Learning Curve Effects¶
Core Idea¶
Learning curve effects name the structural pattern whereby the unit cost, time, or error rate of an activity falls predictably as cumulative experience — total units produced or repetitions performed — accumulates, typically following a power law in which each doubling of cumulative volume yields a roughly constant fractional improvement, a regularity T. P. Wright (1936) first quantified in observing that airframe labor-hours declined a fixed percentage with each doubling of cumulative output. [1] The defining driver is cumulative practice rather than current rate or scale: improvement is paid for in repetitions, accrues to the entity that has done the work, and is largely irreversible once acquired. The pattern answers a recurring question that confronts any system that does the same thing many times — why does performance keep improving long after the obvious process changes have been made, and how can that improvement be forecast rather than merely hoped for? [2]
The prime separates two confounded sources of performance gain. A system can become faster, cheaper, or more accurate because it is operating at a larger current scale (more output this period spreads fixed costs), or because it has accumulated more total experience over its lifetime. Learning curve effects isolate the second mechanism and assert that it has a characteristic shape: monotone decline, decelerating in absolute terms, linear when both axes are logarithmic. That shape is the structural core; the substrate — a factory, a surgeon, an industry, a single nervous system — supplies only the units in which experience is counted.
How would you explain it like I'm…
Getting Faster with Practice
Practice Makes Cheaper
Learning Curve Effects
Structural Signature¶
Learning curve effects encode a structural pattern: cumulative-experience axis → constant-fractional-improvement-per-doubling → decelerating monotone decline toward an asymptote. The pattern reindexes performance away from calendar time or current throughput and onto a single accumulating quantity, then asserts a specific functional form (power law) relating that quantity to the cost, time, or error of the next unit. [3]
Recurring features:
- Cost or error falling as a power of cumulative experience
- Constant fractional improvement per doubling of total volume
- Performance indexed to lifetime repetitions, not current rate
- Improvement bought with repetitions and largely irreversible
- Decelerating, asymptote-approaching decline along a single curve
- Early-mover advantage durable because followers must traverse the same experience
- Forecasting the next unit's cost from accumulated volume
The structural insight is robust across very different carriers: a factory's labor-hours per airframe, a typist's keystroke latency, an industry's cost per watt of installed solar, and a surgical team's complication rate all trace the same downward-bending curve when plotted against cumulative count, even though the underlying mechanisms (process refinement, motor consolidation, supplier scaling, tacit-skill accrual) are unrelated. [4] What travels is not the mechanism but the relationship: a doubling of accumulated experience buys a fixed proportional gain, so the curve looks self-similar at every scale.
What It Is Not¶
Learning curve effects do not claim that performance improves automatically with the mere passage of time. The independent variable is accumulated doing, not calendar age; an idle plant or an out-of-practice skill does not ride down the curve and may slide back up it. Confusing the two leads forecasters to attribute improvement to "maturity" when it is actually the byproduct of volume, and to expect gains during periods of low activity that never arrive. [5]
Nor does the prime assert that improvement is unbounded. A power law decelerates: each doubling of experience requires twice the cumulative output of the last, so the absolute gains shrink and real curves bend toward a floor set by material costs, physical limits, or irreducible variability. The pattern describes how improvement decays, not a promise that cost reaches zero.
The prime also makes no claim about the rate of learning being a constant of nature. The "learning rate" (e.g., 80%) is an empirical parameter that varies enormously across activities, organizations, and even across periods within the same organization; it can be accelerated by deliberate effort or stalled by turnover and forgetting. Learning curve effects name the form of the relationship, not a universal slope.
Finally, it is not a statement about individuals as opposed to systems, or vice versa. The same structural form appears whether the experiencing entity is a single learner, a team, a firm, or an entire industry. The prime is silent on the locus of learning; it only asserts that wherever experience accumulates and is retained, performance tends to decline as a power of that accumulation. [6]
Broad Use¶
Manufacturing and operations: Wright's airframe observation became the canonical "80% learning curve," and progress functions are now standard in cost estimation, production planning, and pricing for aircraft, shipbuilding, semiconductors, and assembly work generally. Managers forecast the labor and cost of future units from cumulative production to date. [7]
Cognitive science and skill acquisition: Reaction time and error rate on a practiced skill decline as a power function of the number of practice trials — the "power law of practice" — observed across motor tasks, perceptual tasks, and complex problem solving, and central to theories of how procedural skill consolidates with repetition.
