2.1 The CCP’s Surveillance State

The Information Age has revolutionized surveillance in the world’s autocracies. In 1998, the CCP launched the Golden Shield Project, which Xu (Reference Xu2021, 316) describes as “a domestic surveillance and filtering system that integrates online government databases with an all-encompassing surveillance network.”Footnote ³ In the first phase, completed in 2005, the CCP built a massive network of population databases, ID tracking systems, and internet surveillance tools, which let it record the movement of potential dissidents as revealed, in part, by their online behavior. In 2017, the CCP announced the completion of its “Sky Net” program, which entails 176 million surveillance cameras across China and plans for 626 million by 2020, nearly one camera for every two citizens (Hersey Reference Hersey2017; Russell Reference Russell2017). The result, Qiang (Reference Qiang2019) writes, is “the largest video-surveillance network in the world.”

Simultaneously, the CCP built a facial database that encompassed every adult citizen (Chin and Lin Reference Chin and Lin2017) and a DNA database that encompassed 54 million citizens and, by 2020, will reportedly reach 100 million (Qiang Reference Qiang2019). The CCP’s facial recognition technology is employed for check-in and security at airports (Dai Reference Dai2018; Yang Reference Yang2018), train stations (Chen, Jing and Dai Reference Chen, Jing and Dai2018), and hotels (Chan Reference Chan2018). In 2017, the CCP applied facial recognition technology to detect jaywalkers, with offenders notified via text message and their pictures displayed at major intersections (Li Tao 2018). This pervasive surveillance apparatus lets the CCP repress dissidents and spend less on public goods (Xu Reference Xu2021). It also complements more analog forms of repression, such as informants and hired thugs (Deng and O’Brien Reference Deng and O’Brien2013; Mattingly Reference Mattingly2020; Ong Reference Ong2022). Digital surveillance is now a conspicuous feature of everyday life.

The CCP’s digital surveillance apparatus is supported by a network of domestic technology firms, which are subsidized by the state and routinely used as instruments of foreign policy. The most general are Huawei and ZTE. Huawei is the world’s largest manufacturer of telecommunications equipment (Chin and Lin Reference Chin and Lin2022, 131), and especially dominant in Africa, where it has provided 70% of the 5G network. Its products span mobile phones and other consumer electronics, telecommunications networks, face and voice recognition technology, and video cameras. China has a number of more focused technology firms that are implicated in surveillance. Several of these specialize in video cameras and facial recognition software: Hikvision, Dahua, CloudWalk, Megvii, YITU, and SenseTime, most notably. Of these, Hikvision is perhaps the most consequential. In 2019, it was responsible for nearly a quarter of the world’s surveillance cameras (Hillman Reference Hillman2021, 98). Dahua has also supplied cameras for Safe City projects, so called for their use of digital surveillance to support the local security apparatus (Chin and Lin Reference Chin and Lin2022). Other firms specialize in still different areas of surveillance. Meiya Pico reportedly built an app used by the Chinese government to extract data from citizens’ smartphones during street checks (Chen and Jing Reference Chen and Jing2019; Cheng Reference Cheng2020). iFlytek develops voice recognition software (Hvistendahl Reference Hvistendahl2020). Each of these firms has been sanctioned by Washington.

As the largest and most general of China’s technology firms, Huawei often collaborates with the more specialized firms, both to develop new technologies and to ensure inter-operability: that the technologies developed by the specialized firms work on Huawei’s infrastructure. Huawei partnered with Megvii to enhance the CCP’s facial recognition technology in Xinjiang (Harwell and Dou Reference Harwell and Dou2020). Huawei teamed up with iFlytek to develop the “iFlytek Voiceprint Management Platform,” which “can identify individuals by comparing the sound of their voice against a large database of recorded ‘voiceprints’” (Dou Reference Dou2021). The voice assistant in Huawei’s smartphones thus features iFlytek’s voice recognition software (Cave, Ryan and Xu Reference Cave, Ryan and Xu2019). Huawei has partnered with other Chinese technology firms to pioneer surveillance tools for prisons, “political persons of interest,” and even employees of private sector firms (Dou Reference Dou2021). Huawei’s Smart City systems routinely incorporate YITU’s facial recognition and traffic monitoring software (Cave, Ryan and Xu Reference Cave, Ryan and Xu2019).

