How Autocracies Disrupt Unsanctioned Information Flows: The Role of State Power and Social Capital in North Korea

ABSTRACT

What determines the flow of unsanctioned information in authoritarian regimes? To answer this question, we explore the use and circulation of illicit foreign information among citizens in North Korea. This paper focuses on how the structure of social relations among citizens determines where and with whom people acquire and disseminate politically unsanctioned and potentially illegal content offline. Based on findings from a choice‐based conjoint administered to 313 former residents of North Korea, we show that social capital and trust-related effects associated with the place where information is acquired and the person who provides it determine information flows.

Introduction

Horizontal information flows between citizens are crucial to the spread of radical or authority-challenging ideas (Kim and Pfaff 2012). In authoritarian regimes, digital surveillance programs are increasingly used to curtail the spread of horizontal connections and information flows, especially between social actors critical of the government (Xu 2021). In the low-level war of attribution between the state and the citizenry for control over information, the state infiltrates society and makes selective use of repression to ensure that information does not lead to organized opposition demanding substantive political change (Hassan, Mattingly, and Nugent 2022).

Under such conditions, what are the determinants of illicit information use and dissemination among citizens? Information environments under autocratic conditions have been widely studied in the past decade. Research has focused on digital media and the internet (Reuter and Szakonyi 2015; Ruijgrok 2021), examining how information spreads (Wolfsfeld, Segev, and Sheafer 2013; Castells 2015; Grönlund and Wakabi 2015), the use of censorship to control the flow of digital content (Chen and Yang 2019), and implications for regime durability (Kern 2011; Kern and Hainmueller 2017; Huang and Yeh 2019). What remains understudied is how citizens living in authoritarian countries obtain illicit information offline.

Drawing on insights from social capital theory, which have yet to be fully leveraged in the comparative literature on authoritarianism, we fill a gap in the existing literature by examining how unauthorized content circulates in North Korea.

Using a choice-based conjoint, we ask former residents of North Korea to choose among pairs of hypothetical information profiles that they would have used themselves and then whether they would have shared that information with others. Focusing on the information provider and the location where the information is acquired, this empirical strategy allows us to infer the relationally patterned and location-specific aspects of social trust in an authoritarian context and consider the broader impact these factors have on information use and sharing.

Using North Korea as a case allows us to examine the determinants of illicit information flows in a highly authoritarian context absent widespread internet use. The limitations of the internet as a tool by which to organize and mobilize in authoritarian settings are well documented, even in states that are less effective in penetrating civil society (Oates 2013). By focusing on a society where the internet is not available for ordinary citizens, we offer a theoretical argument and empirical confirmation of how and why authoritarian states disrupt, co-opt, or entirely block the flow of information they deem threatening.

We find support for social capital and trust-related effects associated with the place where content is provided and the person who provides it. North Korea is believed to be a low-trust society,¹ but our findings show that the distribution of social trust is location- and relation-specific. Content acquisition, use/consumption, and distribution in North Korea are mainly limited to places like the home, which are beyond the structural constraints created by the state, and between close personal ties associated with particular trust (i.e., friends and family). Notably, markets do not support information flows, which runs contrary to expectations that marketplaces have liberalizing effects in North Korea (Kim and Kim 2019). In contested spaces like these, and other outdoor places, there is limited potential for general trust between strangers to facilitate content distribution.

Our findings demonstrate that state domination of social relations in North Korea disrupts weak ties among citizens and limits one’s trust radius to family and friendship networks. Trust, then, remains low among citizens, as social capital theory predicts (Uslaner 1999, 123; Rothstein and Stolle 2008; Freitag and Bühlmann 2009), and the dissemination and flow of information are either restricted or stymied altogether. This research contributes to the literature examining the foundations of political control in authoritarian systems, showing how social trust and relational social capital can either facilitate or constrain politically dangerous content acquisition and distribution in a highly authoritarian system. In doing so, we underscore the important link between the control of civil society through co-optation, infiltration, and selective repression and its effects on authoritarian durability (Hassan, Mattingly, and Nugent 2022). We thus add to the discourse on the “dark side of social capital” (Alcorta et al. 2020), demonstrating how social capital can promote political control in the area of illicit information flows.

The rest of the article is structured as follows. We begin with a brief discussion of the literature on information flows and social control in authoritarian regimes before turning to social capital theory, from which we draw our primary variables of interest and empirical expectations. After the literature review, we examine North Korea as a case and explain its relevance and appropriateness. Next, we connect our theoretically informed expectations to our experimental design and discuss our data and methodology. We then present our findings. The paper concludes with a summary of the research and a discussion of its contribution, broader significance, and implications.

Why Do Autocrats Suppress Information Flows?

Authoritarian information environments have been widely studied. It is not our objective to provide an exhaustive overview of the literature. Instead, we aim to show how social capital theory suggests how and why authoritarian regimes seek to control information flows. We then propose a set of theoretical expectations concerning how social trust and relationship- and location-specific factors affect the choice to acquire information and further disseminate it.