Economics and corporate strategy: The Boston Consulting Group's experience curve treats cumulative-volume-driven cost decline as a source of competitive advantage and a justification for aggressive pricing ahead of cost, on the logic that the firm that accumulates volume fastest secures a durable cost lead. [8]
Energy and technology policy: Solar-panel, wind, and battery cost declines track cumulative installed capacity (Swanson's law and its kin), so analysts forecast future cost from deployment trajectories rather than calendar time, and deployment subsidies are defended as investments that "buy down" the curve for everyone. [9]
Medicine and clinical quality: Surgical complication and mortality rates fall with a surgeon's and a hospital's cumulative case count, which motivates volume thresholds and case-concentration policies for complex procedures and frames the early phase of any new technique as a measurable, expected "learning period." [10]
Clarity¶
Naming this pattern separates improvement-from-experience from improvement-from-scale and from improvement-with-time, three things that routinely coincide and get conflated. It lets a practitioner say precisely: "We are not cheaper because we are bigger this year, and not cheaper merely because years have passed; we are cheaper because we have built more units in total." Those three claims have entirely different strategic and forecasting consequences — the scale claim predicts costs rise again if volume falls, the time claim predicts costs keep falling regardless of activity, and the experience claim predicts costs depend on cumulative output and stick once earned. [11]
The concept also clarifies what kind of evidence settles such disputes: plot cost against cumulative volume on log-log axes. If the relationship is linear there, the experience mechanism is doing the work, and the slope quantifies it. This converts a vague managerial intuition ("we are getting better at this") into a falsifiable, parameterized claim.
Manages Complexity¶
The learning curve compresses a tangled mass of incremental improvements — better tooling, smarter sequencing, accumulated tacit knowledge, supplier maturation, fewer reworks — into a single parameter, the learning rate, and indexes all of it to one cumulative variable. Instead of tracking hundreds of micro-improvements, a planner tracks one number and one curve. [12]
This conversion turns an open-ended, demoralizing question ("how much more can costs possibly fall, and when?") into a bounded extrapolation along a known functional form. It also makes the future tractable for budgeting and bidding: the cost of unit number 1,000 can be estimated from the costs of units 1 through 100 without modeling each intervening improvement, because the curve carries the aggregate effect.
Abstract Reasoning¶
Once the pattern is recognized, it licenses forecasting future cost or error directly from cumulative experience, and — more powerfully — it reframes strategy. Investments that buy cumulative volume, even at a current-period loss, can be rational because they advance the firm down a curve that competitors must also descend; the loss is reconceived as the purchase price of a durable cost position. The same logic explains why early-entrant cost gaps are sticky: a late follower must traverse the same cumulative experience to reach the same cost, and cannot skip the doublings the leader has already paid for. [13]
The pattern also supports counterfactual reasoning of a distinctive kind. One can ask "where on the curve are we?" and "how many more doublings until the asymptote?" rather than "are we improving?" It reframes a slow, expensive early phase not as failure but as the unavoidable upper region of a curve whose lower region is cheap — which changes whether one abandons a struggling new process or persists through its learning period.
Knowledge Transfer¶
The manufacturing learning curve transfers cleanly to technology cost forecasting (cost versus cumulative installed capacity) and to clinical quality (complications versus cumulative cases): in each, the analyst performs the same move, replacing calendar time with cumulative experience as the explanatory axis and fitting a power law whose slope predicts the next doubling. [14] A cost estimator who knows the airframe progress function can recognize the identical shape in a battery-cost-versus-deployment chart; an energy analyst who has fitted Swanson's law can read a surgeon's complication-versus-case-count data with the same tools and the same caveats about saturation and forgetting. The reasoning is grounded in the shared structure, not in surface analogy: in every case the question is how the next unit's cost depends on the total volume already produced, and the answer is a constant fractional gain per doubling. [15]
Examples¶
Formal/abstract¶
Wright's airframe progress function: Plotting labor-hours per airframe against cumulative units produced, Wright found a straight line on log-log axes with a slope corresponding to an "80% curve" — each doubling of cumulative output cut labor-hours per unit to roughly 80% of the prior level. The first airframe might take 100,000 hours; the second, around 80,000; the fourth, around 64,000; the eighth, around 51,000. Note the deceleration in absolute terms: the gain from unit 1 to 2 (20,000 hours) is far larger than the gain from unit 64 to 128, even though both represent one doubling. The relationship is fully captured by two numbers: the cost of the first unit and the learning rate. Mapped back: The example shows the structural core in its purest form — performance indexed to cumulative count, constant fractional improvement per doubling, decelerating monotone decline. The substrate (airframe labor) supplies only the units; the curve's shape is the prime.