2.2 Exporting Digital Surveillance Technology

Governments across the world have sought to acquire Chinese digital technology. Huawei, given its scale, has been central to this, often incorporating the technologies developed by China’s smaller, more focused firms. The anecdotal evidence suggests four broad ways that these exports can be adapted by recipient governments to facilitate digital repression.

The first is facial recognition. Huawei’s Safe City programs generally feature video surveillance cameras, which feed data to police command centers and help detect dissidents. Huawei began marketing its Safe City technology abroad in 2010 (Chin and Lin Reference Chin and Lin2022, 134). In 2017, Huawei identified 40 countries where its Smart City surveillance technology had been introduced. By 2018, its reach had expanded to at least 90 countries and 230 cities (Cave et al. Reference Cave, Hoffman, Joske, Ryan and Thomas2019). By 2019, Huawei had installed Safe City systems in 700 cities across more than 100 countries (Chin and Lin Reference Chin and Lin2022, 135). Huawei pitches its digital surveillance technologies to national security agencies, with subsidized financing provided by China’s Exim Bank (Feldstein Reference Feldstein2019). ZTE provides similar packages, but its scope is much smaller. It began marketing its Safe City solutions several years after Huawei and, by 2019, had built similar systems in 160 cities across 45 countries (Chin and Lin Reference Chin and Lin2022, 135). Governments occasionally advertise these surveillance capabilities to citizens, even emphasizing their Chinese origins. In Cameroon, as the Anglophone Crisis raged in February 2017, the government announced a contract with Huawei on the front page of its flagship propaganda newspaper, the Cameroon Tribune, that provided for “1,500 cameras in regional capitals and certain strategic points in the country,” “2,000 portable listening devices equipped with cameras,” and “the construction of nine command centers” (Carter and Carter Reference Carter and Carter2023).

Second, the telecommunications networks installed by Huawei generally feature surveillance middleboxes with Deep Packet Inspection capabilities. These middleboxes power China’s Great Firewall, the most extensive censorship operation in human history (King, Pan and Roberts Reference King, Pan and Roberts2013). They monitor users’ internet activity, can censor online content, and can block access to virtual private networks (Weber and Ververis Reference Weber and Ververis2021). Huawei’s middleboxes have been documented in 72 countries. In at least 18 of those, the middleboxes have been used for internet censorship (Earp Reference Earp2021; Weber and Ververis Reference Weber and Ververis2021). Middleboxes with Deep Packet Inspection capabilities have been linked to internet and social media shutdowns (Woodhams and O’Donnell Reference Woodhams and O’Donnell2021), an increasingly ubiquitous feature of life in autocracies (Feldstein Reference Feldstein2021), even if such shutdowns undermine a government’s ability to collect high-quality intelligence (Gohdes Reference Gohdes2015, Reference Gohdes2020, Reference Gohdes2024).

Third, China’s technology transfers have been used to create the sort of integrated surveillance system that the CCP’s Golden Shield Project pioneered. This has been especially well documented in Venezuela, where ZTE helped the government create a national identification card – dubbed the “fatherland card” – that records voting behavior, party membership, social media use, personal finances, and medical histories. ZTE provided servers for the database and developed the ID card’s mobile payment application. The Maduro government appears to be using the fatherland card to divert state resources to loyalists and monitor dissidents: “Everybody must get one,” Maduro announced in 2016. Said one Venezuelan technical advisor, who was assaulted and accused of treason after he objected to it: “What we saw in China changed everything. They were looking to have citizen control” (Berwick Reference Berwick2018). As of 2015, ZTE was helping the government build a series of “emergency response centers” in urban areas and centralize its video surveillance capabilities (U.S. Senate Committee on Foreign Relations 2020, 30-31). The broader objective, many citizens believe, is a “social credit system” modeled after the CCP’s (Polyakova and Meserole Reference Polyakova and Meserole2019).

