Doing Ethnography on Social Media: A Methodological Reflection on the Study of Online Groups in China

This article draws on the two authors’ extensive fieldwork experiences in studying Chinese feminists and lawyers on social media to offer some thoughts on how to conduct qualitative research in the digitalized world. We argue that qualitative methods such as participation observation, in-depth interview, and textual analysis can provide thick descriptions and deep, localized knowledge of social processes that go far beyond the sketches of Big Data. Social science data collection and analysis on social media need not only Big Data’s bird’s-eye view, but also the day-to-day ethnographic immersion—“living on the sites” and interacting with research subjects over a long period of time. The rise of social media has not changed the basic principles of doing ethnography, such as the importance of immersion and reflexivity. Nevertheless, ethnography of online groups presents new challenges and opportunities in terms of accessing field sites, analyzing ethnographic data, and research ethics.

Social science is embracing the era of social media. Although gaps of digital devices ownership and internet access still widely exist within and between countries, social interactions in the early 21st century are increasingly mediated by the internet. In addition, many people's social networks are formed and maintained through multiple digital sites. In 2016, 68% of adults in China owned a smartphone and 60% of adults used online social networking sites; in the United States, these two numbers were 72% and 69%, respectively, in the same year (Pew Research Center, 2018, pp. 27-28). Now, social media platforms can trace almost every digitalized aspect of social life: from sending out party invitations on Facebook, tweeting frustration about current events on Twitter or Weibo, setting dating preferences on Ok-Cupid, attending rallies on Zoom, to mobile-payment for groceries through Apple Pay or WeChat, and so on. This online documentation of social interactions is getting more systematic and massive over time as more people actively use social media on a daily basis. 1 Accordingly, how to conduct empirical research on social media has become a principal methodological question for social science disciplines and interdisciplinary fields alike.
The digitalization of social life has generated a new cultural and technological phenomenon: "Big Data." Digitalization is a process of "converting information from analog into discrete units of data that can be more easily moved around, grouped together, and analysed" (Daniels et al., 2017, p. xviii). The "data science" industry prizes massive computer-mediated data and the technological tools for collecting and analyzing such data with an aura of objectivity and accuracy (boyd & Crawford, 2012;Evans & Aceves, 2016;Golder & Macy, 2014). This approach collects and analyzes human actions through synthesizing social media contents, website metadata, logs, cookies, transactions, website analytics, behavioral data with wearable devices, and so on. Comparing with traditional survey methods, some researchers consider the Big Data approach to be advantageous for evading certain biases (e.g., recall bias and nonresponse bias), reducing respondent burden, improving turnaround time, and facilitating serendipitous finding (Callegaro & Yang, 2018). Increasingly, it becomes trendy to use various forms of 1014610Q IXXXX10.1177/10778004211014610Qualitative InquiryWang and Liu Big Data research in today's social science. The quantitative push in social science is certainly not new for qualitative researchers (Small, 2009). Yet, in the light of Big Data's promises of easy access to large populations and capacity to rapidly process data of sizable magnitudes, many researchers feel obligated to incorporate Big Data in their work.
Nevertheless, by simply making it "bigger and faster," the use of Big Data does not guarantee good science. Social scientists' caution and excitement over Big Data (Bail, 2014;boyd & Crawford, 2012;Daniels et al., 2017) are reminiscent of how feminist theorists tackled science and objectivity in the 1980s. When applied to the study of social media, digitized data creates the illusion of a holistic and infinite view of the social world, which Donna Haraway (1988, p. 581) would criticize as playing "the god trick." In her groundbreaking essay "Situated Knowledge, " Haraway (1988) emphasizes the importance of going beyond criticizing "science" as gendered and "objectivity" as manufactured knowledge. Instead, she urges for a strong objectivity which embraces researching as embodied observations at limited locations that produce situated knowledge (Also see Bhavnani, 1993, for the "partiality of theorizing"). Only when reflectively connecting several partial yet accountable perspectives, we as a community of social scientists can get closer to a strong objectivity (Harding, 1987(Harding, , 1992. By the same token, social scientists nowadays must continue to acknowledge the limit and partiality of what we are capable to know about the digital social world. To be sure, it can be a respectable goal to access sizable samples for achieving more comprehensive views as many scholars in Big Data research do. Yet, to achieve strong objectivity, social scientists of all methodological orientations should reflect on how to narrate and theorize our location-and-context-specific knowledge of the internet. One motivation of this article is precisely our caution over the glorification of the Big Data approaches to social media in social science research. This glorification risks "losing the trees for the forest" and contributes to the marginalization of conducting ethnography and qualitative data analysis on social media. As a methodology, ethnography enables us to observe social interaction in specific time and space, as well as to explain the complex mechanisms and processes that connect people and produce institutions (Desmond, 2014;Jerolmack & Khan, 2014;Small, 2009Small, , 2013. Online ethnography brings this long-standing methodological tradition to the age of the internet and social media (Hallett & Barber, 2014;Hine, 2015). Using ethnography, recent qualitative studies have explored a variety of topics such as how people adapt emails into communication (Menchik & Tian, 2008), how identities and intimacy are negotiated in dating chatrooms (Darwin, 2017), and how collective action is mobilized online (Käihkö, 2018;Postill, 2014;Yang, 2003), to name a few. Ethnographers have developed rich theories on how to select research subjects and sites, be it through considering times and spaces (Abbott, 2001;Bourdieu, 1992), intersectional groups, processes and systems (Choo & Ferree, 2010), social relations and interactions (Desmond, 2014;Emirbayer, 1997), and so on. Yet, online ethnography further unsettles these methodological debates within the ethnographic tradition, especially regarding the selection of research sites, the access and analysis of data of various sources, and the reflection on research ethics.
