Increase work and productivity by learning among highly-skilled workers

Increased work and productivity through learning among highly skilled workers: evidence from heart attack treatment

It is widely believed that learning is a major source of economic growth, human capital and comparative advantage (Arrow 1962, Lucas 1988, Romer 1990, Yang and Borland 1991). The amount of learning is also important for understanding the labor market and wage dynamics. As performance increases with experience, it supports the human-capitalist interpretation of the upward-slope experience-wage profile (Baker 1964). If learning is limited, such profiles need to be interpreted by other theories, such as contract-based theories or matching models, with significant policy implications (Lazear and Moore 1984). Learning is also of particular interest to health economists, as adequate practice is often required to acquire skills in medical technology, and since the impact of education has a significant impact on increasing productivity in the healthcare sector.

Despite its fundamental importance, documenting learning on a personal level has proven to be challenging. Although a large body of literature has documented empirical patterns consistent with education, there are a number of challenges that complicate causal interpretation of results. In many contexts, employees have non-random assignments for the job, where more experienced employees usually take on more challenging assignments. Another challenge is that more productive workers are more likely to be employed, creating an irrational relationship between experience and performance. After all, high-quality performance data is often missing, and researchers are forced to rely on measures such as unit costs, quantities, and wages (Thompson 2001). It also makes it difficult to solve the specific processes behind learning and what kind of skills improve.

In our recent work, we overcame these challenges in the context of the treatment of heart attack in Sweden (Lundborg et al. 2021). The setting allows us to make the selection of more experienced staff more commonly observed by focusing on the so-called PCI heart attack treatment performed on-call shifts (nights, holidays and weekends). During this change, only one physician is present and no systematic appointment of physicians can be made for the patients. We use rich data on performance between 2004-2013 – measured by physician speed, use of medical inputs, decision making – and on patient outcomes, which we relate to physicians’ experience. This simplifies a causal explanation of the effect of experience on performance.

Focusing on the treatment of complex heart attacks, we focus on learning-by-doing in a highly-skilled setting, where learning opportunities are much higher, because the work involves non-trivial and non-standard and involves different decisions. Will be taken under the pressure of time.

Our main result

Our results show that work continues through learning for many years. In terms of efficiency, the panel (a) in Figure 1 shows that physicians get 21% faster in treating heart attacks in their first and 1,000th case. This is a significant improvement in productivity, which is associated with a three-minute reduction in arterial obstruction detection and when performing medical procedures. In the first 600 cases the learning speed is the fastest, then slower and in 1,000 cases it stops. We find similar results for other measurements of efficiency, such as the adoption of more advanced technology that requires greater manual efficiency (Panel B).

Figure 1 Physician experience, expertise, decision making and patient health

The learning process for medical decision-making follows a similar pattern, with the invasiveness of the selected medical approach increasing, at least in the first 1,000 cases (panels c and d). We further show that more aggressive, and more time-consuming, treatment by experienced physicians reflects more appropriate treatment of patients. More experienced physicians tend to be more responsive to patient characteristics when making their decisions toward more appropriate treatment.

We found some evidence that learning effects affect patient health, in the case of mortality or other heart attacks, but only in high-risk patients and only in the first 150 cases (Figure 1, panel (e) and (see) f) and Figure 2 , Panels (a) and (b)). In the first 150 cases, high-risk patients have a 40% reduction in risk of another heart attack or death within a year. This step-by-step learning curve adds physicians to the discussion about the amount of training needed before treating their own patients. We found similar results for the risk of complications during treatment (see Figure 2, panels (c) and (d)).

Figure 2 Low level experience, patient health and patient risk

Our results highlight the difficulties associated with studying the effects of experience on performance. We show that it is important to take into account the potential non-random selection of patient physicians when estimating the learning curve. That is, when studying experience, it is very important to take into account the selection of employees for the job. In our setting, we observe a strong positive correlation between physician experience and patient risk prediction during daytime changes, suggesting that hospitals hire more experienced physicians in more complex cases. This correlation disappears when we use data from on-call shifts, which provides reliable variation in our physician recruitment that is necessary to identify the effects of learning-by-doing. We also address a number of other potential empirical concerns, for example, showing that experience is not related to the number of patients treated during night shifts, confirming that our estimates are not influenced by patients’ selected referrals during shifts, and patient outcomes of those first careers. It doesn’t matter if you stay in the job or not.

