Normalization of Deviation: Quotation Error in Human Factors

Method

A study was conducted to compare published journal articles (test articles) to passages in Vaughan’s (1996) book, The Challenger Launch Decision: Risky Technology, Culture, and Deviance at NASA (the control article). Test articles were selected using the publicly available portal called SCImago ¹ Journal & Country Rank (Scimago) that includes the journals and country scientific indicators developed from the information contained in the Scopus database (Elsevier B.V.). Scimago categorizes books, journals, books series, conferences and proceedings, and trade journals into 313 specific subject categories.

Of the 313 specific subject areas, three areas were considered relevant to human factors: human factors and ergonomics (HFE); safety, risk, reliability, and quality (SRQ); and safety research (SR). Within the HFE, SRQ, and SR categories, and when limited to journals only, this provided 51, 29, and 57 journals, respectively, in other words, 137 “searchable journals.” These searchable journals were then used to find test articles that could then be compared to the control.

Searches to find test articles were undertaken within each of the 137 searchable journals using search terms launch decision, Diane Vaughan, and Challenger launch decision. The administrators of journals were contacted to ensure that search terms used would identify articles that cited Vaughan in cases where it was not clear the search function available in the searchable journal would be able to retrieve test articles. Only 14 of the 137 searchable journals contained test articles that cited Vaughan: 6 in HFE, 4 in SQR, and 5 in SR; 1 journal, Safety Science, appeared in both the SQR and SR categories. Searches within these 14 journals revealed a total of 103 test articles. Using these 103 test articles, a forward citation search was conducted. Articles that were not relevant to human factors were removed from the analysis. This produced a further 15 references that cited Vaughan, resulting in 118 articles in total. Citations of Vaughan’s theory were extracted from these articles. Where the same issue was cited multiple times, this was counted as one citation. This resulted in a total data set of 187 extracts.

The 187 extracts were categorized by accuracy (whether assertions expressed in the test article contradicted statements in the control article) and agreement (whether the test article agreed or not with the assertions in the control article).

A random sample of 10% (n = 12) of the articles were independently coded by a second coder for accuracy and agreement with the test article. The second coder was familiar with both Vaughan’s theory and the human factors literature. Intercoder reliability was 100%.

Results

Of the 187 extracts, 87.2% (n = 163) were classed as accurate and 12.8% (n = 24) classed as inaccurate. In terms of agreement, 96.3% (n = 180) agreed with Vaughan, whereas only 3.7% (n = 7) disagreed with Vaughan. Of the 12.8% (n = 24) of extracts that were inaccurate, 83.3% (n = 20) agreed with Vaughan, whereas 16.7% (n = 4) of extracts disagreed with Vaughan’s view. Of the 87.2% (n = 163) of extracts that were accurate, 98.2% (n = 160) agreed with Vaughan’s view, whereas 1.9% (n = 3) disagreed with Vaughan’s view. Table 1 presents a sample of the 187 extracts analyzed that have been thematically categorized.

OPEN IN VIEWER

Table 1:

