These lecture slides were developed by Anna Cox, Sandy Gould and Sarah Wiseman. They form part of human error teaching at UCL which also uses Errordiary exercises.
Please see www.errordiary.org for more info.
1. Human Error
Anna L Cox | University College London | anna.cox@ucl.ac.uk
2. overview
• a basic taxonomy of human error
• theory: memory for goals
• factors that increase error
• reducing error
Anna L Cox | University College London | anna.cox@ucl.ac.uk
3. Why study human error?
• practically: to avoid catastrophic accidents.
• theoretically: to predict when one errs.
• Cog Sci: it‟s the same system that produces
correct behaviour most of the time! Tells us a
lot about human cognition.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
4. Beatson Oncology Case
• LN, age 15
• chemotherapy & 19 of 20 or 21 planned
radiotherapy treatments
• Because the tumor type, location, and extent
and the patient size, age, and medical
condition vary, the treatment for each patient is
unique
• Each dose 58% HIGHER than it should have
been….and no one noticed
• Scaling factor wasn‟t entered
Anna L Cox | University College London | anna.cox@ucl.ac.uk
5. Blame
• "the error ... was procedural and was not
associated in any way with faults or
deficiencies in the Varis 7 computer system"
(Johnston, 2006; ii).
• “... at no point in the investigation was it
deemed necessary to discuss the incident with
the suppliers if this equipment since there was
no suggestion that these products contributed
to the error.” (Johnston, 2006; 2)
Anna L Cox | University College London | anna.cox@ucl.ac.uk
6. Accidents due to Human Errors
Human Error Others
100%
80%
60%
40%
20%
0%
Petrochemical Medicine Worldwide Jet US nuclear Automobiles Air traffic
plants cargo power plants control
Anna L Cox | University College London | anna.cox@ucl.ac.uk
7. Annual Death Rates in US
100000 98000
One jumbo jet
crash every day
50000 43649
14986
3959
329
0
commerical drowning deaths from motor vehicle deaths from
aviation deaths falls deaths medical
deaths errors
Anna L Cox | University College London | anna.cox@ucl.ac.uk
8. Denise Melanson
• infusion pump
containing a four day
dose of two
chemotherapy drugs,
5-fluorouracil and
cisplatin, to administer
to herself at home.
• pump dispensed the
dose in four hours and
not four days
Anna L Cox | University College London | anna.cox@ucl.ac.uk
9. Blame
• The nurse incorrectly calculated the dose
• A second nurse checked the calculation but
didn‟t notice the error
• The drug bottle first displayed the per 24hr
dose, (and then, in brackets the per 1hr dose)
Anna L Cox | University College London | anna.cox@ucl.ac.uk
10. OptiClik
Anna L Cox | University College London | anna.cox@ucl.ac.uk
12. A system fails to achieve the
intended outcome in a planned sequence
of mental or physical activities
Human Errors
Slips Mistakes
Incorrect execution of Correct execution of
a correct action sequence an incorrect action sequence
Anna L Cox | University College London | anna.cox@ucl.ac.uk
13. Human Errors
Slips Mistakes
Execution Evaluation Knowledge Rule
Slips Slips Based Based
•--------- •--------- •--------- •---------
•--------- •--------- •--------- •---------
•--------- •--------- •--------- •---------
Anna L Cox | University College London | anna.cox@ucl.ac.uk
14. Norman‟s Action Theory
Delete file
Goal
Use delete key
Intention Evaluation Form sub-goal
Select file + hit
delete” Action Nothing happened
Interpretation
Specification
Use mouse and Screen doesn‟t
Execution Perception change
keyboard to
perform action User Activities
sequence
Physical System
System Activities
Anna L Cox | University College London | anna.cox@ucl.ac.uk
15. Mistakes
Goal
Intention Evaluation
Action
Interpretation
Specification
Execution Perception
User Activities
Physical System
System Activities
Anna L Cox | University College London | anna.cox@ucl.ac.uk
16. Mistakes
Knowledge-Based Rule-Based
• Faulty conceptual knowledge • Misapplication of good rules
• Incomplete knowledge • Encoding deficiencies in
• Biases and faulty heuristics rules
• Incorrect selection of knowledge • Action deficiencies in rules
• Information overload • Dissociation between
knowledge and rules
Anna L Cox | University College London | anna.cox@ucl.ac.uk
17. Slips
Execution Slips Evaluation Slips
Goal
Intention Evaluation
Action
Interpretation
Specification Interruptions
Execution Perception
User Activities
Physical System
System Activities
Anna L Cox | University College London | anna.cox@ucl.ac.uk
18. Capture Slip: automatic activation of a well-
learned routine that overrides the current
intended activity .
