# v0.2 Cross-Era Structural Analysis — PA001–PA015 Status: **working analysis**. Dataset: `v0.2/micro_map_manifest_v0.2.csv` (15 selected studies). This is a structural comparison of a convenience sample, **not** a systematic review, meta-analysis, prevalence estimate, or causal model. ## Era split used for description - **Earlier set:** PA001–PA010, published 2007–2019 - **Recent refresh:** PA011–PA015, published 2021–2026 The recent set was intentionally selected to add adaptive design, long-term follow-up, JITAI, maintenance, and active-comparator structure. Differences between eras therefore reflect both historical change **and deliberate sampling**. --- # 1. What did not change: self-monitoring remained the backbone All 15 rows contain some form of behavioral observation/self-monitoring in the intervention structure: - smartphone/app step monitoring; - accelerometer or wearable feedback; - diaries/logs; - progress displays; - goal/reference comparison. Across two decades, the recurring control loop is still: `observe behavior → compare with reference/goal → receive feedback/support → act again` The main historical shift is therefore **not** from “no self-monitoring” to “self-monitoring.” The shift is in **what happens around and after self-monitoring**. --- # 2. What changed: the unit of innovation moved from components to adaptation ## Earlier pattern Most PA001–PA010 interventions change a relatively fixed package: - add goal setting; - add points; - add SMS; - add social comparison; - add tailored advice; - add a web/mobile layer. PA003 is the important early exception: goals were algorithmically recalculated from prior behavior. ## Recent pattern The refresh adds two qualitatively different forms of adaptation: - **PA013 — JITAI:** contextual, preference-sensitive coaching delivered just in time; - **PA011 — SMART:** the treatment sequence itself changes based on behavioral response. This creates a three-level distinction that should remain explicit: 1. **Static/fixed package** — assigned components remain broadly the same. 2. **Dynamic parameter adaptation** — e.g. PA003 recalculates goals from prior behavior. 3. **Adaptive intervention architecture** — context-sensitive JITAI or response-triggered treatment sequencing. ### Interpretation Digital PA intervention research has moved from asking: > Which component should we add? toward: > Which support should this person receive, at what moment, and should the intervention change when the person is not responding? That is a structural change in intervention design. --- # 3. But greater adaptive sophistication has not solved the outcome problem The three clearest adaptive examples do not produce one simple story: - **PA003:** adaptive personalized goals produced a relative advantage, but absolute steps declined in both groups. - **PA013:** JITAI coaching produced no significant average incremental step benefit over an active self-monitoring control; perceived usefulness moderated response. - **PA011:** SMART response-adaptive treatment showed favorable trajectories for some groups/responders and follow-up gains. Therefore: `more adaptive technology ≠ automatically larger average behavioral effect` A more plausible working interpretation is: `adaptation quality × experienced relevance × continued exposure` is a useful working interpretation of the observed cross-study pattern, without implying a validated causal model. It matters more for hypothesis generation than the label “personalized” or “adaptive” alone. --- # 4. Active comparators increasingly reveal the incremental-value problem Several studies — old and new — give the comparator substantial behavioral support: - PA003: both groups receive app self-monitoring + notifications + feedback; goals differ. - PA006: both groups receive Fitbit self-monitoring; SMS is the add-on. - PA009: all groups carry the same step-recording phone; feedback differs. - PA010: acquisition package is shared between regular and plus groups before the maintenance contrast. - PA012: every arm receives tracker + PAI app; training and peer support are add-ons. - PA013: both groups receive a step counter; JITAI coaching is the add-on. - PA015: both groups receive lecture, self-weighing, emails and periodic feedback; the app is incremental. This changes the meaning of a null result. A null incremental effect does **not** necessarily mean: > digital intervention does not work. It may mean: > the additional component does not add measurable value beyond an already active behavioral backbone. This is why `shared_backbone_or_active_comparator` is a required v0.2 field. --- # 5. “More digital contact” remains an unsolved problem across eras The same failure pattern appears repeatedly: ## Earlier examples - **PA006:** intensive SMS produced an early response that disappeared; about half considered three messages/day too many and some stopped reading them. - **PA007:** SMS produced a temporary relative advantage, largely by limiting decline; the difference was not significant by week 24. - **PA009:** adding social comparison did not significantly outperform personal feedback. - **PA010:** continuing the app diary during maintenance did not outperform accelerometer-only maintenance. ## Recent example - **PA012:** adding online training and then peer support did not generate a durable long-term advantage across 18 months. ### Cross-era conclusion The recurring problem is not insufficient feature count. A better working rule is: > additional contact has value only when it changes the behavioral decision at the right time without producing burden, habituation, or redundant support. --- # 6. Engagement/exposure is not a side metric; it constrains interpretation The unified manifest makes a distinction that the original v0.1 could not: `intervention assigned → intervention delivered → intervention received/used → behavioral response` Examples: - **PA005:** weekly participation fell from >85% early to ~75% later. - **PA006:** some participants stopped reading automated messages. - **PA010:** acquisition adherence was ~85%, but maintenance diary adherence fell to 68.4%. - **PA012:** only 42.7% supplied accelerometer data at 18 months. - **PA013:** only ~30% of possible daily step observations were available; message reading could not be verified. - **PA015:** daily app checking declined from ~80% early to ~74% later. ### Implication A null or attenuated behavioral result can arise under at least two different evidential states: 1. intervention receipt/exposure is documented, but there is no incremental behavioral effect; 2. intervention receipt/exposure