The Price and Quantity of Time in Digital Consumption: A Field Experiment on Inattention and Time Misperception

This study tests two informational interventions designed to address distinct sources of suboptimal smartphone usage, implemented in a 2x2 factorial design:

INTERVENTION 1: VALUE OF TIME (VOT)

This intervention increases the salience of the opportunity cost of smartphone time. On Day 1 of the treatment week, participants complete a survey that asks: "If you used your smartphone for one hour less today, what would you most likely do with that extra hour?" and "How valuable, in RMB, is doing that activity for one hour to you?" The survey auto-calculates the participant's personalized opportunity cost and displays it on screen. Participants are instructed to save a screenshot of this result. On each data collection day (Wednesday and Sunday), a WeChat group broadcast message reminds participants to recall their personal opportunity cost. The message reads: "Reminder: Think about the opportunity cost of your smartphone time. Recall the personal value you calculated in the Day 1 survey. If you cannot remember, check the screenshot you saved."

INTERVENTION 2: TIME CONSUMPTION (TC)

This intervention corrects the bias in time perception by providing feedback on perceived vs. actual usage. On Day 1 of the treatment week, participants estimate their total smartphone usage in the previous week; the survey then reveals their actual usage (from the pre-treatment StayFree data). On each data collection day, participants complete a brief self-estimation exercise: they estimate how many hours they have used their phone since the last check-in (with a 30-second page timer to prevent checking StayFree before answering), and are then prompted to open StayFree and compare their estimate to the actual figure.

CONTROL:

The Control group receives no informational intervention. They submit their StayFree usage data on the same schedule as all other groups.

VOT+TC:

The fourth arm receives both interventions simultaneously.

All interventions are delivered remotely via Wenjuanxing (online survey platform) and WeChat group broadcasts. No in-person contact is required during the treatment period. Participants receive 3 RMB per data submission via WeChat Pay as compensation for their time.

2027-03-01

2027-03-31

1. Total daily smartphone screen time (hours), measured by StayFree app CSV export. This is the primary dependent variable for the difference-in-differences (DID) analysis.

2. Change in screen time from pre-treatment (Week 1) to treatment (Week 2) period, by treatment arm. The VOT main effect estimates Delta_k (overconsumption due to inattention to opportunity cost). The TC main effect estimates Delta_S (overconsumption due to biased time perception).

3. Interaction between VOT and TC treatments (beta_3 in the factorial regression), testing whether the two biases operate independently.

4. Heterogeneity of VOT treatment effect with respect to the participant's marginal opportunity cost p_i (mechanism test). If beta_1p < 0 (VOT effect is larger for participants with higher opportunity cost), this supports the interpretation that the VOT treatment operates through the opportunity cost channel rather than a generic awareness channel.

5. Heterogeneity of TC treatment effect with respect to p_i. The TC effect should be uncorrelated with p_i, providing a "double dissociation" between channels.

6. Self-estimated vs. actual screen time gap (S), measured from TC arm's twice-weekly self-estimation surveys.

7. Degree of sophistication (lambda), computed as 1 - Delta_S / S_bar.

8. Persistence of treatment effects in Week 3 (short-run, post-treatment) and the 1-month follow-up (long-run).

9. Welfare loss, computed using the sufficient statistics formula: WL = (p / (2 * eta)) * (Delta_k + Delta_S)^2, where p is the marginal opportunity cost of time (measured by survey), eta is the attention semi-elasticity (estimated from the cross-sectional relationship between p_i and screen time in the VOT group), and Delta_k and Delta_S are the treatment effects.

We employ a 2x2 factorial design with four treatment arms:

TC Off TC On
VOT Off Control (n~75) TC Only (n~75)
VOT On VOT Only (n~75) VOT+TC (n~75)

RECRUITMENT:

Participants are recruited from the CeDEx China experimental subject pool at UNNC. CeDEx China is a research centre affiliated with the Department of Economics at UNNC, maintaining a standing pool of students who have voluntarily registered to be contacted about economics research studies. The PI is an External Fellow of CeDEx China with authorized access to this pool. The university email system is not used for recruitment. Supplementary recruitment may include physical campus posters and voluntary sharing through personal WeChat networks.

