1 Introduction
As identities of study participants have proven to influence the uptake, experience, and benefits of technologies, the Human-Computer Interaction (HCI) community has made considerable efforts towards inclusive and diverse research practices over the past few years [
1,
16,
25,
47,
104,
105,
110]. For instance, gender HCI has emerged as a mature subfield of HCI that focuses on how people of different genders interact with technology [
109]. Likewise, HCI for development is a growing subfield that considers how designs and technologies interact with the under-resourced and economically disadvantaged communities [
112,
116]. In this work, we focus on one such diversity dimension — race and ethnicity of participants in HCI — which has remained relatively under-explored in current research. In 2017, Schlesinger et al. [
104] found that less than 0.1% of the papers in the CHI proceedings between 1981 and 2016 engaged meaningfully with race, compared to 0.2% and 0.6% for gender and socioeconomic class, respectively. Similar findings were reported on the basis of a quantitative content analysis of accepted papers in CHI 2006, 2011, and 2016 [
47]. The authors highlighted the importance of intersectionality (i.e., an identity framework that seeks to understand the complexity of multiple, overlapping, intersecting social identities [
17,
28,
95]) when examining the composition of the participants in HCI research. In particular, they emphasized that HCI researchers should take an interest in understanding how various dimensions of participants’ identities (e.g., race, gender, socioeconomic status) interact with each other and provided recommendations for deeper engagements with the resulting complex identities.
Other research has explored the issue of race, ethnicity, and bias in HCI research and the HCI community via a critical race theory lens [
83,
107]. Ogbonnaya-Ogburu et al. [
83] argued that racism is pervasive in social-technical systems and implored that HCI research should be attuned to the issue of race; they suggested that participation of under-represented minorities must be sought after in
all research activities. Concerted efforts among HCI researchers also led to a workshop titled “engaging in race in HCI” in CHI 2020 that aimed to identify better practices for engaging with race and improving racial inclusiveness and equity in the broader HCI community [
107]. This workshop allowed the community to begin assembling recommendations for like-minded researchers to discuss the role and implication of race in HCI, which ultimately led to a series of zines featuring the relationship among race, inclusiveness, and HCI research [
91]. One of the zines in the series, in particular, urged HCI researchers, practitioners, and designers to consider race and the implication of race throughout their research process, and provided an infographic to aid the process of investigating and reporting race in HCI research. In addition, recent works in HCI have highlighted the importance of engaging with traditionally excluded groups in HCI such as Black women [
43,
44,
93,
94,
95]. They collectively demonstrated how current technology marginalized Black people and how race, gender, and economic class influence the design choice of various pieces of technology. Finally, this paper builds directly on the “prequel” on race and ethnicity collection of participants in HCI research [
96] by providing considerations on when, why, and from whom HCI researchers collect study participants’ race and ethnicity.
In addition to HCI, other fields of research have explored the topic of race and ethnicity in their research as well. For instance, in medical sciences, the American Medical Association (AMA) Manual of Style states that “specifying the race or ethnicity of study participants can provide information about the
generalizability of the results of a specific study”; therefore, it recommends reporting aggregate race and ethnicity for all study participants [
33]. The American Psychological Association (APA) [
3] has made similar suggestions for empirical studies in psychology. However, given the breadth of research interests and methodologies in HCI, we should neither thoughtlessly copy existing recommendations nor ignore established practices from other disciplines. In particular, HCI has the tradition of illustrating how groups of users interact with technology, which situates the field in a unique position to narrate and reflect on the lived experiences of racial and ethnic minorities through data [
28].
When put together, the existing work calls for (i) a deeper understanding of the current practice of reporting race in HCI; and (ii) a guideline consisting of considerations and recommendations for when, why, and from whom to collect and report race in research. Therefore, our work makes strides toward this goal by answering the following research questions:
RQ1:
When are the study participants’ race and ethnicity reported in HCI research?
RQ2:
What are some considerations that speak for and against collecting this information?
RQ3:
What are relevant considerations on how to collect, report, and use this information?