Examples of Quotation Error Across Different Themes

Example	Theme	What Other Authors Say	What Vaughan Says
A	Presence of groupthink	“Group think has also been used to understand the processes that led to the Challenger launch decision. Engineers had warned about the risks of a cold temperature launch. A group of four managers was then called on to make the decision. Three were in favour of launching, while one remained undecided. Eventually, the group leader told him that it was ‘time to take off his engineering hat and put on his management hat’. He capitulated and, as a result, the group was able to present the launch recommendation as unanimous (Vaughan 1996).”	“Many post-tragedy accounts concluded that Janis’s theory of groupthink . . . was responsible for the launch decision. But many of the elements of groupthink were missing, and those that were present have explanations that go beyond the assembled group to cultural and structural sources.” “Jack Knapp who opposed the launch said ‘I guess I did not know how the other principals (at the teleconference) there stood, but I did not hear anybody in the engineering complex, the design engineering complex, those ten that weren’t in the upper management position, I did not hear any of them capitulate in any way concerning their recommendation not to fire.’”
B	Engineers and managers	“Decisions about NASA’s ill-fated Columbia shuttle, and the Challenger before it (Vaughan, 1996) were not made by the engineers best equipped to make these decisions, but by senior NASA officials who were protected by NASA’s bureaucratic structure from the debates about the wisdom of the proposed actions.”	“The post-tragedy imagery of solidary opposition of managers versus engineers is not borne out: most participants were both managers and engineers.”
C	Flight delays	“Importantly, a flight was rarely delayed because of FRR deliberations (Vaughan, 1996).”	“Post-tragedy arguments that NASA’s [culture] forced them to push ahead, no matter what, are contradicted by the many launch delays” (Vaughan, 1996, p. 50). Vaughan provides a table listing 25 launch delays between April 12, 1981, and January 28, 1986. “These delays were for safety reasons” (p. 50). “I feel that if Marshall personnel felt that there was unacceptable risk, they did not hesitate to take the steps necessary to correct that, even to the extent of delaying launches and changing hardware.”
D	Redacted quote = redacted meaning	“Leon Ray, Marshall engineer, stated, ‘If a guy is too conservative, he’s “shootin’ down launches”; if he’s too conservative, you will find someone else in that position’ (Vaughan, 1996, p. 230). Hence, scientific methods were used for reasons other than interpreting fact; they were used to enforce the flight schedule.”	“Marshals’ Leon Ray said: . . . ‘If a guy is too conservative, he’s “shootin’ down launches”; if he’s too conservative, you will find someone else in that position. But do they say, “To heck with the risk?” No, they just don’t do it, they just don’t do it.”
E	Hindsight bias	“While the insights into the display of data are instructive, it is important to recognize that both the Vaughan and the Tufte analyses are easier to do in retrospect. In the field of cognitive engineering, this common mistake has been labelled ‘hindsight bias’: it is easy to see what is important in hindsight. It is much more difficult to achieve this goal before the important data has been identified as critical after the accident.”	“To understand the choices open to people of another time, one must limit oneself to what they knew; see the past in its own clothes, as it were, not in ours. My explanation rests on the presentation of details necessary to elicit cultural patterns. . . . As a result, many controversial actions that defied understanding after the accident—including the launch decision—become understandable in terms of the meaning they had for participants at the time those actions were taken.”
F	Management	“Sloppy management fails to put in place adequate information gathering systems. Moreover, sloppy management allows various processes to operate that may nullify warnings: the normalisation of deviance (Vaughan, 1996), group think (Janis, 1982), cultures of denial (Hopkins, 1999b) and so on (Turner, 1994: 218). Good management would have systems designed to override these tendencies and to highlight and respond to warning signs; the failure to establish such systems is a management failure.”	“While good management and organizational design may reduce accidents in certain systems, they can never prevent them. Their potential may vary depending on the characteristics of the technology, the organisation, and the task. We might surmise that the greater the complexity of technology, organisation, and task, the greater the probability of an accident. The causal mechanism in this case (although certainly distinctive on all three scores) suggest that technical system failures may be more difficult to avoid than even the most pessimistic among us would have believed.”
G	Vaughan’s degree	“Diane Vaughan’s Master’s Degree in Sociology does not qualify her to evaluate a scientific experiment.”	“Diane Vaughan is professor of international and public affairs and sociology at Columbia University.”
H	Causes of accidents	“In this view, the causes to accidents in complex systems are ‘embedded in the banality of organizational life’ and facilitated by an environment characterised by scarce resources, competition and incremental changes (Vaughan, 1996).”	“The cause of disaster was a mistake embedded in the banality of organizational life and facilitated by an environment of scarcity and competition, elite bargaining, uncertain technology, incrementalism, patterns of information, routinization, organizational and interorganizational structures, and a complex culture.”
I	Inevitability of disaster	“Errors and other undesired outcomes are often an inevitable product of the structural interactive complexity and tight coupling of most safety–critical systems (Perrow, 1984); they emerge non-randomly as anti-effects from well-organized processes (Pidgeon and O’Leary, 2000) and might well be inevitable (Vaughan, 1996, 2005).”	“THE INEVITABILITY OF MISTAKE The foregoing examples affirm that mistakes are systematic and socially organized, built into the nature of professions, organizations, cultures, and structures. Collectively, they are chilling in their suggestion that the normalization of deviance creates the potential for mistake in organizations large and small.”
J	Framing causal questions	“Books and papers that provide in-depth accounts of major accidents, often combined with analyses that introduce new ways of thinking or provide new points of view. Two recent examples of that are Diane Vaughan’s treatment of the accident with the space shuttle Challenger in 1986 (Vughan, 1996) and Scott Snook’s account of the ‘friendly fire’ shoot down of two Black Hawk helicopters over Northern Iraq in 1994 (Snook, 2000).”	“This book challenges conventional interpretations. I address two questions: Why, from 1977 to 1985, did NASA continue to accept the risk of a design known to be flawed, accepting more and more booster joint damage? And why did NASA launch the Challenger when the engineers who worked most closely on the joint were opposed?”
K	Production of culture	“Vaughn’s immensely detailed account of the accident is probably the best book we have on a single accident. However, I disagree with her interpretation. Along with several other commentators, I see this as an example of a damaged organization that allowed unique production pressures to override safety concerns, but that is not her interpretation.”	“Production pressure had a powerful impact on the teleconference discussion. But to attribute the controversial actions of Marshall managers on the eve of the launch to production pressure alone is to assume, wrongly, that the launch decision was the outcome of that single cultural imperative.”