The supply side of information in authoritarian states and the role of the internet in raising information availability is a major strand of the existing literature. Some of the main works that consider the issue of online information dissemination in authoritarian regimes (Wolfsfeld, Segev, and Sheafer 2013; Castells 2015; Grönlund and Wakabi 2015) focus on the role of the internet as a medium through which anti-regime information is disseminated and by which social connections are formed and nurtured (Reuter and Szakonyi 2015; Ruijgrok 2021). There is also an extensive literature on internet censorship (e.g., King, Pan, and Roberts 2013, 2017), with research focusing on authoritarian states. It finds that censorship may dampen demand for politically sensitive information and temper people’s desire to share such information when it is acquired (Chen and Yang 2019).

The research on online information flows under authoritarian regimes leaves little doubt that maintaining control of information, foreign media availability, and digital tools is important to political rule, but the literature indicates considerable disagreement over whether the internet, information access, and digital tools empower authoritarian rulers or the citizenry (Pearce 2013; Rød and Weidmann 2015; Repnikova and Fang 2018; Weidmann and Rød 2019; Kendall-Taylor, Frantz, and Wright 2020; Choi and Jee 2021). Moreover, these studies yield less insight into how citizens of authoritarian states make decisions to consume and share illicit and potentially subversive foreign information offline and what this can tell us about authoritarian control. We currently lack a theoretically grounded account of how illicit foreign media content circulates in authoritarian states offline.

Much of the comparative literature on information in authoritarian regimes focuses on the role of the internet and censorship. Still, the literature on political control suggests ways in which social capital theory may be helpful to understand the limits of offline information flows better.

Research on social capital points to the integral role that both associational and personal ties play in the rise of authoritarianism (Satyanath, Voigtländer, and Voth 2017) and authoritarian stability (Jamal 2009; Acemoglu, Reed, and Robinson 2014). Social trust and social institutions in authoritarian countries may act to suppress non-institutionalized forms of political action (Roßteutscher 2010, 743), helping to stabilize non-democratic regimes. Mass surveillance and repression are shown to have a negative effect on the formation of general social trust (Lichter, Loeffler, and Siegloch 2015; Xue 2020).

In short, authoritarian states generally discourage people from forging ties that could be used to gather useful, desirable, or potentially politically sensitive information (Narayan 1999, 14–15; Woolcock and Narayan 2000, 237). Under these conditions, social capital is co-produced by state and non-state actors but in ways prescribed by authorities (Fox 1996).

As the literature on social capital makes clear, the stock of trust and relations between people is more than just an individual or collective resource. It is also an instrumentalized source of social control (Portes 1998; Warner and Clubb 2013). Authoritarian states harness the control function, creating associations and other societal organizations that are subservient to the state and used to enhance the state’s capacity to control or co-opt civil society. We leverage the insights from this literature to craft general empirical expectations regarding why citizens in authoritarian regimes will or will not consume and further distribute illicit content.

Social Capital and Information Flows

Social trust and interpersonal ties, both formally embodied in associations and through informal ties, help to create two forms of social capital: bonding capital and bridging capital (Uphoff 2000, 218–219).² Bonding capital represents strong ties and particular trust between people, such as friends, family members, and to a lesser extent, people of the same social background (class, occupation, ethnicity, etc.). Bridging capital involves weak ties and more general trust between groups, essential for the functioning of civil society and many economic interactions (Granovetter 1973, 1985; Putnam 2000, 137; Burt 2001; Delhey, Newton, and Welzel 2011). Relationships beyond family and friends rely more on general trust (Stolle 2002, 404–407).

The “weak ties” that cut across class and other social cleavages (i.e., bridging capital) are widely recognized as crucial for supporting the spread of information (Granovetter 1973; Aral 2016; Burt 2000). As Burt (1995, 26) notes, “the spread of information on new ideas and opportunities […] must come through the weak ties that connect people in separate clusters.” In social life, organizations like political parties, broad-based advocacy groups, and even social clubs create bridges between people in otherwise disparate social groups. These ties form the foundation of civil society institutions (Putnam, Leonardi, and Nonetti 1993; Putnam 2000).

Hence, dictatorships seeking to control and curtail the spread of potentially dangerous information would rationally target potential sources of bridging capital and general social trust, as the existing literature indicates (Hassan, Mattingly, and Nugent 2022, 164–167). In other words, countries like North Korea, China, and other autocracies with sufficient state capacity seek to dominate the spaces where social ties and social trust may be produced, thus blocking the emergence of a potentially anti-regime civil society, primarily through control of bridging social capital ties. Spaces where such ties can be forged include diverse urban neighborhoods (Heberer 2009), public spaces, workplaces (e.g., Shaw 1996; Krzywdzinski 2018), and other major sites of people’s lives where citizens interact with each other and the state.

