Testing Classical Predictors of Public Willingness to Censor on the Desire to Block Fake News Online

Abstract

This study examined media use and attitudinal predictors of public willingness to censor fake online political news among representative samples in Lebanon, Saudi Arabia, and Tunisia (total N = 2880). The study utilized research on the corrective action hypothesis (CAH) and the theory of presumed media influence (TPMI) as frameworks. The CAH holds that an individual’s belief that media are hostile and influential increases the likelihood that the individual will participate in public discourse urging countermeasures. TPMI maintains that the belief that media are influential is associated with attitudes about media, though those attitudes need not be negative. Perceived exposure to fake news online positively predicted willingness to censor fake news in all countries, aligning with some prior research on both the CAH and the TPMI. Facebook use was negatively associated with willingness to censor fake news in two of the countries, while trust in news media was a positive correlate in two countries. Implications for research on both willingness to censor and on fake news are discussed.

Keywords

Arab region censorship corrective action hypothesis fake news misinformation online news survey theory of presumed media influence

Online fake news is believed to be common by large percentages of national populations in several countries. Seventy-four percent of US respondents say they often or sometimes encounter online political news that appears fabricated (Dennis et al., 2018). In that study, the same share or more of nationals in three Arab countries – Lebanon, Qatar, and Saudi Arabia – also said they often/sometimes encounter concocted political news stories online (the figure was slightly lower, 6 in 10, in Tunisia). Numerous governments in the Arab region have criminalized fake news, though the target of those laws is suspected to be dissenting speech, not fabricated information (Funke and Flamini, 2020). Also, fake news sometimes comes from governments in the Arab region themselves, such as when the Saudi government claims that the dismembering of Jamal Khashoggi was merely a questioning session that went too far (see Lamb and Rashwan, 2020).

While the term ‘fake news’ is often applied colloquially to a range of information – sometimes news someone dislikes or does not want to believe – the scourge of online fake news is real, and so are its effects. Pennycook et al. (2018) reported that individuals shown fake news were more likely than people who had not been shown fake stories to believe the falsities, even in experimental conditions in which fake news was labeled false. Individuals sometimes choose to spread and repeat information they know is fake, which may actually create less cognitive dissonance than acknowledging that the information is false. This may be partly why, in the current study, we may see respondents who feel fake news should not be stanched.

Given how often fake news has recently been discussed – it was Collins English Dictionary’s word of 2017 (Abadi, 2017) – there has been a flurry of research on it, including on the term’s definition (Lazer et al., 2018; Tandoc et al., 2018), its origins and reach (Benkler et al., 2018), its perceived prevalence online (Martin and Hassan, 2020), and its effects (Pennycook et al., 2018). In 2019, Americans ranked fake news as a bigger problem than terrorism (Mitchell et al., 2019). The current study examines the extent to which respondents in three countries believe something should be done to stem the spread of fake news online.

The study tests whether variables found to predict willingness to censor in previous research apply to the desire to censor or block the spread of fake news online. For example, prior work has found that people who frequently consume a particular kind media content (video games, pornography, for example) are less eager to see that content censored (McLeod et al., 2001), and this study examines whether perceived exposure to fake news online is associated with eagerness to censor fake news. Prior research has also found that many demographic variables – such as low education, female gender, and conservative ideology – predict willingness to censor media, and these indicators and more are also tested here.

This study is a secondary analysis of data from the Media Use in the Middle East survey and examines predictors of Arab nationals’ willingness to censor fake news – in Lebanon, Tunisia, and Saudi Arabia. The nations are all Arabic-speaking, Muslim-majority countries, and yet represent three different locales the Eastern Mediterranean, North Africa, and the Arab Gulf. Fake news is a timely topic in Arab countries. One of the countries in the current study, for example, Saudi Arabia, led a blockade against its rival Qatar, following the publication of fake news stories that Saudi Arabia’s partner UAE planted on a Qatar news site (Ulrichsen, 2020).

Additionally, several Arab countries have enacted or are contemplating enacting anti-fake news laws, though human rights groups typically see the measures as anti-free speech mechanisms (Funke and Flamini, 2020). So, the current study not only adds to literature on correlates of willingness to censor media by examining their predictive validity vis-à-vis fake news but also examines data from a part of the world where fake news is salient and consequential.

Definition of fake news utilized in the current study

The definition of fake news used in this study is similar to Lazer et al. (2018) and comes from Tandoc et al. (2018): fabrication or news manipulation created to deliberately mislead. Lazer et al. (2018: 1094) defined fake news as ‘fabricated information that mimics news media content in form’. Tandoc et al. identified six categories of or relating to fake news, two of which are relevant to the current study. Fabrication is news with no factual standing, dressed up as news to deceive people, such as deep fake videos – highly persuasive, tech-generated videos of typically public figures uttering things that they never said – and concocted online stories. The questions put to respondents in the survey from which the current data were drawn asked about ‘made up’ news, specifically describing fabrication. News manipulation refers to tampering with visual communication to create a false impression, such as with doctored photographs.

Other categories from Tandoc et al. do not apply to the current study, including official propaganda, native ads, and satire. Willingness to block/censor fake news, this study’s dependent variable, is not likely the same as willingness to censor news generally, as fake news is deemed antisocial and news is typically considered prosocial. Still, as willingness to proscribe news has a partisan political component (Rosen, 2002) so, too, can the perception of where fake news comes from have a partisan element (Newman et al., 2018). For this reason, measures of conservatism and religiosity are among the predictors included in this study’s models.

Predictors of public willingness to censor

Prior research has found that even people in allegedly liberal societies are willing to censor media (Fisher et al., 1994) and that every country censors some things, based partly on shared norms, especially those addressing content like disinformation and propaganda (Paek et al., 2008). Some people in the United States are willing to censor political speech, particularly when they perceive that speech to be both harmful and influential (Salwen, 1998). What qualifies as harmful and influential may differ considerably from person to person. ‘Harmful’ speech may consist of hate speech, for example, in which case some people might say that little is lost in censoring it. ‘Influential’ may mean anything from mildly affecting attitudes about an issue, or, at the other end of a spectrum, inciting receivers of the message to violence. In such studies, like Salwen’s, then, it is important that respondents are given a clear definition of the thresholds for both harm and sway.

In studies in Western locales, positive correlates of willingness to censor are age, being female, low education level, conservative self-identification, and religiosity (Lambe, 2002). Some prior research found that people who use a certain kind of media – porn users or video game players, for example – are less eager to censor that content than people who do not use it (McLeod et al., 2001). Lo and Paddon (2001) found that women with little prior exposure to pornography reported greater perceived adverse effects of porn on others than did women with high exposure to porn and that the gap between the perceived effects of porn on themselves and on others, the third-person phenomenon – the tendency for individuals to believe that media influence other people much more strongly than they affect oneself – was larger (see also Perloff, 2002; Rojas et al., 1996). While the parent survey from which the data scrutinized in the current study were drawn did not ask respondents whether, and how much, they ‘indulge’ in fake news by, say, entertaining the false information communicated in the fake news or by sharing it online, there nonetheless may be a relationship between respondents’ perceived exposure to fake news and their willingness to have governments, laypersons, and tech companies block its spread. Indeed, unlike gaming and porn, fake news may be something that Internet users endure, rather than indulge in.

Still, the finding that consuming media content is inversely correlated with the inclination to censor that content has also been observed in the context of news; people who get news online have been found less supportive of censoring online news. In Russia, for example, people who relied more on state TV networks for news and relied less on Internet sources reported greater willingness to censor Internet content (Nisbet et al., 2017).