In the increasingly digitalized world of the early 21st century, online groups and cyberspace are co-constituted with multiple mediums and complex interconnectivity among users and groups. This makes it even harder for researchers to set a spatial and/or a temporal boundary on her research subjects (Hallett & Barber, 2014;Hine, 2015). In addition, ethnographers sometimes raise concerns of the reliability and validity of social media data. As social media change the nature of people's communication, how sustainable is a researcher-informant relationship built online and how reliable is a person's online profile for predicting her offline actions (Käihkö, 2018)? How should a researcher crosscheck and analyze an online discussion with a face-to-face discussion (Beneito-Montagut, 2015;Lane, 2016)? Finally, how would digitalized communication brought by social media affect the researcher-informant power dynamics, as now informants have greater capability to access and respond to a researcher's findings through the internet (Reich, 2015)? Methodological reflections on the challenges and opportunities of online ethnography remain scarce, particularly in authoritarian contexts. It remains an open question how qualitative inquiry can resist the lure of Big Data and provide alternative approaches to make effective use of social media in social science.
In this article, we draw on our extensive social media fieldwork experiences in studying the activism of Chinese feminists and lawyers to offer some thoughts on how to conduct qualitative research in the digitalized world, especially in politically sensitive contexts. We argue that qualitative methods such as participation observation, in-depth interview, and textual analysis can provide thick descriptions and deep, localized knowledge of social processes (Geertz, 1973) that go far beyond the sketches of Big Data. Social science data collection and analysis on social media need not only the bird's-eye view provided by data scientists, but also the day-to-day ethnographic work of "living on the sites" and interacting with research subjects. China is an excellent site to explore the perils and pleasures of online ethnography for two reasons. The first is its large number of "netizens." The second is its omnipresent internet censorship, which makes qualitative data collection and analysis on social media a delicate, sophisticated, and sometimes risky cat-and-mouse game. This is especially true for researching on sensitive populations such as feminist activists or human rights lawyers.
In the following pages, we first provide an overview of the opportunities and challenges of using social media as ethnographic sites. In the next two sections, we use our own fieldwork experiences to discuss three major issues including collecting and analyzing social media data; accessing the online field sites, ethics, and reflexivity; and reassembling the data in analysis and writing. To collectively examine our research experiences, the two authors interviewed each other as a reflexive methodological experiment in July 2018. Some first-person quotes from the interviews are used to better illustrate our arguments. The concluding section returns to the debate with the Big Data approach and reassesses the value of qualitative research in the age of social media.

Social Media as Ethnographic Sites
Social media-fueled research opportunities come with challenges for social scientists. First, social media constitute a cyberspace with fluid boundaries and fragmented information. Most internet users now have the ability and potential to create a message through user-generated content (UGC) vehicles and distribute it through blogging platforms, social networking sites, video-sharing venues, and photo-sharing sites (Earl & Kimport, 2011;Gerbaudo, 2017;Sobieraj, 2011, p. 169;Teng, 2012;Valenzuela, 2013). On one hand, all those sites empower people to collectively share information to support their demands for social change. This is especially critical for people who have been historically marginalized or left out by data gathering institutions. Empirical studies have shown that marginalized social groups (e.g., African Americans) increasingly rely on social media to identify patterns of inequality, distribute progressive messages, and mobilize for collective action (e.g., the #BlackLivesMatter movement; see Byrd et al., 2017;Stout et al., 2017). On the other hand, it is difficult for researchers to clearly identify the research population or precisely set the boundaries of a research site on social media. Instead of geographic or institutional boundaries, it is hyperlinks and shared user profiles that connect social media platforms. Social media users, as well as their locations and shared contents, can change within seconds.
