Digital Ethnography of A Social Media Accessibility Community
Overview
Accessible imagery creation, the combining of text descriptions with visual content so that screen reader users can consume the visuals, is an important aspect of digital content creation and sharing. From improving SEO to attracting more of the 3.3 billion blind and low vision users to digital platforms, it's important for companies and individuals to consider how to support a culture for creating accessible images. To study how such a culture could be cultivated and the processes used by real accessibility advocates working online today, I conducted a virtual ethnography of a volunteer-based community dedicated to crowdsourcing accessible imagery through Discord and posting the resulting user generated content (UGC) to platforms like Twitter, Facebook, and primarily Tumblr.
Alt = Discord logo
Problem Statement
How can companies and individuals leverage accessibility communities, volunteer crowdsourcing, and social media platforms to create and share more accessible visual content online?
Methods:
I chose to conduct participant observation as it is one of the fundamental methods in digital ethnography. I observed interactions occurring between members of the Discord server over an intensive 3-month period, and on Tumblr for an additional year. I also took part in creating image descriptions and answering questions as a member of the community in order to understand the particular challenges participants faced when conducting their online interactions. I ended up getting into a habit of checking the Discord a few times a week, up to daily at some points, and copying-and-pasting chat logs (anonymized as I went) into my observational field notes spreadsheet.
Using the other fundamental methodological tool in any ethnography, interviews, I spoke to 10 Discord server members recruited through my interactions as a participant observer. Due to the particular interaction norms and communication preferences of this community, most participants chose to have their interviews conducted over Discord’s chat feature. This meant that I had to prepare the exact wording of the questions ahead of time (far more structured than my usual approach) to save time on typing, while also making sure I was keeping the conversation flowing naturally and picking out any interesting potential for follow-up questions. Chat interviews also took significantly longer than verbal interviews I’m more used to, so I set aside two hours for what would normally be one hour’s worth of interview questions.
To address the need for data documenting “typical” interactions, and to complement the particularly interesting interactions I had already noted in my observational data, I reviewed the three-month span of messages that constituted my primary observational period and collected a random sample of 30 days where I documented every interaction. I collected primarily via copying and pasting, making sure to get all potentially relevant data: timestamps, hashed user IDs (so I could track how many people were interacting and how often without collecting any identifiable information), messages, and user profile information. This formed my quantitative dataset that I cleaned, explored, and analyzed entirely through Google Sheets using a combination of formulas, data transformations, and old-fashioned color-coding.
After I had collected approximately 80 interactions over the 30-day sample, I performed content analysis on the messages themselves to pull out information on what the topic of the interactions was, the suggested solution from the volunteer, and the requester’s primary concern. This quick-and-dirty content analysis only took a day and allowed me to return to my spreadsheet and report new quantitative results on the frequency and co-occurrence of different codes in the sample.
Outcomes:
Excitingly, I was able to present and put the findings from this study into practice when teaching a workshop on operationalizing accessibility for the DesignOps Summit 2023 in October of 2023. For more details on the workshop (taking place over two half-days), see the announcement post here. Additionally, the slides used for the workshop can be accessed here, and the recordings of the sessions can be found here.