Approach

To answer this question, I contributed to a three-phase mixed-methods study designed to move from exploratory behavioral observation through rich qualitative insight to statistical validation. Each phase informed the next.

This sequential explanatory design allowed the team to surface hypotheses from behavioral observation, deepen understanding through interviews, and then rigorously test those hypotheses quantitatively, a structure critical for high-stakes domains like healthcare.

Phase 1: Contextual Observation

Methodology

Eight participants (ages 23–34) were recruited through personal networks and university communities across Illinois, California, New York, and Minnesota. Participants completed a scenario task (imagine you are experiencing work-related stress and want to find mental health support) using any website or tool they chose. Sessions were conducted primarily over Zoom with screen recording and think-aloud protocols.

My Role

Analysis with Miro

Session recordings were coded using the AEIOU framework (Activities, Environments, Interactions, Objects, Users). The team consolidated notes in Optimal Workshop’s Reframer, then migrated data into Miro for collaborative affinity diagramming.

💡 Miro served as our shared synthesis workspace across all three research phases.

For the observations, each researcher logged their AEIOU-coded notes independently, then the whole team met virtually to cluster 200+ data points by theme. The visual, drag-and-drop environment let us rapidly iterate on groupings, surface disagreements, and reach consensus on emergent patterns in real time, a workflow far more effective than a shared spreadsheet for interpretive work at this scale.

Key Behavioral Insights

👤 Users began with broad condition-based searches before narrowing to specific providers

🔍 Heavy reliance on filter tools: location, insurance, and specialty were prioritized

📰 Cross-platform validation: participants triangulated information across Psychology Today, WebMD, Reddit, and social media

😞 Visible frustration when profiles were incomplete, generic, or difficult to interpret

Phase 2: In-Depth Interviews

Methodology

Eight participants who had actively used online mental health resources within the previous 12 months were recruited through personal networks and online message boards. This cohort spanned Hawaii, Washington D.C., Maryland, California, and Illinois, and was more diverse in gender identity (6 female, 1 male, 1 nonbinary). Sessions averaged 45–60 minutes.

My Role

Qualitative Coding with Atlas.ti

All interview transcripts were imported into Atlas.ti for systematic qualitative analysis. I took the lead on coding, working through transcripts independently before bringing preliminary codes to the team for review and reconciliation into a shared codebook.

This process involved:

• Open coding to surface initial concepts directly from transcript language

• Code grouping to cluster related concepts (e.g., all codes related to provider trust consolidated together)

• Theme development to identify the four major patterns that emerged across participants

• Quotation retrieval to pull representative verbatim examples for each theme, preserving the evidentiary chain

💡 Atlas.ti provided the rigor and auditability that manual annotation cannot.

By maintaining a codebook with definitions, codes could be applied consistently across transcripts and any theme traced back to its supporting quotations. This is essential when presenting findings to stakeholders or teams who weren’t in the room.

Key Themes from Interviews

Mental Health Literacy Gap

Seven of eight participants reported lacking confidence in their mental health literacy: not knowing what conditions they might have, how to interpret clinical terminology, or how to evaluate provider credentials. This created anxiety layered on top of anxiety during the research process.

Provider Fit Above All Else

Cultural competency was the most emotionally charged theme. Participants described searching specifically for BIPOC or LGBTQIA+-allied therapists, often with great difficulty.

"I do try to look for therapists who are BIPOC and people of color as well, which is really hard to find… I think it’ll make it easier if there were more options for BIPOC therapists, because then I’m connecting with somebody who looks like me."

Information Quality Over Quantity

Participants consistently preferred depth of information over breadth. Generic profiles raised skepticism rather than trust. They wanted treatment approaches explained in plain language, not a list of bullet-pointed specialties.

Scheduling Friction as Abandonment Trigger

Multiple participants described a demoralizing loop: identify a promising provider, call the office, discover they are not accepting new patients, repeat. Administrative friction around scheduling was a significant emotional drop-off point.

Personas

Based on interview and observation synthesis, the team developed three behavioral personas representing distinct user archetypes. The personas were created by a teammate; my contribution was the behavioral and attitudinal data from interviews that informed them.

Phase 3: Survey Research

Purpose and Design

The survey was designed to validate (and where necessary challenge) patterns identified in observations and interviews. Qualitative data is rich but limited in generalizability; the survey added a quantitative layer to assess the prevalence of patterns across a broader sample.

The instrument was deployed via Qualtrics to 30 participants meeting the same screening criteria as the interview cohort (18+, used online mental health resources in the past 12 months). Geographic representation spanned IL, CA, NY, MN, HI, DC, and MD.

My Role

Hypothesis Testing

Primary hypothesis: Users with higher mental health literacy (operationalized as research frequency) would experience less difficulty selecting a provider.

Given the ordinal nature of both variables, I used Spearman’s rank correlation in R Studio, the appropriate non-parametric test for this data type.

💡 The correlation between research frequency and selection difficulty was not statistically significant (rho = −0.085, p = 0.66).

This null result was itself meaningful: the problem is not simply how much users research, but the quality and structure of available information. The barrier is systemic, not individual — and that reframing had direct implications for design.

Therapist Preferences

Participants rated the importance of therapist attributes on a 1–10 scale. Specialization in specific conditions emerged as the top priorities.

Convenience and Access Factors

Insurance acceptance was the single highest-rated criterion overall (median 10/10, mean 8.83). Wait time and virtual session availability followed closely.

Cultural Compatibility Factors

Comfort discussing physical health and disabilities ranked highest among cultural criteria. Comfort discussing LGBTQIA+ issues and cultural identity (BIPOC) were both rated with a median of 6–7, validating the qualitative findings on the importance of cultural fit.

Search and Filtering Features

The ability to search by condition/issue was the highest-rated feature (median 9/10). Scheduling online and connecting with a provider’s biography tied at median 8, confirming qualitative themes around scheduling friction and profile depth.

Biographies vs. Ratings & Reviews

When asked how helpful biographies and ratings/reviews are, biographies emerged as clearly stronger — 73% rated them extremely or somewhat helpful. The likelihood-to-use data reinforced this: 22 of 30 respondents said they were ‘extremely likely’ to use a biography vs. 17 for ratings and reviews.

What Makes Selecting a Therapist Online Difficult?

Therapist fit was the most commonly cited barrier to selection (13 respondents), followed by mental health literacy (9) and scheduling issues (8). This triangulated directly with the top themes from interviews.

What Would Stop You From Scheduling?

Therapist fit and scheduling issues topped the list of abandonment triggers, with money and insurance concerns close behind (9 respondents). Notably, mental health literacy — the #2 barrier to selection — ranked last as a scheduling barrier, suggesting literacy concerns dissipate once a user has narrowed their choice.

Preferred Scheduling Method

Online scheduling via website or app was preferred by 50% of respondents (15 of 30), outpacing email (8), phone (4), and text (3). This directly validated the qualitative finding that phone-based scheduling friction was a key pain point.

Design Implications & Feature Recommendations

Research findings were synthesized into a feature prioritization matrix for the WellMind platform, with each feature grounded in specific evidence from across all three research phases.