Paying per completed item induces fast, low-quality work

What it is

The dominant payment model on crowdsourcing platforms is the piece rate: a fixed fee per labeled item, judgment, or completed survey. This structure removes "being right" from the payment function — income grows only with count, and quality is enforced only indirectly through after-the-fact checks. For a rational contributor with fixed hours, the optimal strategy is to shrink time per item: skim the instructions, skip hard cases, answer by pattern. Piece rates do not doom quality outright, but they put the platform's quality control and the contributor's income strategy into the same arms race — whatever the checks detect, contributors will clear exactly that, and every other dimension yields to speed.

Why it happens

Under a piece rate, quality is a non-priced byproduct. The fixed per-item fee means the gains from better quality accrue to the platform while the cost (time) is borne by the contributor — incentives are split. The market is liquid: contributors flow between tasks by expected hourly wage, abandoning low-wage ones, so the price effectively buys "output per minute," not "correct judgments." Detection probability completes the loop: quality dimensions the checks miss are free to discard, and dimensions they cover (gold questions, spot audits, agreement) contributors clear precisely — holding accuracy just above the dismissal threshold is the income-maximizing optimum. Three factors modulate the chain: the wage floor (when hourly pay is comfortable, the marginal return of rushing falls), the density and consequences of detection, and the task's intrinsic appeal (interesting tasks partly escape the pure economic calculus — but that belongs to the motivation entries). Note the subject of this boundary is the piece-rate structure, not "lazy workers": change the payment function and the same person's behavior changes with it.

Studying it

• Paradigm: payment-structure experiments — the same task randomized across piece rate / hourly / quality-contingent / hybrid (piece base plus quality bonus) conditions, measuring time per item, error rate, expected hourly wage, and retention; re-analyses of platform logs use natural price changes as quasi-experiments to estimate the price elasticity of quality.
• Variables: payment structure, price level, and the density and severity of quality checks as independent variables; response speed (time-per-item distribution), output quality (gold error rate), retention, and task abandonment as dependent variables.
• Use in interface research: payment design parameters in task publishers — unit price tied to estimated completion time, interface presentations of tiered quality pay; coupling configurations between quality checks and pay (dock the rate or dock access on detection).
• Methodological caveat: lab tasks can understate the speed-quality tradeoff (high pay, short tasks); quality-contingent conditions invite self-selection (only high-skill workers remain, so quality gains are confounded with composition change); time-per-item data must separate "skilled speedup" from "corner-cutting" — the former lowers time without raising errors, and gold-based typing is needed to tell them apart.

Where it stops holding

Piece rates are not doomed to low quality: with dense checks (seeded gold throughout, full comparison) piece rate plus strict control delivers stable acceptable output, at the price of pushing control costs into the unit price. For skilled creative work the piece is ill-defined to begin with ("what is one item?"), and payment questions recede to motivation and fairness. The converse also fails: hourly or flat pay only weakens the speed incentive, buying motivational dilution and dawdling instead — not a free lunch for quality. There is a more basic boundary between pay levels and participation — too-low pay hurts more than no pay — which belongs to the motivation group; the piece-rate speed-quality tradeoff is what occupies this entry.

Applying it

• Publish the expected hourly wage, not just the unit price: convert via calibrated completion time; tasks below the market line need their time cost fixed (better instructions, easier items) before their price is even discussed.
• Tier pay by quality: gold-measured accuracy maps to price bands, turning "passing" from a binary dismissal into a continuous price signal — the contributor's optimum shifts from rushing to steadiness.
• Keep detection and consequences moving: vary gold density and thresholds periodically so contributors cannot settle on gaming a fixed line.
• Protect hard cases: items whose time exceeds a percentile cutoff get automatic bonuses or route to a human channel, so speed pressure does not concentrate its damage on the difficult tail.
• Verification: watch the time-per-item distribution over time — a leftward shift (faster overall) with flat gold error rates is skill acquisition; a shift with climbing error rates is corner-cutting, calling for denser checks or repricing.

Related

• Same group: V9.05.1 Having several people repeat the same task and comparing answers is the base quality control · V9.05.2 Redundancy scales cost linearly and must be tiered by task difficulty · V9.05.3 Seeded gold questions continuously estimate contributor accuracy · V9.05.4 Agreement does not guarantee correctness; shared misconceptions err in unison
• Nearby: V9.06 Contributor Motivation and Payment · V9.04.4 Time per task item and abandonment
• Search terms: piece rate · speed-accuracy tradeoff · payment design