DATA COLLECTION:

All participants install StayFree, a free third-party screen time tracking app developed by StayFree Apps (an independent developer), on their smartphones. StayFree records app usage durations locally on the device. The research team has no access to participants' data through the app. Participants manually export their data as CSV files and upload them to Wenjuanxing (a Chinese online survey platform) twice per week (Wednesday and Sunday). The CSV files contain only app names and usage durations; no message content, browsing history, or other personal information is collected.

ANONYMITY:

Each participant is assigned an anonymous ID at enrollment. All survey data is linked to this anonymous ID, not to names or student IDs. A master linking file is stored separately on a password-protected device, used only for payment distribution, and permanently deleted after the study.

TREATMENTS:

Control: No treatment during Weeks 2 and 3. Participants submit StayFree CSV exports only. Opportunity cost survey administered at the end of Week 3 (after all behavioral data is collected), ensuring the Control group remains uncontaminated throughout the treatment and post-treatment periods.

VOT (Value of Time): On Day 1 of Week 2, participants complete an opportunity cost survey via Wenjuanxing. The main measure asks: "If you used your smartphone for one hour less today, what would you most likely do with that extra hour?" followed by "How valuable, in RMB, is doing that activity for one hour?" Wenjuanxing auto-calculates the participant's personalized opportunity cost (p_i) and displays it on the submission confirmation page. Participants are instructed to save a screenshot of this result. On each data collection day (Wednesday and Sunday), a WeChat group broadcast reminds participants to recall their personal p_i. The reminder does not display a group average to avoid distorting individual perceptions. As a robustness measure, participants also value all 10 non-digital activities and report their time shares, yielding a weighted average opportunity cost (p_i_avg).

TC (Time Consumption): On Day 1 of Week 2, participants estimate their total Week 1 smartphone usage; the survey then reveals their actual usage from the Week 1 CSV data. On each data collection day, participants complete a self-estimation survey with a 30-second page timer (to prevent checking StayFree before answering), followed by a prompt to check their actual usage in StayFree. Opportunity cost survey administered at end of Week 3 (same timing as Control).

VOT+TC: Both treatments combined. Morning opportunity cost reminders + twice-weekly self-estimation exercises.

TIMELINE:

Week 0: Recruitment, baseline survey (demographics, self-estimated usage, optional CRT and time preference items), StayFree installation and verification.
Week 1: Pre-treatment baseline observation (2 data submissions: Wednesday and Sunday).
Week 2: Treatment period (2 data submissions + treatment-specific surveys and messages).
Week 3: Post-treatment observation (2 data submissions). End survey and delayed opportunity cost survey for Control and TC Only groups.
Month 2: 1-month follow-up (1 data submission).

MAIN SPECIFICATION:

d_it = mu_i + gamma_t + beta_1 * (VOT_i * Post_t) + beta_2 * (TC_i * Post_t) + beta_3 * (VOT_i * TC_i * Post_t) + epsilon_it

where d_it is screen time for participant i at measurement occasion t (Wednesday or Sunday), mu_i is an individual fixed effect (absorbs all time-invariant characteristics including treatment assignment), gamma_t is a measurement-occasion fixed effect (absorbs weekday/weekend differences and common time trends), and Post_t indicates Week 2. Standard errors clustered at the individual level.

Coefficients: beta_1 = -Delta_k (VOT effect), beta_2 = -Delta_S (TC effect), beta_3 = interaction.

IDENTIFICATION OF eta (attention semi-elasticity):

Within the VOT group (where attention has been raised), we estimate: d_i = alpha - eta * ln(p_i) + X_i' * gamma + epsilon_i, where X_i includes baseline screen time, gender, age, academic year, and major. eta measures how many hours of screen time decrease per unit increase in ln(p_i). Endogeneity is addressed by controlling for baseline screen time and reporting robustness to alternative p_i measures and a range of eta values.

