Longitudinal relationships among problematic mobile phone use, bedtime procrastination, sleep quality and depressive symptoms in Chinese college students: a cross-lagged panel analysis

This was an observational longitudinal study, not an experiment. The researchers tested four directional hypotheses over 12 months:

Does problematic mobile phone use cause bedtime procrastination, poor sleep quality, or depressive symptoms?

Do bedtime procrastination, poor sleep quality, or depressive symptoms cause problematic mobile phone use?

Are these relationships bidirectional (A causes B, and B causes A)?

Does sleep quality mediate the link between phone use and depression?

The four constructs measured were:

1. **Problematic mobile phone use** – compulsive, excessive phone use that interferes with daily life (measured by the Mobile Phone Addiction Index, MPAI)

2. **Bedtime procrastination** – voluntarily delaying going to bed despite no external reason (measured by the Bedtime Procrastination Scale, BPS)

3. **Sleep quality** – subjective sleep quality, latency, duration, efficiency, disturbances, and daytime dysfunction (measured by the Pittsburgh Sleep Quality Index, PSQI)

4. **Depressive symptoms** – mood, anhedonia, energy, self-worth, concentration (measured by the Self-Rating Depression Scale, SDS)

No intervention was applied. Participants simply completed questionnaires at two time points 12 months apart.

**Sample size:** 1,181 Chinese college students (from an initial pool of 1,350; 169 dropped out between Time 1 and Time 2)

**Age:** Mean age 19.8 years (SD = 1.3), range approximately 17–23

**Gender:** 43.9% male, 56.1% female

**Setting:** Two universities in Henan Province, China

**Inclusion criteria:** Enrolled undergraduate students; no exclusion criteria reported for medical conditions, sleep disorders, or psychiatric diagnoses

**Attrition:** 12.5% dropout over 12 months; dropouts had higher baseline problematic phone use and depressive symptoms than completers, which may bias results

All measures were self-report questionnaires administered in paper-and-pencil format in classroom settings:

**Problematic Mobile Phone Use:** Mobile Phone Addiction Index (MPAI) – 17 items, 5-point Likert scale (1 = never to 5 = always). Total score range 17–85. Higher scores = more problematic use. Example item: "I feel anxious if I haven't checked my phone for a while." Cronbach's alpha at Time 1 = 0.91, Time 2 = 0.92 (excellent internal consistency).

**Bedtime Procrastination:** Bedtime Procrastination Scale (BPS) – 9 items, 5-point Likert scale (1 = never to 5 = always). Total score range 9–45. Higher scores = more procrastination. Example item: "I go to bed later than I intended." Cronbach's alpha at Time 1 = 0.86, Time 2 = 0.87.

**Sleep Quality:** Pittsburgh Sleep Quality Index (PSQI) – 19 items, 7 component scores (subjective sleep quality, sleep latency, sleep duration, sleep efficiency, sleep disturbances, use of sleep medication, daytime dysfunction), each scored 0–3. Global score range 0–21. Higher scores = worse sleep quality. A score >5 indicates poor sleep. Cronbach's alpha at Time 1 = 0.73, Time 2 = 0.74 (acceptable).

**Depressive Symptoms:** Self-Rating Depression Scale (SDS) – 20 items, 4-point Likert scale (1 = none or a little of the time to 4 = most or all of the time). Total score range 20–80. Higher scores = more depressive symptoms. Cronbach's alpha at Time 1 = 0.83, Time 2 = 0.84.

**Study design:** Two-wave longitudinal panel study with cross-lagged panel analysis (CLPA). This is an observational design, not an experiment. Participants completed all four questionnaires at Time 1 (baseline) and again 12 months later (Time 2).

**Why this design matters:** Cross-lagged panel analysis is a statistical technique that uses longitudinal data to test directional relationships between variables. It controls for the stability of each variable over time (e.g., how much depressive symptoms at Time 1 predict depressive symptoms at Time 2) and then examines whether Variable A at Time 1 predicts Variable B at Time 2 *above and beyond* that stability. This is stronger than cross-sectional correlation because it establishes temporal precedence (A comes before B), but it is NOT proof of causation.

**What the design can and cannot prove:**

**Can prove:** Temporal precedence – that changes in one variable precede changes in another over a 12-month period. For example, if problematic phone use at Time 1 predicts sleep quality at Time 2 after controlling for sleep quality at Time 1, this suggests phone use may drive changes in sleep (not just the reverse).

