expression and engagement. Taken as a whole, then, our study is important because it provides evidence that individual features pertaining to general patterns of connection with others are associated with increased use of the Internet’s politically connective potentials, even after accounting for potentially confounding variables such as political interest and political information consumption.

More granularly, the currently presented results augment the current body of literature on social capital and online political participation. Prior work by Gil de Zúñiga et al. (2012, 2017) has shown a null relationship between offline social capital and online engagement. Examination of the measures employed in these studies, however, suggests that the authors were generally interested in bridged social capital. With regard to bridged social capital, our findings generally comport with prior work in the sense that we did not find a consistent statistically significant association between bridged social capital and online political engagement in either its general form or as it pertained to observed political talk on Facebook. However, in the case of bonded social capital, we observed a negative and statistically significant association between the construct and online political engagement. Our work shows that distinguishing between types of social capital is important, as the strands of social capital do not have unilaterally positive implications for democratic engagement online.

This work also has significant implications as they pertain to the current scholarly understanding of online political engagement as an operational construct. As noted, most to-date research on online political engagement measures the concept using self-report measures. Our study’s unique combination of self-reported and trace data shows that there exists a fairly robust association between a recollected measure of one’s overall political engagement and a much more specific, observed measure of political expression on Facebook (Table 3). This finding, we believe, supports the validity of self-report measures of political behavior online. Moreover, looking at Table 4, we note that political content creation on Facebook seems to be especially strongly linked to behaviors which result in the production of political messages that oscillate around self-expression. That said, our findings do not address whether social media-based political expression should be considered distinct from or, instead, part of future online political engagement measures, they do suggest that future research would be well-served by the exploration of the degree to which these concepts are/are not distinct.

The data further provided support for models of political communication that have suggested linkages between online and offline engagement (e.g. Dahlgren, 2000). Specifically, our data show that factors such as offline civic engagement and generalized political interest are associated with use of the Internet for political purposes (Table 2), and that using the Internet for political engagement is specifically associated with posting political content on Facebook (Table 3). Having said that, the current data do not speak to whether these associations are supplementary or, alternately, redundant. In other words, do platforms like Facebook help people accomplish tasks they cannot accomplish offline or do these platforms merely afford users the ability to accomplish the same task in multiple ways? Future research should address this question more thoroughly.

This research also suggested that ideological extremity may play an important role in the decision to be politically engaged online. In the presented theoretical rationale, we suggested that Internet-based political communication is attractive to ideologically extreme users because it allows for both the articulation of the self as political entity and, perhaps more importantly, because it allows for the bounding of political discussion on the basis of others’ legitimacy as civic agents. The results shown in Tables 2 and 3 indicate that ideological extremity was fairly robustly linked to online political engagement in both general and specific ways. One interpretation of this finding is that Facebook may be an especially attractive platform for the expression of ideologically extreme political sentiment. Such proposition is generally consistent with research on partisan media, fake news, disinformation, and misinformation, which broadly shows that Facebook is the central marketplace for such content (e.g. Allcott and Gentzkow, 2017).

Interestingly, we observed a lack of directional and associative consistency both across and within the OLS and NBLH models for some variables. For instance, as predicted by H1, offline civic engagement was strongly and positively related to online political engagement (Table 2). In the NBLH models predicting political expression on Facebook, however, the coefficient for offline civic engagement was somewhat paradoxically negative in the logit model component of the model. A similar effect was observed for political news surveillance. The coefficients for offline civic engagement and news surveillance in the NBLH model may be of limited importance for several reasons. The primary intention of H5 was to assess sharing rate. As seen in the count component of the model, both offline civic engagement and news surveillance were positively (albeit non-significantly) related to political expression frequency on Facebook. These coefficients can be contrasted with the positive and comparably robust association between online political engagement and Facebook political expression. This supports the overarching contention that certain individual-level attributes are associated with general use of the Internet for political engagement, and that the decision to engage politically online informs the more specific use of Facebook for ongoing political expression. Moreover, in the bivariate component of the NBLH model, the parameter estimate for online political engagement is considerably more robust than the parameter estimates associated with either offline civic enjoyment or political news surveillance. Considered in conjunction with our theorizing, this suggests that online political engagement is a comparatively more important predictor of political expression on Facebook. Finally, bivariate analyses (not reported due to space limitations) suggested that both news surveillance and offline civic engagement were positively and weakly related to posting at least one instance of political commentary on Facebook. The addition of the online political engagement variable results in both sign reversal and magnitude amplification for these factors. This finding considered in conjunction with the results reported in Table 2 and the fact that there were strong bivariate correlations between online political engagement and both offline civic engagement (r = .70) and news surveillance (r = .65) might suggest that that there exists a subset of civically engaged political news consumers who are politically active in non-Facebook online spaces. While this conclusion is potentially informative, it does not challenge the overarching contentions of this study. Although Facebook may be an especially important platform for political engagement, it is not the only digital space where such interaction occurs.