By answering the three research questions, primarily in the context of race and ethnicity in the United States, we make the following contributions to the literature:
•
We provide an empirical analysis of the frequency of reporting CHI participants’ race and ethnicity, showing that less than 3% of CHI papers in the proceedings from 2016 to 2021 have included such information. Moreover, nearly half of these papers were published in CHI 2021, suggesting a recent increase in research efforts on race and ethnicity data in HCI.
•
Through a survey with authors who published in CHI and were affiliated with a U.S. institution during the time of publication, we summarize the motivations and considerations for and against reporting race and ethnicity in HCI research.
•
We also synthesize existing discussions on racial and ethnic data in related fields to examine the potential benefits and (unintended) consequences of reporting racial and ethnic data in HCI research.
•
Finally, we close with a set of considerations designed for HCI researchers to reason about racial and ethnic data collection and analysis; in particular, we call for careful interpretations of race and ethnicity data, which serves to narrate the lived experiences of racial and ethnic minorities when they interact with the technology under study. While our work does not claim or intend to provide a set of absolute and complete rules, we hope it will spark the discussion on the topic of collecting racial and ethnic information in the broader HCI community. After all, racial and ethnic data collection is a start rather than an end in combating systemic inequality, discrimination, and oppression in HCI, a journey requiring much caution and nuance from researchers to minimize harmful narratives and misrepresentations.
The authors humbly acknowledge that collecting and discussing race and ethnicity data is extremely complex and researchers’ perspectives on this topic are greatly shaped by their own personal experience and research background. Collectively, we are U.S.-based HCI researchers with different research themes (large-scale cross-cultural online studies, narratives and participatory methods for more inclusive design, and statistical models), methodological focuses (quantitative, qualitative, design, and mixed-methods) and career stages (graduate students, early tenure-track assistant professors, and tenured associate professors). While our team has good coverage of HCI research methodologies, we are largely limited to specific social and cultural contexts of race and ethnicity in the U.S., where all authors currently work and live in. Thus, our work might have limited discussion on race and ethnicity in certain non-U.S. cultures and societies. We hope that our work, despite its U.S.-focus, could encourage broader conversations across research domains and cultural contexts (especially non-Western ones) in HCI.
5 Considerations
In the following, we discuss some considerations for collecting race and ethnicity data, synthesized from our multidisciplinary literature review and survey data.
5.1 Why: Whether to collect racial and ethnic data
Answering the question of whether to collect racial and ethnic data of study participants requires substantial care. On the one hand, researchers in HCI and many related scientific disciplines are trained and required to justify the collection and planned analysis of demographic variables such as race. Documented reasons for not collecting this data may be privacy concerns for the study participants [
6], or preventing survey disengagement and fatigue [
38,
51]. Moreover, some fear that collecting and reporting on the findings in different racial groups, when presented without a thorough and nuanced discussion, could reify the racial inequality and stereotype, and even delve into “scientific racism” (i.e., the pseudoscientific belief that empirical evidence justifies racial discrimination) [
4]. When put together, these concerns result in recommendations such as
“Researchers should not collect more information from participants than is needed to answer their research questions” by the Institutional Review Board (IRB), an institutional organization that approves, monitors, and reviews human-subject studies in U.S. academic research institutions [
10]. By contrast, demographics including gender, age, and whether participants are in vulnerable groups (prisoners, minors) are routinely requested in human-subject studies applications to IRB. In practice, researchers often need to cite established differences in racial groups to justify their decisions. On the other hand, one cannot hope to find a documented “established difference” if no study is devoted to documenting and understanding potential differences of the piece of technology across racial groups. This “what we do not measure, we do not understand” argument is especially relevant, given that race has been understudied, and the historic impact of race (and racism) in the U.S. and many countries have been made invisible. For example, in the U.S., the NIH started mandating the recruiting and reporting of racial and ethnic minorities in
all clinical trials after observing significant differences between racial groups across a wide range of health care outcomes [
34,
65,
78]. While federal agencies in the U.S. (such as the NIH and the Census Bureau) have generally leaned towards collecting race and ethnicity of participants, the legality and regulations of the collection and analysis of racial and ethnic data differ greatly around the world. For instance, in the European Union (EU), the General Data Protection Regulation (GDPR) [
18] mandates that “processing of personal data revealing racial or ethnic origin, political opinions, religious or philosophical beliefs... shall be prohibited [in general].” Despite some listed exceptions, e.g., “archiving purposes in the public interest, scientific or historical research purposes or statistical purposes”, the legal and logistical challenges led to another important consideration for participants and researchers based in the EU (and other regions with similar laws and regulations).