Of course, the ability of recipient governments to use Chinese technology transfers for digital repression may be hampered by limited state capacity or other obstacles (Pan Reference Pan2017). Feldstein (Reference Feldstein2021) documents, for instance, how several governments have imported the tools of digital repression, but have had difficulty putting them to use. This may be why Huawei offers direct personnel support to recipient governments. This is nowhere more apparent than in Sub-Saharan Africa, where preexisting digital surveillance capabilities are generally less sophisticated than those provided by Chinese firms. In 2018, for instance, the Ugandan security apparatus began to track Bobi Wine, a musician turned leading opponent to President Yoweri Museveni in the 2021 election. After installing spyware on Wine’s phone, the Ugandan police enlisted Huawei engineers – with whom they shared an office, emblazoned with Huawei’s logo on the walls – to help them operate it. Huawei engineers helped Ugandan security officials intercept encrypted messages, eavesdrop on his phone, and track his location (Parkinson, Bariyo and Chin Reference Parkinson, Bariyo and Chin2019; Woodhams Reference Woodhams2019; Chin and Lin Reference Chin and Lin2022). Huawei engineers provided similar support in Zambia, helping security forces intercept encrypted messages sent by local journalists, track their whereabouts, and ultimately help with their arrest (Woodhams Reference Woodhams2019).

2.3 Differential Effects by Preexisting Political Institutions

To be sure, there are important differences among forms of digital repression (Gohdes Reference Gohdes2015, Reference Gohdes2020, Reference Gohdes2024). Surveillance technologies – such as facial recognition, content monitoring, and location tracking – help governments engage in targeted repression against dissidents and opposition leaders. By contrast, censorship technologies – like internet shutdowns and internet filtering – may provide cover for indiscriminate repression and inhibit collective action, but at the cost of high-quality intelligence that can help governments target dissidents with precision. Technically, shutdown technologies may also be easier to implement than surveillance technologies, which some governments have had trouble using effectively even after adoption (Feldstein Reference Feldstein2021).

Our focus is not on the strategic calculations and technical constraints that give rise to different forms of digital repression. Autocrats, after all, may employ different technologies at different times and against different targets, using Chinese technology transfers for digital surveillance and internet shutdowns alike. Rather, we probe whether Chinese technology transfers have been adapted to facilitate digital repression, broadly construed, and whether there are key differences across regimes. Technology transfers, indeed, are intrinsically dual-use. Recipient governments can use them for digital repression, as the examples in Section 2.2 suggest, but also to expand internet access, provide public goods, and give citizens tools to monitor governments and coordinate collective action against them. Chinese technology firms are attractive suppliers on purely market terms. Huawei’s telecommunications equipment, in particular, is generally high-quality and as much as 30% cheaper than competitors, due in part to financial subsidies from the Chinese government (El Kadi Reference El Kadi2022) and preferential loan terms from the Exim Bank (Feldstein Reference Feldstein2019).

Our basic argument is that dual-use technology transfers are likely to have differential effects based on a country’s preexisting political institutions. The world’s autocracies and democracies likely pursue Huawei technology transfers for different objectives and, once acquired, confront different constraints in their application.

3.1 What, Who, and When

To probe the effects of Chinese technology transfers in recipient countries, we focus on Huawei. We do so principally for reasons of data availability. Since technology transfers are generally covert, we lack systematic data on the range of transfers provided by individual Chinese companies. Huawei, the world’s largest telecommunications provider, is an exception. AidData’s Global Chinese Development Finance Dataset records 153 Huawei projects worth roughly $1.6 billion in 64 countries worldwide between 2000 and 2017 (Custer et al. Reference Custer, Dreher, Fuchs, Ghose, Lin, Malik, Parks, Russell, Solomon, Strange, Tierney, Walsh, Zaleski and Zhang2021). We view this as a relatively low-cost tradeoff. If governments around the world receive Chinese technology transfers, as we discussed in Section 1, they are overwhelmingly likely to come from Huawei. Since technologies developed by China’s smaller, more focused technology firms are routinely embedded in broader Huawei systems (Cave, Ryan and Xu Reference Cave, Ryan and Xu2019), Huawei transfers are a good proxy for technology transfers from other Chinese firms.