Second, social media platforms are products of the social and political status quo and often operated for economic or political interests. As a result, their algorithms can influence or even modify user behavior. Corporations, government agencies, and other resourceful institutions appropriate technological developments, including Big Data, in favor of their own interests. These institutions gradually use Big Data algorithms and models to replace humans in organizational decision-making, from optimizing the highest returns for an advertisement, identifying health insurance fraud, to defining the quality of time spent on social media. In many recent cases, computational sorting and monitoring systems are found to further segregate and marginalize historically disadvantaged communities and reinforce social norms that are inherently racist, sexist, or homophobic (Eubanks, 2018;Scannell, 2018;Tsika, 2016). Furthermore, digital service sites also provide the state more advanced technological instruments (e.g., facial recognition and photo-search technologies) for the purpose of control and repression. For example, such technologies assisted the Iranian government in marking and searching for participants in the Green Movement, a political uprising after the 2009 Iranian presidential election (Morozov, 2011, pp. 153-154).
For China studies, Big Data techniques have brought excitement as both data quantity and quality improve in recent years with China's rapid advancement in web technology, but its application has been complicated by censorship and the difference of language structures between English and Chinese (Zhang, 2018). The Chinese state's information control has intensified the fluid and fragmented nature of social media. The Chinese censorship system not only filters public information through computation and human reading (King et al., 2013;Yang & Wu, 2018) but also fabricates hundreds of millions social media comments every year to distract the public from criticisms of the regime (King et al., 2017). In addition, the internet usage in China is disciplined to be highly individualistic through the state's extensive and repressive responses to any collective action online or offline (e.g., D. Fu & Distelhorst, 2018;Mackinnon, 2011;Tsai, 2016). Consequently, activists in China organize their actions as decentralized and sporadic as possible and avoid posting sensitive messages that are easily detectable by computers. 2 A successful social movement sometimes requires intentional disguise of the organizers behind the scene (D. Fu, 2016). Therefore, even after attaining massive amount of social media content, it remains difficult for researchers to locate key action platforms and identify potential informants.
The Chinese state's capacity of information control makes scholars in China studies intimately familiar with the second challenge of doing research online discussed above, that is, social media platforms are products of the social and political status quo and often operated for economic or political interests. The "China" on the internet is only a title slide of its multilayered, complex reality. What is most accessible online is the state-filtered information, including propaganda materials that the state uses to signal its control over society and to induce self-censorship among citizens (Huang, 2015). For the study of Chinese society, rare cases, which would probably be dismissed as outliers in Big Data research, sometimes are the most valuable data that have successfully escaped the vast net of censorship. By closely observing or actively participating in online actions in real time, researchers can get a good sense of not only notable events but also those events that have been censored or even coercively erased (Yang & Wu, 2018).
Those challenges for social science research brought by the rise of social media are profound. Yet, the Big Data approach still has a long way to go in confronting them, especially if it routinely ignores the inherent ambiguities and inequalities in data collection and analysis in the name of large sample size and high-tech objectivity. By contrast, online ethnography, either used alone or combined with other qualitative or mixed methods, can help address some of the concerns discussed in this section. With its highly reflexive and culturally sensitive nature, ethnography equips researchers with contextual knowledge to every social interaction of interest. Ethnography for the internet should be embodied, embedded, and every day (Hine, 2015). This requires not only deep ethnographic immersion on social media sites but also making delicate methodological decisions to overcome challenges in sampling, interacting with informants, storytelling, or even choosing applications and software. The rest of the article discusses these issues in detail and illustrates them with six stories from our online fieldwork experience.

