Personal Network Composition and Cognitive Reflection Predict Susceptibility to Different Types of Misinformation

While some researchers rely on source-based definitions, that is, defining misinformation according to whether or not the source is fictitious or lacks credibility (e.g. Grinberg et al., 2019; Pennycook et al., 2021), other researchers use content-based definitions, defining misinformation according to whether the content is false or misleading (Roozenbeek et al., 2022; Traberg, 2022; Traberg et al., 2022). However, social scientists generally agree that misinformation represents information that is inaccurate, false, or misleading regardless of the intent (Pennycook & Rand, 2021; Traberg, 2022). The range of misinformation includes misleading hyperpartisan headlines to completely untrue fake news. An example of hyperpartisan news was the headline that Nancy Pelosi vowed to raise taxes if Democrats won the US House in 2018. The reality was more nuanced as Pelosi discussed negotiating taxes with the Republicans at an event, but she did not explicitly say she would raise them (Politifact, 2021). Fake news examples are often more straightforward because there is not even a kernel of truth in them, such as Pope Francis endorsing Donald Trump during the 2016 Presidential election (Factcheck.org, 2016).

Explicitly fake news spreads quickly and pollutes the information ecosystem (Grinberg et al., 2019), and this is especially true for political fake news (Vosoughi et al., 2018), but it is a small portion of the average American’s overall media diet (Allen et al., 2020). Hyperpartisan and misleading news, including unsubstantiated political rumors, account for the majority of misinformation as they are far more common (Bradshaw et al., 2020; Faris et al., 2017). However, both types of misinformation can be dangerous as demonstrated by the harmful false information spread during the COVID-19 pandemic (Roozenbeek et al., 2020). In addition to misinformation’s harmful effects on public health, political misinformation and polarization also threaten the stability of democracies (Graham & Svolik, 2020). To combat misinformation, it is important to identify the factors that help explain what makes people susceptible to it.

Our identities can help us make sense of our social worlds by adhering to their values, roles, and scripts (Burke & Stets, 2009). We also can form identities around groups that we belong to (Tajfel et al., 1979). These identities based on group memberships can inspire us to protect our group while denigrating opposing groups (Stets & Burke, 2000). This protection of our social group and the motivation to follow the roles associated with our identities can also bias information-processing.

Political identities are a common example of this identity bias process. Growing political polarization and partisanship are crucial factors that increase the spread of false information. Negative feelings toward the political outgroup have been steadily increasing over the past few decades in the United States (Finkel et al., 2020). Greater political polarization in our information ecosystems creates a feedback loop where individuals are exposed to political misinformation from their networks or media sources and then they share this misinformation to others in their network, increasing both polarization and the strength of partisan identities (Van Bavel et al., 2021). Research has shown that both Democrats and Republicans are more likely to interpret information in a way that supports their own political group (Ditto et al., 2019). Misinformation was again central in the 2020 Presidential Election and 59% of the Republicans agreed that believing the election was stolen was important to their political identity (CNN, 2021). Additionally, when one’s political identity is made more salient to partisans, they report even stronger partisan beliefs (Unsworth & Fielding, 2014). Research has further found that political biases may impact which sources we accept falsehoods from. For instance, Bauer and Clemm von Hohenberg (2020) found that individuals are more likely to believe fake news from sources that had reported politically congruent news in the past, and Traberg and van der Linden (2022) found that social media users were more likely to perceive headlines from politically congruent sources as reliable due to perceiving those sources as more credible. This may in part be because perceived similarity with sources plays a role in both the persuasion (Chaiken & Maheswaran, 1994; Metzger et al., 2003) and the advice-taking process (Marks et al., 2019). That is, individuals are more likely to be persuaded by or accept information from sources they perceive to be similar on some characteristic, such as sharing similar political attitudes (Marks et al., 2019).

Not only does political identity influence our information processing, but it can motivate individuals to share misinformation to attack their political outgroup (Rathje et al., 2021). The subjective importance of identities also plays a role in how strongly they can influence our behavior (Ervin & Stryker, 2001; Stets & Serpe, 2016). For example, Democrats and Republicans who strongly identified with their political group perceived greater political polarization (Westfall et al., 2015). Additionally, vaccine skeptics who agree that their anti-vaccine identity is very important to them are the most likely to endorse misinformation regarding vaccines (Motta et al., 2021). Given the social nature of misinformation, it is important to understand the specific conditions where misinformation and networks are the strongest.

