Learning Curve (Power Law of Practice)
Also known as: Power Law of Practice, Experience Curve, Wright's Law, Öğrenme Eğrisi
The learning curve models how performance improves predictably as cumulative experience accumulates. Formalized by Theodore Wright in 1936 using aircraft manufacturing data, it expresses the relationship between the number of practice trials (or production units) and the time or cost per unit as a power-law function. It is widely applied in educational psychology, industrial engineering, health professions training, and human factors research whenever repeated task execution is the mechanism of skill acquisition.
Read the full method
Sign in with a free account to read this section.
Method map
The neighbourhood of related methods — select a node to explore.
When to use it
Use the learning curve when you have paired observations of cumulative practice counts and a performance measure (time, cost, error rate) collected from repeated task execution. It assumes the power-law form is the true data-generating process and that observations are independent. It is most appropriate when the performance measure is continuous and positive. It does not apply when learning plateaus abruptly, when the task changes across trials, or when substantial forgetting occurs between sessions. For richer sequential modeling consider knowledge-tracing models; for non-monotone trajectories consider growth-curve or spline regression.
Strengths & limitations
- Parsimonious two-parameter model that is interpretable and easy to fit via OLS after log transformation
- Grounded in a century of empirical evidence across manufacturing, education, and clinical training
- Learning rate provides a standardized, domain-comparable index of skill acquisition speed
- Supports forecasting of future performance or resource requirements given additional practice
- Assumes a strict power-law relationship; fits poorly when learning accelerates mid-practice or plateaus early
- Log transformation compresses variance and can distort residual diagnostics if heteroscedasticity is severe
- Requires a meaningful starting point (trial 1 must be defined consistently across individuals or units)
- Does not account for forgetting, transfer effects, or changes in task difficulty across trials
Frequently asked
Is the learning curve the same as the experience curve used in business strategy?
They share the same mathematical form but differ in scope. The learning curve typically tracks direct labor time or error rates on a specific repeated task, while the experience curve (popularized by the Boston Consulting Group in the 1960s) extends the concept to total unit costs across an entire industry or product line, incorporating capital, overhead, and scale effects beyond pure practice.
How do I choose between fitting the model on raw data versus log-transformed data?
Log transformation enables OLS and is appropriate when measurement error is multiplicative and proportional to the level of Y. Fitting via nonlinear least squares on raw data is preferable when additive error is a more realistic assumption. In practice, both approaches should be tried and compared using residual plots; the log-linear form is far more common in published learning-curve research.
What learning rate values are considered typical?
Empirical learning rates across industrial and educational studies most commonly fall between 0.70 and 0.95. An 80% curve (20% improvement per doubling) was Wright's original finding for airframe assembly and remains a widely cited benchmark. Rates below 0.70 indicate very rapid learning and are unusual outside highly repetitive manual tasks; rates above 0.95 suggest slow acquisition or near-asymptotic performance.
Sources
- Wright, T. P. (1936). Factors affecting the cost of airplanes. Journal of the Aeronautical Sciences, 3(4), 122–128. DOI: 10.2514/8.155 ↗
How to cite this page
ScholarGate. (2026, June 2). Learning Curve (Power Law of Practice). ScholarGate. https://scholargate.app/en/education-analytics/learning-curve
Which method?
Set this method beside its closest kin and read them side by side — the library lays the books on the table; the choice is yours.
- Knowledge TracingEducation Analytics↔ compare
- Learning AnalyticsEducation Analytics↔ compare
- Nonlinear ProgrammingOptimization↔ compare