The data suggest that accuracy is better considered a scale rather than a discrete category as some excerpts appear to be more accurate than others. For example, in Table 1, Excerpt C, which states that a flight was rarely delayed, appears more inaccurate than Excerpt D (which is also inaccurate), given the evidence of delayed flights in Vaughan’s book. In Excerpt D, the author has not provided the whole quote, which seems to change the meaning, but is not directly contradicting Vaughan.

In terms of agreement, some of the examples were found to have a high level of agreement with Vaughan while others a low level of agreement. Like accuracy, there is some suggestion that agreement is also a scale, with several excerpts appearing to disagree more with Vaughan than others. For example, in Table 1, Excerpts C and D are far less critical of Vaughan than excerpts G and E. In Excerpts G and E, the authors are directly and indirectly, respectively, critical of Vaughan and her findings. However, in Excerpts C and D, the authors do not directly criticize Vaughan, only pointing out that they consider their assertions to be different.

The categories of high and low accuracy and agreement form a two by two matrix (see Figure 1).

OPEN IN VIEWER

Figure 1.

Matrix of accuracy and agreement.

Discussion

The data presented here show that quotation error occurred in 13% of extracts. This compares to medicine, which found the frequency of quotation error of between 7% and 26% (Buchan et al., 2005; Davids et al., 2010; Reddy et al., 2008), and marine biology, which was 24.2% (Todd et al., 2010). This suggests that quotation error is occurring in the domain of human factors and needs to be more effectively managed. To our knowledge, our study is the first demonstration of quotation error in the domain of human factors.

Previous research on quotation error has coded data into categorical schemes of: accurate or inaccurate; major or minor; totally, partially, or not accurate; or clear support, no support, ambiguous, or empty citation (Buchan et al., 2005; Davids et al., 2010; Haussmann et al., 2013; Lukić et al., 2004; Reddy et al., 2008; Todd et al., 2007, 2010). This study identified another dimension: agreement. Additionally, the data suggests both accuracy and agreement should be considered scales rather than categories—extracts were found to differ on the extent to which they were accurate and the extent to which they agreed with Vaughan’s assertions. From a theoretical viewpoint, this seems reasonable: Few citations are likely to be completely accurate or inaccurate or completely agree or disagree, especially within a social science discipline such as human factors.

The two dimensions of accuracy and agreement create a matrix on which the examples can be placed (see Figure 1). These have different implications for building a body of knowledge.

