“Let’s Talk about Gender” – Development of a Card Deck on (Gender) Sensitivity in HCI Research and Practice Based on a Contrasting Literature Review

2.1 Identity & Intersectionality

We can broadly identify three approaches to gender focusing on essentialism, performance or identity. The most prevalent popular view on gender within society draws on essentialism. It assumes a strictly binary, immutable, complementary and fixed set of genitalia that neatly align with gender, learned behaviour and social roles [5]. In rejecting this alignment, scholars have started to distinguish between sex (assigned at birth) and gender, with gender becoming a mostly aligned reiterated performance. However, the sexed body still determines how gender is supposedly performed and any performance not following suit marks deviant behaviour [43].

Both of these views provide a biologically deterministic perspective on gender where the assignment of genitalia to a specific sex implicates gender. More recently, theorists have started understanding gender (and sex for that matter) as a social construct going far beyond a simple binary [21]. Assuming a gendered identity consciously constitutes an exercise of self-determination [32]. Due to the “sexing” of the body [21], there is no clear distinction between presumably ‘biological’ sex and ‘performed’ gender; instead, both are continuously and consciously upheld within societal practices and customs. Our work is informed by an understanding of gender as self-determined and independent of genital makeup.

Inequality, though, operates not exclusively on gender. The concept of intersectionality articulates how different marginalised identities might overlap and amplify specific modes of oppression. Crenshaw illustrates this vividly for black women. “Feminist efforts to politicize experiences of women and antiracist efforts to politicize experiences of people of color have frequently proceeded as though the issues and experiences they each detail occur on mutually exclusive terrains” [16]. Hence, beyond gender, further markers of identity are likely similarly relevant and important. If gender and other factors remain implicit, researchers risk operating within unmarked norms of, e. g., masculinity, whiteness, ableness and (often upper) middle-class [7].

2.2 Related Work in HCI

With the recent surge of publications discussing how gendered dimensions shape HCI research, we see increased activity regarding gender recommendations in the forms of analytical reviews, software analysis and guidelines.

4.1 Complexities of Disclosure in Blackwell et al., 2016

Blackwell et al. were interested in social media use patterns of LGBT^[3] parents [4]. They asked which platforms their participants engaged with, which details they shared and how they restricted their audiences. While this work contains no explicit focus on gender (e. g., by choice of keywords), gender is implicitly relevant. In a series of interviews with 28 participants, they identified three themes relating to (1) continuously gauging safety of disclosure around particular audiences (p. 614), (2) the external attribution of activism to lived experiences (p. 615) and (3) detailed deliberations regarding privacy of self and family members (p. 616).

Methodologically, the authors point out the relevance of respecting participants’ choice of self-identification regarding gender and sexual orientation. Their participants were recruited through peer groups on a range of social media platforms (e. g., Twitter, Facebook). Even though the context was social media use, the authors refrained from seeking out their participants’ online profiles to respect their individual privacy and dignify the desired representation towards the researchers. From their results, we additionally take that researchers should communicate a non-judgemental safe environment to participants, create a space in which it is comfortable to share personal experiences and refrain from probing deeper where participants do not readily disclose aspects of their lives.

4.2 Diversity of Embodiments in Karuei et al., 2011

Karuei et al. were interested in the sensitivity of different body locations to vibrations, how they are affected by movement and whether visual workload, participants’ expectations or gender played a role in task performance and preferences during the use of tactile displays [34]. The authors identified differences in the rate of signal detection as well as response time when analysing according to participants’ gender (which they take as an indicator for body fat distribution). In this, they follow a biologically essentialist understanding of gender which becomes further convoluted as the authors point out themselves that the results are neither consistent nor significant.

While the authors attend to gender explicitly, they miss out on fundamentally gendered issues. For example, they identify that the “thigh was among the least effective and least preferred stimulus site we tested; and yet, front pocket is a common location to stow a mobile device, particularly for men” [34, p. 3274]. In this context, differences in gendered clothing (such as the lack of or diminished size of pockets in clothes aimed at women [9]) and their influence on preferences are not reflected upon. As different embodiments are not specifically discussed, we assume that they operate from an unmarked norm of presumably able-bodied participants.