Given the nuanced nature of this topic, we encourage HCI researchers to have an active discussion during the study design phase within the team. Some points of discussion (as framed through quotations from our survey respondents and findings from our review) may include:
(1)
Observed differences in the outcome of interest across racial groups:
-
e.g., Yardi and Bruckman [
123] demonstrated that low-income African-American families share digital devices more often than their middle-class White-American counterparts, motivating future work (e.g., Garg [
35], Pina et al. [
89]) to consider the effects of race and socioeconomic status on technology usage.
-
e.g., One participant studied “the use of [ICTs] within and among online communities engaging in identity work”. Since their outcome of interest is interwoven with race and ethnicity, “collecting the racial and ethnic information of participants contextualizes the results within the perspective of the moderator’s background”.
(2)
Potential consequences and implications of the study:
-
e.g., Collecting and reporting racial and ethnic information could also be useful for “future studies that may want to compare their results with ours ”. While their study mainly examined the effect of gender, they still “wanted to enable any future studies examining race to be able to compare their context with ours.”
(3)
Race and ethnicity of study participants is an important dimension of the external validity of a study:
-
e.g., Some researchers “always collect these data. As a person trained as a quantitative researcher, this seems like basic, key demographic information that one should have on-hand in case it is relevant to the RQs.”
-
e.g., another respondent reflected, “the 2016 paper had a sample that was over 90% white, which is definitely not diverse or representative of the trans population [their population of interest].” Therefore, in their follow-up work in 2020, they made sure to “made sure to prioritize recruiting a diverse group on many dimensions, with a particular focus on race/ethnicity”.
(4)
Concerns for privacy and race “determinism”:
-
e.g., Widespread collection of individual race and ethnicity data may spark privacy concerns. The research team should communicate the intended use of the collected data, and work with the community to understand whether the benefits outweigh the privacy and confidentiality concerns.
-
e.g., Presenting data and findings with racial and ethnic identities may also risk perpetuating theories of racial or genetic determinism (i.e., the belief that genetics or phenotype exclusively account for human behaviour and ability). This is echoed by one of our survey respondents: “I think it’s important to choose to include these variables [race and gender variables] in the analyses very carefully. I think race and gender variables especially are sometimes overused because when differences arise they can give way to post-hoc explanations that reinforce stereotypes. That is, the use of these variables should be thoughtful and intentional.”
Of course, given the nuanced nature of this question, we are not arguing to simply lower the barriers to collecting racial and ethnic data of participants for all HCI studies — as noted above, concerns for privacy and misinterpretation of the results should not be taken lightly. Instead, we advocate that whether or not to collect racial and ethnic data of participants should be a deliberate rather than a default decision.
5.2 How to collect race and ethnicity data
Our findings show that researchers who have committed to collecting race and ethnicity of their participants desire a standardized method of collection and seek out templates and best practices. Because of our restriction to U.S.-based participants and research, the vast majority of researchers in our study used the U.S. census as their references when designing their surveys and questionnaires (see Table
1). A considerable subset of surveyed researchers also used the modified format to ask one combined question about race and ethnicity, rather than two separate questions. While this may be an effective starting place, the U.S. census is only conducted every 10 years, and its categories often represent the political values at the time rather than the best research practice [
5,
69]. Therefore, there is a need for best practice that is research-driven and takes into account how participants actually want to be identified by race. In this section, we offer some specific, practical advice on how to collect participants’ race and ethnicity using surveys.