Whether exposure to media content correlates negatively with willingness to censor that content is tested in the current study; one predictor of willingness to censor fake online political news is respondents’ perceptions of how often they encounter fake news online. Prior research has studied effects of various media use on willingness to censor content such as hate speech and porn. Lambe (2004) found that newspaper use negatively predicted willingness to censor hate speech, but TV news use positively predicted willingness to censor porn; newspaper use was not predictive of willingness to censor porn, and TV news use was not associated with willingness to censor hate speech. Variables similar to those in Lambe’s work, such as print media use and online news use, are also tested in the current study.

The corrective action hypothesis and theory of perceived influence

The corrective action hypothesis (CAH) holds that perceptions of media bias – suspicion of hostile media – and the perceived influence of media increase the likelihood that people will engage in public discussion urging countermeasures. The origin of the CAH (Davison, 1983) described a political campaign worker who found a leaflet for an opposing candidate and then drafted a response. The likelihood that people engage in public discussion increases when they see something they deem harmful (Barnidge and Rojas, 2014). People are more likely to write online posts or comments on a public topic if they feel media coverage of the topic is biased and harmful (Rojas, 2010). In the current study, the dependent variable of ‘action’ is not counteractive speech but a call for censorship – blocking the spread of fake news online – though it is a call for corrective action. The term ‘fake news’ entered the public discourse as a political scourge, which is why the questions that asked respondents about exposure to fake news online in the survey from which the current data were drawn specifically referenced fake political news (Wendling, 2018).

This study examined predictors of an index recommending that three entities (governments, tech companies, laypersons) block fake news online. The three items load together extremely well, suggesting that individuals who have concerns about fake news are eager for any entity to confront the phenomenon. The CAH is relevant in several of the specific relationships examined in the current study. Take trust in news media: low trust in news media is partly a belief that news media produce negative outcomes, trust in media is a correlate examined in this study. Some prior research has found that trust in news media is positively associated with perceived exposure to fake news (Wasserman and Madrid-Morales, 2019), another reason we include trust as a predictor variable.

Meanwhile, with age as one of our demographic control variables, we also contend with prior research that trust differs among age groups, and older news consumers are more likely to get news from mainstream outlets (Tsfati and Cappella, 2003). This study tests other variables relating to corrective action as predictors of willingness to censor fake news, such as support for censorship more generally and support for greater Internet regulation.

Most work on CAH is from Western locales, and predictors of willingness to censor in Arab countries may not covary in the same way. In several Arab countries, Martin et al. (2016) examined correlates of support for censoring entertainment content, finding that predictors of willingness to censor found in prior, mostly Western-centric work – religiosity, conservatism, female gender, and others – didn’t predict willingness to censor in the countries. Most notably, in Saudi Arabia, religiosity negatively predicted willingness to censor.

The current study examined correlates of willingness to censor in Arab countries, but there is an important difference in the dependent variable here: while content like violent video games and misogynistic music has some public supporters, fake news has few advocates, with the possible exception of those arguing that criminalizing fake news has a chilling effect on speech, that is, it encourages self-censorship (Tompros et al., 2017). Both the group of countries studied here and the dependent variable, then, differ from the contexts and outcomes in prior research on willingness to censor.

The theory of presumed media influence

The theory of presumed media influence (TPMI) holds that a belief that media have strong effects is associated with media use and attitudes about media (Tal-Or et al., 2010), including individuals’ willingness to censor. Hong (2020) found that presumed influence of media positively predicted support for regulating fake news about climate change, and it was also associated with support for other corrective action, such as donating money to environmental organizations. Cohen and Weimann (2008) found that the perceived influence of reality TV shows was a positive correlate of willingness to censor them.

We discuss the TPMI here, separately from CAH, because the outcome variable in that body of research on the former is often willingness to censor. Behavior predicted by the CAH is often pro-social: participating in public debate. TPMI, though, focuses on both negative and positive effects of believing media have strong influence. Tsfati et al. (2011) found that scientists who believe news media have powerful effects are more likely to grant interviews about their research to journalists. Censoring fake news is viewed by many people as a pro-social undertaking, though again, human rights advocates express concern about criminalizing it.

The omnibus survey utilized in this secondary analysis did not include explicit measures of the perceived influence of fake news, though implicit in the outcome variable in the current study – respondents’ willingness to ‘prevent made up stories from gaining attention’ – is that fake political news online, if it reaches people, is influential, and negatively so. Some prior research cited above has found that increased exposure to a specific type of media content is associated with reduced willingness to censor that content, but the media use examined in that research, such as viewing violent entertainment or sexual content, is often deliberate and purposeful; while presumably, few people use the Internet to regularly consume information they believe is false.

Media system and regulatory differences among the Arab countries in this study

Although they share a common language, many Arabic-speaking countries have distinct media systems (Rugh, 2004), and some of the most striking differences involve censorship laws and regulations, media and news production, and media consumption. Rugh categorized press systems in Arab states as loyalist (Saudi Arabia, constrained), diverse (Lebanon, vocal), and mobilizing (Tunisia, driving political action, at least since its 2010 revolution), and while a few decades old, Rugh’s typology is still relevant with regard to the Arab countries studied here.

Saudi Arabia has strict censorship mechanisms that limit freedom of speech and expression, and Saudi Arabia ranks nearly last on press freedom, 172 of 178 nations, while Lebanon is 101 and Tunisia 72 (Reporters Without Borders, 2019). Recent cases of Saudi Arabia crushing dissent include the murder of Jamal Khashoggi, arrests and prolonged incarcerations of dissidents, and jailing women’s rights activists (‘World Report,’ 2018). In 2017, according to World Report, Saudi Arabia introduced a ‘counterterrorism law’ that assigns 5–10 years in prison to anyone who criticizes the king or crown prince in a way that ‘brings religion or justice into disrepute’.

Regarding media production, Saudi Arabia produces little compared to several other Arab countries, especially Lebanon (Schoenbach et al., 2016), as legal and cultural restrictions of media in Saudi Arabia are disincentives to artists and investors. That said, media use among Saudis is higher than among nationals in several other Arab countries, including with regard to online gaming, TV use, and binge-watching, and many Saudis use VPNs, which enables them to view content online that is otherwise blocked in Saudi Arabia (Dennis et al., 2018).

Since the Tunisian uprising in 2010 and 2011, there had been striking improvements in free speech in that country, which had previously restricted media more than most countries in the world (Reporters Without Borders). Tunisia is highlighted as a model in the Arab region for protecting freedoms of press and expression (Walters, 2016). In 2018, Tunisia’s Court of First Instance rejected a petition to block the website of Shams Rad, Tunisia’s first LGBTQ+ radio station, and the operation is believed to be the first gay radio operation in the Arab region (Ram, 2018). Schools in Tunisia include sex education for both boys and girls – rare in Arab countries, many of which censor sexual information in science textbooks (Masri, 2019). While media in Tunisia now enjoy greater protections under the law than prior to 2010, the country has recently regressed in its freedom of expression milieu by prosecuting several laypersons for their political or social commentary (Amnesty International, 2020). Tunisia has indicated to the world that, regardless of any regulatory and/or policy protections of free speech, peaceful protesters should be fearful of disappearing into the detention system of the interior ministry (Human Rights Watch, 2021).