Methodologically, the authors ensured that participants wore the same type of close-fitting clothes (sportswear) to reduce the effect of different clothing styles. However, they do not disclose how they inquired into gender and only report it for eight participants as male. From their results, we can take that gender only poorly transfers to a concept of specific embodiment and that body characteristics should be explicitly and directly analysed instead of inferred from identity characteristics.

4.3 Development of Gender Bias in Otterbacher, 2015

To crowdsource textual metadata for images, Von Ahn and Dabbish developed ESP,^[4] a Game With A Purpose (GWAP). Players are randomly paired and challenged to agree (under a time constraint) on as many labels as possible for a given image [62]. The game requires players to create more and more specific labels once a label has reached a threshold of popularity within the broader player community. Hence, players contribute to the accumulation of big database structures, but also introduce their personal and collective biases into the systems and the algorithms learning from them.

Within this context, Otterbacher studied how bias, particularly gender bias, might be introduced and piped into biases in automated decisions drawing on the data [45]. They operate from a binary notion of gender that can be externally assigned and focus on adjectives, especially strongly subjective ones, and their polarity and frequency in over 33.000 images of people [45, p. 1958f.]. They state that descriptive adjectives are commonly used for ‘women’, whereas active nouns refer more to ‘men’, with a heavy tendency to assign heteronormative descriptors in both cases [45, p. 1961].

Methodologically, the author cautions readers towards a pre-existing bias in online pictures of people as they are pre-shaped by gendered societal expectations. Further, Otterbacher reflects on the limits of assigning gender to pictures, particularly reduced to binary labels. From their results, researchers can take guidance on ensuring equivalence in describing people regardless of gender, i. e., using the same type of words (adjectives, nouns etc.) and classifiers (e. g., jobs, colours etc.) to consistently articulate the same information across participants, case studies, personas or stories.

4.4 Agency Around Trauma in Clarke et al., 2013

Domestic violence is often conceptualised as a gendered problem within a binary concept, where ‘women’ are dominantly seen as survivors and men as perpetrators [3]. However, as class [2] and culture [13] play an additional role, disabled people are overall more at risk to abuse [15] and LGBTQ2IA* people face additional barriers to access services [11], researchers increasingly call for an intersectional lens (see Section 2.1) [46].

On that backdrop, Clarke et al. worked with six women who had recently left an abusive partner [12]. Specifically, they investigated the potential of photographs as a means to (re-)gain agency over survivors’ narrative through storytelling. Their work was conducted over the span of several months and design iterations.

Methodologically, the authors opted for building careful relationships with participants. This included attending to participants’ needs and availability, i. e., not holding workshops during times schools were closed as many of their participants had care duties for young dependants. Additionally, they collaborated with topical professionals to ensure that their research was safe and appropriate. They also disclosed their perspective and motivations on the topic during the sessions to create a trustworthy setting. These close relationships were embodied within the prototypes and deemed essential for the designs to be meaningful and relevant.

4.5 Attending to Everyday Violence in Ahmed et al., 2014

Ahmed et al. created a platform where everyday experiences of sexual violence and harassment could be reported, shared and collectively discussed [1]. They employed user-centred design including surveys, interviews and focus groups. Participants indicated a strong need to share their experiences and find community to work through feelings of shame, sadness and regret as well as defiance and anger to facilitate positive change.

Methodologically, the researchers acknowledge the high prevalence of sexual harassment and violence as well as their gendered impact. Understanding the damaging effect repeated violations might have on some participants can contextualise, e. g., work discussed previously [34]. Hence, Ahmed et al. specifically sought out more privileged participants who might feel safer sharing experiences associated with shame. Attending to economic factors shaping research participation allowed the researchers to more critically reflect on the limitations of their work.

4.6 Language on Sexual Health in a Haimson et al., 2014

Haimson et al. studied the language men who have sex with men (MSM)^[5] use to pass on sexual health information in online personal ads [25]. Their intent was to inform linguistic choices within system architectures as well as shedding light on representations of “health conditions, preferences, and activities” [25, p. 1622f.].