For example, low accuracy with high agreement is a discrepancy and contributes to confusion, which can cause incoherence within an academic debate. While there is agreement in arguments, assertions do not contribute to consolidation of the theory or concept because the supporting evidence (i.e., the reference to previous work) is weak.

Low accuracy and low agreement forms a misrepresentation of the work of another researcher. While academic disagreement is a healthy part of science, if one author misrepresents another, then the ground on which the academic debate is based is weak.

High accuracy/high agreement results in confirmation, which provides appropriate support for the arguments proffered. Confirmation builds a greater body of evidence to support the cited author’s arguments or assertions.

The high accuracy and low agreement section produces divergence, which leads to appropriate discussion or academic debate. While there is disagreement, the ground for the difference is more clearly represented, which allows proper debate on the merits of the two sets of arguments.

Within the literature, there are several different theories that have been proposed to account for quotation error (Vicente & Brewer, 1993). These different theories can be classified as either individual focused or system focused. Theories focused on the individual include: authors only reading abstracts or citing secondary references (Masters, 2005); indolence (Lukić et al., 2004); carelessness and misuse of language (Fenton et al., 2000); insufficient time to check citations against the originals (Buchan et al., 2005, Davids et al., 2010); “academic name dropping,” which authors use to show they are aware of relevant literature (Hopkins, 2014, p. 3); and the effect of schema-based reconstructive memory processes (Vicente & Brewer, 1993). Theories focused on systemic explanations include: pressures to publish (Harzing, 2002; Haussmann et al., 2013), increasing work pressure (Todd et al., 2010); lack of training and academic guidance (Harzing, 2002); and Vicente (2000) ² shows how a single mis-citation can be propagated over time as an evolutionary process, as per Hull’s (1988) thesis that science is an evolutionary process.

Another set of theories that could be used to account for quotation error is the literature on human error. Human error is a subject that has been extensively discussed in the human factors literature, yet quotation error has, until now, received little or no consideration. Human factors has proposed numerous theories that move beyond individual causes to consider system-level pressures that lead to error (e.g., Dekker, Siegenthaler, & Laursen, 2007; Reason, 1990, 1997, 2008; Snook, 2000; Vaughan, 1996; Weigmann & Shappell, 2003; Woods, Dekker, Cook, Johannesen, & Sarter, 2010). For example, Vaughan (1996) proposes that “mistakes are systematic and socially organized, built into the nature of professions, organizations, culture, and structure” (p. 415). We suggest that such systems-based human factors theories can and should be used to move beyond simple individual explanations of quotation error. As an example, we discuss quotation error in light of Vaughan’s (1996) normalization of deviance theory since this provides a systems-level account and formed the control article for our analysis. This is intended to show the applicability of human factors theories to quotation error.

As discussed earlier, Vaughan (1996) proposes that NOD has three aspects: production of culture, culture of production, and structural secrecy.

Conclusion

This paper has identified that quotation error is occurring in the domain of human factors, with 13% of citations of Vaughan’s theory being inaccurate. In addition to the dimension of accuracy/inaccuracy identified previously in the literature, a further dimension of agreement/disagreement could be identified in our data. These two dimensions form a matrix, with each quadrant having different implications for building a body of knowledge.

We have discussed some of the reasons why quotation error occurs and suggest that Vaughan’s theory not only provides a useful base for identifying quotation errors but also a plausible systems-level explanation for why it occurs. We propose this theory to show that human factors theories can and should be applied to explain quotation error while acknowledging that NOD is not the only plausible systems-level theory. To determine which human factors theory best explains quotation error requires a program of research that is beyond the scope of this paper.

In conclusion, while it is easy to blame the individual researcher for quotation error, we contend that such errors, like the Challenger launch decision, are embedded in the banality of organizational life (in this case, academic life): culture, production pressures, and structural secrecy. To misquote Vaughan slightly, we suggest that it is the normal everyday systems that promote scholarship that also allow mistake, mishap, and quotation error to occur.

Key Points

A study was conducted that examined quotation error using Vaughan’s 1996 book as the control article.