Living Online, Living Onsite: Access and Ethics of Online Ethnography
Online ethnography is qualitatively different from aimless online browsing, but its beginning is often similar to accessing any website or social media platform. Ethnographic immersion for social scientists involves observing other people as they respond to social interaction, as well as experiencing the events and interactions oneself (Emerson et al., 2011). Thus, the researcher must spend a substantial amount of time on the site on a regular (often daily) basis and familiarize herself with its users and discourses. This process of online ethnographic immersion is crucial for developing the researcher's identity in the online group and getting access to potential informants. It resembles traditional ethnographic immersion in many ways, yet there are also notable differences. In this section, we use our own fieldwork experiences to discuss the issues of access and ethics in "living online." The first author started to observe feminist online actions in China since 2012 and then to participate in some actions since 2013. Through active daily interactions on different social media platforms, her participation as a commenter led her to a chatroom of @FeministVoices (nüquan zhi sheng), a Weibo account established by the Media Monitor for Women Network (MMWN) in 2010. @FeministVoices was the largest grassroots feminist media outlet in China with 181,019 followers when it was forcibly shut down on March 8, 2018, the International Women's Day. 3 Because of the time difference between China and the United States (where the first author was based), she altered her daily schedule to maximize interactions with Chinese feminists in the chatroom, from staying up late at night to checking hundreds of messages every morning. As a result, she was able to form close connections with many activists in chatrooms before her first field trip in China in 2014. The following is her story of how an online chatroom became a network of informants: In the summer of 2013, a Chinese feminist activist I met on Weibo added me to a QQ chatroom for @FeministVoices readers, right around the time when the U.S. Supreme Court ruled the Defense of Marriage Act unconstitutional. Back in 2013, @FeministVoices only had about 16,000 Weibo followers and its reader chatroom had fewer than 300 people, but the account was at the frontline of reporting on feminist activist actions and leading debates on gender issues in China. When I joined, they were having an enthusiastic discussion over law and marriage drawing on personal stories and political campaigns from all around the world. Several prominent feminist activists whom I only knew through reading news and their Weibo posts shared their experiences of campaigning for the anti-domestic violence law in China, as well as their individual politics on marriage and family. To my (joyous) surprise as a young queer woman, many of them were also LGBTQ identified. Over the past five years, my participation in activist chatrooms changed from QQ to WeChat and later to encrypted messenger apps like Telegram. On numberless nights, I fell asleep during a heated debate, waking up still holding my phone and immediately checking the hundreds of messages that I missed. Chatrooms like this became my "neighborhood" where I ran into young Chinese feminists of different genders, sexualities, and geographic locations.
The second author's first encounter with online lawyer activism echoes the first author's account above. As a graduate student interested in studying the Chinese legal profession, he began to regularly visit the All-China Lawyers Association's (ACLA) online forum in 2003, shortly after the forum opened on the ACLA's official website. After actively participating in the forum discussions using a pseudonym for about a year, he became the board manager of one of its discussion boards, "Jurisprudence and Constitutionalism" (fali xianzheng). This new role gave him access to not only regular discussions but also the forum's recycle bin, which contained the deleted messages, including many politically sensitive ones. The following is his account of this early experience and its impact on his research: In 2002In -2003, when I first started to look for information on Chinese lawyers, there were very few online forums focusing on lawyers. The ACLA forum was a rare site where ordinary lawyers across China could exchange ideas and discuss their problems in practice. At that time most online forums and bulletin boards in China were anonymous, which is very different from the age of Weibo and WeChat. Most lawyers used pseudonyms and discussed things quite freely. At first the forum didn't even have a recycle bin for sensitive messages. For me it was an eye-opening experience because the forum allowed me, an overseas student thousands of miles away, to follow what ordinary lawyers in China were talking about on a daily basis. Later on, I used some of the discussion threads that I collected on the forum as empirical data in some of my writings. But now looking back, I think the most valuable part of my participant observation on the ACLA forum is not data collection, but the possibility of immersing myself in the everyday lives and discourses of many ordinary Chinese lawyers. The problems that lawyers on the forum discussed are very different from the experiences of my former classmates at Peking University Law School, most of whom are corporate lawyers working in shiny office buildings in Beijing.
As our early field experiences suggest, social media, be it a chatroom or a forum, can provide a researcher instant access to the social networks of potential informants, as well as details about their demographic information and political views. This information is critical, especially for researching on under-studied or sensitive topics, for which the population parameters and issue areas have not yet been clearly defined. Through the instant access of a network of potential informants rather than "snowballing" from only one or two informants, the researcher can make more informed decisions on which issue areas to focus on and whether to emphasize certain characteristics of research subjects in sampling and case selection. Even so, information from online chatrooms or forums is often incomplete or even scattered. Sometimes basic demographic information such as gender, age, or geographic location can be ambiguous or missing. In comparison with traditional ethnography, access in online ethnography presents a trade-off between widening research population and deepening information for each informant in that population. Only after a long period of immersion can the researcher gradually assemble the basic profiles of her research subjects.