Methodologically, the authors use self-identification for sexual preference (analogous to [4]). As they did not directly work with participants, their positional disclosure of two coders identifying as gay men aids the contextualisation of their findings. Additionally, they categorised age groups along life experiences instead of conventional age clusters. However, they do not explicitly disclose how they operationalise gender. As they only talk about ‘men’ without further specifying, they implicitly rely on the dominant binary model.^[6] Hence, we suggest that even in cases where the focus lies on one gender only, researchers should briefly reflect the inclusion and exclusion criteria for their understanding of which populations they refer to.

4.7 Intersectional Identities in DiSalvo et al., 2011

DiSalvo et al. developed Glitch Game Testers, a program introducing low-income high school students to computing through video games [18]. Their focus lies on African American^[7] men, who are less likely to be interested in technology even though they are passionate about video games. DiSalvo et al. root this in stereotypical representations of “geek” or “nerd” as white young men disregarding personal appearances, whereas African American men tend to construct their identity through athleticism and body work. After participating in a summer programme interning in the games industry, participants were more likely to see themselves and their peers as technically adept.

Methodologically, researchers can take from this research an awareness of how racialised and gendered stereotypes amplify intersectionally (see Section 2.1) [16]. The work (together with more recent examples, e. g., [44]) challenges us to reflect on the unmarked norm of Whiteness in computing. DiSalvo et al. also show how computing professions are not just inaccessible for (white) women. Hence, in thinking about equity and gender, researchers need to critically interrogate their viewpoints and allude to their standpoints [30] to appropriately contextualise the knowledge they produce including the limits thereof.

4.8 Automating Stereotypes in Keyes, 2018

In a theoretically driven analysis, Keyes identified how Automated Gender Recognition (AGR), the “automatic computational identification of a person’s gender from photographs or videos” [35, p88:4] is inherently trans-exclusive due to the presumption that gender could be assigned externally. More recent work has subsequently shown how computers fail to assign gender in a way that aligns with a person’s gender reliably [50]. Keyes “found a remarkably consistent operationalisation of gender within AGR research. Almost every paper with a focus on gender, and many of those without, treated gender in a way aligned with the traditional [exclusively binary, immutable and physiological] view” [35, p88:7]. According to Keyes, such findings likely hold for racialised contexts as well [35], rendering such approaches inherently prone to risk amplifying pre-existing inequalities.

Methodologically, this paper suggests to generally avoid automated identity assignment. Researchers should not externally define aspects of a person’s self-representation. They also suggest to frame gender explicitly and trans-inclusively and call for appropriate resources for researchers. Since then, some such resources have been developed, for example, the HCI Gender Guidelines [51] or specific recommendations for accounting for gender in survey design [55].

4.9 Interfacing in Metaxa-Kakavouli et al., 2018

Metaxa-Kakavouli et al. created two webpages for an introductory course in Computer Science with differing design and layout. In an experimental setup, men did not distinguish much between ‘masculine’ and ‘gender-neutral’ designs, contrary to women [42]. Subsequently, women had lower degrees of belonging in the context of stereotypically masculine designs. The authors explicitly mention that no participants reported to be nonbinary.

Methodologically, the paper operates weakly within a gender model of self-identification. As cited related work predominantly uses a binary model of gender, the authors take care to denaturalise previous work and point out different conceptualisations of gender across cultures, though participants were restricted to US residents only. From their results, researchers can reflect on how the designs of prototypes and software for studies might be perceived as stereotypical within their test population to appropriately control for gendered impressions.

4.10 Identity Disclosure in Fernandez and Birnholtz, 2019

Trans people (particularly trans women of colour [22, p226:6]) can rarely be certain whether it is safe (i. e., not leading to slurs, harassment, hate speech and/or violence by another (cis) person) to disclose their gender in online contexts and have developed strategies that allow them to assess the safety of a given situation. In one-on-one interviews with 20 trans people using dating platforms in the United States of America, Fernandez and Birnholtz identified two forms of disclosure: direct (clear and open) and indirect (with the option of plausible deniability) [22].

