Here are representative quantitative projects where I drew from prominent HCI and social psychology research to identify users' pain points with digital and wearable products, develop prototypes with design solutions, and conduct empirical research to validate prototypes and make actionable recommendations.
Project 1. Using Similarity Cue to Promote Well-being in Social Media Users
Social media users feel depressed after seeing another user who looks better off (i.e., an upward social comparison target). To protect users' well-being, we designed "a similarity cue" to explicitly emphasize the overall similarity between the user and the upward comparison target. A 0% similarity cue emphasized low overall similarity, whereas 50% and 90% similarity cues emphasized high and moderate overall similarity. Our study found that the moderate and high overall similarity cues protected users' well-being after they compared themselves to the upward comparison target. Read more about the study here.
Skills
-
Website prototyping
-
Survey design
-
Multivariate testing (i.e., a cross-sectional between-subjects experiment with 9 versions of a chatbot tutor)
-
Quantitative data analysis (i.e., ANOVA, correlation, exploratory factor analysis, mediation analysis)
Recommendations for Social Media Platforms
-
Make the current features that tacitly emphasize the overall similarity between the user and the upward comparison target to explicitly emphasize it. Facebook's Mutual Friends feature is a good example that shows users the overall similarity tacitly (e.g., users see which group of friends they have in common with other users with the lukewarm phrase "Mutual Friends"). Facebook can adopt the stronger phrase that was used in our study, "You and Taylor are similar to each other by sharing 3 mutual friends").
Project 1. Prioritizing chatbot features and students needs in interventions
Research Question
Which features should we prioritize in a chatbot peer for a growth mindset intervention?
Skills
-
Survey design (Qualtrics)
-
Experiment design (balanced incomplete block design)
-
Data collection via MTurk
-
MaxDiff analysis (R)
Survey Snapshot
-
Make the current features that tacitly emphasize the overall similarity between the user and the upward comparison target to explicitly emphasize it.
They help determine the priority of features to implement and how the importance of a given feature compares to the available engineering resources and schedule.
a two-item survey that promises to indicate the strategic priority of product features)
Note. This is a follow-up project after examining whether undergraduate students prefer chatbot theorist versus chatbot peer in interventions. Read a full research report here.
Project 2. Uncovering optimal chatbot tutor attributes and student segments
Research Question
Which features should we prioritize in a chatbot peer that interacts with students in a growth mindset intervention?
Design Recommendations
-
Make the current features that tacitly emphasize the overall similarity between the user and the upward comparison target to explicitly emphasize it.
Note. This is a follow-up project after examining whether undergraduate students prefer chatbot theorist versus chatbot peer in interventions. Read a full research report here.
Skills
-
Desktop/Secondary research
-
Survey design (Qualtrics)
-
Experiment design (balanced incomplete block design)
-
Data Collection via MTurk
-
MaxDiff Analysis (hierarchical bayes model)
-
R and Excel
They help determine the priority of features to implement and how the importance of a given feature compares to the available engineering resources and schedule.
a two-item survey that promises to indicate the strategic priority of product features)
-
Find my R analysis script
Project 2. Helping Students Accept Negative Feedback through Chatbot Tutor
Undergraduate students don't like to receive negative feedback that is critical for their growth. As a result, they do not accept the feedback and progress slowly to reach their potential. This study examined whether students were more likely to accept negative performance feedback from a chatbot tutor than from a human tutor. I hypothesized that a chatbot tutor's objectivity and lack of a mind would minimize the negative feedback's threat to students' self-esteem and public self-image, which in turn can lead to greater negative feedback acceptance. The quantitative data showed that students did not accept negative feedback from a chatbot and a human tutor. The qualitative data showed that most students preferred to receive negative feedback from a chatbot tutor and offered 2 features for a chatbot tutor. You can read the study's short report here.
Skills
-
Chatbot prototyping and conversational script design
-
Survey design
-
Multivariate testing (i.e., a cross-sectional between-subjects experiment with 9 versions of a chatbot tutor)
-
Quantitative data analysis (i.e., correlation, ANOVA, exploratory factor analysis, path analysis)
Recommendations for AI-based Learning Platforms
-
Include positive emojis or introduce a "sandwich" technique (i.e., negative feedback is placed between two positive feedback) to tone down the negativity of the feedback.
-
User does not want to be reminded of how badly they are performing by receiving negative feedback, and they don't want to know how they are doing in comparison to other users. A learning platform can empower users to opt out from social competition features (e.g., a good example of this is Duolingo's Leaderboards).
Project 3. Increasing User's Attachment to Fitness Trackers through Customization
Problem: There is a high abandonment rate for fitness trackers. One reason for such a high rate is that users do not see the trackers as representing themselves.
Recommendation: Offer many tracker customization options. With customizations, users can modify the tracker to represent them truly and thus feel more attached to it, ultimately lowering tracker abandonment.
Read more about the study findings here.
Project 4. Increasing User Evaluation of Chatbot through Personalized Support
Problem: Many chatbots or AI companions (e.g., Siri) support users without considering their support needs. A user with an intrinsic need for emotional support may get instrumental support from a chatbot, which can dissatisfy the user. Is this current development practice benefitting users?
Recommendation: Our data indicates the answer is 'no.' Users who received personalized support from a chatbot based on their support needs liked and trusted the chatbot more.
Read more about the study findings here.
Project 5. Identifying Attributes of Persuasive Reminders to Motivate Users to Log
Problem: Continuous logging of behaviours on tracking mobile apps using is critical for meaningful self-reflection. This is why the apps send reminders to users, but users do not listen to the reminders and forget to log their behaviours. What makes reminders persuasive?
Recommendation: We found 6 message attributes that can motivate users to log their behaviours on tracking devices. Reminders should look like they come from a close friend, specify users' health goals, be humourous, and more.
Read more about the study findings here.