Although Big Data scientists are usually invisible to their research subjects, online ethnographers are visible and accessible to their informants once they are connected on social media platforms. Consequently, when studying feminist or lawyer activists, social connections and networks are leverages for both researchers and activists. On one hand, researchers can observe more complex dynamics and narratives of the movement through accessing activists' social networks. Snowballing often gets easier through online social networks than in traditional offline fieldwork. On the other hand, when activists mobilize their social connections for collective action, they can also access the researchers' networks for resource mobilization (Cress & Snow, 2000;Edelman et al., 2010;Edwards & McCarthy, 2007). This reciprocal nature of their interaction complicates another classic question for ethnographers-to what extent should you "go native" and become one of them? Both authors faced this question repeatedly in their online fieldwork over the years, which is further complicated by the risky and unpredictable nature of political mobilization in the authoritarian context of China. The first author recounted her participation of combating sexual harassment campaigns in China as follows: As I was packing for my trip to the 2018 winter meeting of Sociologists for Women in Society (SWS), a Chinese feminist activist asked if I wanted to help collect signatures of a petition on combating sexual harassment, which was addressed to members of the National People's Congress, the Ministry of Education, and the heads of all universities in China. I immediately noticed the direct language used in the petition and was even slightly concerned about the risks of using my real name. Unnecessary as it sounds now, at that moment I was back and forth in my head about how my signature would affect my upcoming trip to China. About an hour later, I logged in my school email and sent out the petition recruitment to all the oversea Chinese professors I know. This was not the first time I was approached for such initiatives. During my 2014 summer fieldwork in Beijing, I was asked to take the lead in a campaign to call my college alma mater for accountability over sexual harassment cases. I hesitated-conscious of my responsibility to the cause, but overwhelmed by potential risks. After a long discussion with the campaign organizers, we concluded that it would be better to choose someone who graduated from my college and did not have close family members living in the same city. I am not sure what my decision will be for the next time. But I am grateful for the energy and trust that the activists took to guide me through the complicated decision-making process from security to strategy.
For scholars working on politically sensitive topics, the question of "should you become one of them?" is a constant reminder of how complicated social change and the politics around it are, especially in an authoritarian context. The process of pondering on this question offers the researcher a reflexive experience that helps to contextualize her frontstage action with backstage strategies (Goffman, 1956(Goffman, /1959(Goffman, , 1974(Goffman, /1986) as a participant observer. In traditional ethnography, the in-group and outgroup boundaries are relatively clear. On the contrary, online ethnography can often give the researcher instant access to group membership, but it can also generate an identity crisis. Once a member of a social media group, it is no longer possible to completely retreat to the seemingly objective standpoint of a social science researcher, at least on the frontstage of online interaction. Even after the fieldwork is completed, group membership and solidarity remain, unless the researcher withdraws herself from the social media platform. Some researchers use multiple Facebook or WeChat accounts to separate the identity in the field and the identity in the ivory tower, but this is not always convenient or even feasible. More often, online ethnographers struggle with this question in the whole research process and make decisions on a case-by-case basis, as the first author did in the example above.
The second author's encounter with this reflexive question has an additional layer of complexity. Although he is not a licensed lawyer and thus cannot directly participate in legal cases as the first author did in feminist actions, over the years he has accumulated a reputation as a scholar who studies and writes about the Chinese legal profession. Consequently, his lawyer informants sometimes would approach him to seek his voice as a public intellectual in support of their activism. He gave the following account of how he dealt with such requests: Scholars like me who study politically sensitive topics in China like lawyers and human rights have to walk a fine line between an objective social scientist and a visible public intellectual. For example, in 2012 or 2013, a notable human rights activist in Beijing (who was detained later during the 709 Crackdown on activist lawyers in July 2015 4 ) sent me a private message on Weibo and asked if I would be interested in meeting with him and discussing possible collaboration. It put me in a real dilemma because, while I very much would like to help him, an in-person meeting with him might put both of us in trouble with the state security agents. So I decided to give him a polite reply online but declined the meeting. I was also often asked by activist lawyers to make public statements on Weibo or WeChat to help their ongoing cases, and I did actively make such online statements in a few critical cases like the Li Zhuang case in 2009-2011 5 or after the 709 Crackdown in 2015. But if I made my voices heard in public in every case they were doing, not only would my personal safety be at risk in China, I would also lose my objectivity as a social science researcher. This is a constant struggle for me all these years in studying lawyers and political mobilization. What the second author described above is not only an issue of self-censorship in an authoritarian context, but also the difficult personal struggle of a social scientist who deeply cares about his research subjects yet constantly feels the limit of his own capacity in supporting them in their everyday practice. This leads to a second question: "What would you do if you could not become one of them?" Arguably, researchers can use their scholarly writings to expose the problems and risks that activists face from the authoritarian state. For instance, the second author's book on Chinese criminal defense lawyers  is considered "an act of solidarity" for activist lawyers by one of its reviewers (Stern, 2018: 278), though as a piece of scholarly work it still maintains a disciplined distance required by the objectivity of social science writing.