For a longer time than Tunisia, Lebanon has earned a positive reputation for lax regulation of speech among Arab countries. A paper publication can be censored in Lebanon only if a successful lawsuit is brought against it (‘Cut It Out’, 2012). Lebanon has one of the largest media production landscapes among Arab countries, especially in film, TV, news, and music (Schoenbach et al., 2016). The first Arabic daily newspaper, Hadiqat al-Akhbar, was published in Beirut in 1858. Still today, Lebanon is a greater media producer for its size than nearly all other Arab countries and has dozens of daily and weekly newspapers (Schoenbach et al., 2016). Perhaps, greater barriers to journalism and mass media productivity than regulatory or policy limitations are the chaos within a nearly failed state like Lebanon’s (Gavlak, 2021).

There are also at least nine broadcast TV channels in Lebanon, many of which are not owned by the government. Media production in Lebanon outpaces Tunisia in many media sectors, but production in the latter country has been quickening in the decade since its revolution (see Buckley et al., 2013). What currently threatens media in Lebanon may be not regime censorship but unbridled chaos; the country has been blasted by currency devaluation, broad financial collapse, coronavirus, and the physical implosion of Beirut (Allsop, 2020). Due to several important distinctions between countries regarding media content and customs, separate regression models predicting fake news burden (FNB) are conducted for each Lebanon, Saudi Arabia, and Tunisia, and at the same time this approach enables the identification of predictors of FNB that behave similarly across countries.

Research question and hypotheses

Based on research on public willingness to censor, the CAH, and the TPMI, this study poses a research question and hypotheses relating to willingness to censor fake news online (FNB, for ‘fake news burden'). Some prior research has found exposure to a certain kind of content – pornography, violent video games – is negatively associated with willingness to regulate that content (McLeod et al., 2001), which might suggest that people who are more exposed to fake news are less interested in censoring it. But porn and video games, unlike fake news, are forms of content many people deliberately seek out, and often purchase, while few people consistently seek out information online they believe is false.

Additionally, Jang and Kim (2018) found that a belief that fake news is harmful to others did not predict increased willingness to regulate fake news. Pennycook et al. (2018) identified a mere-exposure effect of fake news; people exposed to fake news stories were more likely than people who had not seen the stories to say the claim of the story was true, even when the fake news items were labeled as untrue. As prior research relating to exposure to content and willingness to censor, then, is mixed, we pose a research question rather than a formal hypothesis:

RQ: How, if it all, does perceived fake news exposure (FNE) online correlate with willingness to censor fake news online (FNB), after controlling for other attitudinal, media use, and demographic variables?

This study also examined a number of media use and media-related attitudinal variables as predictors of FNB. In addition to demographics, such measures include attitudes about censoring online content, trust in news media, support for Internet regulation; online news use, Facebook use, and print media use.

Martin et al. (2016) found that nationals and residents in Saudi Arabia, Qatar, and the UAE who supported banning ‘offensive’ scenes from films were likely to also support entire films being banned. Support for censorship generally should positively predict FNB. Such prior research underlies H1:

H1: Support for Internet censorship generally and for censorship of entertainment media will positively predict FNB, after controlling for media use, additional attitudinal factors, and demographics.

The Internet in Saudi Arabia is heavily censored (‘10 Most Censored Countries’, 2015), while online content in Tunisia and Lebanon is mostly left alone (‘Internet censorship listed’, 2012), thus, Internet users in Saudi Arabia who manage to see fake news online should report a heightened, negative response to it. Additionally, opposition to censorship is higher in Saudi Arabia than in several other Arab countries, including in some cases Lebanon and Tunisia (Dennis et al., 2016). Lebanese and Tunisians, who have a broader range of potentially shocking content in Internet networks in their countries, then, should report less of an adverse reaction to fake news. This brings us to hypothesis 2:

H2: The association between FNE and FNB, while holding the aforementioned variables constant, will be stronger (positively) in Saudi Arabia than in Tunisia or in Lebanon.

Based on frequencies reported in Media Use in the Middle East 2018, fewer Tunisians said that each governments, tech companies, or laypeople bear a great deal/some responsibility to stop the spread of fake news online, while Lebanese had higher FNB scores, which were exceeded by Saudis (see Figure 1, reproduced with permission). While a different pattern could emerge in comparing the scale means across countries, depending on how many people in each country answered ‘great deal’ versus ‘some’ responsibility, such a Simpson’s paradox (Wagner, 1982) is not likely. We thus pose as hypothesis 3:

H3: Tunisians and Lebanese will report lower average FNB scores than Saudis.

Figure 1.

(Reproduced with permission). In most Arab countries and the United States, FNB frequencies were similar. While the Media Use study asked fake news questions in five Arab countries, the combined FNB scale was internally reliable in Lebanon, KSA, and Tunisia. FNB, fake news burden.

Method

This study examined predictors of attitudes among nationals in three Arab countries (total N = 2880 from Saudi Arabia, Lebanon, Tunisia) regarding the responsibility of governments, tech companies, and citizens to stop the spread of fake news online. The three items were combined to form FNB.

Sampling and data collection

This study is a secondary analysis of omnibus survey data from Lebanon, Saudi Arabia, and Tunisia, collected in 2018 as part of the annual Media Use in the Middle East study, published every year since 2013 by Northwestern University in Qatar. Nationally representative samples were collected in each country using multistage randomized probability sampling. The Harris Poll conducted fieldwork in all countries, which involved randomized in-person interviews at respondents’ households. Dates of data collection, all of which occurred in 2018, were Lebanon – July 10 to August 13; Saudi Arabia – July 20 to September 20; Tunisia – August 8 to September 5.

Respondents completed the survey in Arabic, English, or French (Lebanon and Tunisia). Response rates were robust: 48% in Lebanon, n = 991 nationals; 33% in Tunisia, n = 1144; 74% in Saudi Arabia, n = 745. Data were rim-weighted to increase representativeness, and weighting factors were gender and geographic location in a given country. The sample size in Saudi Arabia is smaller than in the other two countries because this study examines nationals in the countries, and in a nationally representative survey of Saudi Arabia, roughly 25% of the sample will be non-citizens. In the Appendix 1, demographic characteristics of the three samples can be compared to census figures from each of the three countries (CIA World Factbook Middle East, 2019), on age, gender, education, and median monthly household income. The weighting of cases in the current data set was light; as shown in the Appendix 1, as sample data from each country closely align with census figures.

Measurement

Dependent variable: FNB

Index of three items developed by Pew Research Center (Mitchell et al., 2016), asking respondents if each governments, the public, and tech companies bear responsibility for stopping the spread of fake news online. The items were replicated in the 2018 Media Use in the Middle East survey, and the three questions were combined as an index.

The prompt was: ‘You may have heard some recent instances of so-called “fake news stories” circulating widely online. How much responsibility does each of the following have to prevent made-up stories from gaining attention?’ Note that the prompt does not include a value judgment on fake news, so respondents should have felt free to say that nothing should be done to impede fake news. The question was asked for each ‘Members of the public; Government/politicians; Social networking sites like Facebook/Twitter and search sites like Google’. 4 = Great deal of responsibility; 1 = No responsibility. Combined, the index is a variable ranging from a low of 3 to a high of 12. Figure 1 shows frequencies for FNB items (respondents who answered ‘great deal’, ‘some’ responsibility) in Arab countries.

