Tax Morale and Institutions: Experimental Evidence from East Africa

2026-10-19

2026-11-19

(1) Tax morale, measured by the endline item asking whether a taxpayer who underreports to pay less tax is doing something not wrong at all, wrong but understandable, or wrong and punishable. (2) Perceived responsiveness, measured by the endline item asking how likely it is (0 to 100) that residents could get together and make their elected councilor listen to a matter of community importance. (3) Perceived spending quality, measured by the endline item asking how well Nairobi County uses tax money (0 to 100).

Reply to the voice invitation in T1 and T3: any reply, and roads/drainage vs. healthcare. Endline items: self-reported payment of property rates in the current period; trust in the President, County Governor, Kenya Revenue Authority and County Assembly (1 to 7); whether local governments serve constituents better than higher levels of government; message recall, attributed sender and perceived truthfulness; ownership of property subject to local property rates; payment of non-tax fees to the county in the last six months; presence of a post office, school, police station and health clinic within walking distance; ranking of county service priorities; and survey response rate by arm.

Nairobi County residents with mobile numbers in county contact records are randomized into five equal groups defined by the content of a single SMS sent by Nairobi County. T1 (Voice): the message states that the county is planning projects that improve Nairobi and invites the recipient to reply 1 for roads/drainage or 2 for healthcare to indicate which is most important. T2 (Information): the message states that the county is planning projects that improve Nairobi, including paving new roads and constructing new hospitals. T3 (Voice and Information): the information in T2 followed by the invitation in T1. P (Placebo): a county reminder of sub-county office and Huduma Centre opening hours, with no reference to spending or taxes. C (Control): no message. The design is a 2 x 2 factorial in voice and information; the placebo arm separates message content from the effect of receiving any message from the county, and the no-message control provides a baseline for all endline outcomes. Project areas and content are drawn from the Nairobi County 2025/2026 Annual Development Plan and were selected from a pilot SMS in which about 1,000 residents chose among ten county service areas. After the treatment round, all five groups receive an identical SMS from a research number inviting them to a five-minute endline survey on county services, incentivized by lottery [ten prizes of KES 10,000, or one prize of USD 1,000, per the approved consent]. The survey covers tax morale, self-reported payment of property rates, property ownership, payment of county fees, whether local government serves constituents better than higher levels, message recall, attributed sender and credibility, perceived spending quality, trust in the President, County Governor, Kenya Revenue Authority and County Assembly, perceived responsiveness of the elected councilor, presence of public facilities within walking distance, and ranking of county service priorities. Pre-specified comparisons are T1 and T3 vs. P (voice), T2 and T3 vs. P (information), the voice x information interaction, and P vs. C (any county message).

Not available

Randomization done by the relevant tax and government authority by computer (in Excel). Individual property owners are assigned to one of three groups (control, reminder, reminder plus participation), with the same number of subjects per arm. Randomization is at the individual level, stratified by electoral ward (85 wards in Nairobi).

Individual taxpayers

No

Errors will be clustered at the ward level, with 85 wards in Nairobi.

Planned number of observations: 471,935 Nairobi County residents with mobile numbers in county records. All outcomes are measured on endline respondents; at the 5 percent response rate assumed in the pre-analysis plan, about 23,600 respondents, roughly 4,700 per arm.

94,387 individuals in each of five arms: 94,387 control (no message); 94,387 placebo (county office-hours message); 94,387 voice; 94,387 information; 94,387 voice and information. Total 471,935 individuals, allocated proportionally across the 85 electoral wards of Nairobi. All outcomes are measured on endline survey respondents; assuming 90 percent SMS delivery and a 5 percent survey response rate, approximately 4,250 respondents per arm, about 21,000 in total.