Disrupting weak ties can result in a mono-organizational society (Rigby 1999; Ward, Lankov, and Kim 2021). We consider marketplaces as a contested space in North Korea’s socioeconomic order, where private actors transact on commercial, private terms, making them relatively freer from state surveillance and control compared to state-owned enterprises and other workplaces in the state sector that are subject to regular surveillance by party officials. Marketplaces are unusual in representing a partially non-state-dominated space, facilitating the creation of both bonding capital between people of similar backgrounds and with similar interests, namely traders,³ and bridging capital between people of different backgrounds and interests, such as consumers and distributors (Meadowcroft and Pennington 2008; Storr 2008; Choi and Storr 2020).

Nonetheless, our first empirical expectation is that public spaces, such as markets, will be less favored as places of illicit information acquisition compared to private spaces because of the threat that state power and repression present and the state’s perceived and real capacity to surveil its citizens (Lankov, Kwak, and Cho 2012). Commercial incentives at play in markets may help forge trust to some extent. Still, without preexisting relations between the trader and the would-be purchaser, such effects are likely to be limited. Hence, people are liable to favor their own homes or those of trusted friends because the family is one of the few redoubts free of state surveillance.

Under an authoritarian regime, the state’s coercive capacity is used to penetrate society through a mix of co-optation, infiltration, and repression (Way and Levitsky 2006; Göbel 2011; Gallagher 2017; Reny 2021; Hassan, Mattingly, and Nugent 2022). The state’s domination of public spaces and social organizations pushes people to retreat into the private sphere of family and friendship ties. The public sphere is then left totally monopolized by the state (Shlapentokh 1989), or at least co-opted and infiltrated to serve state interests (Reny 2021; Hassan, Mattingly, and Nugent 2022). This implies that strong ties between people who know one another privately would be favored as a source of information and for further information dissemination.

Hence, our second empirical expectation is related to the low levels of general social trust in autocratic societies and how these are likely to pattern the relationship between the person acquiring illicit information and the person through whom they acquire it. We expect that in authoritarian regimes, people prefer to acquire information from people with whom they have highly particular, trusting relations (i.e., family and friends). Conversely, due to a lack of general trust, people are unwilling to acquire illicit information through neighbors and strangers. We also expect them to be more willing to distribute such information further to people with whom they have similarly trusting relations.

In summary, we have set out two theoretical expectations informed by the existing literature on social capital theory and comparative authoritarianism: (1) people prefer locations that are less infiltrated by state power and surveillance for information acquisition and dissemination, and (2) people prefer social ties undergirded by particular trust for the acquisition and further dissemination of information. Now, we turn to our case: North Korea.

Information Flows and Social Control in North Korea

We focus exclusively on North Korea in this paper. Given its lack of internet infrastructure and access, dearth of independent civil society institutions, and the state’s notably repressive approach toward foreign media, it represents an ideal case for analysis. It is one of the only countries in the world where internet access is outlawed for all but a small set of senior officials working in areas where it is deemed necessary (Seliger and Schmidt 2014; Gerschewski and Dukalskis 2018). Internet access is thus not a potential confounding element in the analysis and allows us to examine how information circulates in its absence.

North Koreans, like citizens in any authoritarian country, are part of a public sphere (Dukalskis 2017, 37–44), even if it is designed to legitimize dictatorial power. The North Korean information environment is characterized by its remarkable degree of closedness and control. The state uses tools of surveillance and coercion to police market spaces as they do other places (e.g., neighborhoods). It has also sought to raise the transaction costs associated with the distribution of dangerous foreign information, going so far as to execute people for consuming or distributing foreign media content (Transitional Justice Working Group 2021).

Yet, the country is not as closed to information as it was for much of the Cold War (Hunter 1999, 134), with markets and social networks facilitating access to foreign media via DVDs, USBs, memory cards, and other digital devices, including Chinese mobile phones (Kretchun and Kim 2012, 12). Survey data from 2015 indicate that around 90 percent of the population had access to a DVD player, around 80 percent to USB devices, and over 60 percent to a mobile phone (without internet). These devices and media are key vectors by which foreign audio and visual materials are acquired and disseminated (Kretchun, Lee, and Tuohy 2017, 7).

More recent surveys present similar findings of widespread ownership of digital devices (Cheon et al. 2019, 66, 68) that are used to consume and disseminate foreign media (Unification Media Group 2019, 12–15). It is an open question whether general trust between strangers transacting in public spaces is sufficient to facilitate the spread of illicit content in North Korean society. These same surveys indicate that marketplaces, family, and friends are key sources of foreign media for consumption and distribution (Unification Media Group 2019, 19), and that North Koreans generally obtain and share foreign media only from and with family or friends (Kretchun, Lee, and Tuohy 2017, 20, 21; Unification Media Group 2019, 16, 17).⁴ It remains unclear, however, which factors are most important in shaping the decision to consume and then further distribute illicit information or how they might interact. Is it the place, the relationship between those swapping illicit content, or other factors?

