Learning by Doing

August 2008

This paper reviews the theoretical and empirical literature on learning by doing. Many of the distinctive theoretical implications of learning by doing have been derived under the assumption that the cost-quantity relationships observed in numerous empirical studies are largely the result of passive learning, and some further require that passive learning is unbounded. The empirical literature raises doubts about both assumptions. When observed cost-quantity relationships indicate sustained productivity growth, factors other than passive learning are generally at work. When passive learning is the dominant factor, productivity growth is invariably bounded. Thus, empirically-relevant theories incorporating learning by doing are hybrid models in which passive learning coexists with other sources of growth. But in such models, many of the distinctive implications of passive learning become unimportant. Moreover, passive learning is often an inessential component of long-run growth; to the contrary, too much learning can lead to stagnation.

Stars and Misfits: A Theory of Occupational Choice

with Thomas Astebro and Jing Chen

This version: August 2008

In some datasets, the self-employed earn markedly less than wage earners, even though those at the top end of the distribution earn more than their wage-earning peers. This observation is explained by a model of entrepreneurial choice that blends Lazear’s [Journal of Labor Economics, vol. 23, pp. 649-680 (2005)] notion that entrepreneurs must be skilled in a variety of activities with the strong complementarity between skills central to Kremer’s [Quarterly Journal of Economics, vol. 108, pp. 551-575 (1993)] O-ring theory of production. We test some predictions of the model using two datasets, with mixed results.

This paper presents an explicit model of the entrepreneur’s role in organizing the work undertaken by employees. The model assumes that agents vary in their ability to carry out this task, and so in one sense the model is a special case of Lucas’ [Bell Journal of Economics, 9, pp. 508-523, (1978)] span of control framework. However, the model also relates ability to risk. As in Kihlstrom and Laffont [Journal of Political Economy, 87, pp. 719-748 (1979)], the entrepreneur bears all risk. But while Kihlstrom and Laffont assume the entrepreneur bears all risk and show that as a consequence the least risk averse become entrepreneurs, in the present model all agents are risk neutral but entrepreneurs bear all risk as an equilibrium outcome. The model is used to study the consequences of firm growth driven by entrepreneurial learning or by rising demand.

Human capital investment theory suggests that entrepreneurs should be generalists, while those who work for others should be specialists; it also predicts higher incomes for entrepreneurs with generalist skills. An alternative view predicts that those with greater taste for variety are more likely to become entrepreneurs and that entrepreneurs will see their incomes decrease with greater skill variety. Data from a survey of 830 independent inventors and 300 individuals from the general population confirm that inventor-entrepreneurs typically have a more varied labor market experience. However, the more varied their experience, the lower their household income. The results support the interpretation that both choice of entrepreneurship and investment in generalist skills are driven by a taste for variety.

A growing empirical literature on spinoff formation has begun to reveal some striking regularities about which firms are most likely to spawn spinoffs, when they are most likely to spawn them, and the relationship between the quality of the parent firm and its spinoffs. Deeper investigations into the causes of spinoffs have highlighted the importance of strategic disagreements in driving some employees to resign and found a new venture. Motivated by this literature, we construct a new theory of spinoff formation driven by strategic disagreements, and show how the theory can explain the emerging empirical regularities.

The Iron  and Steel Shipbuilding Data Set, 1825-194: Sources, Coverage, and Coding Decisions

This version: August2008

A number of plausible theories offer explanations for the propensity of many young industries to undergo a shakeout phase, during which the number of firms declines sharply in the face of continued rising output. However, none of the theories considers the role of labor market sorting. This paper presents a model in which individual abilities are complements in production, but frictions permit only gradual sorting among firms. The quality distribution of firms becomes wider over time, inducing exit of firms that have ended up with predominantly low-quality workers. The model does not ensure that a shakeout takes place, but when it does it will be characterized by rising output alongside a declining price, an increasing average wage, and a widening of the distributions across firms of employment, output, productivity and average wages.

This paper develops a model of marital dissolution based on communication difficulties. The quality of a marriage depends on the proximity of an action to a target. The target is unknown, and must be learned over time. Each individual receives private signals about the target, and can communicate them only imperfectly to his or her spouse. Because of imperfect communication, spouses may hold different beliefs about the optimal action. The action actually chosen is a compromise of the spouses’ distinct beliefs. If a couple’s beliefs diverge too widely, one or both of them may prefer to dissolve the marriage. The paper explores how poor communication contributes to marital unhappiness, as well as its implications for the dynamics of divorce risk, the welfare properties of divorce decisions, and the role of counseling. When the distribution of decision-making power in the household favors men, wives (but not husbands) can find themselves trapped for prolonged periods in a marriage that leaves them as unhappy as it is possible to be without seeking relief through divorce.

How Much Did the Liberty Shipbuilders Forget?

Management Science, 53(6):908-918 (2007).

This paper produces new estimates of the rate of organizational forgetting in the well-known case study of US wartime ship production. Estimates obtained using data constructed from primary sources at the National Archives yield rates of forgetting that are much smaller than previously reported, and may well be zero. The richness of the data make it possible to control for variations in the product mix, to explore alternative formulations for the learning curve, and to investigate the relationship between organizational forgetting and labor turnover.

Submarkets and the Evolution of Market Structure 

with Steven Klepper

Rand Journal of Economics, 34(4):862-888 (2007).

Private information induces individuals to self-select as subjects into clinical research trials, and it induces researchers to select which trials they conduct. We show that selection can induce ex ante therapeutic misconception and ex post disappointment among research subjects; and it undermines it the rationale of collective equipoise as an ethical basis for clinical trials. Selection provides a reason to make non-trivial payments to subjects and it implies that researchers should not design experiments to maximize statistical power.