Eliciting Preferences for Redistribution in the Presence of Evasion

The survey asks people to make redistribution choices when faced with trade-offs such that redistribution is costly.

Respondents are randomly assigned to one of two framings: (1) a gain-gain framing (T1), where both recipients benefit from the allocation, or (2) a loss frame (T2), where the endowment is initially given to the richer recipient and may be redistributed.

The social welfare weights are subsequently used in optimal tax simulations for Kenya.

Respondents in the survey face a distributional choice resembling Okun’s (1975) 'leaky-bucket' problem. Participants allocate a fictional endowment of 1,000 Ksh between two hypothetical recipients: R1, who is better off, and R2, who is worse off. Allocating more funds to R2 leads to an efficiency loss, reducing the total available resources. Thus, participants must balance equity against overall efficiency. The marginal efficiency loss is increasing in the allocation to R2, introduced through an increasing marginal leakage rate shown numerically during the allocation task, thus efficiently identifying the MRS for consumption between R1 and R2.

Participants choose freely from a continuous range of allocations.
The survey comprises three rounds, each with two questions:
- Both recipients are tax-compliant, but one has higher income (R1 > R2).
- Recipients have identical pre-tax incomes but differ in tax compliance (R2 compliant, R1 non-compliant).

Aside from these conditions, hypothetical incomes vary randomly to capture preferences across different income levels.

The main survey (T1) uses a 'gain-gain' framing, allowing both recipients to benefit from allocation decisions, thereby avoiding biases related to initial ownership. To test for the importance of framing effects, a secondary framing (T2) assigns the initial endowment entirely to the richer recipient, R1, and reframes the problem as a transfer from R1 to R2.

2025-06-18

2026-01-01

Schedules of Marginal Social Welfare Weights (MSWWs), defined over two dimensions: (1) gross income and (2) formality. Gain-gain framing, full sample.

When aggregating from individual preferences to societal preferences we employ a bottom-up aggregation approach which imposes transitivity.

The primary outcome of the survey is the schedule of marginal social welfare weights for formal and informal workers respectively across the income distribution.

These MSWWs are derived from respondents’ continuous allocation choices using a revealed preference approach, estimating marginal rates of substitution between equity and efficiency.
The main outcome will be the weighted sample of all T1 (treatment 1, gain-gain framing) responses. Confidence bounds will be based on 1,000-replication nonparametric bootstrap resampling of the respondent-level preference schedules.

MSWWs for Formal Workers Only

This outcome is MSSWs for T1 (gain-gain) respondents restricted to only those in the formal sector. This is investigated to test the hypothesis that formal workers have more political sway and thus have MSWWs more closely aligned with actual policy.

The survey uses costly redistribution to elicit MSWWs.
Two framings (T1 gain-gain, T2 transfer-loss) are randomly assigned. Respondents make continuous allocation decisions under conditions with income inequality or tax non-compliance.

Respondents in the survey face a distributional choice resembling Okun’s (1975) 'leaky-bucket' problem. Participants allocate a fictional endowment of 1,000 Ksh between two hypothetical recipients: R1, who is better off, and R2, who is worse off. Allocating more funds to R2 leads to an efficiency loss, reducing the total available resources. Thus, participants must balance equity against overall efficiency. The marginal efficiency loss is increasing in the allocation to R2 thus efficiently identifying the MRS for consumption between R1 and R2.

Participants choose freely from a continuous range of allocations.

The survey comprises three rounds, each with two questions:

In the first question, both recipients are tax-compliant, but one has higher income (R1 > R2).

In the second, recipients have identical pre-tax incomes but differ in tax compliance (R2 compliant, R1 non-compliant).

Aside from these conditions, hypothetical incomes vary randomly to capture preferences across different income levels.

The main survey (T1) uses a 'gain-gain' framing, allowing both recipients to benefit from allocation decisions, thereby avoiding biases related to initial ownership. To test for the importance of framing effects, a secondary framing (T2) assigns the initial endowment entirely to the richer recipient, R1 and reframes the problem as a transfer from R1 to R2.

Computer-based randomization. Uses PPS to choose random sub-counties. Within each sub-county a random starting point is assigned. From this point enumerators are assigned a random route to walk, surveying every other household. If a selected household does not participate, enumerators skip to the next household on the route. Assignment of survey (T1 or T2) is handled in the app. T1 is assigned 80 pct. of the time.

Treatment is randomized at the individual respondent level. Sampling is clustered: 10 geographic clusters (starting points). Within each, enumerators conduct individual-level randomization for survey assignment.

No

10 Geographic clusters. 900 households.

900 Respondents

80/20 Split between survey T1 and survey T2. Split is not enforced within clusters and may thus vary across clusters.

Sample size determined by budget and logistical constraints; primary outcomes used for descriptive inference with bootstrapped confidence intervals