Honduras's tax agency found that announcing its own anti-corruption reform made taxpayers cheat more, until it told them the truth about corruption first

the problem

A real institution had genuinely reformed itself after a corruption problem, but announcing the reform made citizens trust it less, not more

background

what everyone would do

Announce the anti-corruption reform directly and clearly -- tell taxpayers the agency has cleaned house, renewed its staff, and tightened its human-resources policies -- the standard, seemingly self-evident move for an institution that has genuinely improved and wants citizens to know it.

what they saw

The researchers saw that a reform announcement is not neutral information -- it doubles as a signal that the underlying problem was serious enough to require fixing. If the audience's prior belief about how corrupt the institution was is already more pessimistic than reality, an announcement that confirms 'yes, we're fixing corruption' doesn't correct that belief, it validates it, and the taxpayer responds to the confirmed-bad institution rather than to the actual improvement.

the move

Ajzenman, Ardanaz, Cruces, Feierherd and Lunghi ran a survey experiment testing three message types against a Honduran taxpayer sample: a 'reform' message describing SAR's anticorruption efforts alone, a 'debiasing' message using an independent, non-government source (Latinobarometro) to show Honduras's tax-agency corruption was actually among the lowest in the region, and a 'sequential' message that showed the debiasing message first and only then the reform announcement. They then validated the approach in a government-run field experiment: SAR itself emailed a representative sample of 45,000 real taxpayers before the 2023 tax-filing deadline, testing the debiasing and sequential messages against a plain control (the reform-only message was deliberately excluded from the field test because the survey experiment had already flagged it as likely to backfire).

why it works

Showing an independent, non-government benchmark first -- Honduras's tax agency was actually among the least corrupt in the region per Latinobarometro survey data -- shifted taxpayers' reference point about the baseline before they ever heard the reform announcement, so the announcement landed as confirmation of an institution already known to be relatively clean, rather than as evidence of one bad enough to need saving. Because the sequential treatment corrected the belief before adding new information that could be misread through it, it avoided the priming/confirmation-bias trap the reform-only message fell into, and the effect showed up not just in self-reported survey intentions but in real declared income on actual government tax filings.

the payoff

In the survey experiment, the reform-only message backfired: it increased taxpayers' self-reported willingness to evade taxes by roughly 11-14 percentage points and increased dishonest behavior in an incentivized dice-game measure by about 4 percentage points over an 8% baseline, while the sequential message reduced perceived corruption and evasion willingness. In the field experiment, of the roughly 45,000 taxpayers emailed, about 33,000 filed returns; the sequential treatment increased declared income by 180,000 lempiras on average (7% over control, p<0.1), rising to 875,000 lempiras among taxpayers already flagged as high non-compliance risk, while the debiasing message alone produced no statistically significant effect on its own.

where it breaks

The approach depends on there being a genuinely favorable independent benchmark available to correct the audience's prior belief -- if the institution really is as bad as people fear, or worse, no credible debiasing message exists to lead with, and the sequencing fix has nothing true to work with. It also requires the debiasing source to be seen as independent and credible rather than another arm of the same institution trying to defend itself; a benchmark that reads as self-serving collapses back into the same confirmation-bias problem the sequencing was designed to avoid, and the debiasing message alone, without the follow-up reform announcement, produced no measurable compliance effect on its own in the field data.

what came after

The study is one of the few anticorruption-messaging experiments to combine a controlled survey design with a real government field intervention showing an effect on actual filed tax returns rather than only self-reported intentions, and it reframes anticorruption communication as a sequencing problem rather than a content or tone problem.

references

1. [1]Unraveling the Paradox of Anticorruption Messaging: Experimental Evidence from a Tax Administration ReformRIDGE-LACEA Working Paper 012 / Inter-American Development Bank (Ajzenman, Ardanaz, Cruces, Feierherd, Lunghi), 2024ridge.org.uy
2. [2]Unraveling the Paradox of Anticorruption Messaging (SSRN working paper record)SSRN / Inter-American Development Bank, 2024papers.ssrn.com

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