ddr-globem-analysis

name

ddr-globem-analysis

description

Analyze a specific participant's longitudinal passive-sensing and psychological data in the GLOBEM digital depression research dataset. Use this skill whenever the task involves: analyzing a user's mental health or behavioral data from wearables/smartphones, generating QA pairs about behavioral/psychological changes over time, working with EMA, depression scores, activity, sleep, communication, location, or phone-usage data, or any user-profile analysis in the DDR/GLOBEM context.

DDR GLOBEM Participant Analysis

Task Overview

Analyze all available data for a specified participant (pid, e.g., INS-W_011) and submit QA pairs covering their behavioral and psychological changes across the observation period. Aim for 28–37 high-quality, distinct QA pairs. Each QA pair covers ONE specific dimension — never bundle multiple modalities or metrics into a single pair.

Critical workflow rule: Always compute all numbers BEFORE calling submit_qa_pair(). Never submit a QA pair with a placeholder, partial, or incomplete answer — doing so produces an unfixable record. Compute first, construct the full answer string, then submit.

Submit incrementally: Submit each QA pair immediately after computing that modality's data. Do not batch all submissions at the end.

Dataset Structure

All CSV files share columns pid and date. Filter every file by the target pid.

Sensor files (92 days per participant, many NaN rows are normal):

• activity_allday_raw.csv — daily step count, active/sedentary bout counts and durations
• sleep_allday_raw.csv — sleep duration (minutes), efficiency, bedtime/wake-time
• communication_allday_raw.csv — incoming/outgoing/missed call counts, durations, distinct contacts
• location_allday_raw.csv — distance traveled, radius of gyration, home time, significant places, circadian routine, location entropy, location transitions, top-location times, flight characteristics, average speed
• phone_usage_allday_raw.csv — unlock episode count, total/avg/std duration; location-context columns
• connectivity_allday_raw.csv — Bluetooth scan count, unique devices

Assessment files (one row per observation per participant):

• ema.csv — negative_affect_EMA score, timestamped
• dep_weekly.csv — weekly feel_anxious, feel_depressed, dep_weekly_subscale, anx_weekly_subscale, BDI2 (endterm only), dep
• pre.csv / post.csv — baseline vs. end-of-study psychological scales (columns use _PRE / _POST suffixes)
• dep_endterm.csv — final depression label and BDI2 score

Schema tip: Use get_field_description(data_file="<filename>") for the six sensor CSV files.

Known column pitfalls:

• platform.csv uses column platform, not os
• Home time: use barnett_hometime (not barnett_homelabel)
• Sleep duration: use summary_rapids_sumdurationasleepmain for daily sleep duration; bedtime/waketime use summary_rapids_firstbedtimemain / summary_rapids_lastwaketimemain
• summary_rapids_* columns repeat period-wide values — use intraday_rapids_* for daily activity variation
• Phone usage location columns: suffixes _locmap_home, _locmap_living, _locmap_study, _locmap_greens, _locmap_exercise
• Proactivity ratio with near-zero incoming: use max(incoming, 0.1) to avoid artifacts; a ratio >10 is likely a near-zero denominator — report raw counts instead
• Pre/post columns use exact names like CESD_9items_PRE, UCLA_10items_PRE, 2waySSS_receiving_emotional_PRE, BRS_PRE, CHIPS_PRE, STAIS_PRE, MAAS_7items_PRE, 2waySSS_social_fit_PRE

Starter Code — Define Once, Reuse Throughout

import pandas as pd
import numpy as np

pid = "INS-W_XXX"  # replace with actual pid

def early_late_thirds(df, pid, value_col):
    sub = df[df['pid'] == pid].copy()
    sub = sub.sort_values('date').dropna(subset=[value_col])
    n = len(sub)
    if n < 2:
        return None, None, None, None, None
    mid = n // 2
    early = sub.iloc[:mid][value_col].mean()
    late = sub.iloc[mid:][value_col].mean()
    third = n // 3
    if third > 0:
        t1 = sub.iloc[:third][value_col].mean()
        t2 = sub.iloc[third:2*third][value_col].mean()
        t3 = sub.iloc[2*third:][value_col].mean()
        if t1 < t2 > t3: pattern = "inverted-U"
        elif t1 > t2 < t3: pattern = "U-shaped"
        elif t1 < t2 < t3: pattern = "progressive increase"
        elif t1 > t2 > t3: pattern = "progressive decline"
        else: pattern = "mixed/stable"
    else:
        t1 = t2 = t3 = pattern = None
    pct = (late - early) / early * 100 if early and early != 0 else 0
    return early, late, pct, (t1, t2, t3), pattern