“I intended to pick up
my prescription on my
way home. I drove
home directly.”
Anna L Cox | University College London | anna.cox@ucl.ac.uk
19. Double Capture Slip: The unintended
activation of a related strong action schema.
“I meant to take off
only my shoes, but
took my socks off as
well.”
Anna L Cox | University College London | anna.cox@ucl.ac.uk
20. Omission Slip: due to interruptions.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
21. Loss of Activation Slip
Anna L Cox | University College London | anna.cox@ucl.ac.uk
22. Description Slip: incomplete or ambiguous
specification of intention that is similar to a
familiar intention.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
23. Associative Activation Slip: activation of
similar but incorrect schemas.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
25. Repetition of Action Slip: repetition of an
correctly performed action.
A nurse repeated
radiation therapy to
a patient three
times in a row, due
to poor feedback.
The patient died
three months later.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
26. Cross-talk Slip (concurrent): action
components are exchanged between two or
more concurrent actions.
English writing Dutch reading
Goal Goal
Intention Evaluaton
i Intention Evaluaton
i
Action Action
Interpret tion
a Interpret tion
a
Specifcation
i Specifcation
i
Executi n
o Perce
ption Executi n
o Perce
ption
Physical
Systems Physical
Systems
Anna L Cox | University College London | anna.cox@ucl.ac.uk
27. Cross-talk Slip (sequential): action
components are exchanged between two or
more sequential actions.
“I had just finished talking on the phone when my secretary
ushered in some visitors. I got up from behind the desk and
walked to greet them with my hand outstretched saying „Smith
speaking‟.”
Talking on the phone (previous) Greeting (current)
Goal Goal
Intention Evaluaton
i Intention Evaluaton
i
Action Action
Interpret tion
a Interpret tion
a
Specifcation
i Specifcation
i
Executi n
o Perce
ption Executi n
o Perce
ption
Physical
Systems Physical
Systems
Anna L Cox | University College London | anna.cox@ucl.ac.uk
28. Applying Norman’s Action Cycle to
Error Classification
• Errors are sometimes made when
programming infusion pumps
• These errors might occur when entering
numbers
• What causes these errors and when to they
occur?
Anna L Cox | University College London | anna.cox@ucl.ac.uk
29. Number Entry Error Taxonomy
Anna L Cox | University College London | anna.cox@ucl.ac.uk
30. #errordiary
• We‟re tweeting when we make an error
• Have a go at classifying these errors using
Norman‟s action cycle
• Tweet when you make an error
Anna L Cox | University College London | anna.cox@ucl.ac.uk
31. issues
• Norman relied on self-report data of errors
- gives a sense of common everyday errors
- no reliable estimate of frequency of error
- cannot determine cause of error
• move to the lab
- difficulty is that errors are normally infrequent
- this brings about issue for how to operationalize design
and conduct statistical analysis - self-monitoring
- slip errors occur with routine, procedural tasks
- critical to train participants to criterion
Anna L Cox | University College London | anna.cox@ucl.ac.uk
32. post-completion error
• routine, procedural task
- goal-directed task
- many subtasks
• post-completion step
- the final step in a procedure
- completed after the goal is achieved
• post-completion error (PCE)
- missing the post-completion step
- US ATM, photocopying, filling petrol, etc
Anna L Cox | University College London | anna.cox@ucl.ac.uk
34. memory for goals
Threshold
Time
• goals are declarative memory representations defined by activation
• most active goal directs behavior
• activation declines over time, rehearsal can strengthen goal activation
• associative links between goals propagate activation through network
• completing ‘make copies’ lessens activation of ‘remove original’
Anna L Cox | University College London | anna.cox@ucl.ac.uk
35. Working memory capacity & workload
Anna L Cox | University College London | anna.cox@ucl.ac.uk
36. experiment
• memory for goals predicts that working
memory load should mediate likelihood of error
• seminal study: Byrne & Bovair (1997)
- working memory capacity: with larger capacity the goal
can be actively rehearsed, making error less likely
- memory load: with increased load the main goal will
decay more rapidly so high chance of error because
post-completion step less likely to reach activation
threshold
Anna L Cox | University College London | anna.cox@ucl.ac.uk
37. Byrne & Bovair’s results
Anna L Cox | University College London | anna.cox@ucl.ac.uk
38. Interruptions
Anna L Cox | University College London | anna.cox@ucl.ac.uk
39. Interruptions in the real world
• Nurses are frequently
interrupted during their duties
• 20% of the time they don‟t go
back to the task they were
working on when they were
interrupted
• Sometimes it‟s because that
task is redundant
• Sometimes it‟s because they‟ve
forgotten
• What are the implications for
patient care?