**Cannot prove:** Causation. There may be unmeasured third variables (e.g., stress, personality, social environment) that cause both phone use and sleep problems. The design also cannot rule out reverse causation within shorter timeframes (e.g., phone use affects sleep the same night, which then affects phone use the next day – these dynamics are invisible at a 12-month lag).

**Statistical approach:** Structural equation modeling (SEM) using Mplus 7.0. They tested a fully cross-lagged model with all four variables at both time points, including:

Autoregressive paths (T1 → T2 for each variable)

Cross-lagged paths (e.g., T1 phone use → T2 sleep quality; T1 sleep quality → T2 phone use)

Cross-sectional correlations at T1 and T2

Model fit indices: χ²/df = 2.89, CFI = 0.99, TLI = 0.97, RMSEA = 0.04 (all indicating good fit)

**Major methodological weaknesses:**

1. **Only two time points** – With only two waves, you cannot model trajectories or test for nonlinear effects. Three or more waves would be stronger.

2. **12-month gap** – This is a very long interval for variables that likely fluctuate daily or weekly. The study may miss short-term bidirectional dynamics.

3. **All self-report** – No objective measures of phone use (e.g., screen time logs), sleep (e.g., actigraphy), or depression (e.g., clinical interview). Self-report bias is likely.

4. **Attrition bias** – Dropouts had worse phone use and depression at baseline, so the sample at Time 2 may be healthier than the general population.

5. **Single cultural context** – Chinese college students in Henan Province. Results may not generalize to other ages, cultures, or non-student populations.

6. **No control for baseline sleep disorders or depression treatment** – Participants with clinical insomnia or depression were not excluded, and treatment status was not measured.

All results are standardized regression coefficients (β) from the cross-lagged model. Values range from -1 to +1; larger absolute values = stronger predictive relationships.

**Bidirectional relationships (A predicts B, and B predicts A):**

**Problematic phone use ↔ Bedtime procrastination:** Phone use at T1 predicted bedtime procrastination at T2 (β = 0.08, p < 0.01). Bedtime procrastination at T1 predicted phone use at T2 (β = 0.06, p < 0.05). Both directions were statistically significant but small in magnitude.

**Problematic phone use ↔ Depressive symptoms:** Phone use at T1 predicted depressive symptoms at T2 (β = 0.06, p < 0.05). Depressive symptoms at T1 predicted phone use at T2 (β = 0.07, p < 0.01). Again, small but significant bidirectional effects.

**Sleep quality ↔ Bedtime procrastination:** Poor sleep quality at T1 predicted more bedtime procrastination at T2 (β = 0.07, p < 0.01). Bedtime procrastination at T1 predicted worse sleep quality at T2 (β = 0.07, p < 0.01).

**Sleep quality ↔ Depressive symptoms:** Poor sleep quality at T1 predicted more depressive symptoms at T2 (β = 0.08, p < 0.01). Depressive symptoms at T1 predicted worse sleep quality at T2 (β = 0.06, p < 0.05).

**One-way relationships:**

**Problematic phone use → Sleep quality:** Phone use at T1 predicted worse sleep quality at T2 (β = 0.07, p < 0.01). But sleep quality at T1 did NOT predict phone use at T2 (β = 0.02, p > 0.05). This was a one-way effect.

**Bedtime procrastination → Depressive symptoms:** Bedtime procrastination at T1 predicted more depressive symptoms at T2 (β = 0.08, p < 0.01). But depressive symptoms at T1 did NOT predict bedtime procrastination at T2 (β = 0.03, p > 0.05). Also one-way.

**Non-significant paths:**

Sleep quality at T1 did not predict phone use at T2 (β = 0.02, p > 0.05)

Depressive symptoms at T1 did not predict bedtime procrastination at T2 (β = 0.03, p > 0.05)

**Stability coefficients (how much each variable predicted itself over 12 months):**

Problematic phone use: β = 0.51 (p < 0.001)

Bedtime procrastination: β = 0.45 (p < 0.001)

Sleep quality: β = 0.43 (p < 0.001)

Depressive symptoms: β = 0.42 (p < 0.001)

These are large effects, meaning these traits are fairly stable over a year. The cross-lagged effects (0.06–0.08) are small by comparison but still meaningful because they predict change *after* controlling for stability.

The cross-lagged effects are small in standardized terms (β ≈ 0.06–0.08). To translate into real-world terms:

A one-standard-deviation increase in problematic phone use at baseline (roughly 12 points on the MPAI, e.g., going from 40 to 52) predicted about a 0.07-standard-deviation increase in sleep quality problems a year later (roughly 0.5 points on the PSQI, e.g., going from 5.0 to 5.5). This is a small shift – equivalent to reporting one additional minor sleep disturbance per month.