As an example, consider the collection of race and ethnicity in the U.S. census — census forms now have two separate questions about race (i.e., what is this person’s race) and ethnicity (i.e., whether this person is of Hispanic, Latino, or Spanish origin). This separation was largely motivated by the growing diversity within the Hispanic-American population. However, in the 2010 Census, 37% of surveyed Hispanic or Latinx chose not to identify with any of the provided race categories [
52], which reveals the gap between the provided categories in the status quo and how people identify themselves.
While there is certainly no one-size-fits-all solution to collecting race and ethnicity of study participants, we hope the following pointers could be of help to researchers during data collection:
(1)
When the race and ethnicity information is collected via multiple choice questions, the research team should consider allowing the option of
identifying with more than one race and ethnicity, and including an
open-ended option for participants to self-describe their race and ethnicity [
20,
98].
(2)
If researchers want to provide racial and ethnic categories in a questionnaire (e.g., in a large-scale online study), the U.S. census is a good starting point for U.S.-based studies (e.g., in the U.S. context, consider asking separate questions for race
and ethnicity). However, depending on the nature of the study, categories used in the census (e.g., Asian Americans and Pacific Islanders) do not necessarily capture the underlying diversity of the group, and more granular choices might need to be included to reflect and communicate participants’ identities [
48,
53,
62]. As one of our survey participants put it,
“highlighting the inner diversity of the sample also communicates that the AAPI [Asian Americans and Pacific Islanders] umbrella should not be construed as a monolith (e.g., the socioeconomic experiences of East Asians vary from Southeast Asians).”(3)
When there are limited resources on racial and ethnic categories of participants for a study, consider asking participants to self-describe and then cluster the responses afterwards, which could serve as a starting point for future studies engaging with similar populations. In resource-limited settings, researchers could employ similar strategies in a smaller pilot study with a subset of participants. If resources permit, we also encourage researchers to solicit feedback from participants on whether the proposed categories capture their identified race and ethnicity.
(4)
In addition to race and ethnicity, consider alternative data that captures a particular dimension of racial differences and could address the research questions of interest more directly: languages spoken at home, household disposable income, access to quality healthcare (or the lack thereof), etc. For instance, Hanna et al. [
42] and Roth [
99] have outlined some of these particular dimensions or “proxies”.
5.3 From whom to collect race and ethnicity data
Even equipped with perfect survey questions, researchers are likely to run into a series of additional practical challenges — Who should be recruited for the study? Should we always strive for racially-representative samples? How to recruit and retain racial and ethnic minorities? While answers to these questions will be context- and study-specific, we briefly summarize some common motivations and barriers for recruiting racially- and ethnically-diverse or homogeneous participants in this section.
Recall that many of our surveyed researchers cited the external validity of their studies as a primary motivation for collecting participants’ race and ethnicity (see Table
1). Indeed, HCI researchers routinely draw inferences about populations by extrapolating findings based on data from small samples of people. Recent years have seen a surge of work on how studies in HCI, along with psychology, might have relied too much on samples from Western, Educated, Industrialized, Rich, and Democratic (WEIRD) samples [
7,
45,
73,
90] and among college students [
41,
46,
90].
Therefore, we encourage researchers to assess the external validity, as it relates to participants’ race and ethnicity, of their specific studies. For instance, Obiorah et al. [
81] proposed and evaluated a novel interactive multi-person tabletop exhibit in museums, and the race and ethnicity of participants they engaged
“is reflective of typical museum visitors’ demographics in the U.S.”, which speaks to the applicability of their findings on the populations they are interested in. By contrast, Hamidi et al. [
40] acknowledged the lack of diversity in their samples as a limitation and stated that
“Our sample includes diverse age and gender perspectives but lacks diversity of race or ethnic identity [with respect to the U.S. demographics].”Of course, researchers do not always need to recruit participants to match national (or regional) race and ethnicity composition — external validity should be assessed with respect to the population from whom we are drawing inferences. For instance, F. Maestre et al. [
30] recruited a majority of White participants, in keeping with the demographics of the rural region of the U.S. Midwest. Other notable examples include studies with a focus on selected racial and ethnic groups: Lee and Rich [
68] recruited primarily Black American participants because of the “substantial history and contemporary issues with medical racism towards the Black community”. Similarly, Dosono and Semaan [
24] were interested in the dynamics of online Asian American and Pacific Islanders communities. In both cases, a lack of racial diversity does not necessarily pose threat to the external validity of the results.