The power law of practice: A subject performing a repetitive cognitive task — mentally rotating figures, reading inverted text, solving a class of puzzles — shows a response time that falls as T(n) = T(1)·n^(−b), where n is the number of trials and b is a positive exponent. Early trials yield large speedups; after thousands of trials the improvement per additional trial becomes nearly imperceptible, asymptoting toward a floor set by perceptual and motor limits. Mapped back: Here the experiencing entity is a single nervous system and the "units" are practice trials rather than manufactured items, yet the functional form is identical to Wright's. This is what justifies treating manufacturing learning and skill acquisition as instances of one prime rather than two coincidentally similar phenomena: the cumulative-experience axis and the power-law decline recur unchanged when the substrate is swapped.
Applied/industry¶
Solar photovoltaics and the energy transition: Analysts forecasting the cost of solar electricity do not extrapolate from calendar year; they plot module cost against cumulative installed capacity worldwide and find a stable learning rate (historically around 20% per doubling). Policy then follows the structure: deployment subsidies are justified not as permanent props but as the price of "buying down" the curve — each gigawatt installed at today's cost lowers the cost of every future gigawatt for every market participant. Forecasters who used cumulative-capacity learning curves repeatedly out-predicted those who used time trends, because the curve correctly attributed cost decline to accumulated deployment rather than to the passage of years. Mapped back: The structure mirrors the airframe case exactly — cumulative experience (installed capacity) on the x-axis, constant fractional cost decline per doubling, an asymptote approached but not reached. The strategic move (subsidize early volume to ride the curve) is the abstract-reasoning consequence of the prime made into industrial policy.
Surgical volume and clinical outcomes: When a complex procedure is introduced, the first cases performed by a surgeon or hospital carry elevated complication and mortality rates that decline as cumulative case count rises and tacit skill, team coordination, and protocol refinement accumulate. Health systems respond structurally: they set minimum annual volume thresholds, concentrate complex cases in high-volume centers, and explicitly budget for an expected "learning period" when evaluating a new technique, rather than condemning it on early results. The same caveats the prime warns of apply — turnover resets the curve (a new team starts higher), and disuse causes forgetting — so volume must be sustained, not merely once-achieved. Mapped back: Complications-versus-cumulative-cases is the error-rate form of the same curve, with cumulative cases as the experience axis. The policy of case concentration is the early-mover-advantage insight applied to safety: the durable performance gap between high- and low-volume centers exists because the laggards have not traversed the same cumulative experience.
Structural Tensions¶
T1: The experience axis is unambiguous in manufacturing but contested everywhere else. In Wright's airframe data, cumulative units produced is a clean, countable quantity. But in an industry with heterogeneous products, in a hospital performing related-but-distinct procedures, or in a learner practicing variations of a skill, "what counts as one unit of experience" is a modeling choice that drives the entire fit. Counting differently — by units, by hours, by dollars of cumulative spend — can flatten or steepen the apparent curve, so the same underlying reality supports rival learning rates depending on how experience is operationalized.
T2: Experience can be retained or forgotten, and the prime's predictions hold only under retention. The clean power law assumes that accumulated experience persists, but organizations forget: when production pauses, when skilled workers leave, when tacit knowledge is not codified, the system slides back up the curve and a restart begins from a worse position than where it left off. A firm that treats its cost position as a permanent asset earned by past volume can be blindsided when depreciation of experience erases gains that the static curve says should be locked in. The tension is between the irreversibility the prime usually assumes and the depreciation real systems exhibit.
T3: Riding the curve rewards specialization, which narrows the space of future curves. The logic of the experience curve counsels concentrating volume on a single product or process to accumulate experience fastest and secure a cost lead. But the deeper a firm descends one curve, the more its tooling, skills, and routines are committed to that specific activity, and the higher the cost of starting a fresh curve when technology or demand shifts. The very specialization that wins the current curve can leave a firm stranded at the top of the next one, out-competed by entrants unencumbered by sunk commitment to the old trajectory.
T4: The curve predicts the shape of improvement but not its cause, which invites false attribution. Because a power-law fit can be drawn through almost any monotone-declining cost series, observers routinely "find" a learning curve where cost is actually falling for unrelated reasons — input prices dropping, scale economies, exogenous technical advance, or accounting changes. Treating cumulative volume as the cause when it is merely correlated leads to overconfident strategy: a firm prices aggressively to buy volume it believes will cut costs, when the costs were going to fall anyway, and the loss-leading was wasted. The descriptive cleanliness of the curve is precisely what makes spurious attribution easy.
