Skip to the main content

Experiment 19 Incentives and Hidden Action

You Get What You Reward

Choose effort and shortcuts under four reward systems, then compare personal pay with real customer value.

5–10 minutes to explore Prototype Updated

A workshop produces many rough cubes on one side and a few smooth cubes on the other, beside a balance and reward counters.
About these models Math step by step For experts Math symbol guide

Loading the interactive experiment…

Start with “Work It Out, Step by Step” below. The optional expert section explains its symbols as you go. For more examples, use the plain-language math guide.

1. Predict, Then Choose Your Action

What Does the Reward System Make Worth Doing?

You are an employee choosing effort and whether to take a shortcut. The organization wants useful customer outcomes, but its contract may reward something else. A contract is the rule for your pay. An incentive is a reward or cost that makes a choice more or less attractive.

The organization is the principal, the person or group asking for the work. You are the agent, the person doing it. Your action is partly hidden: the organization cannot always see a shortcut. The model compares points, not real wages or a forecast of employee behavior.

2. Reveal the Incentives

Personal Rewards and Useful Work

All four possible actions under the same contract. Pay and scores are expected averages when monitoring is imperfect.
ActionUnits completedQualityRewarded measureNet payEffort costPersonal scoreOrganization scoreCombined

3. Change One Assumption

Design the Contract

Every contract starts with 2 points of base pay. The bonus is the rewarded measure multiplied by the bonus rate. A detected shortcut incurs the fine. Monitoring also costs the organization 2 points at 100% checking, 1 at 50%, and zero at 0%. A net pay below zero means the fine exceeds pay.

Imperfect measurement means the rewarded number can differ from actual customer value. At 100% weight, the customer-value contract uses actual value. At 0%, it uses units completed. At 50%, it averages the two. This control affects only the customer-value contract. Detection and fines affect every contract; checks also remove inflated reports when output is rewarded.

If a shortcut is detected half the time and its fine is 8 points, the average fine is 4: half of the checks cost 8 and half cost zero. A chance-weighted average counts each outcome according to how often it happens.

The Work Rules

  • Start with 4 completed units. High effort adds 2; a shortcut adds 4.
  • Start with 80% quality. High effort adds 15 percentage points; a shortcut removes 50. Quality means the useful fraction of completed units.
  • Each useful unit creates 3 customer-value points.
  • Personal effort cost starts at 1. High effort adds 3; a shortcut adds 1.
  • A shortcut inflates the reported output by 6 units when it goes undetected. At 50% detection, the expected inflation is 3.

These numbers are chosen for teaching. All calculations are expected averages; there is no sampled worker behavior. A “percentage point” is a step on the percentage scale: 80% plus 15 percentage points is 95%.

Work It Out, Step by Step

Use high effort without a shortcut, the speed contract, bonus rate 1, and no monitoring.

  1. You complete 4 plus 2 units, so the speed measure is 6.
  2. Your pay is 2 base points plus 6 bonus points: 8.
  3. Your effort costs 1 plus 3 points: 4. Your personal score is pay minus effort cost, so it is 4.
  4. Your quality is 95%, meaning 95 out of every 100 completed units would be useful on average. Customer value is 6 times 0.95 times 3, which is 17.1.
  5. The organization keeps customer value minus your pay: 17.1 minus 8, or 9.1. Together you receive 13.1 points.
\[6\times0.95\times3=17.1.\]

In words: six units, multiplied by the useful fraction, multiplied by three points per useful unit. A fraction of a unit describes a long-run average, not a claim that one task literally breaks into pieces.

For experts: formal model and assumptions

Hidden Actions and Expected Payoffs

A payoff is the score a participant wants to increase. A best response is a choice with the highest personal payoff under the current contract. All ties are retained. Moral hazard is the standard name for this kind of incentive problem when a consequential action is not fully observed; it is not a judgment about a person’s character.

\(U, O, V, W, C, M\)In words: personal score, organization score, customer value, net pay, effort cost, and monitoring cost.
Each capital letter is a short name for the amount listed in the same order. W already includes the expected fine. “Expected” means the chance-weighted average over possible detection outcomes.
\[U=W-C,\qquad O=V-W-M.\]
\[U+O=V-C-M.\]

In words: personal score is net pay minus effort cost. Organization score is customer value minus net pay and monitoring cost. When added together, pay cancels because it is a transfer between the two participants. Fines also move points between them; they are not destroyed. Monitoring and effort are real costs in this model.

The employee values expected points directly, is able to pay fines, and must choose one of four actions. There is no option to quit, no limit on losses, no team interaction, and no uncertainty about the task rules. These restrictions matter when interpreting the result.

For more examples of the notation, use the math reading guide.

4. Transfer the Lesson

A Help Desk Gets a New Target

A help desk starts rewarding the number of tickets closed. What could happen if a difficult ticket can be closed quickly but the customer’s problem remains? What would you need to measure before replacing that target?

Compare your reasoning

The measure can reward actions that improve the count without improving customer outcomes. Better measurement and monitoring can change incentives, but both have costs. You would want evidence about which actions help customers, how reliable the measurements are, and whether staff can influence them fairly.

Concept reference: Stanford economist Susan Athey’s contract theory course description and readings on hidden actions.

A Model Is a Place to Start

These small models make the incentives visible. Their results follow from their stated rules; they are not forecasts of how every person or organization behaves. A simulated strategy is a rule, not a personality.

Scenario links save the controls and random seed. To reproduce an interactive run, make the same choices in the same order. Changing a setting restarts the experiment.