Cronbach αs of FNB were 0.70 in Lebanon, 0.72 in Tunisia, and 0.63 in Saudi Arabia. The α in Saudi Arabia is somewhat low, though a meta-analysis by Taber (2017) reported that αs > 0.60 are deemed acceptable in many publications. In addition to running regression models for each of the three countries separately, given the lower alpha in Saudi Arabia, we disaggregated the three items comprising the dependent variable index for Saudi Arabia and ran separate regression models for each (Appendix 2). The disaggregated regressions for Saudi Arabia did not differ significantly from the aggregated model. The Media Use study initially fielded the fake news items in five countries, those studied here and also Qatar and UAE, though reliability of the scale in those two countries was too low for analysis. As such, this study does not necessarily recommend universal use of FNB and generalizes findings to the three countries studied.

Predictors of FNB: Media use variables

Extending Lambe’s work, numerous media use predictors were scrutinized. Checks news online: ‘How frequently do you check news online?’ 1 = never; 6 = several times a day; Print media use: index of three items, Cronbach’s α = 0.80, ‘How often do you read books/read newspapers/read magazines?’ For each 1 = never; 6 = several times a day; Time spent playing video games: ‘How many hours in a typical week do you spend playing video games, either online or offline?’ Ratio-level measure. Some video game portals and forums are themselves spaces where fake news spreads (see Condis, 2018), and so we include it as a potential correlate of FNB.

As Facebook has been a major vehicle for fake news (Lazer et al., 2018), we include it as a predictor. Uses Facebook daily or more: ‘Do you use Facebook at least once on a typical day?’ 0 = no; 1 = yes; Uses Twitter: ‘Do you use Twitter?’ (too few respondents use Twitter for ‘once a day’ measure) 0 = no; 1 = yes; Uses VPN. ‘Do you use a VPN?’ Clarification by interviewer if needed: ‘VPNs can make it possible to view films or TV programs that aren’t available in your country, and can help protect online privacy’. 0 = no; 1 = yes.

Attitudinal variables

Perceived FNE online: Index of two Pew items (Mitchell et al., 2016), split-half α = 0.70, ‘How often do you come across news stories about politics and government online that you think are not fully accurate?’, and ‘How often do you come across political news stories online that you think are almost completely made up?’ 1 = never; 4 = often. Fake news as fabricated news came into the public discourse via political news that was deliberately false (see Wendling, 2018), which is why the Pew questions on FNE focus on political news. The actual words ‘fake news’ are not present in these questions, though the question refers to the definition of fake news provided by Tandoc et al. – as fabricated or manipulated news. There is theoretical justification for combining these items beyond just the strong split-half reliability: Nielsen and Graves (2017) found that people make little distinction between news they deem partially fake and news they believe is completely fake – partially false and completely false are both fake to many news consumers. This is an attitudinal measure of perceived exposure to fake news, and so the limitations inherent in self-reported data, such as recall error and ideological bias, apply.

Wants more Internet regulation: The Internet in this country should be more tightly regulated. 1 = strongly disagree; 5 = strongly agree; Trusts mass media (Gallup question; Swift, 2016): ‘In general, how much trust/confidence do you have in mass media – such as newspapers, TV, radio – when it comes to reporting the news fully, accurately and fairly?’ 1 = no trust; 4 = a great deal of trust; Desires more cultural preservation: ‘More should be done to preserve cultural traditions.’ 1 = strongly disagree, 5 = strongly agree. Martin et al. (2016) found that a desire for cultural preservation positively predicted willingness to censor media in several Arab countries. Supports Internet censorship: index of two items, split-half α = 0.66, ‘People should be free to criticize governments on the Internet,’ and ‘It’s OK for people to express their ideas on the Internet even if they are unpopular.’ For each, 1 = strongly disagree; 5 = strongly agree. Supports entertainment censorship: four-item index, Cronbach’s α = 0.87, ‘Entertainment content in the [Arab] region should be more tightly regulated for romantic content,’ ‘Entertainment content in the [Arab] region should be more tightly regulated for violent content,’ ‘Films or other programs should be banned if some people find them offensive,’ and ‘It is appropriate to delete scenes some people may find offensive.’ For each, 1 = strongly disagree, 5 = strongly agree.

Demographics

Age: ‘What is your age?’ Ratio-level variable; Gender: 0 = male; 1 = female; Education: ‘What is the highest level of schooling you completed?’ 1 = no formal education; 10 = master’s degree and/or terminal degree; Conservatism: ‘Compared to most nationals in this country, how would you describe yourself?’: 1 = Culturally very progressive; 5 = Culturally very conservative; Religiosity: ‘How often do you attend religious services?’ 1 = never; 9 = once a day or more; Income: ‘What is your total monthly household income?’ Lebanon: 1 = less than 750,000 Lebanese lira; 12 = more than 10 million lira. Tunisia: 1 = less than 100 Tunisia dinars; 12 = more than 5000 dinars. Saudi Arabia: 1 = less than 3000 Saudi riyals; 12 = more than 45,000 Saudi riyals.

Analyses

SPSS 26 was used for analyses. Multiple linear regression models use media use measures, attitudinal variables, and demographics to predict FNB. Regressions were run separately for each country, to allow easy comparisons of predictors across nations. Three multiple regressions were also run separately for the three items in the dependent variable index in Saudi Arabia, where Cronbach’s α for the measure was comparatively low (0.63), and the results of those models are in Appendix 2. Results of these disaggregated models for Saudi Arabia did not differ dramatically either from one another or from the aggregated model. For the main regression models in this study, cases were excluded pairwise. Multicollinearity tolerance was set at 0.20. None of the possible pairs of predictor variables violated that mark. In an additional multicollinearity precheck, we ran correlations among all the media use and attitudinal predictors. Some of the predictors were significantly correlated with others, but none of the associations was greater than r = 0.49.

Results

This study examined predictors of support for censorship of fake news online (FNB) among nationals in Lebanon, Saudi Arabia, and Tunisia.

The RQ asked about the relationship between perceived FNE online and FNB, after controlling for media use, attitudinal measures, and demographics. FNE was strongly and positively correlated with FNB in all countries. In fact, FNE was the single strongest predictor of FNB in each country. Respondents who believe that they encounter a lot of fake news online were much more eager to support censoring fake news than were people who said they infrequently see fake news. Table 1 shows standardized betas from the regression models.

Table 1.

Correlates of fake news burden (standardized βs and p-values).

	KSA	Tun.	Leb.
Media use measures (standardized β; p-value)
Checks news online	−0.010; 0.89	0.032; 0.63	−0.031; 0.42
Print media use	−0.133; 0.052	0.034; 0.64	−0.055; 0.15
Time spends playing video games	0.132; 0.07	−0.053; 0.395	0.069; 0.077
Uses Facebook daily or more	−0.138; 0.038	−0.038; 0.52	−0.073; 0.038
Uses Twitter	0.06; 0.32	0.048; 0.41	0.016; 0.68
Uses VPN	−0.196; 0.002*	0.013; 0.83	−0.086; 0.016*
Attitudinal measures (standardized β; p-value)
Perceived FNE	0.311; <0.001*	0.45; <0.001*	0.26; <0.001*
Wants>Internet regulation	−0.092; 0.154	−0.08; 0.284	0.002; 0.96
Trusts news media	0.209; 0.003*	0.133; 0.039	0.06; 0.12
Desires >cultural preservation	0.074; 0.267	−0.09; 0.26	0.12; 0.001*
Supports Internet censorship	0.003; 0.98	0.09; 0.21	0.20; <0.001*
Supports entertainment censorship	0.115; 0.11	0.217; 0.008*	0.01; 0.87
Demographic measures (standardized β; p-value)
Age	−0.072; 0.19	0.012; 0.86	−0.026; 0.53
Gender	0.044; 0.47	0.036; 0.54	0.02; 0.42
Education	0.033; 0.61	−0.079; 0.24	0.03; 0.46
Conservatism	0.087; 0.19	−0.01; 0.89	−0.023; 0.593
Religiosity	0.085; 0.195	0.07; 0.26	−0.018; 0.64
Income	−0.016; 0.826	0.054; 0.39	−0.014; 0.74
Maximum N; Minimum N	745; 263	1144; 396	991; 777
Adjusted R² (%)	42.8	26.5	13

FNE: fake news exposure. Boldface values signify <.05; *<.01; **<.001.