Methodologically, the authors recruited through community-based platforms (e. g., an LGBTQI2A* community centre, trans-specific mailing lists) but also employed targeted Facebook advertisements. However, this strategy lead to a disproportionate representation of White trans people volunteering. Participants’ preferences were taken into account for the interview setting. Additionally, Fernandez and Birnholtz explicitly asked about the pronouns each participant had and made clear these would be used in publications. The authors also reflect on their status as cis researchers and identify themselves as allies.

6.1 Designing Research Protocols

Before funding can be sought, before research can be conducted, researchers have to design their research protocol. To build a strong basis for all subsequent work, researchers should reflect on gender sensitivity early on and be aware of power mechanics affecting potential participants, target populations or informants.

6.2 Acquiring Funding

To fund research activities, researchers need to engage more and more with funding bodies awarding finances competitively. Especially those tied to political or governmental entities, e. g., the European Union or national science funds such as the Deutsche Forschungsgesellschaft (DFG), distribute available funds according to several criteria. Being themselves funded largely through taxpayer money, they need to attend to the usefulness of research to advance societies to become more just and equal. Hence, they often ask researchers to allude to impacts on gender norms, rules, or stereotypes. Here, researchers have the opportunity to not just push funding bodies into a more progressive position (while simultaneously presenting their projects as progressive as well) but also to extend knowledge about more substantial issues regarding equity to reviewers and boards.

6.3 Conducting Research

Making research practices more inclusive requires passionate labour and comes in many different shapes and sizes. However, by conducting research accessibly and in line with the needs of marginalised participants, the resulting knowledge has the potential to be more thorough and reflected along its limitations.

6.4 Presenting Research

In writing and talking about research, we need to exercise reflexivity and thoughtfulness on the implications of our research as it pertains to gender and other identity factors.

7.1 Design Goals

Given the abstract characteristics of our recommendations above, we expected conversation cards to offer more flexibility for reflection as facilitating the start of a process (instead of merely a static set of suggestions) and being more conducive to a variety of group settings. Individuals could use the card deck by themselves, while groups have several options for engaging with the material ranging from studious to playful. The basic idea for the card deck was for a project team to sit down together and, instead of going through a checklist, try to answer the questions and discuss the concepts presented on the cards. For example, the group organising a university lecture could use the cards to assess how they could include and acknowledge gender dimensions in their class and exercises. We assumed that the cards’ haptic nature would offer an additional “hands-on” aspect to the process of discovering and discussing new concepts and angles.

The card deck intends to offer HCI researchers and practitioners a range of different perspectives to investigate their project from, independent of the specific context of inquiry. When used in a group setting, e. g. distributing one card per participant, each participant will have to assume a different position from which to discuss a given project from. Each card consists of a text snippet offering information and proposing actions to take. To increase the accessibility of the content, we decided to use plain German as the language of the cards, removing technical terms common within Gender Studies as to not erect artificial barriers to said content. A translation to English is currently in progress.

7.2 Card Development

The recommendations with their verbose and brief descriptions were used as the basis for the text and headers for the first draft of cards. During the development of the cards we used Libre Office Writer, choosing an A5 page format and split-page setup to come close to the intended size of cards. The text then was (a) translated to a more vernacular German and (b) cut down so each recommendation would fit on one card. The first author decided to use a less academical German because the workshop context for initial testing took place at a German-only venue,^[8] and because people should be able to use the cards regardless of their level of education or research focus, e. g. facilitating use by non-academic practitioners. This design choice was driven by our commitment to accessibility and a desire to bring the topic closer to those who until now had little to no contact with it, as described in Section 2.2. Subsequently, we planned the initial testing workshop as an opportunity to receive feedback from people who were unfamiliar with our work.

Another difference between the original recommendations, and the deck of cards based on our goal of accessibility for topical novices lies in the amount of footnotes. Many cards feature one or several footnotes which mostly cover recurring concepts and terms (cis, trans, intersectionality) to establish a baseline for everyone engaging with them. We wanted this crucial information to be available on each card, but without cluttering the main text too much. If someone in a group using the cards does not know a term or phrase, they do not have to publicly acknowledge uncertainty to the person holding onto the description. In addition, having the information directly on the card means that readers would not have to handle multiple documents or negotiate the use of the description with other group members.