Data and Methodology

To explore the determinants of foreign information consumption and dissemination in North Korea, we fielded a choice-based conjoint (Hainmueller, Hopkins, and Yamamoto 2014) with 313 North Koreans residing in South Korea. This method is used for causal inference in a multidimensional design. It is an increasingly common instrument used in the social sciences to investigate public opinion on issues as wide-ranging as immigration (Hainmueller and Hopkins 2014; Ward and Denney 2022) and medical preferences (Al-Omari, McMeekin, and Bate 2021), and it has been used in commercial market research for decades (Wind et al. 1989). Given the ability to consider multiple attributes simultaneously, this approach is particularly appropriate for assessing the desirability of foreign information, because that information is defined by multiple dimensions (content, medium, etc.).

In the conjoint experiment, respondents are asked to evaluate two hypothetical foreign information profiles and choose which they would have used (i.e., read, listened to, or watched) in North Korea. They are then asked to answer whether they would have shared the chosen information profile with a stranger, neighbor, friend, or family member. Lastly, respondents are asked to evaluate on 7-point rating scale the two profiles for use preference. Respondents are asked to answer as if they were still in North Korea.

Each profile includes seven attributes with varying values. The attributes include information provider, acquisition place, distribution media, content subject, target age, and type. These attributes were chosen to best approximate types of foreign information that North Koreans might encounter and how they would encounter it. During the design stages of the survey, attributes and their values were generated after consultations with former residents of North Korea and various NGOs that work to provide outside information to North Koreans.

The principal attributes of theoretical interest to us are provider and acquisition place. As discussed above, both attributes measure types of social capital and interpersonal trust that we expect will determine whether the information is used and shared. Bonding capital and particular trust are represented by family and friends, whereas neighbors and strangers are chosen for their association with general rather than particular social trust.

One’s own home implies a level of security that would be associated with particular trust and bonding capital, while someone else’s home would likely be a less risky location than either a market or open spaces. The latter two spaces were chosen as spaces defined by general social trust.

We are also interested in the place of acquisition for what it might say about the spatial limitations of social capital in the context of a heavily surveilled authoritarian state. The market is clearly of interest as a site where bridging capital can be formed and utilized, in principle, but also as a space tightly regulated by the state (Gray and Lee 2021). The role of content type and medium are also considered as potentially important variables that must be accounted for in a closed society where content consumption and sharing are potentially highly dangerous, but the theoretical justification for these attributes is less of a concern.

Our experimental design is, in summary, being used to measure the extent to which the North Korean state is using the structures of social capital in the form of bonding capital (relationships within groups including the family and neighborhood) and bridging capital (cross-cutting associations like marketplaces) in service of social control. The approach also permits us to consider the impact on social trust in the form of both particular social trust (e.g., friends and family) and more general trust in neighbors and strangers (Uphoff 2000; Yamagashi 2001).

The primary outcome variable of interest is determined by the forced choice, a binary variable for whether the respondent would use an information profile (1, else 0). This, then, determines sharing preferences, constituted by four additional dichotomous outcome variables for whether the respondent would share the information they would use with a stranger, neighbor, friend, or family member. The ratings-based use preferences are treated as robustness checks on forced-choice questions. Notably, we do not ask whether the respondent would share the information independently of consuming it. Rather, we ask whether they would share the information they would be hypothetically consuming. The point here is to determine what information that North Koreans consume they would subsequently share, and with whom.

Figure 1 shows an example of the experimental design and an approximation of what the respondents saw in the survey after the introduction to the experiment. The introduction included an explanation of what each question was instructing the respondent to do.

Figure 1. The experimental design.

Table 1 presents the full list of attributes and their values. Across seven attributes with four values each, except for type (which has two), there are 8,192 possible combinations. Each respondent is shown a total of 10 unique profiles, otherwise known as “tasks.” All 313 respondents completed 10 tasks in total and two information profiles per task. This yields 6,260 total observations (313 * 10 * 2). For each task, the order of the attributes and attribute levels are fully randomized. This design permits us to specify the causal effect of attribute level on the probability of a foreign information profile being preferred for consumption and sharing. Varying effects of different attributes on the same scale also allow us to consider the relative importance of any given attribute level. We can, for example, consider the importance of the information provided by a friend or relative to information received at a market. The experiment also lets us consider interaction effects between attribute values, such as whether the impact of place (e.g., a market) is mediated by the provider (e.g., family).

Table 1. Conjoint Attributes and Values

Attributes	Values
Provider	Stranger
	Neighbor
	Friend
	Family
Acquisition Place	One’s own home
	Another person’s home
	Outside (park, street, etc.)
	Market
Distribution Media	North Korean phone/tablet
	CD/DVD
	USB
	Foreign device (phone etc.)
Content Subject	Not country-related
	International content
	South Korean content
	North Korean content
Content Form	Text
	Recorded file (audiobook etc.)
	Video
	Computer program or app
Target Age	Youth
	19–29
	30–49
	50+
Type	Fiction
	Nonfiction

Sample

Survey participant recruitment and implementation were carried out by Woorion, a South Korea–based NGO specializing in North Korean resettlement and survey logistics. The survey design and content were extensively workshopped with NGOs that deal with the North Korean information environment and with focus groups of North Koreans residing in South Korea. To limit the effects of resettlement and exposure to new institutions, we sought to recruit respondents who left North Korea no more than 10 years before the survey was carried out. In total, 313 respondents were recruited in June and July 2021.