is weak or uncertain, limiting what can be inferred about the behavioral pathway. Those states should not be coded as equivalent. Unless a mechanism of action is directly measured, the public interpretation should remain at the level of documented exposure and observed behavioral effect. --- # 7. Perceived usefulness introduces a missing variable: behavioral salience PA013 is especially informative because average JITAI effect was null while **perceived usefulness moderated the response**. This suggests that personalization should not be evaluated only by its technical sophistication. A technically personalized message can still be behaviorally irrelevant. The working construct for v0.2 is therefore: ## Behavioral salience A digital intervention remains behaviorally salient when its feedback/support is: - noticed; - perceived as relevant/useful; - timed to a decision opportunity; - actionable; - not overly repetitive/burdensome; - still connected to a meaningful goal/reference. This construct is not claimed as a validated latent variable in the current dataset. It is a **working explanatory construct** generated by the cross-study pattern. --- # 8. Acquisition and maintenance require different architectures The unified phase coding shows that “long intervention” and “maintenance intervention” are not synonymous. Examples: - **PA015:** 32 weeks of continuous active exposure — long duration, but not a distinct maintenance phase. - **PA004:** 3-month intervention followed by 5 months without intervention before follow-up. - **PA010:** explicit acquisition phase followed by a randomized maintenance contrast. - **PA011:** adaptive active phase followed by self-directed app monitoring. - **PA014:** maintenance support begins after participants finish a prior exercise program. ### Implication A component that is useful for acquisition can become redundant during maintenance. PA010 is the clearest example: continuing app diary/self-monitoring did not improve maintenance over accelerometer-only maintenance. A maintenance architecture should therefore be coded by: `what behavioral problem exists after acquisition?` not simply: `which acquisition components can be kept running?` --- # 9. Preserving behavior is a real outcome and must not be mistaken for failure Two rows illustrate why absolute direction matters: - **PA007:** the apparent intervention advantage largely reflected less decline rather than a clear increase. - **PA014:** tracker and telephone-support groups broadly preserved step levels while usual care declined. For maintenance research, the relevant success state may be: `prevent decline` rather than: `increase above the acquisition peak` Therefore v0.2 should preserve three separate quantities whenever possible: 1. absolute change from baseline/acquisition endpoint; 2. between-group difference; 3. time point / phase. --- # 10. Behavior change still does not guarantee downstream physiological change This pattern survives the era transition. - **PA008:** leisure-time PA/walking improved without an advantage in peak VO₂. - **PA015:** PA improved without additional weight loss. The outcome chain should remain: `digital exposure → behavior → sustained behavioral dose → physiological/clinical adaptation` A positive effect at one node does not imply a positive effect at the next. --- # Candidate explanation model for maintenance failure The 15-row map supports a **working hypothesis**, not a causal estimate: ``` maintenance value ≈ incremental mechanism value × continued exposure × perceived usefulness / fit × phase appropriateness × timely adaptation − burden / habituation − comparator saturation − attrition / missing exposure ``` The terms should be read conceptually, not mathematically. ## Six variables to carry forward ### 1. Incremental mechanism value What does the digital add-on contribute beyond what the comparator already receives? ### 2. Exposure persistence Does use/wear/message exposure continue long enough for the mechanism to operate? ### 3. Perceived usefulness / relevance Does the participant experience the support as useful enough to act on? ### 4. Phase fit Is the component solving an acquisition problem or a maintenance problem? ### 5. Adaptive timing Does support change with context or nonresponse, or simply repeat? ### 6. Burden / habituation Does frequency, automation, repetition, or intervention complexity reduce attention over time? --- # Three testable v0.2 hypotheses ## H1 — Simply prolonging an acquisition package will usually have diminishing incremental value Supporting pattern: PA006, PA007, PA010, PA012. Potential counterevidence: PA014 suggests ongoing support can preserve behavior, but human counseling performed similarly to tracker-based support. ## H2 — Adaptation is more promising when it changes a meaningful decision, not merely when an algorithm is present Supporting pattern: - PA003: adaptive goals affected relative trajectory. - PA013: usefulness moderated JITAI response. - PA011: response-triggered treatment adaptation produced favorable responder trajectories. This hypothesis remains provisional because the adaptive studies differ substantially in population, duration, comparator and design. ## H3 — Maintenance success is more likely when the intervention protects against decline than when it attempts to keep producing acquisition-sized gains Supporting pattern: PA007 and PA014. This reframes maintenance evaluation from: > Did activity continue increasing? to: > Was the acquired behavior preserved better than the relevant counterfactual? --- # Implication for the next research question The original v0.1 question: > Under what conditions does a digital self-monitoring + feedback loop remain behaviorally salient long enough to support physical-activity maintenance? can now be sharpened to: > **Which combinations of continued exposure, perceived usefulness, adaptive timing, and maintenance-phase support preserve physical activity after initial behavior change, above the value already provided by self-monitoring and feedback?** This is a stronger v0.2 question because it explicitly controls for the fact that self-monitoring/feedback is already present in many active comparators. --- # What not to infer from this map Do not infer: - prevalence of any mechanism in the full literature; - superiority of JITAI, SMART, wearables, SMS, apps, or human counseling; - causal mediation by engagement or perceived usefulness across studies; - risk-of-bias rankings; - pooled effect sizes. The current value of the map is **structural hypothesis generation and correction**, not effect estimation.