Authors:
(1) Yuwen Lu, he contributed equally to this work, University of Notre Dame, USA;
(2) Chao Zhang, he contributed equally to this work and work done as a visiting researcher at the University of Notre Dame;
(3) Yuewen Yang, Work done as a visiting researcher at the University of Notre Dame;
(4) Yaxin Yao, Virginia Tech, USA;
(5) Toby Jia-Jun Li, University of Notre Dame, USA.
Table of Links
7 LIMITATIONS AND FUTURE WORK
This presented work has several limitations. First, given the diverse range of dark pattern taxonomies [82], it is difficult to comprehensively cover all types of dark patterns in one work. Our sampled dark pattern instances were limited to three genres of online services: online shopping, video streaming, and social media. However, dark patterns can exist in a wide variety of platforms or task domains. In addition, although we ensured diversity in the sample of dark pattern instances we used, our manual curation process allowed us to reach only a limited sample. These limitations are in line with our above-mentioned scalability challenge; we hope that future studies can scale up our efforts with the research agenda we have discussed.
Also, although our probe Dark Pita only supports web browsers on computers, dark patterns also exist in mobile applications [26, 44]. In fact, many participants in our deployment study expressed their desire to use Dark Pita on smartphones: They feel more vulnerable and less alert to dark patterns on mobile devices, given the often casual usage context. Future work can explore the expansion of our end-user-empowerment approach to mobile platforms. Previous work such as [67, 69] has shown the technical feasibility of a similar approach on Android with the Accessibility API, while the stricter developer permissions on iOS remain a challenge.
The use of in-person co-design workshops, while facilitating more effective and smooth interactions, could have biased our participant pool towards individuals close to the workshop locations. This geographically constrained recruitment approach might limit the diversity of experiences and perspectives contributing to our co-design process (reflected in Appendix A.1). To offset this potential bias, we conducted the probe study online, recruiting participants from a wider range of backgrounds (shown in Appendix A.5). In addition, in-person interaction with participants might make them hesitant to share negative opinions, which could introduce bias to our results. Future studies can improve by recruiting more geographically diverse participants to explore their perspectives based on different cultural and socio-economic backgrounds.
The scale and primary qualitative nature of our two-phase study inherently limit the representativeness of our findings. Despite our best efforts to cultivate a diverse participant pool, our conclusions mainly reflect the perspectives of our specific sample group, failing to fully capture views from homogeneous groups (e.g., experts and non-experts). We recommend future research to consider larger-scale studies for a more comprehensive exploration of this topic.
Despite the promising qualitative results reported in the probe deployment study, the limited duration and scale of our technology probe study did not allow us to track long-term user behaviors to quantitatively examine the efficacy of our approach in behavioral changes. Consequently, the feedback we gathered could be affected by novelty effect [120]. We plan to further develop our probe into a fully functional system and conduct larger-scale longer-term field experiments to measure the impacts on user individual welfare (e.g., financial loss) and autonomy [82]. Furthermore, to measure collective welfare [82] and collect community intelligence, we also plan to release Dark Pita to the general public.
This paper is available on arxiv under CC 4.0 license.