In addition to administering the survey experiment, we collected background information on survey participants, including demographics and socioeconomic information. We also asked a battery of direct questions about information consumption in North Korea. Information about migration and resettlement, background in North Korea, and pre-migration socioeconomic class is provided in Table 2. We see that 73 percent of the sample are women with an average age of approximately 35. The average year of defection from North Korea is 2015, the average time spent in China is less than two years, and the average number of years spent in South Korea is five. A few respondents defected from North Korea more than 10 years before the survey (the earliest in 2004), but we note here that 96 percent of those recruited left no later than 2010.

Table 2. Background on Defector-migrant Sample

Statistic	N	Mean	St. Dev.	Min	Max
Women	313	0.73	0.45	0	1
Age	313	35.05	11.05	20	64
Origins: Borderlands^+	313	0.85	0.36	0	1
Migration and Resettlement^++
Year Defected from North Korea	313	2015	2.69	2004	2020
Years spent in China	313	1.70	1.90	1	19
Years spent in South Korea	313	5.00	2.38	0	17
Background in North Korea
Market Experience	313	0.67	0.47	0	1
Party Background (self)	313	0.09	0.29	0	1
Party Background (family)	313	0.65	0.48	0	1
Army Service	313	0.15	0.36	0	1
Socioeconomic Status in North Korea^+++
Upper	313	0.34	0.47	0	1
Middle	313	0.30	0.46	0	1
Low	313	0.26	0.44	0	1

⁺ Defined as the following jurisdictions: North Pyongan, Chagang, Ryanggang, North Hamgyong, and the special city of Rason.

⁺⁺ As noted, a limited number of participants (n = 7) left North Korea prior to 2010.

⁺⁺⁺ Status is defined by a question about material well-being prior to defection. Those who were able to afford expensive items such as cars and apartments or could easily purchase higher end durable goods like refrigerators are counted as “upper.” Those who could purchase basic goods without struggling are classified as “middle.” Those who can afford mere necessities (e.g., food) or struggle to get by on a day-to-day basis are counted as “low.”

The sample attributes of this survey are consistent with similar surveys, reflecting the background of the migrant community. See Denney (2018), Kim (2017, 41–217), and Green and Denney (2021, 159–162) for a more thorough consideration of the implications of using migrant samples to make inferences about life in North Korea.⁵ Information about the reasons respondents defected from North Korea is provided in Appendix A of the Supplementary Information (SI) document. Given the quality controls applied to this project, there are no missing variables to report. All questionnaire material and answers are provided in English herein, but they were originally administered in Korean and written specifically for North Koreans.

Findings

This section reports the empirical findings from the experiment. First, we report the various effects that foreign information profiles have on the probability that information is taken up and used in North Korea. Then, we examine the open-text answers and what they tell us about the motivation behind foreign information use. The section then concludes with an examination of how attributes impact the likelihood that illicit information is shared.

Foreign Information Use

Figure 2 shows our main results. As per the statistical approach developed in Hainmueller, Hopkins, and Yamamoto (2014), we estimate the average marginal components effects (AMCEs). Given reference category sensitivity, we also report the marginal means (MMs) for each feature level. Regressing the attribute values against the outcome variable (information use), the AMCEs estimate the average difference in the probability of any given profile being chosen across all of the values. Descriptively, MMs measure the mean outcome of each feature level averaged across all others. In a force-choice design, MMs average 0.5 and can be read as probabilities. An attribute level with an MM of .55 indicates that 55 percent of profiles containing that level were chosen (i.e., that attribute level increases profile favorability).

Figure 2. Effects of foreign information profiles on the probability of use.

Note: For average marginal components effects (AMCEs), the estimates show the effects of the randomly assigned information attribute values on the probability of use. The point values from the marginal means (MMs) show the mean outcome of any given attribute level, averaged across all others. Estimates are based on the benchmark OLS model with clustered standard errors; bars represent 95% confidence intervals.

Focusing on the main quantities of interest—providers and place of acquisition—we find that if the information comes from a friend or family member, that has significant and substantive effects on the probability that the foreign information will be consumed. Relative to a stranger (the reference category), information provided by a friend increases the probability of use by 9.5 percent. If information is provided by a family member the effect is similar (8.8 percent), as per the AMCEs. Foreign information that is provided by friends or family increases profile favorability, as evidenced by the MMs (.54 for friend and .53 for family).

Regarding acquisition place, we see that relative to information provided within one’s own home, no other location drives a greater probability of use. It is the only attribute feature that increases profile favorability according to the marginal means. Information provided in an “outside” space, such as a park or in the street, is significantly less likely to motivate use relative to the home (a decrease of 8.3 percent). Information presented in another person’s home or the market is a wash; it motivates neither use nor rejection. More is said regarding markets as spaces of contention below.