This study has its limitations. First, while this study may suggest the presence of ordered/structural relationships, it employed a survey-based approach that is unable to make determinations relating to causality. Indeed, it stands to reason that variables employed in this study reinforce one another. Second, our sample is non-representative in nature, and skewed more liberal, more politically engaged, and more female than the United States as a whole. Third, our data collection approach, which combined survey and trace data, may have incurred systematic sampling bias due to issues related to data privacy. The degree to which this factor impacted our results is unknown. Fourth, as it pertained to the covariate variables, we, in some cases, deviated from measurement approached used by public opinion projects such as the ANES (e.g. political interest). Fifth, while trace data offer a rendering of user behavior online, the current application cannot speak to the quality of such behavior. Finally, not all civic and political engagement behaviors can be captured in a self-report environment and thus the currently employed measures should be considered non-exhaustive.

In conclusion, this study meaningfully contributes to the literature on online political communication. While some studies navigate similar terrain, this study is the first, to our knowledge, to explicitly address the degree to which individual-level patterns of social, civic, and political interaction are associated with the decision to activate Facebook’s politically connective potential. In addition, our combination of survey and trace data is both methodologically novel and facilitates the extension of a theoretical understanding of online political participation.

Footnotes

Authors’ note

The author has agreed to this submission, and this article is not currently being considered for publication by any other print or electronic journal. The first two authors of this study are listed alphabetically as they contributed equally to the manuscript.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

ORCID iD

Patrick Ferrucci

Notes

Author biographies

Patrick Ferrucci (PhD, University of Missouri) is an assistant professor in the Department of Journalism in the College of Media, Communication and Information at University of Colorado Boulder. His research broadly examines various aspects of media sociology, specifically how economics and technology influence digital news. Primarily, his work focuses on how the shifting models of organizations populating the journalism industry and its boundaries affect news construction processes.

Toby Hopp (PhD, University of Oregon) is an assistant professor in the Department of Advertising, Public Relations and Media Design at the University of Colorado Boulder. His research interests include the uses and effects of video games, online political incivility, and user learning processes in digital contexts.

Chris J Vargo (PhD, University of North Carolina at Chapel Hill) is an assistant professor of big data and analytics at The University of Colorado Boulder. He specializes in the use of computer science methods to investigate social media using theories from the communication and political science disciplines. Research methods of specialization include text mining, machine learning, computer-assisted content analysis, data forecasting, information retrieval, and network analysis.

References

Adler

Kwon

S-W

(2002) Social capital: prospects for a new concept. The Academy of Management Review 27: 17–40.

Allcott

Gentzkow

(2017) Social media and fake news in the 2016 election. Journal of Economic Perspectives 31: 211–236.

Ancu

Cozma

(2009) MySpace politics: uses and gratifications of befriending candidates. Journal of Broadcasting & Electronic Media 53: 567–583.

Binder

Dalrymple

Brossard

, et al. (2009) The soul of a polarized democracy: testing theoretical linkages between talk and attitude extremity during the 2004 election. Communication Research 36: 315–340.

Coleman

Williams

(2015) Looking for myself: identity-driven attention allocation. Journal of Consumer Psychology 25: 504–511.

Dahlgren

(2000) The Internet and the democratization of civic culture. Political Communication 17: 335–340

Dozier

Shen

Sweetser

, et al. (2016) Demographics and Internet behaviors as predictors of active publics. Public Relations Review 42: 82–90.

Frederico

Hunt

(2013) Political information, political involvement, and reliance on ideology in political evaluation. Political Behavior 35: 89–112.

Gibson

Cantijoch

(2013) Conceptualizing and measuring participation in the age of the Internet: is online political engagement really different to offline? The Journal of Politics 75: 701–716.

10.

Gil de Zúñiga

Veenstra

Vraga

, et al. (2010) Digital democracy: reimagining pathways to political participation. Journal of Information Technology & Politics 7(1): 36–51.

11.

Gil de Zúñiga

Jung

Valenzuela

(2012) Social media use for news and individuals’ social capital, civic engagement and political participation. Journal of Computer-Mediated Communication 17(3): 319–336.

