Tax Compliance and Social Network

We implement a between-subjects design with three treatments: Control, Chain Network, and Complete Network. Each participant is assigned to one treatment and remains in it throughout both parts of the experiment.

2025-12-02

2026-06-01

Tax Compliance Rate

This study consists of two main parts (Part A and Part B) followed by a survey. In both parts, participants will be randomly assigned to a group of four and will complete 10 rounds of a tax compliance game under a fixed but unknown audit probability. The information they receive about their group members depends on their treatment assignment. The only systematic variation between Part A and Part B is the level of the (unannounced) audit probability.
The experimental treatment differs only in the social network structure that connects the members within each group. We implement a between-subjects design with three treatments: Control, Chain Network, and Complete Network. Each participant is assigned to one treatment and remains in it throughout both parts of the experiment.

While both network treatments provide social information about audit information, the key distinction is the scope of this information: the Chain Network provides localised information about direct neighbours, whereas the Complete Network provides global information about all group members.

In all network treatments and at the beginning of each round, participants are randomly matched into new groups of four and assigned a new position within their group’s network. Participants are first shown their network position and the previous round’s audit information of their observable neighbours; they then receive a random income and decide how much of it to declare for taxation, with a tax rate of 25% applied to the declared income and a fine of 100% of the evaded tax imposed if audited and found non-compliant; subsequently, they report their beliefs about the current audit probability by allocating 100 tokens across ten probability intervals using an incentive-compatible Quadratic Scoring Rule; finally, they are informed of their own audit result and their after-tax income for the round. Following each of the experimental part, participants will complete a scenario-based assessment designed to measure their perceptions of the appropriateness of different tax compliance decisions.

After the experimental tasks, participants will be invited to complete a comprehensive questionnaire that collects demographic information, behavioural preferences, and their perceptions of the experiment.

Not available

Computer-based randomisation in controlled laboratory setting.

1. Session-level randomization: Experimental sessions are randomly assigned to treatment conditions (Control, Chain, Complete) by using a constrained randomisation procedure.
2. Block-level randomization: the ordinal high audit rate and low audit rate are randomly assigned between two parts.
3. Individual-level randomization: Participants are randomly assigned to network positions within a group of four under the compete and chain treatments, or randomly assigned to a group of four without information about their network positions under the control condition.

No

We plan to recruit 540 participants in total.

Each subject will make 20 decisions, yielding 20*540=10800 observations.

We expect 180 participants in the Control treatment, 180 participants in the Chain Network treatment, and 180 participants in the Complete Network treatment.

Primary outcome:Tax Compliance Rate Unit: Proportion of declared income per participant(0-1 scale) 1. Pilot Study Foundation The pilot study was conducted with two network conditions (Control and Complete), providing the crucial baseline data for our power calculations. Under High Audit Probability: Control Group (n=28): Compliance Rate Mean = 0.512, Standard Deviation = 0.299 Complete Network Group (n=24): Compliance Rate Mean = 0.4239, Standard Deviation = 0.367 Under Low Audit Probability: Control Group (n=28): Compliance Rate Mean = 0.416, Standard Deviation = 0.318 Complete Network Group (n=24): Compliance Rate Mean = 0.379, Standard Deviation = 0.327 2. Main Experimental Design The main experiment will employ a 3 (Network Structure: Control, Chain, Complete) × 2 (Audit Probability: High, Low) mixed design. Between-Subjects Factor: Network Structure. Each participant is randomly assigned to one of the three network conditions. Within-Subjects Factor: Audit Probability. Each participant experiences both high and low audit probability conditions. Total Sample Size: 540 participants, with 180 participants in each of the three network conditions. 3. Power Analysis Calculation To ensure the study is adequately powered to detect a true effect, we conducted a power analysis using the pooled standard deviations derived from the pilot data for the Control and Complete Network conditions. Based on a conservative two-tailed test with 80% statistical power and a significance level (α) of 0.05. In the High Audit Probability context, our design can detect a Minimum Detectable Effect Size of 0.098 in compliance rates (0.30standard deviations). In the Low Audit Probability context, our design can detect a Minimum Detectable Effect Size of 0.095 (0.30standard deviations) Based on the effect sizes observed in prior studies where audit probabilities were not explicitly disclosed to subjects (e.g., Choo, 2016; Alm, 1992), our target Minimum Detectable Effect (MDE) can be considered both feasible and substantively meaningful. References [1] Choo C Y L, Fonseca M A, Myles G D. Do students behave like real taxpayers in the lab? Evidence from a real effort tax compliance experiment[J]. Journal of Economic Behavior & Organization, 2016, 124: 102-114. [2] Alm J, Jackson B, McKee M. Institutional uncertainty and taxpayer compliance[J]. The American Economic Review, 1992, 82(4): 1018-1026.