What process do workers learn through?

In addition to documenting learning, our results also provide additional insights into the processes behind the learning we experience. An interesting feature of our data is that we can study how learning varies across complex varied tasks. Treating high-risk patients is reasonably more difficult than treating low-risk patients and our results suggest that physicians learn more from treatment in really difficult cases.

Our data allows us to examine whether the skills of physicians decline over time or whether knowledge is a ‘stick’. We show that more recent field experience is more valuable than more distant experience. This suggests that fine-tuning manual skills decline over time, where there is more intellectual skill.

Our results also highlight the role of peers in the learning process: the learning rate is significantly higher for physicians who have worked with more experienced peers. It suggests increasing productivity by employing inexperienced workers in occupations where jobs are not standardized and the learning curve is long.

Does increasing productivity follow wage growth?

We can associate our experience-related performance growth with wage growth and thus shed light on the general search for wage growth with experience. Conventional human capital theory saves human capital by explaining the upward sloping experience-wage profile, which is partly acquired through learning. To distinguish such interpretations from others, such as a delayed compensation process, we relate the types of learning to observe wage profiles. If the story of human capital is correct, we would expect a strong connection between wages and productivity profiles. But if wages grow faster than productivity, it would be inconsistent with human capital theory but consistent with the theory of delayed compensation as an incentive process.

We see that productivity growth is in line with the increase in wages in the first four years of physicians’ careers. Productivity growth fades when wage growth continues. This suggests that a human capital process may explain the upward slope experience-wage profile of physicians early in their careers, while other processes better explain long-term wage growth.

The comment is final

Our research documenting the presence of a prolonged learning curve in the treatment of heart attack provides new evidence of learning-to-be done and increased productivity in a highly-skilled work. The cardiologists we study continue to learn the first 1,000 cases performed in both skill and decision-making skills. These learning effects translate into significant effects on the patient’s health, but are only performed in the first 150 cases, which are related to one year of experience.

Long Learning Curve Contrasts found in previous studies that focus on potentially non-random work assigned to employees, where performance is easier to measure and where all employees perform more or less the same task (Shaw and Lazear 2008, Haggag et al. 2017). . While these studies may rule out systematic selection of workers for the job, it comes at the cost of studying quality and less-skilled jobs, where learning curves are usually shorter and steeper, such as windshield installation and taxi driving. The opposite result probably reflects the fact that heart attack treatments constitute a more complex task, with the consideration of higher level staff. This is also one of the unique contributions of our paper – documenting learning-by-doing in a highly-skilled profession, using an empirical technique that manages the non-random selection of employees.

We conclude that our findings support the notion that learning-by-doing can be a powerful engine for increasing productivity in highly-skilled occupations. It provides some insight into the learning process in advanced tasks, where both fine-tuned manual skills and quick decision-making are required.


Arrow, KJ (1962), “The Economic Impact of Learning through Taxation”, Review of Economic Studies 29 (3): 155-173

Baker, G. (1964), Human resourcesColumbia University Press.

Haggag, K, B McManus and G Paci (2017), “Learning by Driving: Improving Productivity by New York City Taxi Drivers”, American Economic Journal: Applied Economy 9 (1): 70-95.

Lazear, E and R Moore (1984), “Stimulus, Productivity, and Labor Agreement”, Quarterly Journal of Economics 99 (2): 275-296.

Lucas, RE (1988), “On the mechanics of economic development”, Journal of Monetary Economics 22 (1): 3–42.

Lundborg P, S James, B Lagerqvist and J Vikström (2021), “Learning to work and increase productivity among highly-skilled workers: evidence from the treatment of heart attack”, IZA Discussion Paper No. 14744.

Romer, P. (1990), “Endogenous technological change”, Journal of Political Economy 98 (5): 71-102.

Shaw, K and EP Laser (2008), “Termination and Output”, Labor economy 15: 705-724.

Thompson, P. (2001), “How Much Did Liberty Shipbuilders Learn? New Evidence for an Old Case Study”, Journal of Political Economy 109 (1): 103–137.

Young, X and J. Borland (1991), “A Small Economic Process for Economic Growth”, Journal of Political Economy 99 (3): 460–82.

Leave a Reply

Your email address will not be published.