The models explained substantial amounts of variance in FNB in the respective countries: 43% in Saudi Arabia, 27% in Tunisia, and 13% in Lebanon. These Rs-squared are even more notable, given that not a single demographic variable in any of the countries significantly predicted FNB; predicted variance was mostly explained by media use and media-relevant attitudes. In Appendix 2, the reader will find three additional regression models conducted with the data from Saudi Arabia (as Cronbach’s α for FNB in Saudi Arabia = 0.63), for each of three items comprising FNB. Perceived FNE was a strong, positive predictor of each of these three items.

Hypothesis 1 said support for censorship would positively predict FNB. Hypothesis 1 was only partially supported. Support for Internet censorship positively predicted FNB in Lebanon, and support for entertainment censorship predicted higher FNB scores in Tunisia, but neither of the two censorship measures predicted FNB scores in Saudi Arabia.

Hypothesis 2 said the association between FNE and FNB would be stronger (positively) in Saudi Arabia than in Tunisia and in Lebanon. Hypothesis 2 was partially supported. The association between FNE and FNB was positive in all countries and was strongest in Tunisia, followed by Saudi Arabia and Lebanon. Standardized betas are, by definition, standardized and can be compared across the countries, and Table 1 shows that the association between FNE and FNB in Saudi Arabia was stronger than in Lebanon but not Tunisia.

Beyond results of the specific research question and hypotheses, there are additional noteworthy findings. Differing from prior research, demographic variables like age, female gender, conservatism, and religiosity did not positively predict willingness to censor fake news online. This is interesting, as getting news from mainstream versus nonmainstream outlets differs by many of these factors, notably age (Tsfati et al., 2011). Not one demographic variable in any of the countries was associated, either positively or negatively, with FNB.

Two media use variables predicted FNB scores (negatively): Heavy Facebook users in Lebanon and Saudi Arabia reported less willingness to censor fake news than non-heavy Facebook users, and VPN users in the same two countries also reported lower FNB scores than non-VPN users. Trust in news media positively predicted willingness to censor fake news in Saudi Arabia and Tunisia.

Regarding possible Type I errors, 18 total potential correlates of FNB were scrutinized in each country, a total of 54 associations. As the largest ‘significant’ p-value – that is, of those less than 0.05 – is 0.039, we should expect that 3.9% of the 54 standardized betas, or two of them, are significant due to randomness rather than meaningful covariance. For variables that are predictive of FNB in more than one country, however, the Type I error rate is likely lower than 3.9%. We should be cautious about variables predictive of FNB in only one of the countries, particularly if the obtained p-value approaches 0.039. While willingness to censor both Internet speech and entertainment was significantly associated with FNB in just one country, both betas were positive, and they were each the second strongest predictor in their respective country, with the p-values each <0.01.

Hypothesis 3 said FNB scores would be lower for Tunisians and Lebanese than Saudis. Hypothesis 3 was supported. One-way analysis of variance (ANOVA) compared FNB in the three countries, and significant differences were observed, F[2, 2,462] = 18.23, p < 0.001. Means for FNB and post hoc comparisons of countries are in Appendix 1. Saudis had higher FNB scores than both Tunisians and Lebanese. Means of FNB were also compared within countries. One-way ANOVA and post hoc tests found that means for governments were > members of the public in each KSA (p = 0.02), Lebanon (p = 0.01), and Tunisia (p < 0.001), and means for governments and tech companies also differed in Lebanon (p < 0.01) and Tunisia (p < 0.001). Recall that three additional regression models were run in Saudi Arabia for each of the three items in the FNB measure, as α for FNB in that country was lower than in Tunisia or Lebanon. Results from those three models (Appendix 2) did not give cause to reconsider the findings from the RQ and hypotheses.

Discussion

This study examined predictors of willingness to censor fake news online in Lebanon, Saudi Arabia, and Tunisia, while utilizing the CAH, the TPMI, and research on public willingness to censor as organizing frameworks. The regression models of media use and attitudinal predictors explained substantial amounts of variance in FNB in the countries: 43% in Saudi Arabia, 27% in Tunisia, and 13% in Lebanon.

There are four primary takeaways from this study: (1) People who said they often see fake news online were much more eager to censor fake news than people who said they rarely or never encounter fake news, aligning with TPMI but countering both the CAH and prior research that found people who are exposed to media content are less interested in censoring it. This likely reflects that exposure to fake news, unlike exposure to, say, porn or violent video games, is often not voluntary; (2) Trust in news media was positively correlated with FNB in two countries, so individuals who have confidence in news media want to see actions against fake information intruding in their online environments; (3) Demographics, associated with public willingness to censor in some prior work (Lambe) – like gender, religiosity, and education – did not predict FNB in any country, aligning with some prior research in Arab countries (see Martin et al., 2016); Heavy Facebook users did not want to censor fake online news as much as casual users or non-users of Facebook did, in two of the three countries.

Predictors previously found correlated with support for censorship might not explain eagerness to censor certain content, such as fake news, either in Arab countries, or in new media environments, where menacing content, like the work of online trolls and bots, doxing (publishing, online, personal details about another person), and revenge porn (publishing online sensitive images of an intimate ex-partner), differs from that studied in prior censorship research. In this study, people reporting they frequently encounter fake news online seem to want someone – anyone – to curb it, regardless of who does so. This association might be partly related to the third-person phenomenon. Corbu et al. (2020) found that respondents rated their own ability to detect fake news as better than that of others (see also Ştefăniţă, 2018), and the third-person phenomenon has been found to be associated with willingness to censor.

Respondents who trust news media reported higher FNB, at least in Saudi Arabia and Tunisia. This counters the CAH, which predicts that perceptions of media bias – that news media are hostile – positively predict motivation to counteract perceived negative media effects. The positive association between trust in news media and concern about fake news does, however, align with some prior research, such as Wasserman and Madrid-Morales (2019), who found that trust was positively associated with perceptions of online fake news among respondents in South Africa. Martin and Hassan (2020), though, did not find this relationship in a study of trust in news media and perceptions of FNE in five countries from the Middle East and North Africa. The findings here may lend weight to the argument that fake news is eroding credibility of established, conscientious news organizations (Rusbridger, 2018), as people who trust established media more strongly support regulating rogue media.

Possibly, people who trust news media are supportive of corrective action against fake online news because they perceive audience attention gravitating toward content they deem harmful, and that erodes the standing of news outlets they trust, which could be evidence of the TPMI at work. Also, people who express confidence in news media may be more eager to see action against fake information because they see it eroding the institution of the press, which they value. Paradoxically, while fake news mostly comes from entities other than mainstream news outlets, such outlets have nonetheless suffered hits to their credibility due to sweeping allegations made by certain public figures that they spread fake news.