MECHANISM TEST:

d_it = mu_i + gamma_t + beta_1 * (VOT_i * Post_t) + beta_1p * (VOT_i * Post_t * ln(p_i)) + ... + epsilon_it

If beta_1p < 0, the VOT effect is larger for high-p_i participants, supporting the opportunity cost channel. If the TC effect does not vary with p_i, this provides a double dissociation.

PRIMARY HYPOTHESES (Bonferroni-corrected, alpha = 0.025 each):

H1: beta_1 < 0 (VOT treatment reduces screen time, indicating inattention to opportunity cost)
H2: beta_2 < 0 (TC treatment reduces screen time, indicating biased time perception)

SECONDARY HYPOTHESES (alpha = 0.05):

H3: beta_3 != 0 (interaction between the two biases)
H4: beta_1p < 0 (VOT effect correlates with p_i, supporting opportunity cost mechanism)
H5: TC effect does not vary with p_i (double dissociation)

EXPLORATORY ANALYSES:

Heterogeneity by baseline screen time (above/below median), gender, academic year, major (STEM vs. humanities), weekday vs. weekend. If baseline survey includes optional CRT and time preference items, heterogeneity by cognitive ability and patience. App-level treatment effects (e.g., social media vs. gaming). Persistence in Week 3 and 1-month follow-up.

WELFARE CALCULATION:

WL = (p_bar / (2 * eta_hat)) * (beta_1_hat + beta_2_hat)^2

where p_bar is the sample mean of the marginal opportunity cost (main specification) or the weighted average opportunity cost (robustness). Confidence intervals by bootstrap (resampling participants with replacement, 1000 iterations). Sensitivity: welfare recalculated under +/- 20% variation in p_bar.

DECOMPOSITION:

WL = (p_bar / (2 * eta_hat)) * Delta_k^2 + (p_bar / (2 * eta_hat)) * Delta_S^2 + (p_bar / eta_hat) * Delta_k * Delta_S

The first term is welfare loss from price-side inattention, the second from quantity-side misperception, the third from their interaction.

Not available

Stratified randomization using a computer-generated random sequence (implemented in Python or R). Stratification variables: gender (male/female), academic year (undergraduate year 1-4, postgraduate), and baseline (Week 1) smartphone usage (above/below median). Randomization is performed after the baseline period (Week 1) is complete, ensuring that stratification by baseline usage is based on observed data from StayFree rather than self-reports.

Individual

No

N/A (individual-level randomization, not clustered)

Target recruitment: 300 participants (approximately 75 per arm). With expected attrition of approximately 15-20%, approximately 250 participants analyzed (approximately 62 per arm). Each participant provides 6 observations (twice-weekly measurements over 3 weeks: 2 in Week 1, 2 in Week 2, 2 in Week 3), yielding approximately 1,500 participant-period observations for the main analysis. An additional observation per participant is collected at the 1-month follow-up.

Control: ~75 (recruited), ~62 (analyzed after attrition)
VOT Only: ~75 (recruited), ~62 (analyzed after attrition)
TC Only: ~75 (recruited), ~62 (analyzed after attrition)
VOT+TC: ~75 (recruited), ~62 (analyzed after attrition)
Total: ~300 (recruited), ~250 (analyzed after attrition)

Minimum detectable effect size: 0.3 standard deviations. With approximately 62 participants per arm (after 15-20% attrition), twice-weekly DID with individual fixed effects, and within-person correlation rho = 0.7, expected power is approximately 0.80 for each primary hypothesis at alpha = 0.025 (Bonferroni-corrected for two primary tests). Power for the interaction effect (beta_3) is approximately 0.40-0.45; this test is therefore designated as secondary/exploratory. Power for the mechanism test (beta_1p) depends on the variance of p_i in the sample and will be assessed ex post.