THE GOLD STANDARD OF INTELLIGENCE · SPECIALIST STUDY
Treat an incentive change as a study intervention, not a fieldwork convenience
Respondent-driven sampling uses primary compensation for participation and often a secondary incentive for successful peer recruitment. Changing either amount can alter redemption, speed, recruiter effort, who is approached, who accepts, and how much peer pressure enters the process. A defensible experiment therefore estimates more than one recruitment rate: it traces mechanisms, protects voluntariness, tests heterogeneous effects and shows how the resulting recruitment tree changes the population estimate.
This guide belongs to the Intelligence subject root, the Gold Standard of Education, and the How X Works hub. It begins with a concrete decision, then tests whether the method still holds when the context changes.
Choose a reading route
Define the ethical and causal question · chapters 1, 2, 3
Design the amount experiment · chapters 4, 5, 6
Measure mechanisms and effects · chapters 7, 8
Protect inference · chapters 9, 10, 11
Decide, transfer and govern · chapters 12, 13
The whole problem at a glance
| Outcome | What it answers | Why it is insufficient alone |
|---|---|---|
| Coupon redemption | Whether issued coupons produce enrolments | Does not reveal pressure or composition |
| Time to recruit | Whether chains move faster | Can improve while subgroup reach worsens |
| Recruitment productivity | Recruits per eligible recruiter | Mixes effort, opportunity and acceptance |
| Estimator sensitivity | Whether prevalence estimates move | Needs a causal account of how recruitment changed |
No single “recruitment success” number captures the consequences of changing money in a peer network.
CHAPTER 1 OF 13 · Define the ethical and causal question
1. Separate primary and secondary incentives
Primary compensation recognises the participant’s own time, inconvenience or expenses. A secondary incentive rewards a successful peer referral. They act at different points and on different people, so changing one cannot be interpreted as changing “the incentive” in the abstract.
Draw the sequence: recruiter receives coupons, chooses whom to approach, a peer decides whether to contact the study, eligibility is assessed, participation occurs, and the recruiter may be paid. Place each amount and safeguard on that sequence.
State whether the estimand is the effect of offering an amount, receiving it, or operating under a full incentive policy. Actual payment depends on recruitment success and is post-treatment; analysing only paid recruiters creates selection.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Replace cash with transport reimbursement or a raffle; mechanism, certainty and timing change even if expected monetary value looks similar. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 1 check · what should survive an audit
- Name the actor and stage affected by each payment.
- Define offer, receipt and policy estimands separately.
- Avoid conditioning the main effect on successful recruitment.
CHAPTER 2 OF 13 · Define the ethical and causal question
2. Define recruitment elasticity without borrowing a market metaphor too literally
Recruitment elasticity can describe the proportional change in a recruitment outcome for a proportional change in incentive amount. It is a useful summary only over the tested range and policy context. It is not a universal property of a population.
For example, raising a secondary incentive from $5 to $10 is a 100% increase. If coupon redemption rises from 20% to 26%, the arc elasticity is positive, but the causal interpretation still depends on assignment, clustering, timing and concurrent fieldwork changes.
Report absolute effects beside elasticities. A six-percentage-point change can matter operationally even when the percentage calculation looks dramatic because the baseline amount was small. Include confidence intervals and cost per additional eligible enrolment.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Repeat the same nominal amounts during a transport disruption or inflation shock; real value and recruitment barriers change, so the earlier elasticity should not be copied. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 2 check · what should survive an audit
- Specify the tested amount range.
- Report absolute and relative effects.
- Tie elasticity to a dated fieldwork context.
CHAPTER 3 OF 13 · Define the ethical and causal question
3. Distinguish coercion, undue influence and ordinary motivation
Coercion involves a threat, while an incentive is an offer. That distinction does not end the ethical analysis: an offer can still raise undue-influence, exploitation, privacy or peer-pressure concerns, especially in vulnerable or highly connected communities.
RDS adds a peer recruiter who is not study staff. Pressure may occur through repeated requests, debt, household authority, disclosure of group membership or attempts to capture another person’s compensation. Recruiter training and confidential reporting therefore belong to the intervention.
Measure voluntariness rather than asserting it. Ask recruits privately how they were approached, whether refusal felt safe, whether compensation was misrepresented, and whether anyone requested a share. Define escalation and support procedures before data collection.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Move from individual coupons to household-linked recruitment; power relations may intensify even if the amount stays fixed. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 3 check · what should survive an audit
- Do not label an offer coercive without analysing the mechanism.
- Measure peer pressure privately.
- Predefine responses to threats, payment claims or disclosure.
CHAPTER 4 OF 13 · Design the amount experiment
4. Choose a randomisation unit that respects network interference
Individual recruiters can be randomised to different amounts, but their recruits and network neighbours may exchange information. One person’s assignment can affect another’s behaviour, violating a simple no-interference assumption.