12.

Gil de Zúñiga

Molyneux

Zheng

(2014) Social media, political expression, and political participation: panel analysis of lagged and concurrent relationships. Journal of Communication 64(4): 612–634.

13.

Gil de Zúñiga

Valenzuela

(2011) The mediating path to a stronger citizenship: online and offline networks, weak ties, and civic engagement. Communication Research 38: 397–421.

14.

Gil de Zúñiga

Barnidge

Scherman

(2017) Social media social capital, offline social capital, and citizenship: exploring asymmetrical social capital effects. Political Communication 34: 44–68.

15.

Gil de Zúñiga

Jung

Valenzuela

(2012) Social media use for news and individuals’ social capital, civic engagement and political participation. Journal of Computer-Mediated Communication 17: 319–336.

16.

Gil de Zúñiga

Valenzuela

Weeks

(2016) Motivations for political discussion: antecedents and consequences on civic engagement. Human Communication Research 42: 533–552.

17.

Gil de Zúñiga

Veenstra

Vraga

, et al. (2010) Digital democracy: reimagining pathways to political participation. Journal of Information Technology & Politics 7: 36–51.

18.

Kahne

Bower

(2018) The political significance of social media activity and social networks. Political Communication 35: 470–493.

19.

Kim

Russo

Amnå

(2017) The longitudinal relation between online and offline political participation among youth at two different developmental stages. New Media & Society 19: 899–917.

20.

Lane

Lee

Liang

, et al. (2019) Social media expression and the political self. Journal of Communication 69: 49–72.

21.

McFarland

Thomas

(2006) Bowling young: How youth voluntary associations influence adult political participation. American Sociological Review 71: 401–425.

22.

Milbrath

Goel

(1977) Political Participation: How and Why Do People Get Involved in Politics? New York: University Press of America.

23.

Papacharissi

(2011) A Networked Self: Identity, Community, and Culture on Social Network Sites. New York: Routledge.

24.

Pasek

Romer

(2009) Realizing the social Internet? Online social networking meets offline civic engagement. Journal of Information Technology & Politics 6: 197–215.

25.

Pew Research Center (2016a) Liberal Democrats Most Likely to Have Learned About Election From Facebook. Pew Research Center. Available at: http://www.pewresearch.org/fact-tank/2016/05/12/liberal-democrats-most-likely-to-have-learned-about-election-from-facebook/

26.

Pew Research Center (2016b) Public Attitudes Toward Political Engagement on Social Media. Pew Research Center. Available at: http://www.pewinternet.org/2018/07/11/public-attitudes-toward-political-engagement-on-social-media/

27.

Pew Research Center (2019) 10 Facts About Americans and Facebook. Pew Research Center. Available at: http://www.pewresearch.org/fact-tank/2019/02/01/facts-about-americans-and-facebook/

28.

Putnam

(2000) Bowling Alone: The Collapse and Revival of American Community. New York: Simon and Schuster.

29.

Quan-Hasse

Young

(2010) Uses and gratifications of social media: a comparison of Facebook and instant messaging. Bulletin of Science, Technology & Society 30: 350–361.

30.

Sajuria

VanHeerde-Hudson

Hudson

, et al. (2015) Tweeting alone? An analysis of bridging and bonding social capital in online networks. American Politics Research 43: 708–738.

31.

Schneider

(2007) Connections and disconnections between civic engagement and social capital in community-based nonprofits. Nonprofit and Voluntary Sector Quarterly 36: 572–597.

32.

Sommerfeldt

(2013) The civility of social capital: public relations in the public sphere, civil society, and democracy. Public Relations Review 39: 280–289.

33.

Uslaner

(1999) Democracy and social capital. In: Warren

(ed.) Democracy and Trust. Cambridge, MA: Cambridge University Press, pp. 121–150.

34.

Verba

Schlozman

Brady

(1995) Voice and Equality: Civic Voluntarism in American Politics. Cambridge, MA: Harvard University Press.

35.

Vitak

Zube

Smock

, et al. (2011) It’s complicated: Facebook users’ political participation in the 2008 election. Cyberpsychology, Behavior, and Social Networking 14: 107–114.

36.

Williams

(2006) On and off the ’net: scales for social capital in an online era. Journal of Computer-Mediated Communication 11: 593–628.

37.

Yoo

Kim

Gil

Zúñiga

(2017) Cognitive benefits for senders: antecedents and effects of political expression on social media. Journalism & Mass Communication Quarterly 94: 17–37.