But researchers can certainly do more than scholarly writing and publishing, especially in the age of social media. Posting on public platforms like Twitter and Weibo or semi-private platforms like Instagram and WeChat is often an effective means to assist activists in their collective action. In the authoritarian context of China, such posts require a combination of courage and delicacy, and it is ultimately the personal choice of every researcher on the best way to handle it. Both authors have written many online posts and essays to support their research subjects over the years, but they have also refrained from doing so in many challenging situations. As the Chinese state increases its repression on social media in recent years (Cairns & Carlson, 2016;Pan, 2017;Tan, 2017), public support of social and political movements have become more precarious for social science researchers than before. Even researchers who do not write in Chinese could face potential backlashes and legal consequences (Greitens & Truex, 2018). How to engage with research subjects via social media in this new era of internet censorship in China? It remains an evolving tough question.

Reassembling the Data: Analyzing Fragmented Online Information
From the first research idea to the final book publication, an ethnographic project can take many years to complete. Researchers and informants have gradually established relationships, sometimes weak ties, through researchrelated interactions, and these ties would evolve over the period of the research project and beyond. The rise of social media makes it easier to sustain interactions between researchers and informants beyond a one-time interview or limited face-to-face interaction. Whenever connected online, the researcher and the informant can read and comment on each other's life events instantly and usually across different social media platforms. In this manner, social media help researchers to stay informed about the development of social issues of interest and provide them opportunities to have interpersonal interactions with their informants beyond traditional research settings.
Although Big Data analysis provides a bird's-eye view of the macro patterns of social media interactions with great efficiency, online ethnography allows the researcher to get to know an online group from day-to-day digitalized communication. It is through this communication that a person or a loosely connected online community comes to understand themselves and form their group identity. It is precisely through being in constant communication with Chinese feminist activists since 2012, the first author came to understand and experience how repression and resistance work for activists on a personal level. Social media enable her to observe these challenges activists face while she is not present in the same geographic location. Ethnographic immersion becomes even more critical for accessing these accounts of challenges when they are the target of censorship. As the first author recalled how she learnt about a lesbian activist couple's eviction stories: It was a hot summer day in 2016. I was on a bus from the Qingcheng Mountain back to Chengdu where I took a short vacation trip to escape from the heat in Guangzhou. Unexpectedly, a WeChat message popped up on my cell phone screen that two of my activist friends, a lesbian couple, were evicted by their landlord in Guangzhou. In their late 20s, these two young women were leading advocates against sexual harassment nationally and community organizers locally. Every time I visited Guangzhou, they opened their home to me. I was right there about two weeks ago, and I still remembered they talked about using their big living room space to host local LGBTQ and feminist events. The WeChat post I received was written in a lighthearted tone but detailing their interactions with the police and the landlord. They were asking on WeChat for donation as well as selling their stuff to downsize for future "moving." In less than a day, the post was censored. About a year later, at a small celebration for my birthday in the United States, I saw another message of their eviction. It was the third time in the past five months that they were asked to move out of Guangzhou by the police. One of them said it was because of their involvement in anti-sexual harassment campaigns. Frustrated as one could imagine, these two activists glued articles of China's Constitution on their front door to remind the police of their rights. Of course, this did not stop the eviction. They were soon forced to move again.