We also want to acknowledge that relying on convenient (and, often WEIRD and non-racially-diverse) samples has its practical advantages, because recruiting and retaining racial and ethnic minorities presents logistical, financial, and sometimes even legal barriers [
6,
15,
18,
85,
90]. Examining the effectiveness of sampling and recruitment strategies tailored to specific ethnic minority groups (e.g., F. Maestre et al. [
30], Sadler et al. [
100]) in HCI is an important thread of future work. In practice, the following questions summarized and adapted from IRB forms at various universities might be an effective starting point for researchers who want to reflect on their study design and sample collection [
54,
55,
56,
97]:
(1)
What is the estimated male-to-female ratio in the study? Does this reflect the distribution of the local population?
(2)
Is there any target population in terms of gender, race, ethnicity, sexual orientation, literacy level, health status, and economic class?
(3)
Is there any vulnerable population in your study who might be disproportionally affected by the research (e.g., indigenous people, minors, students)?
(4)
Will your study participants be representative of the demographics in the study region? Include an estimate of the percentages that will be from minority groups, or the lack thereof.
(5)
In addition, identify any racial, ethnic, or gender groups that will be specifically excluded from this research study. Consider providing a compelling justification for such exclusion (in addition to convenience samples).
(6)
If the research involves the collection of sensitive, potentially identifiable information (e.g., sexual orientation, race, gender, age, and a combination of the aforementioned factors), consider describing what information will be obtained and included in the final research output, and how permission will be sought.
While we focused on practical suggestions for collecting race in simplified settings such as surveys in this section, there is an array of important and exciting work dedicated to collecting complex, qualitative racial data through interviews, focus groups, and case studies [
29,
83,
84,
93,
94,
120]. We encourage researchers to explore listed papers and references therein to engage with qualitative racial data collection.
5.4 What to report on racial and ethnic data
HCI researchers might also find themselves analyzing race and ethnicity data of their participants, such as incorporating participants’ race in a regression model. While a well-conducted analysis will enrich our knowledge of the technology under study, such analyses, if done poorly, might give way to
post hoc explanations that reinforce racial stereotypes and inequality [
27,
32]. Below, we outline a few considerations for reporting and communicating the findings based on the race and ethnicity of study participants. These considerations are largely inspired and adapted from existing discussions in medical sciences [
33,
59,
64], fair machine learning [
6,
8,
50], HCI [
29,
43,
93,
95,
102], and our own experience as researchers in the field.
Firstly, we encourage researchers to acknowledge that race is a social construct and proxy, rather than an objective measure of underlying traits [
8,
50,
59] when presenting observed differences in racial groups. In particular, we caution
against languages that suggest the differences in the measured outcomes can be attributed to participants’ race, a slippery slope to furthering the racial disparity and stereotypes. Instead, authors should provide qualifying statements when presenting their results across racial categories. As an example, in CHI 2021, K. Chua and Mazmanian [
61] stated that
“We recognize that the terms ‘Asian,’ ‘White,’ and ‘Hispanic’ are broad and homogenize the experiences of people from various races, ethnicities, and national backgrounds. We chose to use these terms because they are emic terms used by the vast majority of our participants in describing themselves.”Second of all, we urge researchers to investigate the potential basis of the observed differences across racial and ethnic groups to the extent feasible, e.g., conducting a follow-up study or drawing from existing literature. For instance, when Dillahunt et al. [
23] found that participants of different races had different levels of engagement and outcomes with online employment resources, they drew from the existing literature on Black-White income inequality [
37,
118] and hypothesized that the observed differences across race could be attributed to the design of online resources websites with the “Us versus Them” thinking that marginalized the racial minorities.
