Appendix 1: Methodology

Data selection

To identify a representative subset of apps indicative of the implementation of China’s RNR-related access barriers and their impact on information exchange between users in China and overseas communities, we relied on a dataset by app store intelligence provider Appmagic. Appmagic’s data covers monthly downloads of apps across 60 diverse economies, encompassing 93% of global app downloads.¹ Our data ranges from January 2015 to December 2023. Recorded downloads contain both the Google Play Store and Apple’s AppStore downloads.²

Appmagic manually categorizes apps into 15 main categories, including “Social,” “Games,” and “Finance,” as well as a plethora of subcategories, such as “Chats,” “Puzzles,” or “Money Transfer.”³ We used the labels provided by app stores in a minority of cases where Appmagic labels were missing.

Our query returned a subset of 247 relevant apps, which:

• were developed and operated by a company headquartered in the People’s Republic of China;
• were ranked within the top 10,000 in terms of national downloads in of any of the 58 app stores (filtering out the app stores of China and India);
• were in the main category “Social”;
• were in one of the sub-categories “Chats,” “Messengers,” or “Social Network” within the main category “Social” (omitting sub-categories, such as “Dating,” “Account Status App,” and “Other”).

Appmagic’s download numbers are based on a relational approximation of ranking data. Appmagic assumes this technique provides an average discrepancy of about 10%, with a decline in accuracy at lower download numbers. To maintain accuracy in our quantitative analysis, we dropped apps below a suggested threshold of less than 30,000 global downloads per month.⁴

This curated dataset of apps was further refined to generate a representative group of apps which allowed for the recording of access barriers. We dropped 31 apps, such as “Tencent News” or “Himilaya Life,” as they had ceased operations. Another 26 were also disregarded as they only offered social media adjacent services, such as “5000+ Emoji” or “Sparkling Heart Keyboard Theme.” This curation resulted in a final list of 62 apps.

App walkthroughs to record type #1 and type #2 access barriers

Data on app store availabilities by AppleCensorship.com (a project that monitors and provides a publicly accessible database of the availability of apps across app stores worldwide) provided evidence for type #1 access barriers.

App walkthroughs on apps that were still in operation revealed the extent of type #2 barriers. In a first round of app walkthroughs, we determined which method of account registration was required by each app. We tested on both iOS and Android devices, as well as the nascent Play Store and App Store. To circumvent local biases, we tested on changing locations in EU, North America, and Africa. As noted in Appendix 4, signing up for or engaging on an app was only possible after providing a phone number and subsequent verification codes in 39 instances. In 22 instances, setting up an account was permitted based on providing an email address, which allows for anonymous use and access.⁵ In one instance (“MOGU – Fashion Destination”) the provision of further identification was also required.

To record phone-number-related access barriers, we conducted a second round of walkthroughs on the remaining 39 apps, which took place in February 2024. To minimize human error, we automated the process of recording available phone number prefixes by running an optical character recognition program on screen recordings of the app walkthroughs. Overall, we found 664 instances where an app was inaccessible to local users out of a total of 1,914 tested instances. This figure includes 375 phone number barriers, 231 app store barriers, and 58 instances of both.

App ID-resolution

In most instances apps had one universal ID across app stores. Seven apps had more than one ID, where we used the ID of Chinese original app (QQ邮箱, Tencent Conference, Baidu Tieba, DingDing, Kuai Shou, QQ Browser, and Weibo).

Survey

In March 2024, we conducted an online survey to gather user perceptions of foreign-directed access barriers for Chinese-designed apps. The survey was available in both English and Chinese, and participation was voluntary and anonymous. We specifically targeted Chinese overseas communities for dissemination. The survey received 65 responses, with a high response rate of 98%.

Notably, the survey is non-representative and may come with inherent biases given that it was only disseminated to members of the at-risk community. How accurately the results reflect the experiences of other overseas users is not reflected in the survey (it could be argued that members of this community are more sensitive to access barriers than others). Members of the at-risk community may also face barriers due to individual exclusion. The results of this survey should thus be treated as indicative evidence, not as fact.

Data analysis

To gauge the impact of access barriers on the download trajectories of affected apps, we tagged all instances where, as of March 2024, an access barrier was recorded. A data series was therefore only tagged when the access barrier matrix displayed a value of “1,” “2,” or “3” in a given country. In the time series data of app downloads, we then separated those instances from instances where no access barrier was recorded. This allowed us to isolate the different groups of restrictedness. We extended our analysis beyond the 33 apps within the controlled grey zone internet and included six additional apps in our analysis which permitted no exchange of information (see in Appendix 3). This alteration did not measurably skew the data (<4% of downloads of the entire sample), but rather provided more analytical depth.

Statistical testing

To examine whether the proportion of apps facing access barriers (either from app stores or phone compatibility issues) is appreciably different between countries with varying Chinese diaspora populations, we used a one-tail significance test to determine if larger diaspora countries have a higher proportion of apps with access barriers compared to countries with smaller diaspora populations.

The dataset for this analysis included annual app download data, which was merged with diaspora sizes for each country. Apps were categorized by the type of barrier they faced, and we calculated the proportion of apps with barriers in each country.

Using these proportions, we performed a one-tail Z-test to statistically test the hypothesis that larger diaspora countries have a higher proportion of blocked apps. This involved comparing the mean proportion of blocked apps in countries above and below the median diaspora size.

The results showed a Z-score of 6.227 and a P-value of 0.000, indicating a statistically significant difference. This implies that countries with larger diaspora populations do indeed have a higher proportion of apps facing access barriers compared to countries with smaller diaspora populations.

Global population data

For global population data figures, we primarily relied on the World Bank’s population data.⁶ An average annual growth rate for each country was calculated based on yearly percentage changes from 2015 to 2022. The estimated population for 2023 was then calculated using the 2022 population and this average growth rate, with the estimated values added to the dataset. Data for Taiwan, which was missing in the original data, was pulled from the National Statistics Bureau of the Republic of China (Taiwan).

Diaspora data

For diaspora data, we used a dataset that was initially populated with available data from the OECD, transferring data for each country and year directly from the OECD records. For seven countries with incomplete OECD data (Canada, Chile, Greece, Ireland, Mexico, New Zealand, and Portugal), averages were calculated using both OECD data and UN data from 2015 and 2020, filling missing values with these averages. For 26 other countries, missing data was estimated based on the rate of change between UN data from 2015 and 2020. The 2015 values were filled with UN data, and yearly changes were calculated to interpolate values from 2016 to 2020, extending this trend to estimate values for 2021 and 2022. Values for 2023 were estimated using a Simple Moving Average of the previous three years (2020, 2021, 2022) to ensure consistency across the dataset.

Notably, this data represents a low-end estimate of diaspora communities since the UN data only refers to first-generation migrants. The actual diaspora groups at risk may be four times as large or more, if second and subsequent generations of the original cohorts of migrants from China are included. In particular, Indonesia, Malaysia, the U.S. and Thailand may host considerably larger groups of individuals identifying as Chinese or dependent on communication with people in the PRC.

Appendix 2: Regulatory and legislative framework of China’s Real-Name Registration system (实名制[登记]).

Relevant laws, regulations, speeches, and other party documents issued by institutions such as the CAC, the State Council, or the National People’s Congress

Appendix 3: Extended Access Barrier Matrix

Appendix 4: Recorded account registration requirements by app