Although they are not considered of primary theoretical interest in this paper, the other attributes provide additional and somewhat important insight into foreign information use that is at least worth reporting. We observe that, relative to a North Korean phone or tablet, information on a USB or CD/DVD is more favored (approximately 3.5 percent increase in the probability of use). However, the only finding here is that foreign devices do not motivate information use (a marginal mean of less than 0.5).

Among subject material, we note specific content preferences relative to the reference level (“not country-related”). Foreign information that contains South Korean content increases the probability of use by 5.9 percent, whereas information with North Korean content decreases the probability by the same amount. As the marginal mean for the “South Korean content” feature level shows, respondents are motivated to consume this information.

Regarding content form, there is a modest preference for video over other types. There is a 4.5 percent increase in the probability of information use relative to the text. And for the last two attributes—target age and information type—we see a slight preference for content targeting a 19–29 age group and clearer disapproval of content that targets older audiences (−7.5% for information targeting 50+). As for type, we see that nonfiction is preferred over fiction. Content with fictional content is 4.6 percent less likely to be used. The marginal means show that nonfiction video content form targeting younger viewers indicates to potential viewers that the content is something they want to consume and motivates use.

Robustness checks and additional analysis are provided in Appendix B of the SI. There, we consider whether answers change substantively using the ratings-based answers and look at conditional average treatment effects by select subgroups. Findings do not substantively change using an alternative outcome measurement (ratings-based), nor do we observe any considerable differences in opinion by relevant subgroups (e.g., gender or previous market experience).⁶

To further explore the importance of person and place to foreign information flows in North Korea, we examine the interaction between information provider and place of acquisition from the baseline model. As evidenced above, we know that information provided by a friend or family member is most likely to be used. We also know that places outside the home are either unfavored or at least unlikely to motivate information use. But what if a family member (i.e., someone trusted) provides the information somewhere outside the home? The main effects models cannot answer this question. Figure 3 presents the marginal means of the interactions.

Figure 3. Marginal means of the interaction effects (provider * place).

Note: The point values from the marginal means (MMs) show the mean outcome of any given attribute level, averaged across all others. Estimates are based on the benchmark OLS model with clustered standard errors; bars represent 95% confidence intervals.

As already established, we see that no matter the location, information from a friend or family member makes it more likely that foreign information will be used. But the effect is not equal for all places. The most notable finding is the market location. The marginal means for a friend (.55) and family (.56) indicate that information provided by trusted sources in a marketplace makes that information more likely to be used than not. The effects are not as strong as a family member providing it in one’s own home (.59), but they are close.

It is not as notable a finding, but the interactions model also indicates that foreign information provided by a neighbor within one’s own home also makes that information more likely to be used. This suggests that if one is comfortable enough to have a neighbor in one’s own home, then they might be a trust-worthy contact, rather than simply someone known.

To provide a more substantive presentation of the conjoint findings, we show the predicted probabilities that various foreign information profiles would be used (Figure 4). Estimates are plotted across the distribution. The minimum profile (i.e., least likely to yield use) is information provided by a stranger, delivered outside, and provided on a foreign device with North Korean content in text format. It is fictional (e.g., a novel) and targets an older age group (50+). This information profile has a 20 percent chance of being consumed—in other words, very unlikely. It is hard to say what this information might be exactly, but it may strike the respondent thinking about what it could represent as something boring or not valuable, or even overtly political, given the content is about North Korea. Perhaps it is both and thus absolutely not worth the risk.

Figure 4. Estimated probability of foreign information use for selected profiles across the distribution.

Note: The figure shows the predicted probability of information profiles being preferred for use. Estimates show the minimum, 25th, 50th, 75th, and maximum percentiles of the distribution and are based on the benchmark OLS model with clustered standard errors; bars represent 95% confidence intervals.

At the 50th percentile, where the chance of information use could go either way, we find information coming from neighbors in another person’s home on a North Korean device showing something nonfictional and unrelated to any country in particular. Although it is foreign-sourced content and thus illicit, we can surmise this hypothetical profile reflects something rather mundane but perhaps of interest.

The information profile with the highest probability of use (76 percent) is delivered by a friend in one’s own home on a USB drive containing nonfictional video content about South Korea targeting a younger audience. This could be anything from a documentary to South Korean news; it is certainly illicit but of considerable interest and thus probably used. Notably, if this content were fictional (e.g., a drama), it would lower the probability of its being used, but not by that much.

Open-Text Answers

Next, we use a machine learning technique to find abstract “topics” within the 313 open-text responses regarding the motivation behind choosing an information profile for use. This method serves two purposes. First, as an alternative outcome variable, analysis of the open-text answers works as a robustness check on the conjoint-based analysis. Second, it allows us to assess how respondents express their preferences and opinions qualitatively.