T5: Early-mover advantage and rapid imitation pull in opposite directions. The prime says the leader's cost gap is durable because followers must traverse the same cumulative experience. But experience often spills over — through hired-away workers, reverse engineering, published methods, or shared suppliers — so a follower may descend the curve faster by absorbing the leader's hard-won learning without paying for it. Whether early cumulative volume yields a defensible moat or an expensive gift to imitators depends on the appropriability of experience, which the bare curve does not specify, leaving the central strategic recommendation conditional on something outside the prime.
T6: A slow, costly early phase reads simultaneously as a learning period to persist through and as evidence of a doomed effort. Because the upper region of every learning curve is expensive and shows little per-unit improvement at first, the prime counsels persistence: this is what the start of a curve looks like, and abandoning it forfeits the cheap lower region. Yet a genuinely doomed process — one with no real curve, only sustained losses — produces identical early data. The same observations license both "stay the course, we are learning" and "cut our losses, this is not working," and the prime alone cannot distinguish a curve's expensive beginning from a flat line of failure.
Structural–Framed Character¶
Learning Curve Effects sits at the structural end of the structural–framed spectrum, close to the pole: it names the lawful regularity whereby the unit cost, time, or error rate of an activity falls predictably as cumulative experience accumulates, typically as a power law in which each doubling of cumulative volume yields a roughly constant fractional improvement.
The pattern is value-neutral, took shape in operations and manufacturing rather than any norm-setting institution, and is recognized as a regularity already present in the data — Wright quantified it in declining airframe labor-hours. What keeps it from the very edge of the pole is a faint human texture: its vocabulary of "learning" and "practice" leans on a practicing agent accumulating repetitions, so it cannot be stated quite without reference to someone doing the activity. That mild dependence aside, it reads structural.
Substrate Independence¶
Learning Curve Effects is a narrowly substrate-independent prime — composite 2 / 5 on the substrate-independence scale. The pattern — cumulative experience driving power-law improvement — is real and recurs across manufacturing efficiency, cognitive skill acquisition through the power law of practice, and the experience curve in economics and strategy. But all of those are really the same cognitive-economic-operations family: agents or systems getting better with repetition, not genuinely distinct substrate types. There is no physical, formal, computational, or true biological-evolutionary instantiation beyond loose analogy, so this is a domain-flavored empirical regularity that stays tethered to its origin family — honestly a 2.
- Composite substrate independence — 2 / 5
- Domain breadth — 2 / 5
- Structural abstraction — 3 / 5
- Transfer evidence — 2 / 5
Relationships to Other Primes¶
Parents (2) — more general patterns this builds on
-
Learning Curve Effects is a kind of Increasing Returns
Learning curve effects are a kind of increasing returns specialized to cumulative practice as the state variable: each doubling of cumulative output yields a roughly constant fractional improvement in unit cost, time, or error rate. They inherit the general pattern that the marginal benefit of additional accumulation rises rather than falls, producing self-reinforcing advantage and lock-in for the practitioner, and supply the specific case where the accumulating variable is repetitions performed and the increasing-returns mechanism is acquired skill rather than scale, demand-side adoption, or expectational feedback.
-
Learning Curve Effects presupposes Learning
Learning curve effects describe the empirical regularity that unit cost, time, or error rate falls predictably with cumulative experience, often following a power law. The pattern is parasitic on actual learning happening: without durable experience-driven self-update of the producing agent's internal capability, repetition would yield no improvement. Learning supplies the underlying durable update mechanism; the learning curve is its aggregate quantitative signature visible in production data. So learning curve effects presuppose learning as the underlying capability change that produces the observable cumulative-volume improvement.
Path to root: Learning Curve Effects → Increasing Returns
Neighborhood in Abstraction Space¶
Learning Curve Effects sits among the more crowded primes in the catalog (13th percentile for distinctiveness): several abstractions describe nearly the same structure, so a description that fits it will tend to fit its neighbors too — transporting it usually means disambiguating within this family rather than landing on it exactly.