When major Chinese firms strike agreements with foreign governments, they often insist on keeping their details secret. For example, contracts between the Chinese and Ecuadorian governments only became public after the 2016 Panama Papers leak. Huawei’s contract with the Mauritius Safe City Project required the Mauritian police and Mauritius Telecom to invoke confidentiality when questioned by civil society groups (Walker Reference Walker2023). Inferentially, this secrecy creates measurement error; its principal effect is to include in the control group some countries that were actually treated with Chinese digital technology transfers. This biases against the article’s key results. Our estimates below, therefore, should be regarded as a lower bound on Huawei’s effects on digital repression in recipient countries.

AidData classifies Huawei’s projects into 14 sectors, which appear, by shares, in the top left panel of figure 1. The top right panel gives the leading sectors by project value. For both metrics, the most common sector for Huawei transfers is communications, which, by value, accounts for 90% of all projects worldwide. Projects in the communications sector generally focus on the provision of telecommunications and surveillance equipment to recipient governments. By frequency, the second leading sector is education, but, by value, this constitutes just 0.01% of Huawei’s projects. Many of these are government training programs that support equipment transfers or various programs at educational institutions. By value, Huawei’s second leading sector is energy, accounting for roughly 10% of transfers. These projects include transfers to Cameroon and Ethiopia, among the world’s more repressive governments, that strengthen electricity provision, which is key to powering cameras and other digital surveillance technologies. The center left panel displays the share of Huawei transfers by world region. Together, Asian and African countries account for more than 85% of total Huawei transfers. In Africa, Huawei alone is responsible for as much as 70% of the telecommunications network (Woodhams Reference Woodhams2019). Latin American countries account for roughly 13% of transfers. Transfers to European and Middle Eastern countries are negligible.

Figure 1

Descriptive statistics

The center right panel presents a histogram for the volume of each of the Huawei transfers included in the dataset. For each total transfer value along the $ x $ -axis, the $ y $ -axis gives the number of countries that received it. The single largest recipient is Indonesia, with transfers amounting to $391 million. The largest of these occurred in 2009, when Huawei extended a $300 million supplier credit to PT Telkom, a state-owned telecommunications company. By 2022, Huawei had emerged as the largest provider of telecommunications equipment to the Indonesian market, propelled both by its low-cost hardware and its instructional programs to Indonesia’s universities and government agencies (Herscovitch, van der Kley and Priyandita Reference Herscovitch, van der Kley and Priyandita2022; Priyandita, van der Kley and Herscovitch Reference Priyandita, van der Kley and Herscovitch2022).

The median transfer value is $1.2 million, just greater than the value of the surveillance equipment that Uganda’s Kampala city government was promised in 2014 to “track illegal activity.” In 2015, Museveni formally received from the Chinese ambassador the “Safe City Integrated Communication Platform,” which, Custer et al. (Reference Custer, Dreher, Fuchs, Ghose, Lin, Malik, Parks, Russell, Solomon, Strange, Tierney, Walsh, Zaleski and Zhang2021) note, entailed “20 monitor cameras in 10 key monitoring points in Kampala,” similar to what had “improved the administrative capacity of the Urumqi Municipal Administration and the efficiency of Uighur police.” Huawei was nearing completion of the platform as of September 2016. By 2018, Ugandan security services were using Huawei technology to track Bobi Wine and other opposition activists (Woodhams Reference Woodhams2019).

The bottom left panel gives recipients of Huawei technology transfers worth at least $1 million, shaded according to regime type at the time of the transfer.Footnote ⁵ The top recipient, again, is Indonesia. Other leading recipients include several of the world’s most repressive governments: Uzbekistan under Islam Karimov, Cameroon under Paul Biya, Chad under Idriss Déby, Democratic Republic of Congo (DRC) under Joseph Kabila, Ethiopia under Meles Zenawi, Tunisia under Ben Ali, and Burundi under Pierre Nkurunziza. The fourth leading recipient, however, is Costa Rica, one of Latin America’s most vibrant democracies, which has been credited for its efforts to protect digital privacy (Alvarado, Fernando Martínez de Lemos and Weal Reference Alvarado, de Lemos and Weal2022).