In addition to the categories we had already established, we identified further categorizations by taking on a closer look at thematic rather than temporal connections. The guiding question here was “How, apart from seen over the life cycle of a project, could these cards be connected?”. The temporal categories in the initial design were color-coded into the title of each card, and later in the background of the title to increase legibility. The thematic categories are displayed using icons on the top left of each card (see also, Figure 2).

Figure 2

Schematic of the cards detailing different structuring elements (icons and colours) as well as text built up from high level recommendations via guiding questions and explanations to footnotes.

On the technical side, the base document for the cards is a Libre Office Writer document. Writer was chosen to ensure ease of future editing and potential community contributions. The WYSIWYG character of the program makes alterations like a switch from single-page A6 to double-page A5 immediately visible and better comparable. The initial online version functioned as a proof-of-concept and consisted of screenshots from the PDF generated by Writer. A PHP script would return a random image drawn from a folder with the screenshots. In this form, the card drawing tool was not accessible for people who use screen readers, and the cards did not scale on different screen sizes. To increase the accessibility of the website, we created a script that would convert markdown files to HTML files. Hence, in the final version, the cards are screen reader accessible, and mobile-friendly – and additionally, each card on the online version is located within a separate HTML file with a designated URL, so people can link to and share specific cards. The online version can be accessed (so far, only in German) at https://div.uber.space. To assess the suitability of the cards as physical materials, the first author verified font and overall size through printouts to make it feasible to provide the PDF online for interested parties.

Figure 3

Selection of cards for testing font size and suitability for physical workshops.

7.3 Initial Testing

In September 2020, we had the chance to present this work in the form of a half-day workshop at the ditact Women’s summer IT studies. Usually, these courses take place in Salzburg, Austria- due to COVID-19, however, most (including our workshop) were moved online.

Initially, we planned to evaluate the card deck, and with it our results, with focus groups of research and teaching staff at TU Wien, but after talking about this work to general and informed audiences (e. g., at 36C3 in Leipzig^[9]), a member of the ditact organisational team encouraged the first author to hand in a corresponding proposal. We opted for the workshop format rather than a talk, because we saw the potential of evaluating the cards and gathering critical feedback on our findings and inferences. This would provide us with the option to test our main assumption: that the cards can be used in an interactive setting by encouraging discussion and reflection on (gender) sensitivities in HCI research and practice. As the target group for courses at ditactare both students and professionals working in IT-related sectors, we kept both the title and the course description practical and pragmatic: “Inclusive project design – diversity beyond personas”.^[10]

Due to the change to an online format, we prioritized the development of the online version of the card deck. For the workshop itself, we used Zoom. The workshop attracted five participants from various backgrounds. For introductions, the first author asked each participant what their current connections to diversity and inclusion were, and whether they had any current projects where they would like to apply the card deck. We discussed what their experiences with diversity and inclusion measures were in the past, and followed that with a presentation and discussion of the development and content of the card deck, and offered space and time for questions. Thereafter, each participant had the task to draw a card for themselves and (on their own) try to find ways of applying it to their current project (or their occupation in general). Most participants used the first card the tool presented them with. One person reported looking at multiple cards before starting on the task, using the card they were “handed” initially. At the end of the workshop, we discussed each card participants drew, and what their ideas on applying it were. At that point already one workshop participant asked for a PDF file containing the whole card deck to be put up on the website so they could print it and use the card deck at work. This affirmed our assumption that there is a need for this kind of guidance, and that the card deck seemed to work well for (at least) this participant. Hence, their feedback indicates the potential applicability of the card deck for practitioners.

Participants were informed of the exploratory nature of this workshop and the associated intent of testing the card deck with others unfamiliar with the work. The workshop was documented (mostly by the first author) in a collaborative document^[11]. We further sent out an online form for feedback via email, though the form was only filled in by one person.