Family — Learning & Foresight Capacity (14 primes)
Nearest neighbors
- Learning — 0.84
- Experimental Design — 0.82
- Recurrence — 0.82
- Antifragility — 0.82
- Bias — 0.82
Computed from structural-signature embeddings · 2026-05-29
Not to Be Confused With¶
Learning curve effects must be distinguished from the Dunning-Kruger Effect, their nearest catalogued neighbor. The two are easily conflated because both concern the relationship between experience and competence, but they describe orthogonal phenomena. Dunning-Kruger is a claim about metacognitive miscalibration: it concerns the gap between how skilled someone is and how skilled they believe themselves to be, with low performers systematically overestimating their own competence and the most skilled sometimes underestimating theirs. Its object is self-assessment, a subjective judgment about ability. Learning curve effects make no claim about belief, confidence, or self-perception whatsoever; their object is actual measured performance — cost, time, error rate — and its lawful decline as cumulative practice accumulates. One could ride all the way down a steep learning curve while remaining badly miscalibrated about one's own skill at every point, and one could be perfectly calibrated about a skill that is not improving at all. The prime describes how real performance changes with experience; Dunning-Kruger describes how perceived competence diverges from real competence. They can both operate on the same learner at the same time without interacting, which is exactly why they are distinct primes rather than two descriptions of one underlying fact.
Learning curve effects are also not diminishing returns or convexity more broadly. Diminishing returns describes how output responds to additional units of a fixed-period input — hold everything else constant, add one more worker or one more hour this month, and the marginal product of that input falls. Its independent variable is the current quantity of an input applied within a single period; its claim is about the curvature of a production function at a moment in time. Learning curve effects, by contrast, describe improvement driven by cumulative experience accumulated over time, with the independent variable being the lifetime total of units produced or repetitions performed, not the current-period input level. The two even point in opposite directions in a revealing way: diminishing returns says that piling more input into a fixed period yields progressively less extra output, a pessimistic curvature, whereas the learning curve says that accumulating more total experience yields progressively cheaper output, an optimistic one — though it too decelerates. A factory can simultaneously exhibit diminishing returns to this month's overtime hours and ride a falling learning curve in its lifetime cost per unit; the first concerns the intensity of current input use, the second concerns the depth of accumulated experience. Collapsing them obscures that one is a same-period input-output relationship and the other is a cross-period experience-performance relationship.
Finally, learning curve effects are distinct from economies of scale, which is the relationship most often confused with them in strategic and managerial settings precisely because both predict that "bigger" firms have lower unit costs. Economies of scale tie unit cost to the current output rate or size of operation: at a larger current scale, fixed costs spread over more units, specialized equipment becomes worthwhile, and bulk purchasing lowers input prices — but the advantage exists only while the high current rate is maintained and evaporates if output contracts. Learning curve effects tie unit cost to lifetime cumulative volume, an irreversible stock rather than a current flow: a firm that has produced a million units cumulatively retains its experience-based cost position even if it is currently producing slowly, whereas a firm currently producing at high rate but with low cumulative history enjoys scale economies without learning economies. The diagnostic test separates them cleanly. If cutting current output raises unit cost back up, the advantage was scale; if unit cost stays low even when current output falls, the advantage was learning. A solar manufacturer might enjoy both — large current plants (scale) and vast cumulative production (learning) — but the two would respond differently to a demand slump, and conflating them would mislead any forecast of how costs behave when volume changes. Economies of scale are about the size of the spigot now; learning curve effects are about how much has already flowed through it.
Solution Archetypes¶
No catalogued solution archetypes reference this prime yet.
Notes¶
The "learning rate" conventionally cited (80%, 85%, 90%) refers to the fraction of unit cost retained per doubling of cumulative volume, so a lower percentage means faster learning — a common source of confusion, since an "80% curve" learns faster than a "90% curve." Care with this convention is essential when comparing rates across studies, which sometimes report the complementary "progress rate" or the slope exponent instead.
Two distinct curve families appear in the literature and are often confused: the unit learning curve (cost of the nth unit) and the cumulative-average curve (average cost of the first n units). They imply different slopes and different forecasts from the same data, and Wright's original formulation used cumulative average while many later treatments use the unit form. A forecast can be materially wrong simply from applying one family's parameter to the other's equation.
The prime is honestly substrate-narrow. Its instances — manufacturing process improvement, the power law of practice, the experience curve, technology cost-down, surgical volume-outcome — are all variants of "an agent or system improving with retained repetition." There is no physical, biological-evolutionary, formal, or computational instantiation that is not a loose analogy, which is why its substrate-independence is scored low despite the genuine cross-domain recurrence within the cognitive-economic-operations family.