Anna L Cox | University College London | anna.cox@ucl.ac.uk
40. • Could argue that she wasn‟t paying attention
• She was blamed for not looking at the device
• But what about when we do direct our attention
to the device again
• Does an interruption have any effect on what
happens next?
Anna L Cox | University College London | anna.cox@ucl.ac.uk
41. Anna L Cox | University College London | anna.cox@ucl.ac.uk
42. effect of interruption
• Li et al. (2008):
multi-step
procedural task
making doughnuts
with purposeful
post-completion
step
• interrupted to switch
tasks to pack
doughnuts
Anna L Cox | University College London | anna.cox@ucl.ac.uk
43. effect of interruption
error rate
Interruption position
Anna L Cox | University College London | anna.cox@ucl.ac.uk
44. Reducing errors (I)
• cues
Anna L Cox | University College London | anna.cox@ucl.ac.uk
45. reducing errors
• system redesign
- post-completion errors can often be designed out
- e.g., ATM: card then cash vs. cash then card
- but not always feasible for large complex systems
• provide explicit cues
- must be visually salient, just-in-time, and meaningful
- cues that are not specific are ineffective
- habituation is a concern
Anna L Cox | University College London | anna.cox@ucl.ac.uk
46. visual cueing: Chung& Byrne (2008)
Anna L Cox | University College London | anna.cox@ucl.ac.uk
48. cue effectiveness
Anna L Cox | University College London | anna.cox@ucl.ac.uk
49. But…
• Not necessarily the perfect design solution
• The „just-in-time‟ aspect can make it hard to
design in
• And people become habituated to cues that
are in the world all the time and so they stop
working
Anna L Cox | University College London | anna.cox@ucl.ac.uk
50. Reducing errors (II)
• Slow down!
Anna L Cox | University College London | anna.cox@ucl.ac.uk
51. effect of interruption
Error Rate (%)
Interruption Position
data from Li, S.Y.W., Blandford, A., Cairns, P., & Young, R.M.
(2008). Journal of Experimental Psychology: Applied, 14, 314-
Anna L Cox | University College London | anna.cox@ucl.ac.uk
328.
52. distribution of errors
• working on routine task
• interruption occurs
in between subtasks
• on resumption most
common error is to
repeat or skip a step
• this is because of
competition between
goals at time of
retrieval
data from Trafton, J. G., Altmann, E. M., & Ratwani, R. M., (2009). A memory
for goals model of sequential action. International Conference of Cognitive
Anna L Cox | University College London | anna.cox@ucl.ac.uk
Modeling, 2009.
53. Speed-accuracy tradeoff
• some errors are caused by failure of memory
• memory is sensitive to changes in speed-
accuracy tradeoff criterion
• error rate should be reduced if people invest
time in thinking about where they were before
resuming
• we use an enforced lockout procedure
Anna L Cox | University College London | anna.cox@ucl.ac.uk
54. Anna L Cox | University College London | anna.cox@ucl.ac.uk
55. lockout reduces error rate
Error rate (%)
Condition
Anna L Cox | University College London | anna.cox@ucl.ac.uk
57. find out more
Anna L Cox | University College London | anna.cox@ucl.ac.uk
58. Lockouts in design?
• What would YOU do if you were locked out of
a system after an interruption?