The bottom right panel displays the cumulative value of Huawei technology transfers, by year, across autocracies and democracies. The aggregate data suggests that Huawei technology transfers have flowed slightly more to the world’s democracies than the world’s autocracies: roughly 55% to 45%, respectively. This is somewhat misleading, however, as the light gray line suggests. Excluding Indonesia, Huawei transfers have flowed disproportionately to the world’s autocracies, accounting for nearly 70% of all transfers. Indonesia also distorts how we understand the regional allocation of Huawei transfers. Excluding Indonesia, the African continent accounts for 55% of all Huawei transfers, with Asia and Latin America accounting for just 22% and 16%, respectively.

3.2 Correlates of Huawei Transfers

To better understand which countries receive Huawei transfers, we move to a simple regression framework. We estimate several models of the form:

(1) $$ {Y}_{it}=\alpha +\beta {X}_{it}+{\gamma}_t+\unicode{x025B} $$

where $ i $ indexes country, $ t $ indexes year, and $ {\gamma}_t $ gives year fixed effects. The outcome $ {Y}_{it} $ measures the logged value of Huawei commitments in year $ t $ . The vector $ {X}_{it} $ includes country-level characteristics that may condition the government’s interest in securing Huawei technology and Huawei’s interest in providing it.

The results appear in figure 2. The $ x $ -axis reports point estimates for the standardized coefficients, surrounded by 95% confidence intervals. The results are identical across specifications.Footnote ⁶ Huawei transfers, we find, are driven chiefly by demand in recipient countries. More populous countries are more likely to receive Huawei transfers, as are countries with lower GDP per capita. This is intuitive. More populous countries represent more attractive markets for Huawei, while less affluent countries are more likely to be attracted by its lower price point. Huawei transfers are also more likely if the recipient government has a preexisting relationship with the CCP, as measured by its aid allocation in year $ t $ .

Figure 2

Correlates of Huawei technology transfers

Equally interesting are the country-level characteristics that have no effect on Huawei transfers. Notwithstanding the cumulative transfer data in the bottom right panel of figure 2, we find no evidence that Huawei transfers are more likely in autocracies than democracies. The correlation that appears in a bivariate setting is not robust to the inclusion of other controls. We also find no evidence that Huawei technology transfers flow disproportionately to countries that are rich in natural resources or politically unstable, as measured by coup attempts in year $ t-1 $ . In this respect, Huawei transfers are distinctive relative to transfers of facial recognition technology from Chinese firms, which flow disproportionately to non-democracies with histories of political unrest (Beraja et al. Reference Beraja, Kao, Yang and Yuchtman2023b).

4.1 Challenges of Inference

Estimating the effect of Huawei technology transfers on digital repression is complicated for three reasons. First, although Huawei allocates commitments in a given year, the implementation generally occurs over several. Consequently, there may be a lag of several years before the impact of the transfer is realized. Put differently, a recipient country is “treated” at the moment of the transfer, and this transfer persists, though perhaps with heterogeneous effects due to an initial installation period or some other change in the recipient country.

Second, as the bottom right panel of figure 1 made clear, Huawei transfers were staggered over time. In Sub-Saharan Africa, the most significant roll-out followed the expansion of the BRI between 2014 and 2015, which accounted for 36% of the region’s transfers during the sample period. The next most significant expansion followed the Global Financial Crisis of 2009, when the CCP sought to generate foreign demand for Chinese goods – and hence reduce the likelihood of a domestic recession – partly by having state-owned enterprises undertake major construction projects abroad (China News 2008; Yu Reference Yu2010; Lardy Reference Lardy2012; Wallace Reference Wallace2014).

Third, from the coefficient plots in figure 2, the countries that received Huawei technology transfers are systematically different than those that did not. They are, on average, more populous, poorer, and more likely to receive Chinese aid.

4.2 Estimation Strategy

To accommodate these features, we employ the GSC estimator developed by Xu (Reference Xu2017). The GSC estimator is also appealing because it is robust to violations of the parallel trends assumption, which is required by conventional DiD estimators but inherently untestable.