Forgetting and turnover deserve special emphasis as the most common reason real curves depart from the idealized power law. Organizational learning depreciates: studies of production interruptions find that experience has a "half-life," and a restart after a long pause begins above where the curve left off. Any application of the prime to forecasting or strategy that assumes accumulated experience is a permanent asset should be checked against the system's actual retention.
References¶
[1] Wright, T. P. (1936). Factors affecting the cost of airplanes. Journal of the Aeronautical Sciences, 3(4), 122–128. Original empirical derivation of the learning curve: unit cost in airplane manufacturing falls as a log-linear function of cumulative output, the canonical reference for learning_curve_effects as a sibling specialization of increasing returns under the experience channel. ↩
[2] Newell, A., & Rosenbloom, P. S. (1981). Mechanisms of skill acquisition and the law of practice. In J. R. Anderson (Ed.), Cognitive Skills and Their Acquisition (pp. 1–55). Lawrence Erlbaum. Establishes the power law of practice: performance time on a skill declines as a power function of accumulated practice trials. ↩
[3] Yelle, L. E. (1979). The learning curve: Historical review and comprehensive survey. Decision Sciences, 10(2), 302–328. Comprehensive survey of empirical learning-curve effects across manufacturing and service industries, documenting the cumulative-experience channel of increasing returns and its log-linear regularity across substrates beyond aerospace. ↩
[4] Argote, L., & Epple, D. (1990). Learning curves in manufacturing. Science, 247(4945), 920–924. Reviews productivity gains with cumulative output across organizations and documents the recurrence of the curve along with organizational forgetting and transfer. ↩
[5] Dutton, J. M., & Thomas, A. (1984). Treating progress functions as a managerial opportunity. Academy of Management Review, 9(2), 235–247. Distinguishes autonomous from induced learning and emphasizes that progress depends on accumulated activity, not the passage of time. ↩
[6] Lapré, M. A., & Van Wassenhove, L. N. (2001). Creating and transferring knowledge for productivity improvement in factories. Management Science, 47(10), 1311–1325. Examines how learning occurs and transfers within and across organizational units, underscoring that the experiencing entity can be team, plant, or firm. ↩
[7] Alchian, A. (1963). Reliability of progress curves in airframe production. Econometrica, 31(4), 679–693. Empirical validation of progress-curve forecasting in airframe manufacturing; foundational for using cumulative production to estimate future unit costs. ↩
[8] Boston Consulting Group. (1972). Perspectives on Experience. Boston Consulting Group. Articulates the experience curve as a strategic doctrine: cumulative-volume-driven cost decline as competitive advantage and rationale for pricing ahead of cost. ↩
[9] McDonald, A., & Schrattenholzer, L. (2001). Learning rates for energy technologies. Energy Policy, 29(4), 255–261. Compiles learning rates across energy technologies, supporting cost forecasting from cumulative installed capacity. ↩
[10] Ramsay, C. R., Grant, A. M., Wallace, S. A., Garthwaite, P. H., Monk, A. F., & Russell, I. T. (2000). Assessment of the learning curve in health technologies. International Journal of Technology Assessment in Health Care, 16(4), 1095–1108. Frames the clinical learning curve and motivates volume thresholds and expected learning periods for new procedures. ↩
[11] Hirsch, W. Z. (1952). Manufacturing progress functions. Review of Economics and Statistics, 34(2), 143–155. Early econometric estimation of progress functions; sharpens the separation of experience-driven cost decline from scale and time effects. ↩
[12] Anzanello, M. J., & Fogliatto, F. S. (2011). Learning curve models and applications: Literature review and research directions. International Journal of Industrial Ergonomics, 41(5), 573–583. Reviews learning-curve model families and parameter (learning-rate) estimation that compress many improvements into one curve. ↩
[13] Spence, A. M. (1981). The learning curve and competition. Bell Journal of Economics, 12(1), 49–70. Game-theoretic treatment of how cumulative-volume cost advantages create durable early-mover positions and justify loss-leading to buy experience. ↩
[14] Thompson, P. (2012). The relationship between unit cost and cumulative quantity and the evidence for organizational learning-by-doing. Journal of Economic Perspectives, 26(3), 203–224. Reviews evidence and methods for inferring learning-by-doing, including cautions about attributing cost decline to cumulative volume. ↩
[15] Benkard, C. L. (2000). Learning and forgetting: The dynamics of aircraft production. American Economic Review, 90(4), 1034–1054. Documents both learning-by-doing and organizational forgetting in aircraft production, grounding the cumulative-experience-to-cost relationship in structure rather than surface analogy. ↩