For a period of 15 years beginning in the mid-1930s, the Horndal steelworks plant in central Sweden had been neglected. Except for minor repairs and replacement of broken equipment, no new investments were made to modernize the plant. Despite this apparent neglect, output per worker at the plant rose steadily at about 2 percent per year. Erik Lundberg, the Swedish economist who first observed the "Horndal effect" called it a case of "pure productivity."
One oft-cited source of productivity is learning by doing, which is the ability of workers to raise productivity through experience. In fact, economists have credited the Horndal effect to learning by doing. The longer workers do the same type of job the better they get. The result is higher production without having to put in new machines or hire more workers.
Several studies have looked into the overall dynamics of the learning process - how fast productivity gains accrue, and whether knowledge acquired from experience can be forgotten over time and if it spills over to other areas of production. But partly because of lack of data, these studies reveal little about how learning occurs at a plant, making it seem as though productivity improvements from learning by doing arise spontaneously as production increases, without any scope for managers to affect outcomes.
To find out the specific mechanisms through which learning takes place, Chicago Booth professor Chad Syverson, with Steven D. Levitt and John A. List, professors in the University of Chicago's Department of Economics and at the College, analyzed detailed production records from a major carmaker's assembly plant. Beyond showing evidence of rapid learning by doing, their study, "Toward an Understanding of Learning by Doing: Evidence from an Automobile Assembly Plant," provides insights into how workers' experiences at a plant can lead to greater productivity.
The study finds that the knowledge individual workers gained while working at the plant was quickly incorporated into the production process. Workers, together with plant managers, made adjustments to the assembly line based on what they learned, changes that benefited the next batch of workers and boosted overall productivity. "It's instructive as to how piecemeal, even mundane changes can add up to substantial improvements in the production process," says Syverson.
Looking Under the Hood
The authors measured productivity increases from learning by doing by looking at the assembly plant's defect rates over the course of a year. The plant assembled three variants of a model built on a common midsize-car platform. The shared platform means the three variants had similar body frames and powertrains but required different parts and assembly procedures.
Immediately before the authors started their study, large changes were made at the assembly plant. The platform had just undergone a major redesign that included both mechanical and aesthetic changes. The automaker also altered the assembly line's physical layout, brought in new machines and equipment, and modified the production process to emphasize that teams, rather than individual workers, would carry responsibility for a particular task in the line.
In essence, the plant and its workers were starting over; they were about to make new products in a new way. Previous studies have found that large efficiency gains from learning by doing are realized quickly as a plant ramps up production, so observing the plant at this early and crucial stage allowed Levitt, List, and Syverson to closely scrutinize how learning by doing occurs.
The authors' analysis began with the first shift, which was the first week that more than 100 cars were produced. They excluded the first few weeks when a small number of prototypes were made to iron out any major problems, to train line workers in their new tasks, and to familiarize them with the new team-based production process. A second shift began seven weeks later. Secondshift workers were trained by observing the first shift in action the week prior to the start of their own shift.
Levitt, List, and Syverson find clear evidence of learning by doing at the car assembly plant. Eight weeks after the first shift started, the average number of defects per car dropped by more than 80 percent. Quality improvements continued beyond the eighth week but were much smaller. First-shift defect rates fell by another 10 percent until the end of the observation period.
Unlike the first shift, the second shift did not initially experience a high defect rate. In fact, when the second shift started, the group's defect rate was much lower than that of the first shift, which already had been running for several weeks. "The second shift was able to come in and do a pretty impressive job from the get-go," Syverson says. Throughout the production year, the second shift continued to have fewer errors than the first shift.
Syverson says that second-shift workers were probably on average less experienced, because shift assignments are largely based on seniority at the plant and most experienced workers choose to work the first shift. Still, the study finds that second-shift workers were at least as efficient as those in the first shift. These results suggest that the knowledge gained by first-shift workers through learning by doing was somehow passed on to the next round of workers.
Syverson explains a key learning mechanism at the plant: workers were assigned to a team that would be responsible for five consecutive operations on the assembly line. Whenever there was a problem with a particular operation, workers would go to their whiteboard and write down the problem. Plant managers would walk around the assembly plant visiting teams about every two weeks to help address issues on the teams' whiteboards.
Together they would decide which problems they could fix and how to address them. "Whatever the workers were learning went up on the whiteboard, and it was the managers' and workers' jobs to come up with ways to build that knowledge into the production process in a more permanent way," says Syverson. Whether it was altering the layout of tools at the workstations or regrouping the sequence of operations, workers' suggestions quickly turned into new ways to raise productivity. By the time the second shift started working seven weeks later, improvements to the assembly line already had been made.
The study also finds that worker absences caused only a slight increase in defect rates. This suggests that, even when experienced workers were absent, the learning accumulated at the plant was not lost, since whatever was learned had already become part of the plant's operations and practices.
There are hundreds of processes involved in assembling a car, and some processes on the assembly line are more prone to defects than others. In fact, the study finds that the top 20 percent of the most troublesome stations accounted for 90 percent of all defects. Still, defect rates improved at about the same rate across all operations.
In addition, stations prone to errors in the first shift likewise tended to be prone to defects in the second shift, even though different workers were completing the same task. That these stations were equally prone to errors in both shifts suggests that there was something about the process itself that was responsible for creating defects; that is, the workers' level of experience did not matter as much. This is further evidence that any know-how workers gained through learning by doing was immediately transferred to the plant.
New Models, New Challenges
After focusing for some time on making just one type of car, the automaker introduced the second and third models to the assembly line, several weeks apart.
Unlike the transition from the first shift to the second shift, making a new model always begins with an intense period of learning, the study finds. The number of defects incurred in making a new model was initially very high, certainly higher than that of other cars in production at the time.
Even though the models shared the same platform, each car was sufficiently different, so the workers' experiences in making a particular model could not be fully transferred to producing other types of cars. New designs posed new challenges to both workers and plant managers. In fact, whenever the plant ramped up the production of a new model, defect rates on existing vehicles rose briefly, because the plant's resources had to be diverted to addressing problems with the newer cars.
Resolving new models' production problems came at a cost. But managers could decide that the value of reducing errors in making a new car, particularly at an early stage of production when productivity gains were likely to be largest, was greater than the temporary loss of quality on the older models.
By understanding the sources of learning by doing, managers can make a plant operate more efficiently. Far from playing a passive role, managers can make deliberate efforts to capture what workers learn on the job in order to increase productivity, the study shows.