 ():
     pd.isna(mins):  
    h = (mins // ) % 
    m = (mins % )
     

Analysis Workflow

1. Orient to the participant

Check data availability across all files first:

files = ["ema.csv", "dep_weekly.csv", "activity_allday_raw.csv",
         "sleep_allday_raw.csv", "communication_allday_raw.csv",
         "location_allday_raw.csv", "phone_usage_allday_raw.csv",
         "connectivity_allday_raw.csv"]
for f in files:
    df = pd.read_csv(f)
    sub = df[df['pid'] == pid]
    n_valid = sub.select_dtypes('number').notna().any(axis=1).sum()
    print(f"{f}: {len(sub)} rows, {n_valid} with any numeric data")

Sparse coverage is normal for some modalities (communication can be <30% valid). Always extract what data exists rather than skipping modalities. If valid days <5, note sparsity in the QA answer but still report available trends.

2. Temporal segmentation — apply to ALL modalities including derived metrics

Always dropna() on the target column before splitting. The early_late_thirds() function handles this automatically.

Early/late split (primary comparison unit): first vs second half of valid observations.

Thirds segmentation — mandatory for every quantitative column, including derived ones:

• Ratios: active/sedentary ratio, proactivity ratio, scan efficiency (scans-per-device)
• Timing: first-use timing, avg flight duration, avg speed
• Counts: active bouts, sedentary bouts, network diversity, significant places
• Scores: sleep efficiency, circadian routine

Compute T1/T2/T3 and name the pattern for every such metric. Thirds reveal non-linear patterns (U-shaped, inverted-U, progressive, stable) that early/late splits hide.

Thirds NOT required for (early/late only is sufficient):

• EMA spike analysis (per-date comparisons, not aggregated thirds)
• Sleep timing in HH:MM (shift direction is the key insight)
• Communication and phone first-use timing windows (window shift, not thirds)
• Cross-modal correlations (no temporal split)
• Weekday vs. weekend pairs (no temporal split)

Weekday vs weekend — compute for every modality with ≥10 weekday and ≥5 weekend valid days. Check at minimum: steps, home time, distance, phone unlocks, sleep duration, Bluetooth scan count, active duration. When difference >15%, create a dedicated QA pair for that specific modality (e.g., "How did the user's step count differ between weekdays and weekends?"). Do NOT bundle all modalities into one catch-all QA pair.

3. Compute sub-dimension insights per modality

Sleep timing phase shift: Convert bedtime/wake-time from minutes-from-midnight to HH:MM using mins_to_hhmm(). Report shift direction and magnitude in minutes.

Active/sedentary ratio: Compute active_duration / sedentary_duration for early vs late, plus T1/T2/T3 and named pattern. Also report avg sedentary bout duration with T1/T2/T3.

Communication count vs duration dissociation: If outgoing call count increases but mean duration decreases (or vice versa), name this explicitly. Compute mean_duration_per_call = sumduration / count. Apply same check to incoming calls.

Proactivity ratio: outgoing_count / max(incoming_count, 0.1) early vs late + T1/T2/T3 + named pattern. Report raw counts if ratio >10.

Communication timing window: From rapids_outgoing_timefirstcall and rapids_outgoing_timelastcall (minutes from midnight → HH:MM), report: (1) first call time shift, (2) last call time shift, (3) window duration = last - first, and whether the window expanded or contracted. This is early/late only, no thirds needed.

Phone session variability: Compare rapids_stddurationunlock and rapids_maxdurationunlock early vs late. Increasing variability indicates more extreme phone sessions.

Phone first-use timing: Compute rapids_firstuseafter00unlock early/late plus T1/T2/T3 with named pattern. Earlier values = more disrupted sleep/wake.

Phone by location context: Compare home vs study vs other contexts early vs late using _locmap_home, _locmap_study columns.

Top-location time distribution: Extract doryab_timeattop1location, doryab_timeattop2location, doryab_timeattop3location early vs late. Report % change and concentration shift.