Figure 5 shows the results from a three-cluster implementation of Latent Dirichlet Allocation (LDA) (Chang 2010).⁷ Each topic contains clusters of words that were likely to co-occur among the answers provided, with the most commonly occurring words listed in ascending order. The value expressed on the x-axis indicates the probability of the word occurring, given the topic. We can classify the cluster topics by the meanings they convey. We read curiosity and a desire to know about the outside world in Topic 1, the importance of social trust in Topic 2, and a demand for entertainment in Topic 3.

Figure 5. Word-topic probabilities based on open-text answers.

Note: The figure shows the outcome of a three-cluster implementation of Latent Dirichlet Allocation (LDA). Values represent the probability of the word occurring given the topic (top 15 words shown).

The second topic corroborates our theory-based interpretation of the conjoint findings regarding determinants of foreign information use in North Korea. It indicates that family/relatives and friends, in particular, are safe and trustworthy sources. These are people from whom one is willing to receive illicit content. Neighbors, too, are possibly trusted sources and even more so when information is provided in the safety of one’s own home (see Figure 4 above). The first and third topics reflect the more mundane but individually important motivations for wanting and using foreign information.

To better illustrate the substantive meaning behind the open-text responses, we selected quotes corresponding to each topic. Regarding worldly curiosity (Topic 1), two respondents had the following to say about their motivation behind choosing the information profile that they did:

Because it is the only way to see and learn about the world in isolated North Korea.

I was very curious about the ways of life in capitalist or democratic countries, which I could not see in North Korea.

Regarding Topic 2 (social trust), the following rationales were provided:

There is a tendency to watch South Korean dramas and films with friends you can trust and with family because you can go to prison if caught.

North Korean media is largely unpopular, while media from South Korea and other states inspires great enthusiasm. […] Because it is so dangerous to share with relatives or neighbors, people share with close friends.

Lastly, for topic 3 (entertainment), the following was written:

Media are a way to access outside information. When I was young, I watched and listened to a lot of South Korean and other foreign films and music.

Foreign Information Sharing

Having established the effects that various attribute values have on the probability of foreign information being used in North Korea, we turn now to how these same attributes affect foreign information sharing. Based on the information profile chosen for use, the survey respondents were asked to indicate whether they would share that information with a stranger, a neighbor, a friend, or a family member.

First, we establish the proportion of respondents who said they would share the information they used with the specified recipients. In Figure 6, we see that more than three-quarters of respondents would share with a family member (83 percent) or a friend (82 percent). Sharing with a neighbor is significantly less likely, with just over half answering in the affirmative (54 percent). Only 21 percent of respondents were willing to share with a stranger. As we found for information use, the greatest amount of trust resides among friends and family.

Figure 6. Proportion who said they would share foreign information (per target recipient).

Next, we consider the effects of the randomly assigned information attribute values on the probability that the chosen information profile would be shared across the four target recipients (stranger/neighbor/friend/family). Do the information profile attributes impact the sharing of foreign information? As before, we focus on the effects of the providers and the acquisition place. For the sake of space and in order to focus on our main quantities of interest, Figure 7 only shows the AMCEs for provider and place-based effects on the likelihood of sharing information. Full model estimates are provided in Appendix B.

Figure 7. Effects of foreign information attributes on the probability of sharing.

Note: Estimates show the effects of the randomly assigned information attribute values on the probability of the information profile selected being shared with the target recipient. Estimates based on the benchmark OLS models; bars represent 95% confidence intervals.

Regarding providers, we find some evidence that the relationship between the information user and the provider has an impact on the decision to share the information. Relative to a stranger, information obtained from a neighbor, friend, or family member is less likely to be shared with another stranger. In other words, although strangers are the least favored recipients of foreign information, respondents are more likely to share information obtained from a stranger with a fellow stranger. We see similar effects for the attribute levels of neighbor, friend, and family, although the effects are modest. A neighbor providing foreign information increases the probability that a respondent would share the information with another neighbor by 5 percent.

Additionally, we find a few notable place-based effects. Relative to the reference category (one’s own home), if the information was shared outside in a park or on the street, it increases the probability that the information will be passed along to a fellow stranger, perhaps indicating the effect of anonymity. We note that information acquired in another person’s home also increases the chances that it will be shared with a friend, indicating that friends and other’s homes are likely associated and trustworthy spaces. Lastly, we note that people are less likely to share information acquired at a market with family, relative to information acquired at one’s own home. Again, these effects are not particularly strong, but they do indicate where trust resides and how that might influence one’s decision to share information in North Korea.

Conclusion and Discussion

In North Korea, the state continues to control the spread of information through its domination and control of social institutions, especially those outside the home. As findings from our choice-based conjoint show, North Koreans have relatively high levels of particular trust in their friends and family. These are two types of social ties and bonding capital that are not directly targeted by institutions of daily social surveillance in North Korea. It is from these ties that people will likely acquire and consume foreign information. They are also most likely to disseminate further the information they acquired to the same people who provided it.

While the social capital that binds non-state relationships is largely co-opted by the North Korean state, there is evidence that less intimate but more common spaces are locations of contention. Markets, for example, neither motivate nor discourage foreign information use on average. They are, however, favorable places to exchange information with people highly trusted (e.g., family). Content acquired from strangers is also more likely to be shared in markets.

