Investigating the Role of Context in the Delivery of Text Messages for Supporting Psychological Wellbeing

The paper addresses the problem that push-based digital mental health (DMH) tools, such as text messaging services, often deliver interventions that are poorly matched to users’ momentary needs, risking disengagement. Incorporating user context may help deliver timely, appropriate, and relevant support, yet many systems rely on top-down decisions rather than users’ lived experiences. The study focuses on young adults (18–25) in North America, a group with high mobile phone usage and increasing mental health concerns. The research investigates which contextual factors influence experiences with text messaging for psychological wellbeing and what messaging elements should be tailored to users’ dynamic contexts. The authors pose two research questions: RQ1: Which contextual factors are perceived to influence the user experience of a text messaging service for psychological wellbeing? RQ2: What specific elements of text messaging interventions need to be tailored to reflect the users' dynamic contexts? The work comprises a formative qualitative study (interviews and focus groups) and a deployment study using two text message dialogues centered on daily schedules and affective state, respectively.

The related work reviews: (1) Text messaging for behavior change and mental wellbeing: SMS-based interventions have supported reductions in alcohol use, smoking cessation, medication adherence, physical activity, weight management, and patient engagement. In mental health, message content ranges from CBT/DBT/ACT-based psychoeducation and reminders to motivational quotes, exercises, EMA queries, peer narratives, and chatbot conversations. Evidence shows supportive texts can reduce anxiety and depressive symptoms, with engagement often influenced by message frequency and content type. (2) Contextual factors in DMH tools: Context-aware computing views context as information that characterizes a user’s situation. Dynamic factors (time of day, activity, location, mood/energy) can be sensed, computed, or self-reported and are central to JITAIs that adapt intervention timing and content. Prior JITAIs leverage sensors (heart rate, accelerometer, phone usage), time, and self-reports to personalize support. However, many systems select contextual variables from theory rather than from users’ lived perspectives, potentially missing nuanced, individualized needs. Literature also links mental health with circadian rhythms, physical activity, social interaction, and specific stressors; symptoms often vary by time of day. The paper positions its contribution as user-centered identification of contextual variables and tailoring needs for text messaging-based DMH tools.

The research comprised two studies. Formative study: Participants and recruitment: 36 young adults (mean age 21.7; 30 women, 4 men, 1 non-binary, 1 undisclosed; diverse racial identities) were recruited via Mental Health America (MHA) screening pathways (PHQ-9 or GAD-7 ≥10) and via a North American university (6 students via snowball sampling; CS/cognitive science/psychology). Procedures: Semi-structured individual interviews (n=30; 15–30 minutes) and 5 Zoom focus groups (2–5 participants each; 60–75 minutes), with 9 interviewees also in focus groups (some attending multiple sessions). Topics covered prior DMH tool use and how contextual factors (e.g., schedule, affect) should shape messaging content, timing, frequency, and follow-up. Participants were compensated ($20/hour). Analysis: Thematic analysis with open coding by two coders; iterative codebook refinement; focus on dominant themes around daily schedule and affective state. Ethics: Informed consent, option to skip questions, and C-SSRS risk protocol readiness; no crises occurred. Dialogue design for deployment: Based on formative findings, two dialogues were created: (1) Daily Schedule Dialogue—twice-daily sequences at 9:00 AM and 4:30 PM delivering either brief activities (e.g., short breathing/mindfulness) or reflective questions, followed by a feedback question one hour later; messages randomized from a bank; no duplicate sequences in a day. (2) Affective State Dialogue—morning check-in at 9:30 AM assessing energy (high/low), mood (high/low), then a discrete emotion from a list based on the circumplex model (valence/arousal). Users then received either passive supportive texts for low mood or if no suitable emotion was found (messages labeled as professional/peer/general/self-written), or an active writing task if mood was high (compose a supportive message for others; option to contribute to a message bank; option to see one’s own message later). Deployment study: Participants and recruitment: 42 young adults (mean age 22.0; 29 women, 13 men; multiple racial groups), living in North America; 6+12 via snowball/word-of-mouth (DP1–DP6; DP30–DP42) and 23 via targeted ads on MHA (DP7–DP29) with PHQ-9/GAD-7 ≥10; five overlapped with the formative study. Protocol: Participants received daily messages for 1–2 weeks as part of a broader project; in this paper, DP1–DP11 received the affective state dialogue on one day; DP12–DP42 received both the daily schedule dialogue (one day) and affective state dialogue (another day). Messages were delivered via Twilio using a Wizard-of-Oz approach; team members followed a branching script and handled unanticipated responses. Post-study semi-structured interviews (10–30 minutes) captured feedback on timing, busyness, mood/energy interactions, and perceived helpfulness. Compensation: No payment for message interaction; interview compensation $20/hour. Analysis: Mixed methods—engagement quantified as response rates and qualitative thematic analysis using a separate codebook. Ethics: Participants informed the system was not a crisis service; crisis resources provided; daily monitoring and C-SSRS escalation protocol in place; no risk events occurred.

