Randomized incentives to increase participation in COVID testing in rural Kenya

This study in rural Kenya demonstrates that while cash incentives and revisits significantly increase participation in door-to-door COVID-19 testing and help reduce nonresponse bias, offering a lower incentive of KSh 200 proved more cost-effective than higher amounts despite slightly lower detection rates among the hardest-to-reach individuals.

The Gist

Imagine you are trying to take a census of a small, rural village to see how many people are sick with a virus. But there's a catch: the villagers are busy, they might be shy, or they just don't want to stop what they're doing to talk to you. If you only count the people who happen to be home and willing to talk, your numbers will be wrong. You might miss the sickest people because they are too sick to come out, or the busiest people who can't afford to lose a day of work.

This paper is about a team of researchers in rural Kenya who tried to solve this problem during the COVID-19 pandemic. They wanted to know: How do we get everyone to come out and get tested, and does paying them help?

Here is the story of their experiment, broken down into simple parts:

1. The "Bribe" Experiment (The Incentive)

The researchers went door-to-door in 12 villages. They wanted to test everyone for COVID-19. To encourage people to open their doors and take a test, they offered a cash reward (via mobile phone) for every family member who agreed to be tested.

But they didn't just offer one amount. They played a game of "Goldilocks" with three different price tags:

• The Small Change: 200 Kenyan Shillings (about $1.85).
• The Medium Meal: 350 Kenyan Shillings (about $3.25).
• The Big Payday: 700 Kenyan Shillings (about $6.50).

They randomly assigned these amounts to different households, like dealing cards from a deck, to see which "price tag" got the most people to say "Yes."

2. The Results: The "Busy Bee" Problem

The results were surprising and taught them a valuable lesson about human behavior.

• Most people said "Yes" anyway: Once a researcher actually found someone at home, almost 96% of them agreed to take the test, no matter how much money was offered. The money wasn't the main reason they said yes; it was just a nice bonus.
• The real problem was "Not Home": The biggest hurdle wasn't people refusing; it was people being unavailable. About 17% of people were never home, even after the researchers came back a second time. They were at work, at the market, or out in the fields.
• The "Pay More" Trap: Offering the big $6.50 reward did get a few more people to say yes compared to the $1.85 reward, but it was very expensive to get those extra few people. It was like paying a premium for a tiny upgrade. The researchers realized that the cheapest option ($1.85) combined with a second visit was actually the most cost-effective way to get the most data.

3. The "Hidden Virus" Discovery

Here is the most critical part of the story. The researchers found that the people who needed the biggest bribe (the $6.50 group) were actually more likely to be sick than the people who said yes for the small bribe.

Think of it like this: Imagine you are fishing.

• The small bait catches the fish that are swimming right near the surface (the easy-to-reach, healthy people).
• The expensive bait is needed to catch the fish that are hiding deep underwater or are very cautious (the people who are busy, sick, or worried).

Because the people who required the highest payment were slightly more likely to have the virus, the researchers realized that if they had only offered the small payment, they would have underestimated how much the virus was spreading. The "hard-to-reach" people were the ones carrying the most risk.

4. The Takeaway: Why This Matters

This study is like a map for public health officials. It tells us:

1. Don't overpay: You don't need to offer a fortune to get people to participate in health checks. A small, guaranteed cash reward is often enough to get high participation rates.
2. Keep trying: If you can't find someone the first time, go back. A second visit is often cheaper and more effective than just raising the reward money.
3. Watch out for the "invisible" sick: If you only test the people who are easy to find, you might think the virus is less dangerous than it really is. The people who are hardest to reach might be the ones who need help the most.

In a nutshell: The researchers found that a little bit of cash goes a long way in getting people to get tested, but you have to be persistent (go back for a second visit) and realize that the people who are hardest to reach might be the ones who are actually the sickest. This helps governments get a truer picture of the disease so they can fight it better.

Technical Summary

1. Problem Statement

Public health surveillance relies on high participation rates in testing to generate accurate estimates of disease prevalence. Low participation leads to non-response bias, resulting in underestimates of disease burden and inefficient resource allocation. While cash incentives are known to improve response rates in high-income settings, there is limited evidence regarding their cost-effectiveness and impact on non-response bias in low-resource environments. This study addresses the gap in understanding how monetary incentives affect participation in door-to-door SARS-CoV-2 testing in rural Kenya and whether higher incentives can mitigate non-response bias.