The bidirectional effects between phone use and depressive symptoms are similarly small: a one-SD increase in phone use predicted about a 0.06-SD increase in depressive symptoms (roughly 1.5 points on the SDS, e.g., from 40 to 41.5). This is not clinically meaningful for an individual but could be significant at a population level.

**Important caveat:** These are effects over 12 months. The daily or weekly effects are likely larger but were not measured. The 12-month lag may underestimate the true short-term impact because people fluctuate and recover.

**Acknowledged by authors:**

Only two time points, limiting ability to model complex change patterns

All self-report measures, subject to recall bias and social desirability bias

Sample limited to Chinese college students, limiting generalizability

Attrition bias (dropouts had worse baseline scores)

Cannot establish causation despite longitudinal design

**Additional critical concerns:**

**No objective phone use data:** Self-reported problematic use may not correlate well with actual screen time. People who feel guilty about phone use may over-report, while heavy users may under-report.

**No sleep physiology data:** PSQI measures perceived sleep quality, not objective sleep parameters (e.g., actigraphy-measured sleep onset latency, total sleep time, or sleep efficiency). Perceived and objective sleep often diverge.

**No daily diary:** The study cannot capture within-person day-to-day dynamics. A person might have a bad night of sleep, use their phone more the next day due to low mood, which then worsens sleep that night – all invisible at a 12-month lag.

**No control for confounders:** Academic stress, social support, personality traits (e.g., neuroticism, conscientiousness), and baseline mental health treatment were not measured or controlled.

**Small effect sizes:** All cross-lagged effects were β ≤ 0.08. While statistically significant due to the large sample (N = 1,181), these effects may have limited practical significance for individuals.

**Single cultural context:** Chinese college students face unique pressures (e.g., gaokao exam culture, strict academic schedules, living in dormitories) that may amplify or alter these relationships compared to other populations.

For someone running their own n=1 experiment:

### What to test

Test the hypothesis that **reducing problematic phone use in the hour before bed** will improve sleep quality and reduce depressive symptoms over 2–4 weeks. The key intervention is not just reducing total screen time, but specifically reducing *problematic* use – compulsive checking, social media scrolling, and gaming that triggers emotional arousal or delays bedtime.

**Specific intervention to try:**

Implement a "phone-free last hour before bed" rule. Place your phone in another room (or a drawer) 60 minutes before your target bedtime.

Alternatively, use app blockers (e.g., Freedom, Cold Turkey, Screen Time) to block social media, news, games, and messaging apps from 9 PM to 7 AM.

Replace phone use with a non-screen wind-down routine: reading a physical book, light stretching, journaling, or listening to an audiobook/podcast (not on your phone).

### Minimum meaningful duration

Run the experiment for **at least 3 weeks**, ideally 4 weeks. The study found effects over 12 months, but daily dynamics likely emerge within 1–2 weeks. Three weeks allows you to:

Overcome the initial novelty/hawthorne effect (first week)

See stable changes in sleep patterns (second week)

Observe any downstream effects on mood (third week)

### What to measure

Track these metrics daily (morning and evening) using a simple spreadsheet or app:

**Primary outcomes (measure daily):**

1. **Sleep quality:** Use the PSQI's single-item global sleep quality rating each morning (1 = very good to 4 = very bad). Or use a sleep tracker (e.g., Oura Ring, Fitbit, or even a simple sleep diary recording: bedtime, wake time, estimated sleep onset latency, number of night awakenings).

2. **Bedtime procrastination:** Each evening, rate on a 1–5 scale: "To what extent did I delay going to bed tonight compared to my intended bedtime?" (1 = went to bed exactly when planned, 5 = delayed by 2+ hours).

3. **Depressive symptoms:** Use the PHQ-9 (Patient Health Questionnaire-9) once per week. This is a validated 9-item depression screener (score 0–27). Or track daily mood on a 1–10 scale.

**Secondary outcomes (measure daily):**

4. **Problematic phone use:** Track actual screen time (use your phone's built-in screen time feature) and number of phone pickups in the hour before bed. Also rate daily: "How much did I feel compelled to check my phone today?" (1 = not at all to 5 = extremely).

5. **Daytime sleepiness:** Rate on a 1–5 scale each afternoon: "How sleepy did I feel today?"

**Baseline (measure for 1 week before starting):**

Collect all the above for 7 days without changing your phone habits