Movement characteristics: Extract barnett_avgflightdur (seconds) early vs late plus T1/T2/T3. Also compute barnett_stdflightdur for variability. Shorter avg flight = shift to local movement.

Average speed: doryab_avgspeed (km/hr) early vs late plus T1/T2/T3 and named pattern. Report as transport mode shift when applicable.

EMA spike analysis — compute participant means for ALL 4 behavioral metrics FIRST, then identify spike dates:

# Pre-compute participant means BEFORE writing the QA answer
ema_sub = pd.read_csv("ema.csv"); ema_sub = ema_sub[ema_sub['pid'] == pid]
loc_sub = pd.read_csv("location_allday_raw.csv"); loc_sub = loc_sub[loc_sub['pid'] == pid]
phone_sub = pd.read_csv("phone_usage_allday_raw.csv"); phone_sub = phone_sub[phone_sub['pid'] == pid]
sleep_sub = pd.read_csv("sleep_allday_raw.csv"); sleep_sub = sleep_sub[sleep_sub['pid'] == pid]

mean_dist = loc_sub['barnett_totalpath'].dropna().mean()  # participant avg distance
mean_home = loc_sub['barnett_hometime'].dropna().mean()   # participant avg home time
mean_sleep = sleep_sub['summary_rapids_sumdurationasleepmain'].dropna().mean()  # avg sleep
mean_unlocks = phone_sub['rapids_countuniqueunlocksepisodes'].dropna().mean()   # avg unlocks

top3 = ema_sub.nlargest(3, 'negative_affect_EMA')
for _, row in top3.iterrows():
    date = row['date']
    # Look up each metric for this date, compare to participant mean

For each spike date, include in the answer: actual_value vs participant_avg, deviation% for ALL 4 metrics. Conclude with the common behavioral pattern across all 3 spikes (e.g., "Social isolation strongly correlates with peak negative affect episodes").

Bluetooth scan efficiency: scans / unique_devices early vs late plus T1/T2/T3 with named pattern. Rising ratio = fewer devices each scanned more repeatedly.

Cross-modal correlations: Merge EMA with each behavioral modality on date (inner join), then dropna on both columns together, compute Pearson r. Require ≥10 shared valid days. For every |r| > 0.3 found, create a dedicated QA pair: "What is the relationship between the user's [EMA metric] and [behavioral metric]?" If two closely related columns yield similar correlations (e.g., phone unlock count and phone total duration both ~r=-0.5), report only the one with higher |r| — avoid near-duplicate cross-modal pairs.

# Correct cross-modal correlation pattern
merged = ema_sub[['date','negative_affect_EMA']].merge(
    sensor_sub[['date', col]], on='date', how='inner')
merged = merged.dropna()
if len(merged) >= 10:
    from scipy import stats
    r, p = stats.pearsonr(merged['negative_affect_EMA'], merged[col])
    if abs(r) > 0.3:
        print(f"r={r:.2f}, n={len(merged)}")

Self-report vs behavioral discrepancy: When survey scores contradict behavioral signals (e.g., perceived support falls but outgoing calls rise, or weekly depression remains low while mobility declines substantially), name this explicitly as a separate QA pair.

4. Formulate and submit QA pairs

Mandatory answer format — every answer must include ALL of the following that apply:

1. Early and late values with units (e.g., "9.3 (early) to 10.4 (late)")
2. % change and direction (e.g., "+12.0%")
3. T1/T2/T3 values when applicable (all quantitative modality QAs)
4. Named pattern when thirds are computed (inverted-U, U-shaped, progressive increase, progressive decline, mixed/stable)
5. Brief interpretation — what the numbers suggest behaviorally or clinically

Preferred compact answer format: "X.XX (early) to X.XX (late), +Y.Y%. T1=a, T2=b, T3=c — pattern_name. [Interpretation sentence.]" Keep answers information-dense. Avoid restating the question or padding with phrases like "This indicates that the trajectory showed..."; instead state conclusions directly.

If you realize you are missing a needed number while writing the answer, compute it before calling submit_qa_pair().

One dimension per QA pair — never combine two modalities or two metrics in one question:

• ✗ "How did the user's home time AND location entropy change?" → split
• ✗ "How did the user's proactivity ratio AND network diversity change?" → split
• ✗ "Were there weekday vs weekend differences?" (all modalities at once) → one dedicated pair per modality with >15% difference

Prioritize checklist coverage — aim to generate QA pairs for every checklist item before adding novel sub-metrics. If approaching 37 pairs, skip niche additions rather than skip a checklist item.