Homophily is the phenomenon that describes how individuals who share characteristics are more likely to associate with each other (McPherson et al., 2001). Both Democrats and Republicans prefer homogenous networks (Pew Research Center, 2014) and prefer to watch media that supports their politics (Pew Research Center, 2020). As mentioned above, it has been theorized that network composition creates a feedback loop that exacerbates misinformation and partisan identity (Van Bavel et al., 2020). Individuals are also motivated to share misinformation in order to attack their outgroup and signal ingroup identity (Rathje et al., 2021), which further polarizes their networks. This is consistent with previous work that has shown that people will adapt their politics to match their personal networks overtime (Lazer et al., 2010) and that a group discussion with those who supports one’s politics will increase polarization (Keating et al., 2016). However, it appears that the type of network tie matters for determining these effects. Superficial exposure to opposing views will not break the echo chamber effect. Bail et al. (2018) paid Democrats and Republicans to follow social media accounts that shared opposing political information for 1 month. Neither group reduced their polarization, and Republicans were even more polarized afterward. Consistent with these findings, partisans are often motivated to attack their political outgroup on social media instead of having nuanced discussions of policy (McPhetres et al., 2021). When investigating non-social media connections, Robison et al. (2018) found that increased diversity in one’s network who they discuss political matters with did not reduce polarization either. Importantly, political discussants or social media accounts are not particularly close to the individual.

Facciani and Brashears (2019) evaluated if the political diversity in one’s group who they discuss “important matters” with would influence political beliefs. Those who we discuss “important matters” with tends to record people in our networks who we have particularly meaningful social interaction with (Brashears, 2014). Facciani and Brashears (2019) found that having just one person with different politics in one’s “important matters” network was associated with significantly less extreme political beliefs for both Democrats and Republicans. In a subsequent networks study also using the “important matters” item, Facciani et al. (2023) found that the political composition of one’s network predicted their attitudes toward vaccination. Specifically, Facciani et al. (2023) found that more positive attitudes toward vaccination were predicted by having a greater proportion of Democrats in one’s network and more negative attitudes toward vaccination were predicted by a greater proportion of Republicans in one’s network. Thus, personal network structure does appear to influence how an individual processes political information, but only when the network is especially meaningful to them. This is because the closeness of social ties is crucial for identity formation and belief commitment (Cooley, 1902; Smith & Emerson, 1998).

Stets et al. (2021) argue that shared identity meanings help foster homophily, and interactions with our network reinforce and perpetuate network homogeneity. As described above, individuals have a motivation to maintain consistency between their identities and their behaviors. When individuals are in a network of close social connections who share their identity, those individuals further verify one’s identity, and the individual also verifies the identities of people in their network. As these social bonds strengthen among people with a shared identity, so does the sense of “we-ness” and the social distance from other people who do not share their identity (Stets et al., 2021). This identity may also become more personally important to an individual when more people in their network share that identity with them.

For example, if an individual identifies as a Democrat, they are often motivated to support their Democrat identity with their behaviors. This could involve supporting Democratic politicians, sharing pro-Democrat news in their networks, and sharing positive stories about Democrats (and negative stories about Republicans) with their close associates. If this individual’s personal network is entirely comprised of other Democrats, those Democrats are also sharing pro-Democrat information with the individual. Crucially, the social foundation of their identity is constantly being supported by those in their network, potentially increasing the importance of that Democratic identity. Thus, both the identity verification feedback loop for the individual and the identity support from their homogenous network may compound each other.

In addition to political identities and networks, the Dual-Process Model of Reasoning has been consistently linked to belief in false information (Pennycook & Rand, 2019, 2020). Dual-Process Models of Reasoning assert that there are two main categories of cognitive processes, which are often differentiated between Type 1 and Type 2. Type 1 involves a cognitive process that is fast and intuitive, whereas Type 2 refers to thinking that is slower and more deliberate (Evans & Over, 1996; Sloman, 1996), with work showing that some individuals have a tendency to engage in one style of thinking more than the other (Liberali et al., 2012).

