Rwanda's tax authority tested threatening letters against friendly ones and found the threats were backfiring on its best-paying taxpayers

the problem

corporate taxpayers under-declaring, and deterrence letters not moving them

background

what everyone would do

The standard tool for chasing under-declared corporate tax, and Rwanda's own existing practice, was a deterrence letter warning of audit risk and penalties, on the assumption that a threat of consequences must be the strongest lever for pushing taxpayers to correct their filings.

what they saw

The Rwanda Revenue Authority saw that no evidence actually confirmed deterrence letters were the most effective option in a low-income country's tax administration specifically, where audit capacity was thin and taxpayers might reasonably doubt any threat would actually be followed up, meaning the deterrence approach was untested assumption, not proven practice. Rather than continuing to trust the deterrence letter because it was what every revenue agency sent, the RRA tested it directly against alternative message content, explaining what tax revenue funds or a plain reminder, and alternative channels, letting real taxpayer behavior settle which combination actually worked rather than which one seemed intuitively strongest.

the move

The RRA measured which combination actually raised declared revenue against the no-message control group, rather than trusting the deterrence letter on the assumption that a warning must outperform a mere reminder.

why it works

Randomly assigning 9,000 real taxpayers to nine different message-and-channel combinations plus a no-message control let the RRA measure genuine causal effects on declared revenue rather than relying on assumption or anecdote, revealing that messages explaining what tax funds increased compliance by roughly 12.3% versus control, meaningfully ahead of deterrence's 7.2%, and that deterrence actually backfired among higher-income taxpayers even while working as intended for everyone else. Because the experiment tested real behavior at scale rather than trusting theory, the RRA discovered a result no internal debate would likely have surfaced, that its best-paying taxpayer segment responded worse, not better, to threats, a finding that could only come from measuring actual outcomes across a genuinely varied taxpayer population. The intervention's cost was minimal relative to its return, close to $9 million in additional declared revenue against a few thousand dollars in messaging costs, making the experiment itself, not just its winning message, a disproportionately high-value one-time investment.

the payoff

Messages explaining what tax funds increased compliance by roughly 12.3% versus the control group, ahead of deterrence messaging's 7.2%, and deterrence produced a documented backfiring effect among higher-income taxpayers even as it worked as intended for the rest. Across the full set of interventions the RRA reported close to $9 million in additional declared revenue against a few thousand dollars in messaging costs. A companion study found the effect faded within a year, so the RRA treats it as a cheap, repeatable nudge rather than a one-time fix.

where it breaks

The mechanism depends on the tested population being large and varied enough to reveal genuine subgroup differences, a smaller or more homogeneous taxpayer pool might not have surfaced the specific finding that deterrence backfires on higher-income filers even while working for others, meaning the value of large-scale randomized testing scales with how heterogeneous the actual population's response turns out to be. It also depends on the effect being durable enough to justify the ongoing cost of sending messages, and the RRA's own companion study found the effect faded within a year, meaning nudges like this function as a cheap, repeatable intervention requiring periodic refreshing, not a permanent one-time fix, a distinction that matters for planning long-term compliance strategy. And a result discovered in one country's specific administrative and cultural context doesn't automatically transfer to a different tax authority or population, the RRA's finding shaped how ICTD advises other African revenue authorities, but each context still benefits from its own testing rather than assuming Rwanda's specific winning message and backfire pattern will replicate identically elsewhere.

what came after

The Rwanda experiment, run with the International Centre for Tax and Development and the African Tax Administration Forum, is cited in tax-administration literature as one of the first large-scale randomized compliance-message trials run inside an African revenue authority, and the finding that deterrence backfires on high-income filers has shaped how ICTD advises other African tax authorities designing their own nudge letters.

references

1. [1]Tax Compliance in Rwanda: Evidence from a Message Field ExperimentInternational Centre for Tax and Development, 2021ictd.ac
2. [2]Experimental research on tax compliance in RwandaInstitute of Development Studies, 2016ids.ac.uk
3. [3]Communicating to Improve Compliance: Taxpayers' Feedback on Message and ModeRwanda Revenue Authority, 2016rra.gov.rw

keep it

same kind of clever

• 2010 · American ExpressAmex didn't advertise its card on the biggest shopping weekend of the year — it invented a whole new day inside it
• 1962 · AvisAvis told customers outright it was only #2, and the confession made everything else it said believable
• 2025 · AXA France (with Publicis Conseil)AXA covered domestic violence like a house fire by changing three words in a contract millions of people already had