Possible units include recruiter, seed chain, time block or site. Cluster randomisation can reduce contamination but often provides fewer independent units. Time-block designs risk confounding amount with epidemic, weather or operational changes.
Map plausible spillovers before choosing the unit. Record what participants are told about amounts, whether assignments are visible, and how cross-arm referrals are handled. Analyse at the level supported by assignment and dependency.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Run the study across several sites with different cash-handling rules; stratify or randomise within site rather than treating site policy as noise. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 4 check · what should survive an audit
- Name the randomisation unit.
- Draw likely information and recruitment spillovers.
- Use uncertainty estimates that respect clustering.
CHAPTER 5 OF 13 · Design the amount experiment
5. Use a factorial design only when both mechanisms matter
A factorial design can vary primary and secondary amounts independently, estimating participation, recruitment and interaction effects. It is efficient when all combinations are ethically and operationally acceptable.
The fictional four-arm table separates whether more compensation changes a person’s own participation from whether more referral reward changes recruiter behaviour. The interaction asks whether one amount changes the effect of the other.
Pre-specify marginal contrasts and protect against sparse arms. If seed chains are the randomisation unit, four conditions may require many chains. Do not trade interpretable replication for an elaborate design with one or two clusters per arm.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Add a non-cash support arm, such as extended appointment hours; decide whether the question remains factorial or becomes a multi-component policy comparison. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 5 check · what should survive an audit
- Ensure every combination is acceptable.
- Pre-specify marginal and interaction contrasts.
- Count independent clusters, not only participants.
Worked randomised incentive schedule
| Arm | Primary compensation | Secondary incentive | Pre-specified purpose |
|---|---|---|---|
| A | $25 | $5 per eligible recruit | Reference condition |
| B | $25 | $10 per eligible recruit | Recruitment-amount contrast |
| C | $35 | $5 per eligible recruit | Participation-amount contrast |
| D | $35 | $10 per eligible recruit | Interaction and combined cost |
Amounts are illustrative, not recommendations. Local purchasing power, burden, law, ethics review and community input determine plausible levels.
CHAPTER 6 OF 13 · Design the amount experiment
6. Set stopping and adaptation rules before chains diverge
Higher amounts may accelerate one arm, creating different calendar exposure, staff load and remaining network composition. A fixed total sample can then compare arms observed over unequal periods and stages of depletion.
Define recruitment-rate monitoring, safety signals, minimum representation, budget ceilings and maximum waiting time. An independent reviewer should be able to pause an arm for pressure reports or operational overload without using unblinded outcome preferences.
If the design adapts amounts mid-course, log the rule, decision time and eligible population. The effect of a static assignment differs from the effect of an adaptive policy that responds to chain performance.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Introduce a site closure halfway through fieldwork; preserve assignment history and separate the disruption response from the amount effect. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 6 check · what should survive an audit
- Predefine ethical and operational stopping signals.
- Track calendar exposure and network depletion.
- Distinguish static and adaptive policy estimands.
CHAPTER 7 OF 13 · Measure mechanisms and effects
7. Measure the recruitment funnel, not only enrolment
Count coupons issued, contacts attempted where privacy permits, recruit interest, appointments, eligibility, enrolment and successful payment. Different amounts may affect several stages, and the same enrolment rate can conceal opposite mechanisms.
Use denominators consistently. Coupon redemption per issued coupon differs from recruits per recruiter and from eligible enrolments per contact. Show all three where possible, with missing-contact information acknowledged.
Preserve zeroes. Recruiters who accept coupons but recruit nobody are part of the outcome. Excluding them inflates apparent effectiveness and can exaggerate the return to a higher secondary incentive.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Switch from paper coupons to electronic referrals; delivery and observability change, so rebuild the funnel and privacy analysis. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 7 check · what should survive an audit
- Define every funnel denominator.
- Retain zero-recruitment recruiters.
- Mark unobserved contact attempts as missing, not zero.
CHAPTER 8 OF 13 · Measure mechanisms and effects
8. Estimate heterogeneous effects without turning subgroups into stories
An average amount effect can hide differences by network size, income, gender, age, site, baseline chain depth or access cost. Some differences may reflect opportunity to recruit rather than motivation.
Pre-specify a small set of plausible modifiers and interactions. Report uncertainty and shrink noisy estimates where appropriate. A visually larger subgroup bar is not evidence of real effect modification without a supported contrast.
Pair quantitative heterogeneity with mechanism evidence. If a remote subgroup responds more to transport reimbursement, travel burden is a plausible pathway. If a higher referral reward changes only well-connected recruiters, network opportunity may dominate.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Transport the policy to a population with different network density; effect modification by degree can change the overall average even when within-degree effects persist. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 8 check · what should survive an audit
- Pre-specify modifiers and contrasts.
