Building Tax Compliance from the Start: Evidence from a Large-Scale Experiments among New Taxpayers in Indonesia

This project investigates whether early behavioural interventions can establish tax compliance habits among newly registered individual taxpayers in Indonesia. Using a large-scale RCTs involving approximately 1.1 million taxpayers who registered within the past three months and have never filed a tax return, the study tests whether targeted communication can encourage taxpayers to file for the first time and develop persistent compliance behaviour. Eligible taxpayers are randomly assigned to a control group (standard communication) or one of six treatment arms that vary message content (deterrence-focused vs. information-focused), tone (formal vs. casual), and, for two of the arms, the addition of personalised feedback comparing the taxpayer's reported business or financial indicators against industry benchmarks. Taxpayers receive messages via WhatsApp or email across six intervention cycles from October 2026 to March 2027, allowing the study to examine whether repeated exposure strengthens compliance through learning and habit formation. Outcomes include whether and when taxpayers file their return, tax payments, and subsequent engagement with digital tax services (Coretax).

2026-10-01

2027-03-15

A dummy variable indicating whether the taxpayer submitted their annual income tax return (SPT Tahunan) for tax year 2026 by the statutory deadline of 31 March 2027. This is designated in advance as the single outcome on which the trial's success or failure is judged, and is the outcome the power calculation is based on.

Co-primary outcomes (four additional administrative variables, tested against control across all six treatment arms with multiplicity correction via a step-down procedure controlling the family-wise error rate):

Annual tax return filing, irrespective of timeliness
Monthly tax payment, and whether that payment was timely
Amount of tax paid
Coretax application and reporting-token activation

Secondary outcomes are not corrected for multiplicity and are interpreted as evidence on mechanism rather than as tests of treatment effectiveness. They are intended to capture the learning process and the formation of compliance routine:

Click on each embedded link, by taxpayer and by intervention cycle
Elapsed time between message dispatch and first click
Delivery failure, and (WhatsApp channel only) blocking of the sender by the recipient, recorded cumulatively by arm and cycle
Length of the consecutive run of on-time monthly payments
Time until the first missed monthly obligation
Compliance in periods in which no message is dispatched
Response and engagement trajectories across the six intervention cycles
Quiz participation, completion, and score, and the trajectory of scores across cycles

This is a large-scale randomised controlled trial testing whether the style and content of digital tax communication affect compliance among newly registered individual taxpayers in Indonesia. Randomisation is conducted at the individual taxpayer level. Eligible taxpayers are assigned to a control group (standard DGT communication) or one of six treatment arms, which vary along two dimensions: message content (deterrence-focused vs. information-focused) and communication style (formal vs. casual). Two of the six arms additionally include personalised information comparing the taxpayer's reported business or financial indicators against relevant industry benchmarks.

Randomisation is stratified on registration cohort, age, gender, income level (where available), economic sector, geographic location, and communication channel. Messages are delivered via WhatsApp or email, depending on the availability of valid contact information in DGT's administrative records; channel assignment follows contact-information availability rather than independent randomisation, so within-channel treatment-control comparisons are causally identified, while across-channel comparisons are descriptive only. Treatment is delivered across six repeated intervention cycles from October 2026 to March 2027, allowing the study to examine whether treatment effects strengthen, weaken, or remain stable with repeated exposure.

Not available

Randomisation is conducted at the individual taxpayer level, using Stata's randtreat command, stratified on registration cohort, age, gender, income level (where available), economic sector, geographic location, and communication channel

Individual taxpayer. Randomisation is conducted at the individual level, with each eligible taxpayer independently assigned to the control group or one of six treatment arms.

No

Not applicable — randomisation is at the individual taxpayer level; there is no clustering above the unit of randomisation.

Approximately 1.1 million newly registered individual taxpayers.

To be determined. Randomisation will allocate the eligible sample (approximately 1.1 million taxpayers — pending confirmation, see note on prior field) across the control group and six treatment arms; the final allocation ratio will be documented prior to treatment assignment.

PAP