Anna L Cox | University College London | anna.cox@ucl.ac.uk
59. implications for design
• errors will occur, even with skilled users
• errors reduced by good system design
- well structured tasks
- few interruptions
- low memory demand
- easy to select buttons
- salient display (particularly mode indicators)
- certain visual cues can be useful
• aid recovery from errors
- make actions reversible
- make the results of each action apparent
Anna L Cox | University College London | anna.cox@ucl.ac.uk
60. Summary
• Slip errors occur
infrequently – but are
persistent
• Increasing speed => • Cues: have to be “just-in-
increasing error rate time” and very aggressive
• Interruptions => increasing • Take your time – slow
error rate down!
• …err what‟s it like on a
hosptial ward??
Anna L Cox | University College London | anna.cox@ucl.ac.uk
61. reading
• core
- Byrne, M. D., & Bovair, S. (1997). A working memory model of a common
procedural error. Cognitive Science, 21, 31-61.
- Norman, D.A. (1981). Categorization of action slips. Psychological Review, 88,
1-15.
• supplementary
- Chung, P.H., & Byrne, M.D. (2008). Cue effectiveness in mitigating
postcompletion errors in a routine procedural task. International Journal of
Human-Computer Studies, 66, 217-232.
- Li, S.Y.W., Blandford, A., Cairns, P., & Young, R.M. (2008). The Effect of
interruptions on postcompletion and other procedural errors: An account based
on the activation-based goal memory model. Journal of Experimental
Psychology: Applied, 14, 314-328.
- Ratwani, R.M., McCurry, J.M., & Trafton, J.G. (2008). Predicting
postcompletion errors using eye movements. In Proceedings of CHI’08 (pp.
539-542). New York, NY: ACM Press.
- Reason, J. (1990). Human Error. Cambridge University Press.
Anna L Cox | University College London | anna.cox@ucl.ac.uk
62. Microwave Racing video
http://www.youtube.com/watch?v=Bzy5hVvbei8
Anna L Cox | University College London | anna.cox@ucl.ac.uk
Notes de l'éditeur
Complexity of taskWorkloadPerhaps variability in what has to be entered for each patient in terms of types of info and values
#804Colleague forgot her suitcase on the train! Non-routine having suitcase with her, but she followed normal routine on exiting #errordiary#787 Keep coffee at home in freezer. Made coffee @uclic, put coffee back in freezer. @uclic coffee doesn’t live in freezer. #errordiary
#806twitfortune: the art of brewing coffee includes POURING THE WATER into the coffee maker. Not just placing the pot on the hot pad #errordiary#786Filled in a postgraduate application form and then forgot to actually click submit! Realised two days after the deadline :-/
#650Forgot to tweet my error to #errordiary & then couldn’t remember what the error was
#842 - Realise I just moisturised my face with hair cream. My eyebrows look fab. #errordiary#844 - Reached into my bowl of cashews to grab another handful, only to realize I had shoved my hand into a hot cup of coffee. Awesome. #errordiary#700 - The other day I mistook a packet of twiglets for my iphonehttp://t.co/CYDSfXV7http://t.co/Bcv8k66I#errordiary#yummymistake
#537@LAURAbrap: just went a bit mad and started taking plates from the cupboards and putting them in the dishwasher #seniormoment” #errordiary
Real world error classification
Causes use Norman’s Action CycleBut you can see that some errors could have multiple causesError classification in the real world isn’t straight forward!
For those with high capacity, there’s no effect of load on error rate. They have the capacity to deal with the two concurrent tasks.For those with low capactiy, increased load results in large increase in error rate.
No differences between the conditions
Note the CUE is 0% in both conditions!!!
When people resume quickly there is a 50:50 chance of doing the wrong thing!Slowing down increases the probability of being correct!