- Dominant contextual factors (RQ1): Daily schedule and affective state (mood and energy) were perceived as the primary factors shaping receptivity and preferences for text messages. Participants varied on preferred times: some favored mornings for inspirational tone-setting and quiet time; others preferred late afternoon/evening due to post-work loneliness or rumination; mid-day/work hours were less preferred unless messages required minimal effort. - Tailoring needs (RQ2): Participants identified tailoring dimensions including message volume, required effort (passive vs active), and time sensitivity/urgency. They recommended low-demand reminders during busy periods, more passive support during low mood/energy, and acknowledging that tasks could be completed later. - Affective state dialogue acceptance: Emotion labeling increased self-awareness and perceived personalization; passive supportive texts were appreciated during low mood/energy; active writing was welcomed during high mood/energy as a meaningful break and way to help others. Opinions diverged on the number of labeling questions (some found three questions too many when fatigued; others desired more reflective prompts). - Schedule dialogue acceptance: Preferences for morning vs afternoon remained highly individualized. Contrary to formative expectations, some participants welcomed reminder-type messages during work/school if they did not demand immediate action. High-effort prompts during busy times were often postponed or ignored; participants requested options to snooze or receive follow-ups later, but cautioned against over-notification. - Associations between context variables: Participants linked time-of-day, busyness, and social proximity with mood/energy fluctuations (e.g., low energy in early mornings; anticipatory anxiety before busy days; reduced need for support when socially engaged). - Engagement metrics: Daily schedule dialogue response rate 72.6% (45/62); morning messages 67.7% (21/31); afternoon messages 77.4% (24/31); 25/31 responded at least once. Affective state dialogue response rate 80.3% (106/132); 37/42 responded at least once; 19 participants did not complete the three required responses to reach the supportive text or writing suggestion.

Findings address RQ1 by confirming daily schedule and affective state as key contextual variables shaping receptivity to DMH text messaging. They address RQ2 by specifying messaging elements to tailor: (1) message volume (moderate follow-ups/reminders, minimize interruptions during busy periods), (2) required effort (favor passive support during low mood/energy, reserve active tasks for when capacity is higher), and (3) time sensitivity (normalize delays and reduce urgency when users are busy). Participants’ experiences revealed complex, personal interactions between schedules and affect, with temporal patterns (morning low energy, evening anxiety) and situational factors (busyness, social proximity) influencing receptivity. These insights advance user-centered design for JITAIs by grounding selection of tailoring variables and decision rules in users’ lived experiences. The study proposes design opportunities such as integrating digital calendars and passive sensing (sleep/activity) to infer availability and context; enabling snooze and reminder mechanisms; and supporting diffuse sociality (peer messages, sharing user-contributed supportive texts) to maintain connection without high interaction demands. Overall, context-aware adaptations can improve perceived timeliness, appropriateness, and relevance, potentially enhancing engagement and outcomes in DMH tools.

The paper demonstrates that young adults’ daily schedules and affective states are central to how they receive and act on DMH text messages. It identifies concrete messaging dimensions to tailor—volume, required effort, and time sensitivity—and highlights the nuanced, personal ways context shapes intervention needs. The work contributes user-centered evidence to inform JITAI tailoring variables and decision rules for mental health support. It recommends integrating multiple contextual data streams (e.g., calendars, sleep/activity sensing), offering flexible engagement (snooze, reminders), and facilitating diffuse sociality. Future research should examine broader populations and contexts, explore alternative emotion models and message types, and evaluate clinical outcomes alongside engagement to ensure symptom reduction and behavior change.

- Population and context: Young adults (18–25) in North America; findings may not generalize across cultures, ages, or phone usage habits. - Contextual scope and message types: Deployment examined each contextual factor with bespoke dialogues and limited message varieties; other factors (e.g., busyness, social proximity) surfaced but were not the primary focus. - Emotion labeling constraints: Some participants felt options were insufficient; reliance on the circumplex model may omit nuances; alternatives (e.g., Plutchik, PAD) could be explored. - Engagement vs efficacy: What users prefer at a given moment may not maximize symptom reduction; future evaluations should measure clinical outcomes. - Interaction load: Affective state dialogue’s multi-step questions could be burdensome during low energy; tension exists between brevity and richer personalization.