The Cognitive Reflection Task (CRT) is a popular method for determining one’s preferred cognitive style (Frederick, 2005), with higher scores corresponding with more Type 2 thinking patterns and lower scores with Type 1 styles (i.e., lower cognitive reflection). Pennycook and Rand (2021) have consistently found that individuals scoring low on the CRT are more likely to believe and share false information. Additionally, Pennycook et al. (2020) found that asking people to judge the accuracy of headlines can decrease people’s tendency to share fake news. Pennycook and colleagues (2021) also found that employing simple accuracy nudges as people use Twitter made them less likely to share false information. However, subsequent studies have found such accuracy nudges have only worked among Republicans (Rathje et al., 2022).

Pennycook and Rand (2019) found that lower CRT scores predicted the perceived accuracy of fake news even when the fake news headlines aligned with participant’s politics. However, Batailler et al. (2022) reanalyzed Pennycook and Rand’s (2019) study and separated the general ability to identify real vs. fake news and the partisan biases in evaluating news as real or fake. Batailler et al. (2021, p. 7) found “participants were better in discriminating between real news and fake news when the headlines were congruent with their political ideology than when they were incongruent with their political ideology.” When the data was analyzed in this fashion, it was clear that partisan bias played a major role in how participants evaluated the misinformation stimuli in the study.

The debate between partisan identity bias and cognitive reflection may benefit from comparing how both factors predict different types of misinformation stimuli. Cognitive reflection may better explain overall truth discernment and partisan bias may be more indicative of overall biased beliefs related to one’s political identity (Gawronski, 2021). Thus, the present study aims to evaluate the impact both partisan bias and cognitive reflection have on different types of misinformation.

Our study aims to test the influence of network homogeneity and cognitive reflection on judgments of two different types of misinformation: vague political rumors and published fake news headlines. We propose that network homogeneity and the subsequent identity verification process should create bias for any type of information that provides an opportunity to support one’s identity. We further propose that this identity protection should have a broader influence on processing political information than cognitive reflection processes that are more focused on discerning truth from falsehoods.

Facciani (2020) found that Democrat and Republican participants judged brief political rumors as more likely if their own political group was engaging in a positive behavior (i.e., helping) and their opposing group was engaging in a negative behavior (i.e., bullying). As this identity bias was exacerbated by having a completely homogenous network, we predict that this bias may also apply to processing both unverified political rumors and fake news headline that either support one’s own group or attack one’s opposing group. However, we predict that cognitive reflection should only predict bias in judging the likelihood of fake news headlines, but not political rumors. Both the political rumors and fake news headlines provide an opportunity for identity verification. However, only the fake news headlines provide an opportunity for truth discernment as the social scenarios are purposely vague and allow the participant’s identity bias to impact judgment.

Stets et al. (2021) argue that shared identity within one’s network increases the sense of “we-ness” and the social distance from other people who do not share their identity (Stets et al., 2021). The sense of community among those who share one’s identity is a central factor for how important that identity can be to them (Brenner et al., 2014). We test this idea by applying it to political identities and networks. We hypothesize that participants with a completely homogenous network will have higher ratings of the subjective importance of one’s political identity. Research has shown that partisans exhibit both bias for their ingroup and against their outgroup (Amira et al., 2021). If network homogeneity bolsters this effect, we should observe that both Democrats and Republicans will rate their own group more positively than the outgroup when they have a politically homogenous network.

Beyond evaluation ratings, we also predict that network homogeneity will influence the likelihood judgments of social scenarios involving political groups. Having a politically homogenous network should be positively associated with the likelihood of believing one’s own group is engaging in a positive behavior as well as believing one’s outgroup is engaging in a negative behavior. We also predict that participants with homogenous networks will be more likely to want to share the politically biased rumors as well. This leads us to the following formal hypotheses:

In addition to homogeneity influencing the likelihood of political rumors, we also predict that homogeneity will be positively associated with judging fake news headlines as more likely when they support one’s political ingroup as well as being more likely to indicate sharing politically congruent fake news headlines:

We do not predict cognitive reflection will be associated with political rumor likelihood. Political rumors provide an opportunity for partisans to support their political ingroup. Cognitive reflection is particularly relevant for the perceived accuracy of fake news (Gawronski, 2021) and should not be relevant for a broader bias that involves estimating the likelihood of political rumors or a desire to share such rumors.

We do assert that cognitive reflection should still predict believing and sharing fake news headlines for both Democrats and Republicans as it involves discerning truth from falsehoods (Pennycook & Rand, 2019).