To estimate the ATT population, the GSC estimator imputes a counterfactual outcome for treated unit $ i $ at time $ t $ . It does so in three steps. First, using only data from the control group,Footnote ⁷ it estimates a model using interactive fixed effects (IFE) or the matrix completion (MC) method proposed by Athey et al. (Reference Athey, Bayati, Doudchenko, Imbens and Khosravi2021).Footnote ⁸ Second, it uses these results from the control group to estimate factor loadings for each treated unit by minimizing the mean squared error of the predicted treated outcome in pre-treatment periods. Finally, it constructs counterfactual outcomes for each treated unit in each post-treatment period and estimates the ATT based on the differences between the observed and these counterfactual outcomes. In this sense, the GSC estimator is akin to an out-of-sample prediction method.

Implementing the GSC estimator requires specifying some minimum number of pre-treatment periods, which are used by the GSC estimator to impute counterfactual outcomes. Intuitively, specifying a higher number of minimum pre-treatment periods increases the information with which the GSC estimator imputes counterfactual outcomes, but at the expense of discarding countries that were treated early, at least relative to the beginning of the data. We specify a five-year pre-treatment minimum as a compromise. As we discuss below, our protest and repression variables begin in 1995; Huawei transfers begin around 2001. The five-year minimum lets us maximize country coverage while also providing ample information to the GSC estimator. Our country sample is global.

The basic functional form is:

(2) $$ {Y}_{it}={\beta}_{it}\left({\mathrm{Treated}}_{it}\right)+\delta {X}_{it}+{\lambda}_i{f}_t+\unicode{x025B} $$

where $ i $ indexes country, $ t $ indexes year, and $ {X}_{it} $ is a vector of time- and country-variant controls. The explanatory variable of interest, Treated $ {}_{it} $ , equals 1 if country $ i $ received Huawei transfers greater than some threshold $ T $ during year $ t $ or in some year since. We let this threshold $ T $ vary: from a commitment of just $250,000 in year $ t $ to transfers of $500,000, $1 million, $5 million, and $10 million in year $ t $ . From Section 3, these treatment thresholds are substantively meaningful; the median transfer value is just greater than $1 million. During the sample period, 40 countries received a transfer worth at least $250,000, 34 received a transfer worth at least $1 million, 28 worth at least $5 million, and 19 worth at least $10 million.

Our theory identifies four likely effects of Huawei technology transfers: internet shutdowns that block collective action by citizens, internet filtering that censors online content, digital surveillance, and the sort of targeted repression for online content that digital surveillance facilitates. The V-Dem project includes variables that capture each of these effects. These variables are summarized in table 1 (Coppedge et al. Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Andersson, Michael Bernhard, Glynn, Hicken, Knutsen, Marquardt, McMann, Mechkova, Paxton, Pemstein, Saxer, Seim, Sigman and Staton2022).Footnote ⁹ We rescale these variables to make them more intuitive for readers, such that a higher value indicates more digital repression.Footnote ¹⁰

Table 1

Measuring digital repression

The vector $ {X}_{it} $ includes potential country- and time-varying confounders. We control for a range of features that may reflect underlying political instability, which could be associated with both Huawei transfers and domestic repression: the number of protest and repression events as recorded by the Integrated Crisis Early Warning System (ICEWS) dataset (Boschee et al. Reference Boschee, Lautenschlager, O’Brien, Shellman, Starz and Ward2015),Footnote ¹¹ coup attempts (Powell and Thyne Reference Powell and Thyne2011), successful coups, and whether country $ i $ witnessed a presidential election, as recorded by Coppedge et al. (Reference Coppedge, Gerring, Lindberg, Skaaning, Teorell, Altman, Andersson, Michael Bernhard, Glynn, Hicken, Knutsen, Marquardt, McMann, Mechkova, Paxton, Pemstein, Saxer, Seim, Sigman and Staton2022). We control for economic features that might be associated with Huawei transfers and domestic repression: country $ i $ ’s GDP per capita and GDP in year $ t $ . We also control for country $ i $ ’s electrification rate, which might condition the capacity of recipient governments to use Huawei technology transfers for digital repression or citizens’ ability to engage in collective action. The term $ {\lambda}_i{f}_t $ represents the factor component of the model: $ {f}_t $ is a vector of unobserved common factors and $ {\lambda}_i $ is a vector of unknown factor loadings.Footnote ¹²

We estimate separate models for autocracies and democracies, which let the effect of Huawei technology transfers be a function of country $ i $ ’s preexisting political institutions.