- Report uncertainty, not subgroup rankings alone.
- Connect heterogeneity to measured mechanisms.
CHAPTER 9 OF 13 · Protect inference
9. Trace how incentives reshape the recruitment tree
RDS estimators rely on information about network degree, recruitment links and sampling assumptions. An amount change can alter who recruits, whom they approach, chain length, bottlenecks and cross-group recruitment.
Compare branching, wave depth, homophily, recruiter productivity and subgroup transition matrices by arm. Do not treat these as decorative diagnostics; they show whether the intervention changed the sampling process that underlies estimation.
A faster arm may reach equilibrium-like composition sooner, or it may amplify a well-connected subgroup. Preserve seed identity and calendar time so the same pattern is not misread as either amount effect or seed fortune.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Change the number of coupons per recruiter while holding money constant; branching opportunities change independently of incentive motivation. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 9 check · what should survive an audit
- Compare tree structure by assigned policy.
- Retain seeds, waves and calendar time.
- Do not infer representativeness from speed alone.
Decision table for amount changes
| Signal | Interpretation to test | Possible response |
|---|---|---|
| Faster redemption, stable composition | Lower delay without detected selection shift | Consider continuation with safeguards |
| Faster redemption, more pressure reports | Benefit paired with ethical harm | Pause or redesign recruiter process |
| Different subgroup composition | Heterogeneous participation or referral effect | Report stratum effects and estimator movement |
| No gain, higher cost | Low marginal recruitment elasticity | Retain lower amount or test non-cash barriers |
A pre-specified decision rule prevents a faster chain from automatically winning.
CHAPTER 10 OF 13 · Protect inference
10. Carry assignment through the estimator
The field experiment and population estimator answer linked but distinct questions. First estimate the policy effect on recruitment outcomes under the assignment design. Then examine how each policy changes weighted population estimates and their uncertainty.
Report estimates using the planned RDS estimator, alternative degree handling and sensitivity to recruitment assumptions. If the policy changes degree reporting or recruiter choice, a stable unweighted mean does not establish estimator stability.
Avoid choosing the amount that produces the preferred prevalence estimate. The decision should balance recruitment, ethics, cost and robustness under pre-specified criteria, with estimate movement treated as evidence about selection.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Use the same experiment for a binary and a continuous study outcome; estimator sensitivity and variance may change even when the recruitment tree is identical. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 10 check · what should survive an audit
- Separate intervention-effect and population-estimation analyses.
- Run pre-specified estimator sensitivities.
- Never select an arm because its prevalence looks desirable.
CHAPTER 11 OF 13 · Protect inference
11. Handle time, censoring and unequal opportunity
Recruiters assigned late have less time for coupons to be redeemed. Chains in a faster arm can also enter later waves while a slower arm remains near seeds. Simple end-of-study proportions therefore mix amount with exposure time.
Use time-to-recruit analyses with clear origins, coupon expiry and censoring. Consider competing outcomes such as ineligible contact or study closure. Cluster uncertainty by the assignment and recruitment structure.
Display cumulative recruitment with risk sets rather than only totals. A high early slope followed by rapid exhaustion tells a different operational story from sustained moderate recruitment.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Extend fieldwork for only the slower arm; the intervention now includes unequal follow-up, so estimate the policy actually implemented and retain the original fixed-window comparison. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 11 check · what should survive an audit
- Define time origin and censoring.
- Use comparable observation windows.
- Show risk sets and cumulative curves.
CHAPTER 12 OF 13 · Decide, transfer and govern
12. Convert results into a multi-criterion policy decision
The preferred amount is not necessarily the one with the most recruits. Build a decision table covering eligible enrolment, time, cost, subgroup reach, pressure reports, estimator stability, staff load and payment integrity.
Set unacceptable conditions before examining results, such as a credible rise in threats or payment capture. Among acceptable arms, compare incremental cost per additional enrolment and the uncertainty around all decision criteria.
Retain community and ethics-review input as evidence, not as a ceremonial appendix. A statistically efficient amount can still be rejected if the mechanism conflicts with local dignity, privacy or fairness.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Move the policy to a jurisdiction with different payment regulation and banking access; legal and delivery feasibility can dominate the earlier statistical optimum. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 12 check · what should survive an audit
- Use more than recruitment count.
- Apply pre-specified unacceptable conditions.
- Document community and ethics reasoning.
CHAPTER 13 OF 13 · Decide, transfer and govern
13. Publish the experiment so another team can reproduce the policy
Report amount, currency, timing, eligibility, payment method, primary and secondary components, assignment unit, coupon limit, expiry, recruiter script, safeguards and fieldwork dates. “Participants were incentivised” is not reproducible.
Publish the flow diagram by arm and the analysis code or formulas used for elasticity, time-to-recruit and estimator sensitivity. Describe deviations, cross-arm contamination and missing pressure interviews.
