The Cry-Wolf Phenomenon and Multiple-Alarm Confidence

emerging Updated: 2026-06-13
alarm-management cry-wolf false-alarm-effect alert-fatigue supervisory-control naturalistic-decision-making trust-in-automation display-design remote-operations

The Cry-Wolf Phenomenon and Multiple-Alarm Confidence

Status: emerging
Last updated: 2026-06-13
Sources: 154193129604001612.Pdf
Tags: [alarm-management, cry-wolf, false-alarm-effect, alert-fatigue, supervisory-control, naturalistic-decision-making, trust-in-automation, display-design, remote-operations]

Summary

As operators move from controlling systems to supervising them, much of what they do is judge alarms: deciding whether a warning is real and how to respond. McDonald, Gilson and Mouloua (1996) tested how an operator's confidence in one alarm of known reliability is swayed by the other alarms around it. Thirty participants rated their confidence that a designated "test" alarm — stated to be true only 50% of the time — was genuine, while the number and the spatial closeness of nearby active alarms varied. Both factors raised confidence in the test alarm even though its stated validity never changed, and an alarm shown alone drew confidence far below its true 50% rate. The study shows that how alarms are arranged on a display, not just whether they are valid, shapes the judgements an operator makes — a direct concern for alarm management in any supervisory or remote control centre.

Body

Context

McDonald, Gilson and Mouloua (1996) report a paper-based, two-factor within-subjects experiment (30 undergraduates) framed around the shift of the human operator from system controller to system supervisor and diagnostician (PDF p. 1, orig. p. 850). The lens is naturalistic decision making (NDM): rather than applying the stated probabilities, operators under uncertainty are expected to fall back on relational heuristics — treating alarms that are close in time, space, or concept as parts of one underlying event, after Gestalt grouping (Wertheimer, 1923). The work sits in this knowledge base as its alarm-management anchor, the display-design counterpart to the trust and reliance themes in Use Misuse Disuse Abuse Of Automation (where false alarms drive automation disuse) and to the broader supervisory-monitoring problem in Active Control Vs Passive Monitoring Atc.

Key Points

The design isolated two presentational factors. A column of seven alarm lights was shown on paper, with a "TEST" alarm at the top declared true on 50% of occasions; the number of other active alarms varied from one to three, and their distance from the test alarm from immediately adjacent to three inactive lights away (PDF pp. 2–3, orig. pp. 851–852). Participants rated percent confidence (0–100%) that the test alarm was true for each arrangement, recording each answer on a turned-over card so prior judgements could not be consulted, and afterward recalled the 50% validity figure to rule out misinterpretation (PDF p. 3, orig. p. 852).

Both number and proximity raised confidence in an alarm whose validity was fixed. A within-subjects ANOVA found a significant main effect of the number of additional alarms on confidence in the test alarm (F(2,58)=39.87, p<.0001), with a positive linear trend — more collateral alarms, higher confidence (F(1,29)=45.14, p<.0001) (PDF p. 3, orig. p. 852). Spatial proximity produced a smaller but significant main effect (F(3,87)=6.85, p<.0003): confidence rose linearly as the collateral alarms were placed closer to the test alarm (F(1,29)=9.38, p<.0047) (PDF pp. 3–4, orig. pp. 852–853). The two factors did not interact (F(6,174)=1.36, p=.2234), and mean confidence across all multi-alarm trials sat near the true rate at 49.91% — the effects moved confidence around 50%, not away from it overall (PDF p. 4, orig. p. 853). The confidence values rose monotonically on both axes, from 31.73% (one distant alarm) to 68.80% (three adjacent alarms) (PDF p. 3, orig. p. 852).

The cry-wolf effect appeared when an alarm stood alone. For the single-alarm condition the mean confidence was 17.47% — far below the stated 50% validity and significantly so (t(29)=−7.89, p<.0001) — with 14 of 30 participants reporting 0% confidence and only 8 giving 50%, even though 27 of 30 correctly recalled the 50% figure (PDF p. 4, orig. p. 853). Knowing the probability did not protect the lone alarm from distrust. The authors read both results through the false-alarm effect (Breznitz, 1983): a warning system that produces false alarms loses credibility, and increasingly sensitive systems generate more false alarms and so more distrust (PDF p. 1, orig. p. 850).

Conclusion

McDonald, Gilson and Mouloua (1996) conclude that presentational characteristics — how many alarms are active and how close together they appear — change how an operator relates alarms to one another and judges their validity, producing decisions that conflict with known probabilistic information whenever the situation is uncertain. A lone alarm of moderate reliability is distrusted (the cry-wolf effect), while clustered alarms reinforce one another regardless of their individual validity. The practical reading for alarm management is twofold: display layout is a design variable that can bias diagnosis, and training or interfaces that work with operators' natural grouping heuristics may help most in the unanticipated situations for which no procedure exists. The findings are bounded — undergraduate participants, a static paper display, and a 50% validity that maximises ambiguity — and the authors flag higher-validity and mental-model conditions as the next tests.

References

Breznitz, S. (1983) Cry wolf: The psychology of false alarms. Hillsdale, NJ: Lawrence Erlbaum Associates. To be validated.

McDonald, D.P., Gilson, R.D. and Mouloua, M. (1996) 'Spatial Proximity of Multiple Alarms and the Cry-Wolf Phenomenon', Proceedings of the Human Factors and Ergonomics Society Annual Meeting, 40, pp. 850–854. doi: 10.1177/154193129604001612. mcdonald1996spatial

Stanton, N. (1994) Human factors in alarm design. Bristol, PA: Taylor & Francis. To be validated.

Wertheimer, M. (1923) 'Untersuchungen zur Lehre von der Gestalt: II', Psychologische Forschung, 4, pp. 301–350. To be validated.

Open Questions

  • The effects were measured at exactly 50% validity, the point of maximum ambiguity. The authors ask whether raising the stated validity (e.g. to 70%) would weaken the presentational effects in a less ambiguous situation — unresolved.
  • All collateral alarms were displayed as a controlled block, so the study isolates proximity to the test alarm but not grouping among the collateral alarms themselves. How richer spatial structures map onto assumed relatedness is open.
  • Whether the relational heuristics persist when the operator holds a genuine mental model of the underlying system — rather than the system-blind paper task used here — is raised by the authors and untested.