4.4 Robustness Checks

The Online Appendix includes a series of robustness checks. First, we employ two alternative approaches to regime classification. In Sections 3 and 4, we classified regime types using the V-Dem Polyarchy index. Following Kasuya and Mori (Reference Kasuya and Mori2019), we defined the cutpoint between autocracy and democracy as 0.42. In the Online Appendix, we show that the results are substantively unchanged with cutpoints of 0.33, 0.5, and 0.6, as well as a Polity score of 0. We also employ Boix, Miller and Rosato (Reference Boix, Miller and Rosato2007)’s regime classification, a dichotomous coding of democracy based on contestation (free and fair elections) and participation (a suffrage threshold). Again, the results are substantively unchanged.

Second, we employ two alternative estimation strategies. Although we view OLS and staggered differences-in-differences as less suited than the GSC method to the characteristics of Huawei transfers, we confirm that the results are broadly similar. Following Baker, Larcker and Wang (Reference Baker, Larcker and Charles2022), we implement the staggered DiD estimator with heterogeneous treatment effects in three steps. We begin by estimating a variant of the standard staggered DiD estimator with treatment effects for each country:

(3) $$ {Y}_{it}=\alpha +{\beta}_i\left({\mathrm{Treated}}_{it}\right)+\delta {X}_{it}+{\gamma}_i+{\gamma}_t+\unicode{x025B} $$

where $ i $ indexes country, $ t $ indexes year, $ {X}_{it} $ is a vector of time- and country-variant controls, $ {\gamma}_i $ gives country fixed effects, and $ {\gamma}_t $ gives year fixed effects. The explanatory variable of interest, Treated $ {}_{it} $ , equals 1 if country $ i $ received Huawei transfers greater than some threshold $ T $ during year $ t $ or in some year since. Again, we let this threshold $ T $ vary: from a commitment of just $250,000 in year $ t $ to transfers of $500,000, $1 million, $5 million, and $10 million in year $ t $ . The subscript $ i $ on $ {\beta}_i $ makes clear that equation (3) estimates a treatment effect for each treated country. Next, we construct a contrast matrix that counts the number of periods during which each country in the treatment group is treated. This lets us estimate the cumulative treatment effect of Huawei transfers. Last, we use this estimated cumulative effect to calculate the average treatment effect per country, per period. We then use these per country, per period average treatment effects to compute aggregate average treatment effects.

Third, notwithstanding the journalistic accounts in Section 2, readers may be concerned that there is nothing distinctive about Huawei technology transfers: that the general expansion of digital infrastructure induces digital repression. To be clear, we regard this as unlikely. Pan (Reference Pan2017), for instance, finds that Beijing’s censorship capabilities are heavily driven by its ability to pressure domestic firms to remove content. The social media landscape in most autocracies is dominated by Western firms, which typically refuse government censorship and surveillance requests. Still, to ensure this general infrastructure effect is not driving the results in table 2, we identified three measures of digital infrastructure provision from the World Bank’s World Development Indicators with the same temporal scope as our sample: internet penetration, broadband subscriptions, and fixed telephone subscriptions. We find no evidence that these general measures of digital infrastructure have an effect on digital repression, which suggests that Huawei transfers are indeed distinctive.

Fourth, readers may be concerned that V-Dem’s measures of digital repression are based on expert codings, rather than direct behavior. If expert coders observe a substantial Chinese presence in country $ i $ and year $ t $ , then perhaps they assume digital repression has increased as well. To check for this, we probe whether other forms of Chinese transfers – including development aid, weapons, and overseas official finance – have similar effects on digital repression. All are drawn from AidData’s global dataset over the same time span (Custer et al. Reference Custer, Dreher, Fuchs, Ghose, Lin, Malik, Parks, Russell, Solomon, Strange, Tierney, Walsh, Zaleski and Zhang2021). We find no evidence of this, which suggests that expert coders do not observe a substantial Chinese presence in country $ i $ in year $ t $ and assume an increase in digital repression. We also confirm that our results are robust to using a different set of outcome variables: the Freedom House Freedom on the Net data, which measures obstacles to access, limits on content, violations of user rights, as well as an overall digital freedom score by country. Freedom House codes a smaller sample of countries over a shorter time span, which reduces the size of our autocracy and democracy samples. Although this forces us to shorten the pre-treatment matching period, the results with the Freedom House measures are substantively unchanged. In democracies, Huawei transfers improve digital access and content availability, leading to gains in overall digital freedom country scores. In autocracies, Huawei transfers make obstacles to access and violations of user rights more severe and overall digital freedom country scores worsen.