State the scope: the effect belongs to tested amounts, real purchasing power, population, network and implementation. Recommend a monitoring plan for any adoption rather than declaring one universal best payment.
Evidence discipline: record the observation, version or time boundary that supports this chapter’s conclusion. Preserve the rejected alternative as well as the selected action. That makes a later review able to distinguish a changed world from an undocumented change of mind.
Changed-context transfer: Repeat after inflation or a shift to digital wallets; treat delivery and real value as new intervention components. The method transfers only if the decision question and measurement meaning survive the change. If either changes, rebuild the comparison rather than copying the earlier threshold or weight.
Chapter 13 check · what should survive an audit
- Make the complete policy reproducible.
- Publish deviations and contamination.
- Limit claims to the tested context and amounts.
Contents · Previous chapter · Return to the Intelligence guide
Verification lab: work the decision twice
Fictional study: 24 seed chains are randomised within site to a $5 or $10 secondary incentive. The higher arm raises 30-day coupon redemption from 20% to 27%, but the confidence interval is wide after chain clustering. Report the absolute difference, an arc-elasticity summary, cost per additional enrolment and why the chain count—not participant count—drives independent replication.
Pressure interviews find 2 reports among 180 recruits in the lower arm and 9 among 190 in the higher arm. Do not reduce this to a p-value. Review severity, mechanism, clustering by recruiter, missing interviews and the pre-specified safety rule. Explain how an ethical stop can be warranted despite uncertain average recruitment benefit.
Change the context: electronic referrals double successful delivery in both arms, while the money contrast shrinks. The revised conclusion is that referral logistics, not amount, removed the dominant barrier. The earlier elasticity does not transfer unchanged.
Limits and misconceptions
More recruits is not automatically better sampling
The new recruits may come from a narrower set of recruiters or subgroups, changing selection and estimator behaviour.
An incentive is not identical to coercion
Threats, undue influence, exploitation and peer pressure need distinct concepts and evidence.
A payment effect is local
It depends on tested amounts, purchasing power, burden, network structure and delivery.
RDS estimation does not repair a poorly documented intervention
The experiment must preserve assignment, recruitment links, degree information, time and safeguards.
Questions readers ask
Should primary and secondary incentives be equal?
- There is no universal rule. They compensate different activities and should be justified separately through burden, local context, ethics review and evidence.
Can a larger payment be tested ethically?
- Potentially, with acceptable arms, informed review, community input, private monitoring of pressure and pre-specified stopping rules.
What is recruitment elasticity?
- A context-specific summary of how a recruitment outcome changes proportionally when amount changes proportionally over a stated range.
Why randomise by chain instead of person?
- Chain assignment can reduce contamination between recruiters and recruits, but it requires enough independent chains and cluster-aware analysis.
What if the higher arm recruits faster but changes prevalence estimates?
- Treat that movement as evidence that the recruitment process changed. Examine tree structure, composition, degree handling and estimator sensitivity before adoption.
Should zero recruiters be excluded?
- No. They are a genuine outcome of the offered policy and excluding them inflates apparent recruitment effectiveness.
Practise, verify and continue
Design a two-amount secondary-incentive experiment with 20 seed chains. Specify the assignment unit, primary outcome, exposure window, pressure measure, subgroup modifiers, cluster-aware uncertainty, cost calculation and decision rule. Draw the recruitment funnel and mark every denominator.
Replace the amount contrast with a guaranteed small payment versus a lottery of equal expected value. Explain why certainty, timing and risk preference change the mechanism, so the earlier amount-elasticity interpretation is not enough.
Practical evidence check
Before implementation, write a one-page evidence map with the decision, unit of analysis, version or time boundary, inputs, outcomes, uncertainty, authority and stop rule. Attach one worked case that passes and one that fails. A reviewer should be able to reproduce the reasoning without relying on the memory of the person who built the system.
Run one changed-context rehearsal before approval. Alter a mechanism rather than a cosmetic detail, then identify which calculation, safeguard or governance gate must change. Record any conclusion that does not transfer. This short rehearsal is often the fastest way to expose a hidden assumption before it becomes an operational dependency.
Related eduKate reading
RDS recruitment effectiveness, coupon nonresponse and branching
RDS adaptive fieldwork governance and estimator validity
RDS recruiter choice and differential recruitment
RDS incentive fraud and payment reconciliation
RDS coupon expiry, time-to-recruit and stopping rules
Authoritative sources and scope
Ethical and regulatory considerations in HIV prevention studies using RDS
Qualitative study of peer recruitment pressures in RDS
RDS recruitment-challenge case study and incentive adaptation
Experimental evidence on survey incentives and undue influence
The cited studies inform mechanisms, ethics and field practice; they do not establish a universal optimal RDS payment. All amounts and results in the worked examples are fictional and must not be used as local compensation recommendations.