These insights show the potential utility of markets and the importance of relationships between traders and customers. Similar dynamics have been observed in illegal markets and criminal networks elsewhere (May and Hough 2004; Malm, Bichler, and Van De Walle 2010). In the absence of property rights and formal contract-enforcement mechanisms, relational contracting and trust are crucial for the functioning of institutions specific to capitalism in North Korea (Lankov et al., 2017). These spaces may not be bridges that forge civic ties across groups in society yet, but commercial spaces are less dangerous as sources for illicit content than other places.

Overall, our findings point to the promise and pitfalls of social capital and marketization as potential sources of social change in North Korea and other authoritarian regimes. Clearly, some forms of social capital facilitate access to illicit information and encourage dissemination (e.g., bonding social capital between friends and family members), but the limits to general trust and weak social ties mean that illicit content is unlikely to circulate among people who are less well known. Information flows reflect the limited trust radius for any given person in North Korea.

There are broader implications to the research presented here. Strategies that seek to dominate bridges between citizens, such as those used in North Korea, curtail access to information and buttress state capacity and its durability. These findings are consistent with the literature on the role of social ties in authoritarian states (Völker and Flap 2001). The co-optation of social relations is a continued source of regime stability and social control in authoritarian regimes.

Given the pervasiveness of internet censorship in many authoritarian settings, offline content acquisition and dissemination, and the barriers thereto, will become increasingly important. As our research findings show, illicit information flows offline are heavily curtailed due to the structural dominance of the state over society. With the rise of sophisticated surveillance technologies, state control over public space in authoritarian countries is liable to become even more pervasive than it is now. Censorship and information control online may be complemented offline by distrust of strangers and fears about public and commercial spaces, such as internet cafés, as locations in which one would contemplate illicit information and content. The location-specific aspects of surveillance and social capital appear to be rather understudied in the literature on authoritarian survival, but the research presented here clearly indicates that location may be important for offline information flows.

Authoritarian state infiltration of civil society is a general phenomenon in much of the non-democratic world (Hassan, Mattingly, and Nugent 2022, 164–167). It is plausible to infer that the dynamics explored in North Korea exist in other consolidated authoritarian regimes with high levels of coercive state capacity. Given autocratizing global trends (Lührmann et al. 2019), research on state–society relations and information flows and control will clearly be needed.

Data Availability

The data used in this research can be accessed at: https://github.com/scdenney/Information-Flows-in-North-Korea.

Ethics

Research for this report was approved by the Research Ethics Board at the University of Toronto (human protocol 32781).

Acknowledgments

We thank the NGOs that provided feedback on our research design at the early stages of the research. We also wish to acknowledge the important input from Nat Kretchun, Sarah Yun, and Park Dae-hyeon and the programming and design assistance from Viet Vu. We also thank the valuable feedback provided by two anonymous journal reviewers.

Disclosure Statement

The authors were consultants to and received funding from the National Endowment for Democracy (NED) to conduct research in collaboration with the NGO Woorion.

Supplementary Information

Supplemental data for this article can be accessed on the publisher’s website at https://doi.org/10.1080/10758216.2023.2180038.

Additional information

Funding

This research was supported by a grant from the National Endowment for Democracy (NED Grant No. 2020-0799).

Notes

1. Kim (2018) finds low levels of trust among North Korean refugees. It stands to reason that this is reflective of North Korean society writ large, as we evidence and explain in this paper.

2. On the relationship between bonding and bridging capital, see Schuller (2007).

3. On the role of bonding capital in marketplaces see Geertz (1978), and for the commercial effects of social capital in marketplaces somewhat analogous to North Korean markets see Gabre-Madhin (2001). For more recent and theoretically focused discussion on social capital and markets see Storr (2013). It should also be noted that marketplaces are a workplace, and as such potentially a site for cross-cutting political conversation (Mutz and Mondak 2006). But in the North Korean context, most other workplaces are dominated by highly intrusive systems of state control and surveillance, whereas many market participants are private actors operating private businesses for their own profit, free from regular monitoring by party officials and other agents of the surveillance apparatus and more difficult to monitor at scale.

4. For a broader review of the communications literature dealing with North Korea, see Seo and Nah (2020).

5. North Koreans living in South Korea can be considered refugees, but their co-ethnic status differentiates them from others. They are afforded near-unconditional right to citizenship in South Korea (cf. Greitens 2021). Given the political act it involves, North Korean refugees/migrants are sometimes referred to as “defectors,” which is written and expressed in Korean in several different ways (e.g., talbukcha or talbukmin).

6. There are a number of warranted heterogeneous treatment effects to consider by subgroups (e.g., gender, previous market experience). However, due to length considerations and, especially, statistical power concerns, we do not analyze these results in the manuscript. The analysis presented in the appendix is no more than exploratory and suggestive of what future research might more carefully and accurately explore.

7. Three clusters were determined optimal based on the coherence score, which is a measure of semantic interpretability among words in a topic.