Thirdly, discussions about the potential impact of the study should be sought after. This effort could take many forms: for example, the research artifact could include a paragraph on the societal impact of the research. Such practice would be widely applicable to research artifacts that propose new systems or software, paralleling efforts in the artificial intelligence community [
2]. As an example, if a paper proposes a new piece of virtual reality technology, and researchers observe that the accuracies of the technology differ across racial groups, they should discuss the potential impact when the technology is deployed at a large scale. Similarly, for research outputs that incorporate machine learning models as part of their systems, potential disparate impacts on minority groups, as well as a toolkit for mitigating such ramifications in fair machine learning, could be discussed [
6,
8].
Additionally, we want to call attention to the default practice of treating racial groups as categorical variables with “White” as the reference group (e.g., by classifying participants as “White” versus “Non-White”) in quantitative analyses in HCI. First of all, while easy to the implement, the grouping approach treats race as a non-overlapping attribute and ignores the heterogeneity within each racial and ethnic category [
26,
42]. This could contribute to further reification of the groups under study, especially when the groups exhibit large internal differentiation such as the “Asian Americans and Pacific Islanders” category in the U.S. census [
26]. As an alternative, we encourage quantitative analysis to account for the “multi-dimensionality” of race and carefully examine which dimension or representation is the most appropriate [
42,
99]: for instance, researchers might adopt phenotypes such as skin type in investigating computer vision applications Buolamwini and Gebru [
11]; or geography-based information such as those proposed in Benthall and Haynes [
8] when looking at access to public resources. Furthermore, even when census-like racial categories are suitable, researchers should reflect whether they are defaulting to a culturally-dominant group (e.g., male, White, high income) as the reference could subtly establish and reify the notion that culturally-dominant groups are the most “normal” and “interesting”, thereby creating a nested system of importance over time [
27,
32,
58]. Moreover, using the culturally dominant group as the reference does not always lead to the most interpretable results — for instance, if we are really interested in how different racial and ethnic groups compare against the average in the study population, coding the differences against a specific reference group is not the most informative approach (see more technical details and alternative modelling strategies in Dupree and Kraus [
27] and references therein).
Last but not least, drawing from the unjust history of the U.S. census, as well as the thread of excellent work exposing the oppressive nature of current technology design in HCI [
28,
43,
44,
94,
95,
103], we would like to emphasize that data collection and reporting is
not sufficient in itself — researchers should always (i) consider narrating lived experiences of the racial and ethnic minorities; (ii) explore how oppression from race, gender, and class operates under the current technology design; and (iii) actively resist the current design and technology that perpetuates racism, whenever applicable.
5.5 Moving beyond the U.S.
In this paper, we primarily engage with race and ethnicity in the U.S. context. However, focusing on the North American experience does not fully embrace the issues of race and racism globally. Therefore, in this section, we describe how our analysis and reflections could shed light on how HCI researchers can consider why, when, and how to collect data about race and ethnicity globally:
•
Why and from whom: our considerations for why and from to collect race and ethnicity in Sections
5.1 and
5.3 generalize well to non-U.S. contexts. For instance, socioeconomic and health disparities, and their subsequent impact on HCI research, remain persistent across different parts of the world [
31,
111,
119]. In addition, the external validity of a study is increasingly important in a global context, as the field of HCI continues to expand beyond its focus on WEIRD samples and discovers more country-to-country, region-to-region differences [
72,
73].
•
How: as shown in Table
1, most U.S.-based researchers turn to the U.S. census for categorizing their study participants’ racial identities. While other countries and regions might not have such clear-cut categories, our recommendations in Section
5.2 still point to a general path forward: researchers could start with existing resources such as the national or region census [
106] and scholarly work in related disciplines [
22,
111,
119,
124]; pilot studies could be especially helpful in a historically under-explored community in HCI.
•
What: in addition to the considerations in Section
5.4, we want to highlight that collecting and reporting ethnic identities of participants around the world is merely a start to more equitable, global, and diverse HCI research. Especially with a sample of non-U.S. participants, researchers need to pay particular attention so that the research outputs and insights do not remain WEIRD-focused [
77,
122].