Fifth, we explore alternative interpretations for our results. One alternative is that democracies, which tend to be wealthier than autocracies, enjoyed better access to surveillance technologies prior to Huawei transfers, and that Huawei’s observed effect in autocracies reflects this initial difference in surveillance capacities. We show that this is unlikely to be the case. Autocracies engaged in far more digital surveillance than democracies before Huawei transfers began in the early 2000s. Another alternative interpretation is that Huawei, by providing coordination technologies, stimulates anti-regime protests, which, in turn, beget repression. We find no evidence that Huawei transfers stimulate protests in democracies or autocracies.

Finally, we employ several robustness checks tailored to the GSC estimator. Our baseline results require treated units to have at least five years of pre-treatment observations, which are used by the GSC estimator to impute counterfactual outcomes. In the Online Appendix, we specify alternative minimum pre-treatment periods: of three years, four years, six years, seven years, and eight years. The results are substantively unchanged for each. We also conduct placebo tests developed by Liu, Wang and Xu (Reference Liu, Wang and Xu2022) that are appropriate for the matrix completion method. These tests hide some observation periods before the treatment for treated units, and predict the untreated outcomes of the hidden periods with a model trained on the untreated units. There should be no differences between observed and predicted outcomes if the identifying assumptions are valid.Footnote ¹³ This is precisely what we find.

4.5 Extension: Guardrails Against Democratic Backsliding

Although we find no clear or consistent evidence that, on average, Huawei transfers have amplified digital repression in democracies, several democratically-elected governments have used Huawei transfers to erode the institutions that brought them to power (Feldstein Reference Feldstein2021). This has been especially well-documented in Zambia, where the security forces under former president Edward Lungu used Huawei technology to intercept encrypted messages sent by local journalists, track their whereabouts, and ultimately secure their arrest (Parkinson, Bariyo and Chin Reference Parkinson, Bariyo and Chin2019; Woodhams Reference Woodhams2019). In Section 2.3.2, we identified several constraints that may prevent democratically-elected governments from using Huawei transfers for digital repression: countervailing political institutions, free and fair elections, independent media, and vibrant civil societies. Ascertaining which of these is most important is beyond the scope of this article, but our data and estimation strategy let us suggest possibilities for future research. We exploit the fact that V-Dem includes measures of the subcomponents of democracy. We then estimate a series of staggered differences-in-differences estimators,Footnote ¹⁴ each of which includes an interaction term that lets the Huawei effect vary according to a V-Dem subcomponent.

The results, which appear in the Online Appendix, are noteworthy for two reasons. First, three potential guardrails – a vibrant civil society, the rule of law, and a free and independent media – appear to offer protection from digital repression in the world’s democracies. When citizens are organized, enjoy access to well-functioning courts, and are kept informed by a free media, democratically-elected governments are less likely to use Huawei transfers for digital repression. Second, across potential guardrails, weakly constrained democracies are more likely to impose internet shutdowns and monitor social media than arrest citizens for online content or filter the internet. Put differently, internet shutdowns and social media monitoring are the most common forms of digital repression where the guardrails against democratic backsliding are relatively weak. This, we believe, makes sense. Social media monitoring is less invasive, less conspicuous than arrests and internet filtering, and so easier for democratically-elected governments to employ. Likewise, the taboo against internet shutdowns seems to be weakening, which may free democratically-elected governments to employ them as well. In 2023, for instance, Access Now documented 283 internet shutdowns across 39 countries, up from 78 shutdowns across 27 countries in 2016.Footnote ¹⁵ In many years, India, the world’s most populous democracy, has been the most frequent offender.