Avoid metric duplication — when barnett and doryab cover the same behavioral concept (e.g., radius of gyration, home time, distance), use the barnett or checklist-specified column. Do not generate separate QA pairs for doryab-specific equivalents (e.g., doryab_totaltime, doryab_radiusofgyration, barnett_maxdiam) when a checklist column already covers that concept.

Valid QA pair types:

• Early/late change: "How did X change between early and late periods?"
• Trajectory-only: "What was the trajectory pattern of X across the three study periods?" (when the thirds pattern is more informative than early/late)
• Sub-dimension: "Was there a count/duration dissociation in X?"
• Cross-modal: "What is the relationship between the user's X and Y?"
• Meta-pattern: "Is there a discrepancy between self-reported X and behavioral Y?"

Good QA pair examples:

Early/late change with dissociation:

Q: "How did outgoing call frequency and duration change, and was there a count/duration dissociation?" A: "Outgoing calls increased from 1.24 to 3.67/day (+196%), with a U-shaped pattern: T1=1.57, T2=1.43, T3=4.13. Mean duration per call fell sharply from 305s to 58s (-81%) — count/duration dissociation: more frequent but much shorter conversations in the late period."

Trajectory with thirds (mandatory named pattern):

Q: "What was the trajectory pattern of the user's location entropy across the three study periods?" A: "Location entropy showed a U-shaped pattern: 0.449 (T1) → 0.485 (T2) → 0.238 nats (T3). Early mean 0.467, late mean 0.362 (-22.5%). The sharp 51% drop in the final third indicates substantially reduced spatial diversity toward study end."

EMA spike analysis (include absolute value, participant avg, % deviation for ALL metrics):

Q: "Did the user's peak negative affect episodes coincide with specific behavioral events?" A: "Highest EMA day (June 10: 9.0) showed minimal travel (1.12 km vs 39.59 km avg, -97%), elevated home time (1424 min vs 1029 avg, +38%), above-average sleep (406 min vs 425 avg, -5%), reduced phone use (20 vs 54 avg, -63%). Second peak (April 22: 8.0)... Third peak (April 29: 8.0)... Social isolation appears most strongly linked to peak negative affect."

Weekday vs weekend (separate pair per modality):

Q: "How did the user's step count differ between weekdays and weekends?" A: "Step count showed a substantial weekday/weekend difference: weekday average 9,899 steps/day vs weekend 5,803 steps/day (-41.4%), indicating significantly lower physical activity on weekends."

Cross-modal:

Q: "What is the relationship between the user's daily step count and time spent at home?" A: "Steps and home time were strongly negatively correlated (r=-0.56, p<0.001, n=60 days). Higher step counts coincided with less home time, indicating more active days involved greater time away from home."

Submit: submit_qa_pair(q="...", a="...")

5. QA coverage checklist

Aim to cover all of these (skip only if data is entirely NaN):

Psychological — assessment

• EMA negative affect: early/late change + named thirds pattern
• EMA spike analysis: top 3 specific high-affect dates with multi-dimensional behavioral comparison (distance, home time, sleep, phone use) vs participant average — include actual_value vs participant_avg, deviation% for each metric; conclude with common pattern across spikes
• Weekly depression: feel_depressed trajectory (early/late + named thirds) + final BDI2 score
• Weekly anxiety: feel_anxious trajectory (early/late + named thirds) — separate pair from depression
• Weekly depression subscale: dep_weekly_subscale early/late + named thirds — report as continuous mean value, not as binary endorsement count or percentage; separate pair
• Weekly anxiety subscale: anx_weekly_subscale early/late + named thirds — report as continuous mean value; separate pair
• Pre/post: psychological state (depression CESD, anxiety STAIS, stress PSS, loneliness UCLA — all in one pair)
• Pre/post: social support (all 4 dimensions + social fit 2waySSS_social_fit — all in one pair)
• Pre/post: emotion regulation / coping / resilience / mindfulness (ERQ reappraisal, ERQ suppression, BRS, CHIPS, MAAS — all in one pair)

Activity

• Steps: early/late + named thirds pattern
• Active/sedentary duration ratio (early/late + T1/T2/T3 + named pattern) + avg sedentary bout duration (early/late + T1/T2/T3 + named pattern) — combine in one pair
• Active bout count: early/late + named thirds — separate pair
• Sedentary bout count: early/late + named thirds — separate pair