6 Discussion
The primary goal of our work is to understand the race and ethnicity data in HCI from the following aspects: (i) when are the HCI study participants’ race and ethnicity collected (RQ1); (ii) why are race and ethnicity collected (RQ2); and (iii) how is the collection administrated, as well as curating a list of considerations and recommendations for collecting and reporting race and ethnicity in the future (RQ3). Our analysis revealed that for studies published in CHI between 2016 and 2021, less than 3% included detailed race and ethnicity information about their study participants. Among those studies that are based in the United States, about 64% of total participants identified as Non-Hispanic White. By contrast, 9% and 10% identified as Hispanic and Non-Hispanic Black, respectively. Regarding “why”, our participants cited the themes and historical contexts of their research as the primary reason. Other motivating factors to collect this information are to increase a study’s external validity, achieve a more representative sample, or allow future research. Lastly, regarding how the racial data were collected, we found that the U.S. census, as well as identified groups in prior work, were some of the most commonly cited sources.
Motivated by the literature review and results of our survey, we also compiled a list of considerations — corresponding to the three RQs — for when to include study participants’ race and ethnicity (RQ2), and if so, who to include (RQ1) and how to collect this information (RQ3). Our considerations expanded on the pioneering recommendations by Ogbonnaya-Ogburu et al. [
83] and Race in HCI Collective et al. [
91], and are inspired by efforts in data collection and report of gender, which led to a crowd-sourced, pragmatic working document [
102] serving to approach gender inclusively in HCI research. In particular, we advocate for
deliberate decisions when determining when and from whom to collect race and ethnicity information. Furthermore, while the U.S. census, as well as its foreign counterpart, might serve as a good starting point for collecting this information, HCI researchers should stay attuned to the complexity of race and ethnicity — more granular categories, obtained through related works or pilot studies, could lead to historically-neglected insights and minority-empowering research [
106].
We also want to emphasize the intention of this paper: instead of a panacea for engaging with racial data, or the lack thereof, in HCI, this paper is meant to encourage more conversations on collecting and reporting race and ethnicity among HCI researchers, practitioners, and users including the participants in HCI research. Even within the team of authors, we have divergent opinions on some of the considerations, partly due to our diverse research themes, methodological focuses, and first-hand experience with collecting and reporting racial and ethnic information in HCI research. When it comes to when to collect race and ethnicity from study participants, one author who has extensive expertise in engaging with race in HCI immediately took a strong stance and argued that race and ethnicity of study participants should always be collected for a complete contextualization and narrative of the study. However, another author on the team challenged this stance by citing their own experience in collecting race and ethnicity for large-scale, multi-national online studies — a lack of community-driven and research-driven guidance on collecting racial and ethnic data, especially in non-Western contexts, poses a substantial logistical and methodological challenge in their quantitative work. In addition, the team’s collective experiences with IRBs at different institutions also revealed a lack of standardized institutional guidance on collecting this information.
While the team of authors cannot and, certainly, does not claim to represent all research areas in HCI and identities experienced by HCI researchers, we see the tension among us as an indicator of the potential tensions in other research teams and the HCI community, highlighting that race and ethnicity in HCI is a nuanced and highly personal topic that needs to be handled with care. We believe that tension and discussion on collecting and reporting racial and ethnic data should be welcomed: the tensions among researchers and practitioners today could give rise to safe, inclusive, and equitable resolutions for the HCI community tomorrow.
7 Limitations
Our work is subject to several limitations. One limitation is the scope of the discussion of race and ethnicity: the paper and existing work surveyed herein are based on the racial and ethnic context of the United States. In part, this is due to the vast collection of existing research on race and ethnicity in the U.S. However, given the high research output of U.S.-based HCI researchers [
71], we hope that our work will serve as a proof-of-concept piece for future conversations about race and ethnicity in a global context.
