Learning from complexity: effects of prior accidents and incidents on airlines' learning

Administrative Science Quarterly, Dec, 2002 by Pamela R. Haunschild, Bilian Ni Sullivan

For all the scientific pizzazz [involved in airline accident investigations], unraveling the subtle, complex chain of events leading to aviation deaths is proving more elusive than ever.

--"Why more plane-crash probes end in doubt," Wall Street Journal, March 22, 1999

Organizations like airlines try to learn from experience, understanding what went wrong so that it won't go wrong next time. But if, as the quote above suggests, the causes are often left in doubt, such learning is likely to be difficult. Learning is also likely to vary across firms, despite industry regulation that should affect all airlines equally. Investigators of the 2000 Air France Concorde crash discovered that British Airways had recommended changes to the Concorde's water deflector in 1995 but that Air France had not made those changes (Phillips, 2000). As Donoghue (1998: 36) explained, "... any safety initiative has an unequal effect on the carriers and becomes an issue to be promoted or fought ... seeking the path that best suits [the airline] individually." Other heavily regulated industries, such as nuclear power, also show substantial variance in incident rates among firms (Morris and Engelken, 1973), which indicates that firms vary in how effectively they learn from their experience. Despite much discussion and analysis of aviation errors (airline accidents and incidents), there has been little work investigating the role of organizational learning and none examining variation in learning across firms in the industry.

Learning from experience has been shown to have important effects on such varied outcomes as manufacturing plant productivity (e.g., Argote, Beckman, and Epple, 1990), service timeliness (Argote and Darr, 2000), and hotel survival (Baum and Ingram, 1998). If firms learn from experience, then the attributes of this experience are likely to affect the rate and effectiveness of learning. Some firms have heterogeneous experience in that their accidents and incidents ("errors") are caused by a large number of different factors, which are likely to interact in complex ways. Some firms have more homogeneous experience, with errors caused by a small number of similar factors. It is likely that the complexity of prior experiences, as well as characteristics of the firms themselves, affect how well airlines can learn from that experience. We investigate these issues in the context of airline accidents and incidents to explain variation in learning among firms in the airline industry.

According to the NTSB (2001) Code of Federal Regulations (49CFR830.2, p. 1195), an accident "means an occurrence ... in which any person suffers death or serious injury, or in which the aircraft received substantial damage." An incident is "an occurrence other than an accident, which affects or could affect the safety of operations." Accidents and incidents are the error experiences from which airlines have the potential to learn.

EFFECTS OF PRIOR EXPERIENCE ON LEARNING

In the literature on organizational learning there is a large body of work on the learning curve. The learning curve is an empirical finding showing that, in general, experience produces improvement. Early empirical work on the learning curve showed that the log of unit costs tends to decrease linearly with the log of cumulative production volume. So, for example, cumulative production experience tends to lower costs in shipbuilding and automotive production (Argote and Epple, 1990), nuclear power plant production (Zimmerman, 1982), and coal generation (Joskow and Rose, 1985). More recent work has moved away from a focus on cost reduction and productivity improvement to other outcomes of learning. These studies have shown that experience improves customer service and product quality (Dart, Argote, and Epple, 1995; Lapre, Mukherjee, and Van Wassenhove, 2000) and increases the survival rates of hotels (Ingram and Baum, 1997; Baum and Ingrain, 1998) and banks (Kim and Miner, 2000).

In the context of airlines and their errors, it may be that airlines learn from error experience and are able to improve performance over time, reducing subsequent errors (i.e., accidents and incidents). If we look at the airline industry over long time periods, this seems to be the case. Figure 1 plots the accident rate (accidents per 100,000 hours flown) for all U.S. airlines from 1955 to 1997 and exhibits a characteristic learning curve, i.e., as experience accumulates with the passage of time, the error rate declines.

[FIGURE 1 OMITTED]

When individual airlines' accident rates are broken out, as they are in table 1 for some of the larger U.S. airlines, we see the same general decrease in accidents over time as in figure 1, but there is also a fair amount of variance across airlines. For example, from 1957 to 1986, American Airlines had an average of 10.3 accidents per million departures and US Air had 6.6. Variation in airline error rates could come from many sources. One obvious source is the characteristics of the individual airline, e.g., whether it is large or small, the age of its fleet, characteristics of its corporate culture, its management team, and its training procedures. Another possible source of variation, however, is differences in the characteristics of the accidents and incidents experienced by these different airlines. Because experience affects organizational learning, different types of experiences are likely to produce variation in learning rates. One source of differences in experience is whether that experience has homogeneous or heterogeneous causes.