Experimental Design
In this design we randomize at the property-owner level with 3 treatment arms (a control group with business-as-usual paper bills, a digital billing group (D) and a digital billing plus payment group (DP)). Randomizing at the property-owner level minimizes SUTVA violations and ensures that the same property owner does not receive two different treatments. Randomization is stratified at the property-type and formality-level, ensuring balanced representation of taxpayers across key predictors of compliance and supports precise estimation of heterogeneous effects. Property-owners who hold multiple properties across these 4 strata types will need a pre-specified rule or assignment. The rule is as follows: (i) if the owner holds at least one Residential/Formal property, they are assigned to the Res/Formal stratum; (ii) if not, but they hold at least one Residential/Informal property, they are assigned to the Res/Informal stratum; (iii) if not, but they hold at least one THA/Formal property, they are assigned to the THA/Formal stratum; and (iv) otherwise, they are assigned to the THA/Informal stratum. This rule reflects the policy priority of residential properties over THA properties, and formal neighborhoods over informal ones within each property type, and ensures a consistent and fully pre-specified assignment for all cross-stratum owners without requiring case-by-case judgment. If multiple properties fall within the same stratum, the property with the highest tax liability will be selected.
The preferred design is structured in two parts: (i) a pooled analysis with a Bonferroni correction and (ii) a pre-specified within-stratum analysis with no correction (single analysis). The first part estimates the overall average treatment effect using a pooled regression with stratum fixed effects, specified as:
Y= α+β1D+β2DP+si+ε
Y
=
𝛼
+
𝛽
1
D
+
𝛽
2
D
P
+
s
i
+
𝜀
where Y is a binary indicator equal to 1 if the property owner complied with the land and improvement tax in the study period and 0 otherwise, D and DP are indicators for the Digital Bill and Digital Bill + Payment treatment arms respectively, and stratum fixed effects (sᵢ) absorb baseline compliance differences across strata, improving precision beyond what a simple two-sample calculation assumes. The pooled baseline compliance is computed as a weighted average across the four strata, with weights reflecting each stratum’s share of total properties in the sub-sample of the Blantyre tax registry for which phone numbers have been collected.
This specification from the pooled data allows us to test two hypotheses:
H1 — the full intervention increases compliance relative to control (DP vs. C, corresponding to
β2
𝛽
2
), tested at α* = 0.025; and
(ii) H2 — the payment feature adds incremental value beyond digital billing alone (DP vs. D, corresponding to
β2−β1
𝛽
2
−
𝛽
1
), also tested at α* = 0.025. There is a Bonferroni correction applied to control the family-wise error rate (FWER) at α = 0.05 Treating both comparisons as co-primary reflects the study’s dual policy objectives: establishing the overall effectiveness of the integrated intervention and isolating the specific contribution of the payment mechanism. The comparison of digital billing against control (D vs. C, corresponding to
β1
) is retained as an informative decomposition estimate and does not constitute a pre-specified hypothesis; it is tested at the full α = 0.05 and reported for interpretive purposes only.
For the second, within-stratum analysis, we primarily test DP vs C. The secondary comparison (DP vs. D) and informative comparison (D vs. C) are reported within each stratum as exploratory. The final required sample size based on power calculations, is the larger of the pooled co-primary analysis and the pre-specified subgroup, single hypothesis per stratum grand, since both constraints must be satisfied. The pooled co-primary analysis is the binding constraint, requiring 1,028 per arm and 3,084 in total, compared to 2,127 for the pre-specified subgroup design alone.
The final sample size used for the randomization is however, based on properties that satisfy these criteria: (i) properties must be matched to the 2021 tax roll with available locational information, and (ii) property owners must be charged land and improvement taxes in 2024 and (iii) their phone numbers should have been collected, through a pilot phone number collecting exercise by the BCC in 2026. This leaves us with a sample of 6,659 owners. We then randomly sample 1/3rd observations from the four strata, Residential/Formal, Residential/Informal, THA/Formal, THA/Informal, into each treatment arm.
To ensure that those receiving the digital bill view it as authentic, we design the digital bill to closely resemble physical bills. The digital bill will therefore include the charges and valuation amounts for all three tax categories: land, improvements, and sewerage. It will also include a surcharge notice, the Blantyre City Council header, and the Council’s motto.