We recruited 107 Democrats and 107 Republicans to complete our online study. Participants were recruited from Prolific, which is a website that connects participants to studies and pays them for their completion (Palan & Schitter, 2018). Participants were 62% female, 67% white, M_age = 34.06 (SD = 12.71), 49% Christian identified, 52% had a bachelor’s degree or higher, and 39% had an income of 75,000 USD per annum or higher.

After providing informed consent, participants listed up to six names with whom they discuss “important matters” with. After providing each name, the participant was asked if this name was a Democrat, Republican, Independent, or other. Next, participants completed three cognitive reflection tests, followed by a rating of Democrats and Republicans using Affect Control Theory’s procedure described below. Then participants judged the likelihood of political social scenarios happening, rated the accuracy of fake headlines as well as their likelihood of sharing both. Finally, they answered demographic and political ideology questions.

We first evaluated if having a completely homogeneous network predicted the importance of political identity and ingroup and outgroup evaluation. Combining both samples, 42% of the participants had a completely homogenous network, which was consistent with previous research (Facciani, 2020; Facciani & Brashears, 2019). There was not a significant difference in network homogeneity between Democrats and Republicans (b = 0.10, t (213) = 1.53, p = 0.128). Our ordinary least squared regression model (Table 2) found that network homogeneity significantly predicted higher importance of political identity (b = 0.527, t (213) = 3.28, p < 0.001). This supports H1: Network homogeneity increases participants’ self-reported importance of political identity.

Our next ordinary least squared regression model (Table 3) found that participants with a completely homogenous network had significantly higher ingroup evaluation (p < 0.01) and lower outgroup evaluation (p < 0.01). This supports H2: Democrats and Republicans with homogenous networks will rate their ingroup more positively than their outgroup.

After observing that network homogeneity influences evaluations toward political groups, we analyzed if network homogeneity also predicted judgments and sharing of our two types of misinformation. We first evaluated whether network homogeneity predicted belief and sharing of political rumors. We found that (Table 4) network homogeneity predicted a higher likelihood of believing the helping rumor (b = 0.19, t (213) = 1.74, p = 0.083) as well as the desire to share it (b = 0.25, t (213) = 2.52, p = 0.013). However, network homogeneity did not have any significant association with the bullying rumor or the desire to share in our sample. While the main homogeneity variable was dichotomous, we also measured the impact of proportional homogeneity as a robustness check (see Figure 1).

Figure 1:

Thus, these findings partially support H3a: Both Democrats and Republicans with a politically homogeneous network will be more likely to rate politically congruent rumors as true and H3b: Both Democrats and Republicans with a politically homogeneous network will be more likely to report sharing politically congruent fake headlines.

Next, we evaluated whether network homogeneity predicted belief in fake news headlines (Table 5). We found that network homogeneity did predict belief in conservative fake news headlines for Republicans (p < 0.05; one-tailed), but not in Democrats. Figure 2 illustrates how a greater percentage of Republicans in our Republican participant’s network predicted higher belief in conservatively biased fake news.

Figure 2:

Additionally, network homogeneity did not predict the likelihood to share fake news headlines for either Democrats or Republicans. Thus, we find partial support for H4a: Both Democrats and Republicans with a politically homogeneous network will be more likely to rate politically congruent fake headlines as true, but fail to find any support for H4b: Both Democrats and Republicans with a politically homogeneous network will be more likely to report sharing politically congruent fake headlines.

An OLS regression failed to find any significant difference between CRT scores between Democrats and Republicans (p > 0.05). Homogeneity predicted belief and sharing in half of our political rumors. We predicted that cognitive reflection should not predict belief in political rumors or sharing at all. This is precisely what we found as cognitive reflection was not related to any political rumor belief or sharing (Table 6), which supports H5a: Cognitive reflection not related to rumor belief and H5b: Cognitive reflection not related to rumor sharing.

We found that cognitive reflection predicted belief in fake news (see Table 7), but this was only the case for Republicans (p < 0.05; one-tailed). Figure 3 illustrates how higher CRT scores (combined versions) predict lower belief in conservative fake news.

Figure 3:

However, cognitive reflection did not predict belief in fake news headlines for Democrats or sharing behaviors for either political group. Thus, we found partial support for H6a: Cognitive reflection predicts politically congruent fake news belief and failed to find support for H6b: Cognitive reflection predicts politically congruent fake news sharing. Finally, we ran an exploratory regression model to see if cognitive reflection could predict network homogeneity and found a nonsignificant relationship between the two variables (see Supplementary Materials).