Kolkata experiment tied referral bonuses to performance, curbing nepotism

Firms without formal screening infrastructure lean on employee referrals to fill casual and informal jobs, on the theory that a worker knows who in their own network is actually capable. But the standard referral bonus is a flat, one-time payment for producing someone who gets hired — identical whether the new hire turns out excellent or useless. That structure asks a worker to spend their own social capital vouching for a near-stranger's competence with nothing riding on whether the vouch turns out true.

Dropping referrals isn't an option — informal labor markets like Kolkata's run on exactly the trust networks a formal, stranger-facing interview can't replicate, and firms have no other cheap way to learn who is skilled before hiring them. Beaman and Magruder built a real-money, real-network field experiment there to test what a firm could change about the referral CONTRACT itself, without giving up on referrals as a channel.

Tied part of the bonus to the referred person's performance, so recommending the most capable person pays more than cashing in a favor for a relative.

Flat fees made referring a relative the rational move, since pay didn't depend on the hire's ability.

Performance-contingent bonuses aligned the referrer's payoff with the firm's, rewarding accurate vouching.

High-ability referrers held private knowledge about capable candidates, which the new contract activated.

Low-ability referrers lacked that knowledge, so the incentive change only worked where information existed.

When you pay a fixed fee for a referral, you buy a body; tie part of the pay to the referred person's performance to buy a good recommendation.

Referrals made under performance-contingent pay outperformed those made under flat fees, but only among referrers who were themselves high-ability: those participants started acting on information about who in their network was genuinely skilled once it paid them to, and their referrals' quality rose accordingly. Low-ability referrers showed little capacity to identify strong candidates under any contract — the incentive redesign fixed a willingness problem, not a knowledge problem, and only where accurate knowledge already existed.