Support for censorship – both online speech and entertainment – was mostly not predictive of support for corrective action against online fake news, with the exception that each variable positively predicted FNB in one country. Some respondents who do not support censoring, say, sexual or violent entertainment media nonetheless support corrective action against fake news. In a sense, the desire to censor online political speech or sexual content is the desire to censor content that is, to some people, too real and too graphic, while fake news is crafted to deceive. The qualitative differences in the dependent variable in the current study compared to previous research on willingness to censor, then, may illuminate findings here that misalign with prior work on corrective action and presumed media influence.

Results suggest that support for censoring fake news is predicted more by attitudinal variables than by media use (recall that FNE is still a measure of perceived exposure, and so the benefits and drawbacks of self-reported data apply, just as they apply with the other variables measured in this study reported by respondents). However, a few media use measures explained FNB variance.

Heavy Facebook use negatively predicted FNB in two of the three countries: Saudi Arabia and Lebanon. Given the well-documented volume of fake news disseminated on Facebook (Lazer et al., 2018), this may be cause for concern; people who are exposed to fake news via Facebook are less likely to see fake news as a problem. For Facebook itself, this finding seems like a good thing: Heavy Facebook users may not be interested in holding the tech giant to account for spreading false information – motivation to bring evermore users into Facebook’s fold.

Respondents in the three countries seem to similarly favor each government, corporate, and laypeople’s censoring fake news online, as shown both in the data in Figure 1, in Appendix 1, and in the Cronbach’s αs for the FNB index, though respondents were slightly more likely to agree that governments should block fake news than to say laypeople and tech firms should. In some prior work, scholars distinguished between governments censoring media, which is usually viewed as antisocial, and censorship by companies or other private-sector actors (Cook and Heilmann, 2013), which is sometimes deemed the pro-social result of corporate or organizational social responsibility. The means and Cronbach’s αs for the FNB items, though, suggest people tend to invite censorship of fake online news by governments/corporations/laypeople, or they like oppose it collectively, without a large difference in attitudes on who specifically does the censoring.

FNB was higher among Saudis than among Tunisians or Lebanese. In Saudi Arabia, legitimate news is routinely censored by the government (‘Saudi Arabia bans Al Jazeera in hotels,’ 2017), and fake news is sometimes disseminated by the regime in Riyadh, especially news stories about Iran or Qatar (Ulrichsen, 2020), or fake news about specific abuses like the murder of Jamal Khashoggi. Tunisia and Lebanon have much freer press systems than Saudi Arabia, and their Internet environs are less censored by their governments, and thus citizens in Tunisia and Lebanon may feel less like sitting targets for fake news than Saudis, whose government is, itself, occasionally a source of fake news.

The Saudi-led blockade of Qatar was predicated on fake news that Saudi and the UAE arranged to have posted on a Qatar government news site and represents the first major geopolitical conflict ignited by our contemporary definition of fake news (Ulrichsen). It’s possible some of the heightened eagerness to censor online fake news among Saudis is partly a response to dubious news from their country’s government.

Limitations and subsequent research

The inability to choose which questions are asked is a limitation of any secondary survey analysis. Future research can refine some of the independent variables used here, for example, and ask questions explicitly about presumed influence of media and the third-person phenomenon. Also, as with any surveys, data studied here are self-reported. The variable FNE may be limited, then, as respondents provided their self-reported assessment of FNE, and some respondents may not be able to identify fake news when they see it. While we define fake news with clarity in the current article, respondents of the survey were not presented with that same definition, which leaves open the possibility that respondents define fake news differently. However, given that the data collection was in-person, respondents were able to ask interviewers about what fabricated news, or fake news, means, and interviewers had been trained to explain that fake news meant outright fabrication or manipulation created with the intent to mislead.

And of course, the two-item index on exposure to inaccurate and fabricated news does not actually use the words ‘fake news’, for the very reason that we define it as false and fabricated news, and this is what was communicated to respondents. Also, the FNE items mention fabricated ‘political’ news, so it is implicit that the material is fabricated for political gain. Regarding the FNB questions, which did use the words ‘fake news’, respondents were not presented with a value judgment suggesting that fake news is harmful, benign, or beneficial. This would have left respondents feeling free to say no blocking of fake news is needed.

While the FNB index was reliable in three countries – not bad given reliability problems some scales developed in one country often exhibit elsewhere (see Claes et al., 2005) – it should not be assumed that the FNB scale is internally consistent in any or all other countries. The original Pew Research Center questions used to create the FNB index were constructed as separate questions, not a compound scale. Happily, we found acceptable reliability in the measures in the three countries. While Cronbach’s α in Saudi Arabia was modest, the three additional regression models conducted using the data from Saudi Arabia (Appendix 2) differed little from each other or from the aggregate regression model analysing Saudi data. The FNB scale could be expanded to five-point Likert items in future work – it is currently a four-point index – which may edge the reliability alphas higher.

Quantitative studies can only teach us so much about news consumers’ attitudes about fake news, and even though numerical studies afford large sample sizes, some of the nuances in attitudes about survey subject matter are invariably lost. Future studies should include more qualitative examinations of attitudes surrounding fake news, via in-depth interviews, focus groups, and other qualitative means.

Future research on corrective action against fake news might also focus on attitudinal variables – like perceptions of exposure to fake news, trust in news media, and attitudes about other forms of censorship – given that attitudinal measures accounted for most of the variance in FNB explained in this study. Future work might also consider additional social media and privacy-related behaviors as explanatory variables, given that in the current study VPN use and heavy Facebook use were the only significant media use predictors of FNB. And given findings from previous research on the TPMI, which holds that the perception that media are harmful increases willingness to censor, future research on eagerness to censor fake news should ask respondents if and how harmful they believe fake news is.

Footnotes

Appendix 1

Appendix

Descriptives for FNB and tested predictor variables

	KSA	Tun.	Leb.
FNB (M (SD))	10.11 (1.83)	9.64 (2.43)	9.46 (2.26)
Individual FNB items:
Members of the public	3.35 (0.81)	3.07 (1.08)	3.09 (0.98)
Governments	3.41 (0.80)	3.38 (0.96)	3.29 (0.97)
Tech companies	3.39 (0.79)	3.07 (1.08)	3.08 (0.92)
Post hoc comparisonsKSA – Tunisia, p < 0.001;KSA – Lebanon, p < 0.001;Tunisia – Lebanon, p = .086
Media use measures
-Checks news online (M (SD))	3.79 (1.62)	3.58 (1.99)	3.55 (1.85)
-Print media use (M (SD))	9.44 (2.98)	7.79 (4.34)	7.45 (3.83)
-Time spends playing video games (M (SD))	6.17 (8.28)	2.38 (7.57)	5.45 (12.19)
-Uses Facebook daily or more (%)	26.2	30.3	41.1
-Uses Twitter (%)	51.4	1.4	17.2
-Uses VPN (%)	49.8	7.0	17.0
Attitudinal measures
-Perceived fake news exposure online (M (SD))	6.59 (1.21)	5.99 (1.69)	6.13 (1.21)
Encounter items not fully accurate	3.28 (0.713)	2.83 (1.01)	3.14 (0.655)
Encounter items appear made up	3.28 (0.734)	3.09 (0.940)	2.99 (0.719)
-Wants >Internet regulation (M (SD))	3.82 (0.92)	3.49 (1.20)	3.88 (1.09)
-Trusts mass media (M (SD))	3.05 (0.88)	2.29 (0.87)	2.58 (0.89)
-Desires > cultural preservation (M (SD))	3.96 (0.93)	3.39 (1.16)	3.61 (1.09)
-Supports Internet censorship (M (SD))	4.57 (1.94)	5.44 (2.09)	4.89 (2.06)
-Supports entertainment censorship (M (SD))	16.61 (13.25)	13.53 (4.22)	14.86 (4.06)
Demographic measures
Age (M (SD))	33.19 (11.57)	36.74 (13.01)	35.88 (13.41)
Gender (% Female)	50	50	49.9
Education (M (SD))	7.52 (1.49)	6.44 (2.29)	6.79 (2.09)
Conservatism (M (SD))	3.82 (1.03)	3.46 (1.09)	3.41 (1.11)
Religiosity (MID)	6.00	3.00	4.0
Income (MID)	5.00	6.00	5.0
Demographics from censuses in each country
Age (median)	31	32.7	33.7
Gender (% Female)	45	50.25	49.7
Education^a	17	15	11
Income (median monthly household in US$)	4000	950	1200

FNB: fake news burden; M (SD): mean (standard deviation).

