Learning from Mistakes Is Easier Said Than Done: Group and Organizational Influences on the Detection and Correction of Human Error

Edmondson · Team Learning, Safety & Error · The Journal of Applied Behavioral Science · 1996 · Open access

Open this neighbourhood on the live map →

The study that found the construct before it had a name. The words psychological safety appear nowhere in this paper; what it describes is a climate of openness or of fear about discussing drug errors, produced by how nurse managers handle mistakes, and the 1999 paper names it three years later and cites this study as its origin. It opens with a case: a bag of lidocaine hanging where heparin should be after cardiac surgery, passing six professionals across more than a day before an evening nurse noticed. Edmondson uses it to argue that drug administration is a collective undertaking with ten handoff points, that secondary failures to notice are as critical as the primary slip, and that the question is not who erred but whether some work groups are better than others at catching and correcting error before it becomes consequential.

The design puts three independent measures beside each other. Eight nursing units in two urban teaching hospitals were studied. Over six months, trained medical investigators from a larger Harvard study tracked adverse drug events by daily chart review, daily unit visits and confidential reporting, entirely independently of this research. A survey adapted from Hackman's cockpit-crew instrument went to 289 nurses, physicians and pharmacists and came back from 159, measuring nurse manager direction setting and coaching, unit relationships, perceived performance and three single items about how often errors occur, whether they are reported and whether a mistake is held against you. And a second researcher, blind to both the error data and the survey, observed each unit for several days and ranked all eight on openness.

The expected result was that better-led, better-performing units would have lower error rates. The opposite appeared. Detected error rates ran from 2.3 to 23.7 per thousand patient days and correlated positively with nurse manager coaching (.74), direction setting (.74), perceived unit performance (.76) and quality of unit relationships (.74), each significant at p below .03. Intercepted errors, those caught before reaching a patient, correlated the same way. The discriminating test is what did not correlate: nonpreventable adverse drug events, the allergic reactions no one caused, showed no relationship with any unit variable. Edmondson's reading is that detected error is a function of at least two things, errors made and willingness to report them, and that in a setting where a Salt Lake City hospital found forty times more events once it stopped relying on people to volunteer them, willingness may overwhelm the underlying differences. The blind qualitative ranking matched the error-rate ranking almost unit for unit, with one exception. In the most open unit the nurse manager worked in scrubs, called a nonpunitive environment essential, and a nurse had telephoned a physician at home to halve a dose she thought too high. In the most closed, the manager wore a business suit, held conversations behind a closed door, required visitors to wear badges, and a nurse who had hurt a patient drawing blood described being treated like a two-year-old and expected trouble for saying so.

The paper is more careful than its retellings. It does not show that open units made fewer errors; it says plainly that actual and detected errors are confounded inextricably in this data, that lower actual error in better-coordinated units is a speculation the study cannot support, and that what it does show is willingness to report varying systematically with the perceived openness of the leader. The footnote everyone skips is the strongest part: adjusting for medication exposure, since intensive care patients receive more drugs, strengthened the correlations rather than weakening them, and events that could only be captured by voluntary report correlated more strongly with unit climate than events also visible in charts. Eight units is eight data points, the survey response was 55 per cent, and the outcome is rank-order correlation at that N. What it leaves for the field is a management dilemma stated without comfort: the teams that most need to improve are the ones least likely to surface the errors that improvement would need.

Edmondson, A.C. (1996) 'Learning from mistakes is easier said than done: Group and organizational influences on the detection and correction of human error', The Journal of Applied Behavioral Science, 32(1), pp. 5–28.

Neighbours (20)

View the source paper →