Sleep

• Sleep duration: early/late + named thirds pattern
• Sleep efficiency: early/late + named thirds — separate from duration
• Sleep timing phase shift: bedtime and wake-time in HH:MM, shift direction and magnitude in minutes

Communication

• Outgoing call count + mean duration: early/late, note dissociation if count/duration diverge
• Incoming call count + mean duration per call: early/late, note dissociation
• Missed call trend: early/late + named thirds
• Proactivity ratio: early/late + T1/T2/T3 + named pattern (use raw counts if ratio >10)
• Network diversity (distinct contacts): early/late + T1/T2/T3 + named pattern
• Communication timing window: first/last outgoing call in HH:MM + window duration (last − first); report shift direction and whether window expanded or contracted

Location

• Mobility: distance traveled early/late + named thirds
• Radius of gyration: early/late + named thirds
• Home time: early/late + named thirds
• Circadian routine: early/late + named thirds
• Location entropy: early/late + named thirds
• Location transitions: early/late + named thirds — separate pair from entropy
• Top-location time distribution (top-1, top-2, top-3) shift
• Movement characteristics: avg flight duration (early/late + T1/T2/T3) + std flight duration
• Average speed: early/late + T1/T2/T3 + named pattern
• Significant places visited: early/late + named thirds

Phone usage

• Unlock count + total duration: early/late + named thirds, note dissociation if count/duration diverge
• Session length shift (avg duration) + variability (std, max): early/late
• First-use timing: early/late + T1/T2/T3 + named pattern
• Phone by location context (home vs study): early vs late

Connectivity

• Scan count + unique devices: early/late + named thirds (combine in one pair)
• Bluetooth scan efficiency (scans-per-device ratio): early/late + T1/T2/T3 + named pattern

Cross-modal & meta

• Cross-modal correlation: one dedicated QA pair for EACH distinct |r| > 0.3 found (≥10 shared days); skip near-duplicate correlations between closely related columns
• Weekday vs weekend: one dedicated QA pair per modality with >15% difference (check at minimum: steps, home time, distance, phone unlocks, sleep duration, Bluetooth, active duration)
• Self-report vs behavioral discrepancy (if survey direction contradicts behavioral signal)

Common Pitfalls

• Submit only complete answers — never call submit_qa_pair() with "let me compute" or a partial placeholder. Compute all values, form the full answer, then submit.
• dep_weekly_subscale and anx_weekly_subscale are reported as continuous means — these may be 0/1 binary flags, but always report their mean value (e.g., "0.12") with T1/T2/T3. Never report as endorsement counts (e.g., "1/7 = 14.3%").
• Thirds for derived metrics are mandatory — ratios (active/sedentary, proactivity, scan efficiency), timing (first-use, avg flight duration, avg speed), and counts (active bouts, network diversity, significant places) all require T1/T2/T3 and named pattern, not just early/late.
• Named pattern is always required when thirds are computed — state inverted-U, U-shaped, progressive increase, progressive decline, or mixed/stable explicitly.
• Never split raw arrays without dropna — always dropna() on the target column before early/late or thirds.
• EMA spike analysis requires participant means — compute the participant's personal mean for distance, home time, sleep, and phone unlocks BEFORE describing each spike date. Report as "actual_value vs participant_avg, deviation%" for each metric.
• summary_rapids_* are period-wide summaries — use intraday_rapids_* for daily activity variation; summary_rapids_sumdurationasleepmain is the exception (valid daily sleep duration).
• get_field_description won't work for ema.csv or dep_weekly.csv — infer from column names.
• BDI2 appears in dep_weekly.csv only in the final row — it's the endterm score.
• Pre/post column names use suffixes _PRE / _POST: CESD_9items_PRE, UCLA_10items_PRE, 2waySSS_receiving_emotional_PRE, BRS_PRE, CHIPS_PRE, STAIS_PRE, MAAS_7items_PRE, 2waySSS_social_fit_PRE.
• Pre/post scale directions vary: higher UCLA = more loneliness (bad), higher ERQ_reappraisal = better (good), higher PSS = more stress (bad), higher BRS = better resilience (good), lower CHIPS = lower impulsivity (good).
• does not exist — use .