^a Average number of years of primary through tertiary education (CIA World Factbook, 2019).

Appendix 2

Appendix

Correlates in Saudi Arabia for each separate FNB item (standardized β; p values).

	Public	Government	Tech companies
Media use measures (standardized β; significance)
Checks news online	−.042; .89	.034; .67	.006; .94
Print media use	−.103; .21	−0.08; .29	−.14; .052
Time spends playing video games	−.148; .09	.06; .48	.11; .14
Uses Facebook daily or more	.101; .20	−.185; .017	−.23; .001
Uses Twitter	.001; .99	.10; .16	.01; .85
Uses VPN	−.084; .27	−.22; .003	−.11; .105
Attitudinal measures (standardized β; significance)
Perceived FNE	.264; .001	.22; .003	.244; <.001
Wants > Internet regulation	−.048; .54	−.02; .84	−.11; .09
Trusts news media	.011; .90	.14; .09	.34; <.001
Desires > cultural preservation	.038; .63	.08; .30	.06; .42
Supports Internet censorship	−.048; .53	.075; .31	−.001; .99
Supports entertainment censorship	.079; .36	.074; .37	.10; .20
Demographic measures (standardized β; significance)
Age	−.09; .18	−.03; .64	−.069; .24
Gender (female)	.02; .79	−.04; .58	.11; .10
Education	.02; .81	.01; .95	−.01; .85
Conservatism	.03; .75	.06; .45	.145; .036
Religiosity	.144; .07	−.04; .59	.09; .18
Income	.02; .80	.03; .74	−.13; .09
Maximum N; minimum N	745; 263	745; 263	745; 263
Adjusted R ²	17.8%	23.5%	37.6%

FNE: fake news exposure; FNB: fake news burden.

ORCID iD

Justin D Martin

Author biographies

Justin D Martin, PhD, is an associate professor in journalism and strategic communication program at Northwestern University in Qatar.

Fouad Hassan is a social media specialist at Qatar Foundation.

References

10 Most Censored Countries (2015) Committee to Protect Journalists (CPJ) Available at: https://cpj.org/reports/2019/09/10-most-censored-eritrea-north-korea-turkmenistan-journalist.php (31 October 2019).

Abadi

(2017) The words that defined 2017. Business Insider, 12 December Available at: https://www.businessinsider.com/word-of-the-year-2017-12 (accessed 25 April 2021).

Allsop

(2020) A mammoth explosion, tears, and resilient journalism in Beirut. Columbia Journalism Review Available at: https://www.cjr.org/analysis/beirut_explosion_media.php (accessed 25 April 2021).

Amnesty International (2020, May) Tunisia: End prosecution of Emna Chargui, and investigate alarming death and rape threats. Available at: https://www.amnesty.org/en/latest/news/2020/05/tunisia-end-prosecution-of-emna-chargui-and-investigate-alarming-death-and-rape-threats/ (accessed 25 April 2021).

Barnidge

Rojas

(2014) Hostile media perceptions, presumed media influence, and political talk: Expanding the corrective action hypothesis. International Journal of Public Opinion Research 26(2): 135–156.

Benkler

Faris

Roberts

(2018) Network Propaganda: Manipulation, Disinformation, and Radicalization in American Politics. Oxford: Oxford University press.

Buckley

Chaabi

Ouarda

(2013) Assessment of media development in Tunisia. Available at: file:///Users/jdm711/Downloads/222701eng.pdf (accessed 25 April 2021).

CIA World Factbook Middle East (2019) Available at: https://www.cia.gov/library/publications/the-world-factbook/geos/sa.html (accessed 25 April 2021).

Claes

Beheydt

Lemmens

(2005) Unidimensionality of abbreviated proactive personality scales across cultures. Applied Psychology 54(4): 476–489.

10.

Cohen

Weimann

(2008) Who’s afraid of reality shows? Exploring the effects of perceived influence of reality shows and the concern over their social effects on willingness to censor. Communication Research 35(3): 382–397.

11.

Condis

(2018) Gaming Masculinity: Trolls, Fake Geeks, and the Gendered Battle for Online Culture. Iowa City, IA: University of Iowa Press.

12.

Cook

Heilmann

(2013) Two types of self-censorship: Public and private. Political Studies 61(1): 178–196.

13.

Corbu

Oprea

Negrea-Busuioc

, et al. (2020) ‘They can’t fool me, but they can fool the others!’ Third person effect and fake news detection. European Journal of Communication 35: 165–180.

14.

Cut It Out: Censorship in Lebanon (2012, July 12) Retrieved from Carnegie endowment for international peace. Available at: https://carnegieendowment.org/sada/48831 (accessed 25 April 2021).

15.

Davison

(1983) The third-person effect in communication. Public Opinion Quarterly 47(1): 1–15.

16.

Dennis

Martin

Hassan

(2018) Media use in the Middle East, 2018: A Seven-nation Survey. Doha: Northwestern University in Qatar. Available at: www.mideastmedia.org/survey/2018 (accessed 25 April 2021).

17.

Dennis

Martin

Wood

, et al. (2016) Media Use in the Middle East, 2016: A Six-Nation Survey. Doha: Northwestern University in Qatar. Available at: www.mideastmedia.org/survey/2016 (accessed 25 April 2021).

18.

Fisher

Cook

Shirkey

(1994) Correlates of support for censorship of sexual, sexually violent, and violent media. Journal of Sex Research 31(3): 229–240.

19.

Funke

Flamini

(2020) A guide to anti-misinformation actions around the world. Poynter. Available at: https://www.poynter.org/ifcn/anti-misinformation-actions/ (accessed 25 April 2021).

20.

Gavlak

(2021, 25 January) Lebanon must enact drastic reforms to survive, experts say. Voice of America. Available at: https://www.voanews.com/middle-east/lebanon-must-enact-drastic-reforms-survive-economists-say (accessed 25 April 2021).

21.

Hong

(2020) Presumed effects of “fake news” on the global warming discussion in a cross-cultural context. Sustainability 12(5): 2123–2133.

22.

Human Rights Watch (2021, 5 February) Police use violent tactics to quash protests. Available at: https://www.hrw.org/news/2021/02/05/tunisia-police-use-violent-tactics-quash-protests (accessed 25 April 2021).

23.

Internet censorship listed: How does each country compare? (2012, 16 April) The Guardian. Available at: http://www.theguardian.com/technology/datablog/2012/apr/16/internet-censorship-country-list (accessed 25 April 2021).

24.

Jang

Kim

(2018) Third person effects of fake news: Fake news regulation and media literacy interventions. Computers in Human Behavior 80: 295–302.

25.