2. Methodology

• Study Design: A cross-sectional, randomized controlled trial (RCT) conducted in 12 rural villages across Vihiga and Bungoma Counties in Western Kenya.
• Timeline: February 12, 2021, to April 25, 2021 (coinciding with Kenya's third wave of the pandemic).
• Population: 7,049 individuals from 1,345 households (2,291 adults, 3,368 minors). Eligibility included all residents over 3 months of age.
• Intervention (Randomization): Households were randomly assigned to receive a conditional cash incentive via M-Pesa (mobile money) for each household member who consented to testing. The incentive levels were:
  • Group 1: KSh 200 (~$1.85 USD)
  • Group 2: KSh 350 (~$3.23 USD)
  • Group 3: KSh 700 (~$6.47 USD)
  • Note: The highest amount was estimated to compensate for approximately one day of lost wages.
• Procedure:
  • First Visit: Enumerators visited households. If no one was home, a revisit was scheduled. If adults were home, the randomized incentive was announced, and consent was sought.
  • Revisits: Targeted households where individuals were initially unavailable or had declined.
  • Data Collection: Participants provided nasal swabs (for qPCR) and dried blood spots (DBS) (for ELISA antibody testing). Surveys collected data on symptoms, risk behaviors, and financial stress.
• Statistical Analysis:
  • OLS regression models with village and enumerator fixed effects to estimate the impact of incentives on availability and consent.
  • Marginal cost analysis calculated the cost per additional consent for different incentive tiers and revisit strategies.
  • Infection rate comparisons between incentive groups and visit types (first vs. revisit) to assess non-response bias.

3. Key Contributions

• Cost-Effectiveness Analysis: Provides empirical data on the marginal cost of increasing participation through higher incentives versus revisits in a rural African setting.
• Non-Response Bias Quantification: Investigates whether individuals who require higher incentives (or are harder to reach) have systematically different infection rates than those who participate easily.
• Operational Insights: Demonstrates the efficacy of Community Health Worker (CHW) models in rural Kenya for public health surveillance, leveraging existing trust structures.

4. Key Results

• Participation Rates:
  • Overall Consent: 80.3% of all recorded individuals consented.
  • By Incentive:
    • KSh 200: 78.2%
    • KSh 350: 80.6%
    • KSh 700: 81.9%
  • Availability vs. Consent: The primary barrier was availability (17.2% of individuals were never home), not refusal. Among available individuals, consent rates were extremely high (95.1%–98.2%) across all groups.
  • Impact of Incentives: Increasing the incentive from KSh 200 to KSh 700 increased overall consent by 3.8 percentage points. This was driven by a slight increase in availability (2.2 p.p.) and a slight increase in consent among those available (2.4 p.p.).
• Cost Efficiency:
  • The marginal cost per additional consent was significantly lower for revisits at the lowest incentive (KSh 200) compared to first visits at higher incentives.
  • Increasing incentives from KSh 200 to KSh 350 cost ~KSh 9,581 per additional first-visit consent, whereas increasing to KSh 700 cost ~KSh 14,928.
  • Conclusion: Higher incentives were not cost-effective because the lowest incentive already achieved very high participation rates.
• Infection Prevalence:
  • Active Infection (qPCR): 0.3% (95% CI: 0.2, 0.5).
  • Cumulative Infection (Antibodies/ELISA): 8.6% (95% CI: 7.7, 9.6).
  • Non-Response Bias Evidence: Individuals in the KSh 700 group had a seroprevalence roughly 2.5 percentage points higher than the KSh 200 group. This suggests that those requiring higher compensation (or harder to reach) were more likely to be infected, meaning lower incentives would have led to an underestimation of community prevalence.
  • Revisit Bias: No significant difference in infection rates was found between first visits and revisits, suggesting that "hard-to-reach" individuals (due to schedule) were not necessarily more infected, but those who refused lower incentives were.

5. Significance and Implications

• Policy Recommendation: In rural settings with high baseline trust in community health workers, low-to-moderate cash incentives (e.g., KSh 200) combined with revisit strategies are the most cost-effective method to maximize participation.
• Bias Mitigation: While low incentives are cost-effective, the study warns that they may introduce non-response bias by missing subpopulations with higher infection rates who require higher compensation to participate. To obtain accurate prevalence estimates, public health campaigns may need to offer higher incentives to specific hard-to-reach groups or use alternative data collection methods (e.g., phone surveys, self-sampling).
• Contextual Factors: High participation was likely driven by the familiarity of the CHW model and social norms, rather than fear of infection (which was reported as low).
• Limitations: The study lacked a zero-incentive control group, making it difficult to determine the absolute baseline participation rate. Additionally, supply chain issues prevented ELISA testing for ~5% of samples.

Conclusion: The study demonstrates that while cash incentives effectively boost participation in rural Kenya, the marginal benefit of increasing the incentive amount diminishes rapidly. However, to ensure epidemiological accuracy and avoid underestimating disease burden, public health strategies must account for the fact that the "hardest to reach" (or those needing higher incentives) may represent a higher-risk population.