The Big Data approach can obtain massive information that exists on the internet, but it can easily fail to capture stories and their social contexts which are actively "evicted" from the online space by powerful institutions and state agents, such as the story quoted above. On the contrary, qualitative methods equip researchers with delicate skills in connecting dots of critical information. This is particularly helpful when the digitalized communication of interest is filtered and intentionally isolated by state censorship. By immersing in online activist communities over a long period of time, the first author becomes an agent of memory (Neiger et al., 2011;Yang & Wu, 2018) who not only remembers these stories but also situates the decentralized, sporadic, and often disguised accounts back into the social context as she witnesses them. In this sense, a social scientist is a reflexive agent of memory for online groups because she not only documents their social media interactions but also analyzes and reflects upon them by linking these interactions to her own personal experiences online and offline.
In another chatroom in which the first author participates, feminist activists literally share their dreams as collective-care practices. During a long period after the Feminist Five detention in March 2015 (Fincher, 2016;Z. Wang, 2015), all that they shared were nightmares about the police, eviction, censorship, family pressure, and so on. Since 2012, the first author has witnessed what being a feminist activist means for young Chinese people of her generation, the "One-Child Policy Generation" (Greenhalgh, 2008;D. Wang, 2020), from multiple aspects of their lives. To some extent, her own experience as a young Chinese feminist has always been a part of this larger narrative about young people and social change in early 21st-century China. Guided by feminist strong objectivity (Bhavnani, 1993;Haraway, 1988;Harding, 1987Harding, , 1992, the first author embraces her embodied observation of feminist activist communities and reflects on her own sociopolitical positions as she continually peruses and reflects upon feminist and other social theories-ones that are informed by indepth understandings of on-the-ground practices.
Nevertheless, having personal experiences as an insider does not guarantee that a researcher would produce the best theory about the community of interest or beyond. The art of doing social science requires the researcher to find her best position to observe and participate in the online community as well as a particular writing style to tell the story. The fragmented and often unbounded nature of social media interactions raises ethical questions of confidentiality and accuracy in representing each informant's stories. On one hand, the pain and pleasure of sharing personal details characterizes social media interactions, which can help ethnographers understand their informants from multiple aspects and through crisscrossing boundaries. For example, a researcher may share membership with an informant in one chatroom for a rescue campaign for a detained activist lawyer, as well as another chatroom for lesbian parents of rescued cats. On the other hand, when online activities over time are intentionally collected and documented, a researcher can have a detailed profile of an informant's life, which can be consequential for this informant's privacy and safety. To be sure, one of the most compelling components of qualitative research is telling stories, especially ones with vivid details and characters. Although it is tempting to do so, a researcher should always consider the social and political consequences of her writings and strictly protect the informant's personal information according to the ethical requirements of social science research.
To deal with such ethical challenges, in the process of his data analysis and writing on activist lawyers, the second author often adopts an analytical way of storytelling, which integrates interviews, online observation, and other empirical evidence from multiple informants in different geographic locations to make one analytical point. This writing style allows him to fully protect the identities of informants while presenting a relatively comprehensive picture of their experiences. The following is an example that he gave on how to weave the online and offline data into a web of anonymized yet analytically interconnected accounts: When I interview lawyers, I always try to ask their biographies in detail, because I find the early life history of a lawyer not only fascinating in itself but also very helpful for understanding her law practice. A lawyer once told me, the reason he became a human rights activist was that his parents were persecuted during the Cultural Revolution. When others helped their family in that difficult time, as a child his heart felt warm. So now he hopes to use his activism to warm the hearts of others. This kind of in-depth personal stories are hard to get from the fragmented interactions on social media. Some activist lawyers have a tough and courageous image online but, if you get to know them offline, you see a totally different side of them. They would play soccer with their children or make dumplings with their spouses. When I was writing my book, I seriously thought about presenting a few lawyers' life stories in a holistic fashion as case studies, because the stories were so powerful and the lawyers could speak for themselves, but then I quickly realized that it would not be possible to do that without revealing their identities. Finally, I decided to put their biographical accounts from interviews and the ethnography of their online interactions into different sections and chapters. Within each section, I used the similar or comparable experiences of several lawyers anonymously to make the same analytical point, say, how they were harassed by the state security or disbarred by the justice bureau. As a result, readers do not get the coherent life history of any of the lawyers I discussed in the book, but adding the analytical points together, they can still get a pretty good picture of what happened to them as a group. All the bits and pieces were reassembled in writing.