Lamb

Rashwan

(2020) Saudi court issues final verdicts in Khashoggi killing. The New York Times, 7 September. Available at: https://www.nytimes.com/2020/09/07/world/middleeast/jamal-khashoggi-sentence.html (accessed 25 April 2021).

26.

Lambe

(2002) Dimensions of censorship: Reconceptualizing public willingness to censor. Communication Law & Policy 7(2): 187–235.

27.

Lambe

(2004) Who wants to censor pornography and hate speech? Mass Communication & Society 7(3): 279–299.

28.

Lazer

DMJ

Baum

Benkler

, et al. (2018) The science of fake news. Science 359 (6380): 1094–1096.

29.

Paddon

(2001) Third-person effect, gender differences, pornography exposure and support for restriction of pornography. Asian Journal of Communication 11(1): 120–142.

30.

Martin

Hassan

(2020) News media credibility ratings and perceptions of online fake news exposure in five Arab countries. Journalism Studies 21(16): 2215–2233.

31.

Martin

Martins

Wood

(2016) Desire for cultural preservation as a predictor of support for entertainment media censorship in Saudi Arabia, Qatar, and the United Arab Emirates. International Journal of Communication 10: 23.

32.

Masri

(2019) Tunisia: An Arab anomaly. Reprint edition. New York, NY: Columbia University Press.

33.

McLeod

Detenber

Eveland

Jr (2001) Behind the third-person effect: Differentiating perceptual processes for self and other. Journal of Communication 51(4): 678–695.

34.

Mitchell

Barthel

Holcomb

(2016) Many Americans believe fake news is sowing confusion. Pew Research Center, 15 December. Available at: https://www.journalism.org/2016/12/15/many-americans-believe-fake-news-is-sowingconfusion/ (accessed 25 April 2021).

35.

Mitchell

Gottfried

Fedeli

, et al. (2019) Many Americans say made-up news is a critical problem that needs to be fixed. Pew Research Center, 5 June. Available at: https://www.journalism.org/2019/06/05/many-americans-say-made-up-news-is-a-critical-problem-that-needs-to-be-fixed/ (accessed 25 April 2021).

36.

Newman

Fletcher

Kalogeropoulos

, et al. (2018) Reuters Institute Digital News Report 2018. Available at: http://media.digitalnewsreport.org/wp-content/uploads/2018/06/digital-news-report-2018.pdf (accessed 25 April 2021).

37.

Nielsen

Graves

(2017) News you don’t believe: Audience perspectives on fake news. Reuters Institute for the Study of Journalism Available at: https://ora.ox.ac.uk/objects/uuid:6eff4d14-bc72-404d-b78a-4c2573459ab8 (accessed 25 April 2021).

38.

Nisbet

Kamenchuk

Dal

(2017) A psychological firewall? Risk perceptions and public support for online censorship in Russia. Social Science Quarterly 98(3): 958–975.

39.

Paek

Lambe

McLeod

(2008) Antecedents to support for content restrictions. Journalism & Mass Communication Quarterly 85(2): 273–290.

40.

Pennycook

Cannon

Rand

(2018) Prior exposure increases perceived accuracy of fake news. Journal of Experimental Psychology: General 147(12): 1865–1880.

41.

Perloff

(2002) The third-person effect. In: Bryant

Zillmann

(eds) Media Effects: Advances in Theory and Research (2nd ed). Mahwah, NJ: Lawrence Erlbaum.

42.

Ram

(2018) Inside Tunisia’s Shams Rad – the Arab world’s only ‘gay radio station’. BBC, 18 June. Available at: https://www.bbc.com/news/world-africa-44137901 (accessed 25 April 2021).

43.

Reporters Without Borders (2019) 2019 World Press Freedom Index. Available at: https://rsf.org/en/ranking (accessed 25 April 2021).

44.

Rojas

(2010) “Corrective” actions in the public sphere: How perceptions of media and media effects shape political behaviors. International Journal of Public Opinion Research 22(3): 343–363.

45.

Rojas

Shah

Faber

(1996) For the good of others: Censorship and the third person effect. International Journal of Public Opinion Research 8(2): 163–186.

46.

Rosen

(2002) High school confidential: In their efforts to suppress negative news, administrators are increasingly apt these days to censor student newspapers. American Journalism Review 24(5): 56–61.

47.

Rugh

(2004) Arab Mass Media: Newspapers, Radio, and Television in Arab Politics. Westport, CT: Greenwood Publishing Group.

48.

Rusbridger

(2018) Breaking News: The Remaking of Journalism and Why It Matters Now. New York, NY: Farrar, Straus and Giroux.

49.

Salwen

(1998) Perceptions of media influence and support for censorship: The third-person effect in the 1996 presidential election. Communication Research 25(3): 259–285.

50.

Saudi Arabia bans Al Jazeera in hotels (2017, 9 June) Al Jazeera. Available at: https://www.aljazeera.com/news/2017/06/saudi-arabia-bans-al-jazeera-channels-hotels-170609141041079.html (accessed 25 April 2021).

51.

Schoenbach

Wood

Saeed

(2016) Media industries in the Middle East, 2016. Published by Northwestern University in Qatar and Doha Film Institute Available at: http://www.mideastmedia.org/industry/2016/ (accessed 25 April 2021).

52.

Ştefăniţă

(2018) Fake news and the third-person effect: They are more influenced than me and you. Journal of Media Research 11(3): 5–23.

53.

Swift

(2016) Americans’ trust in mass media sinks to new low. Gallup. Available at: https://news.gallup.com/poll/195542/americans-trust-mass-media-sinks-new-low.aspx (accessed 25 April 2021).

54.

Taber

(2017) The use of Cronbach’s alpha when developing and reporting research instruments in science education. Research in Science Education 48: 1273–1296.

55.

Tal-Or

Cohen

Tsfati

, et al. (2010) Testing causal direction in the influence of presumed media influence. Communication Research 37(6): 801–824.

56.

Tandoc

Jr Lim

Ling

(2018) Defining “fake news”: A typology of scholarly definitions. Digital Journalism 6(2): 137–153.

57.

Tompros

Crudo

Pfeiffer

, et al. (2017) The constitutionality of criminalizing false speech made on social networking sites in a post-Alvarez, social media-obsessed world. Harvard Journal of Law & Technology 31: 65–110.

58.

Tsfati

Cappella

(2003) Do people watch what they do not trust? Exploring the association between news media skepticism and exposure. Communication Research 30(5): 504–529.

59.

Tsfati

Cohen

Gunther

(2011) The influence of presumed media influence on news about science and scientists. Science Communication 33(2): 143–166.

60.

Ulrichsen

(2020) Qatar and the Gulf Crisis. London: Hurst Publishers.

61.

Wagner

(1982) Simpson’s paradox in real life. The American Statistician 36(1): 46–48.

62.

Walters

(2016) Censorship as a Populist Project: The Politics of Managing Culture in Egypt (Thesis, Georgetown University). Available at: https://repository.library.georgetown.edu/handle/10822/1041837 (accessed 25 April 2021).

63.

Wasserman

Madrid-Morales

(2019) An exploratory study of “fake news” and media trust in Kenya, Nigeria and South Africa. African Journalism Studies 40(1): 107–123.

64.

Wendling

(2018) The (almost) complete history of ‘fake news’. BBC, 22 January. Available at: https://www.bbc.com/news/blogs-trending-42724320 (accessed 25 April 2021).

65.

World Report 2019: Rights trends in Saudi Arabia (2018, 17 December) Human Rights Watch. Available at: https://www.hrw.org/world-report/2019/country-chapters/saudi-arabia (21 October 2019).