In terms of research methodology, our sampling could be subject to selection bias — while CHI covers a broad range of research topics, published papers in CHI proceedings are a small subset of the broader HCI research outputs. For instance, we might expect conference proceedings with a more focused theme (e.g., user experience conferences such as NN/g) and conferences with a more explicit international, non-Western focus (e.g., CLIHC and Asian CHI) to have different approaches to collecting and reporting race and ethnicity of study participants in HCI research. In addition, even within CHI proceedings, limiting publications to 2016–2021 also potentially confounds our findings with the longitudinal trend of research on race and ethnicity in HCI. For instance, more recent research outputs might have more discourse on race and ethnicity [
47], and the year 2021 consists of the largest amount of papers in our final corpus. Therefore, if we repeat our methods on the CHI proceedings from 2022, there is likely going to be a substantial amount of papers we would include. Our approach to corpus curation can also lead to an undercount: there is an array of excellent work in CHI on race and ethnicity in HCI without explicitly recruiting participants and collecting their race and ethnicity. These papers are likely to be omitted in our curation process. Furthermore, as the primary intention of this work was to start a discussion on race and ethnicity data collection, we did not prioritize an exhaustive, iterative refining of our corpus. Therefore, putting all these factors together, we postulate that the reported results on the final corpus of papers are likely an underestimate of the CHI publications that collected and reported participants’ race and ethnic information.
Regarding the survey results, the 15 researchers who provided prompt responses to our inquiries might not be a representative sample of HCI researchers. For instance, 47% of the researchers who participated in our survey used separate questions to collect race and ethnicity, as opposed to less than 20% of the researchers in the entire final corpus. In addition, the collection and report of racial and ethnic data without a strong justification may be discouraged by regulatory agencies. Therefore, the information curated from published studies does not necessarily represent the initial study design or intentions of the researchers, but is likely a combination of the research agendas and constraints imposed by external resources and regulations. As one surveyed researcher noted in their response, “ We ran into challenges and limitations in mapping the participant responses to our grant reporting requirements. Therefore, we ended up collecting the racial/ethnic information based on US Census Bureau categories.”
8 Conclusion AND Future Work
As HCI continues to engage with a racially- and ethnically-diverse population of users, understanding the current practice of collecting the race and ethnicity of participants in HCI research takes on high importance. Through a systematic review of published CHI papers and follow-up surveys with selected authors, we found that less than 3% of CHI papers from 2016 to 2021 collected and reported their participants’ race and ethnicity. Among those authors who did collect this information, the primary motivations include (i) strengthening the external validity of the study, and (ii) addressing the established disparities in the uptake and use of technologies between different racial groups. Most surveyed authors mentioned the U.S. census as their reference for designing the questionnaires for collecting participants’ race and ethnicity.
Our findings reveal several important directions of future work. Firstly, CHI is a global community and reporting on the ethnicity of participants outside of the U.S. has been steadily increasing (for instance, David Bowman et al. [
21], Koushki et al. [
66], Randhawa et al. [
92]). Extending our studies to a more global context will champion the call for inclusiveness and representation of non-Western samples in the HCI research community. In addition, even in the U.S. context, the nuance of the racial groups is not necessarily captured by the established categories used in the U.S. census. For instance, although Middle Eastern and North African Americans are classified as White in the U.S. census, a sizable subset of the community believes that they are not treated or perceived as White, and that such classification might even perpetuate further harm [
75,
76,
101]. Moreover, depending on the nature of the study, categories used in the U.S. census do not necessarily capture the underlying diversity within the group, and researchers have called for more granular categories might need to be included to reflect and communicate participants’ identities [
48,
53,
62].
Another avenue of future research is to investigate the challenges encountered in decisions around race and ethnicity data collection and analysis, especially among the researchers who decided not to collect and report such data. For instance, a systematic summary of the primary barriers (e.g., privacy and legal concerns, lack of systematic categories for large-scale international studies) could inform future efforts on providing resources and designing tools to overcome these barriers.
In this paper, we also outlined a few recommendations which serve to further the conversations on whether this data should be collected, and in what circumstances. Critically, we are not proposing that every HCI study should simply collect race and ethnicity of its participants. Rather, we want to highlight the importance of a deeper and broader consideration of racial and ethnic data collection and analysis in HCI, and certainly within the research team — as long as racial and ethnic categories continue to govern social, political, and cultural interactions, collecting and analyzing racial and ethnic data fits squarely within the agenda of HCI.