There are disadvantages of "reassembling the social" (Latour, 2005) in this manner, however. The beauty and liveliness of narratives are often lost in the pursuit of analytical rigor. To mitigate this problem, the second author uses extended quotes from interviews and online ethnography in his writings to give readers more original discourses from the informants. This method worked effectively in the earlier periods of online ethnography in China. This was because the most popular online platforms back then were online forums and Weibo, which were considered as in the public domain. However, the situation changed as WeChat replaced Weibo as the dominant form of online interactions in China in recent years and it has become more challenging to collect and make use of online ethnographic data. WeChat requires its users to register with a cellphone number, which is linked to one's national identity card number. Furthermore, WeChat also restricts its users from publicly searching posts outside one's existing contacts. This leads to the non-anonymous and semi-private nature of WeChat-based online interactions (Tian, 2021;Tian & Guo, 2021). Consequently, researching online groups at the WeChat era has increased not only the ethical burden of researchers but also the risks of surveillance from the state authorities. Although Weibo posts can be deleted, their public nature enables some evidence to be preserved in the public domain not only through any individual user's actions of screenshots and reposts but also through organized efforts such as FreeWeibo.com, which actively monitors and makes available censored Weibo content. In contrast, a WeChat discussion or even an entire WeChat group can be removed by the state censorship without generating much public awareness, because the semi-private interactions within the group had never entered the public domain.
Therefore, as state censorship forces social media interactions out of public spaces in China in recent years, it is even more important for a researcher to immerse in online groups and become a reflexive agent of memory of their "disappeared" stories. It also makes the combination of online and in-person interviews and observation a more effective and desirable methodology for collecting and preserving data. Otherwise, without taking into account "disappeared" information, using a Big Data algorithm or a set of keywords for data collection and analysis would be like typing a story on a keyboard with "an unknown set of keys disabled" (Groves, 2011, p. 869).

Conclusion
Doing social science on social media is a delicate empirical art. For qualitative researchers, the emergence of online ethnography has not revolutionized traditional ethnographic methods as the Big Data approach has done to some areas of quantitative research. The basic principles of doing ethnography, such as the importance of immersion and reflexivity, largely remain the same. Nevertheless, online ethnography does present new challenges and opportunities in terms of accessing field sites, analyzing ethnographic data, and research ethics. The instant access to a large number of potential informants and the highly interactive nature of social media lead to the blurring of field boundaries, the complexity of ethnographic immersion, and the stronger responsibility for the researcher to protect the identities of her research subjects. In this article, we have used the case of China and our own experiences of researching on feminist and lawyer activists to demonstrate the delicate techniques and ethical dilemmas of doing ethnography on social media, particularly on politically sensitive issues in an authoritarian context. Although our dialogue with the Big Data approach runs throughout the article, we have no intention to advocate for any competitive or adversarial relationship between quantitative and qualitative approaches to the study of contemporary China. On the contrary, integrating both approaches in mixed-methods research can be quite beneficial for many research topics, ranging from examining news reporting (Lei, 2016) to studying workplace disputes (Gallagher & Yang, 2017). Neither of the two authors adopted mixed methods mainly because the political sensitivity of our research projects makes quantitative data collection nearly impossible in the Chinese context. For less sensitive topics, however, mixed methods could generate great research potential by combining the bird's-eye view of Big Data with contextualized and interactional online ethnography. The sensitivity of our research makes us extra-cautious of the possible challenges and risks of doing online ethnography in China. As the Chinese state keeps a highly ambiguous and flexible boundary of internet censorship (Han, 2015;King et al., 2017;Stern & O'Brien, 2012;Tsai, 2016), even non-sensitive research topics today could become risky tomorrow-this is precisely what happened to the feminist movement after the detention of the Feminist Five in March 2015. 6 The experiences of online ethnography that we have discussed in the article constitute a firsthand methodological guide for all social science researchers who are interested in incorporating qualitative methods when researching social media in China. We do not believe that China is atypical in any sense regarding the problems of access, ethics, and reflexivity. Instead, we consider China's vast internet population and the rapidly changing landscape of its social media scene great assets for exploring new possibilities of ethnographic research. Whereas the lure of Big Data lies in its scale and efficiency, the enduring attraction of qualitative research lies in its intimacy with social facts and its reflexivity between researchers and informants. The rise of social media as field sites is not going to change that. What it has changed, however, is the spatiality and temporality of ethnography, as thousands of miles are bridged by a few chatrooms and months of intensive fieldwork are transformed to years of online ethnographic immersion. As social science researchers, we have no choice but to critically engage with this new sense of space and time in the social world, online and offline.

Declaration of Conflicting Interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.