Latest / Safe As Podcast / E40: Types and mechanisms of investigator bias

Transcript

E40: Types and mechanisms of investigator bias · Safe As Podcast

• 0:07
  Why do safety investigations so often point to human error?
• 0:11
  This episode uncovers the hidden biases in our brains that
• 0:14
  unconsciously steer experts towards blaming people, leaving
• 0:18
  systemic risks unaddressed. It's a follow up from the last
• 0:22
  episode by the same primary author.
• 0:25
  Good everyone, I'm Ben Hutchinson and This is Safe as
• 0:28
  podcast dedicated to the thrifty analysis of safety, risk and
• 0:32
  performance research. Visit safetyinsights.org for
• 0:35
  more research. This paper examined how bias and
• 0:38
  noise affect workplace safety investigations and has some
• 0:42
  proposed solutions. The existing investigation
• 0:46
  literature this paper notes, does not comprehensively cover
• 0:49
  this topic. The papers by MacLean 2027
• 0:54
  titled Cognitive Bias in Workplace Investigations,
• 0:57
  Problems, Perspectives and Proposed Solutions in Applied
• 1:01
  Economics. So bias in this paper is defined
• 1:05
  as the systematic deviation from evidence based objective
• 1:09
  judgement. I have a few little quirks with
• 1:12
  this idea of objective judgments, but let's just put
• 1:14
  that aside. All individuals, including
• 1:17
  experts, are susceptible to biassed decisions, often unaware
• 1:22
  of how or when these biases influence their judgments.
• 1:27
  So bias is a systematic deviation.
• 1:29
  It's not random error, meaning it consistently skews
• 1:32
  perceptions and judgments, sometimes counter to the
• 1:35
  evidence. While not always leading to
• 1:38
  incorrect conclusions, it directs interpretations in
• 1:42
  specific ways. This paper highlights that
• 1:45
  workplace information is frequently ambiguous and
• 1:48
  incomplete, which can encourage the use of mental shortcuts or
• 1:52
  heuristics, and use of these heuristics can sometimes, under
• 1:57
  certain conditions, influence bias.
• 2:00
  This malleability of what a piece of information means
• 2:02
  fundamentally challenges the notion that the facts are the
• 2:05
  facts. Understanding bias begins with
• 2:08
  understanding how humans process information and how expertise
• 2:12
  effects decision making. Humans rely on automatic mental
• 2:15
  shortcuts or heuristics that often operate without conscious
• 2:19
  awareness. Expertise on enhancing
• 2:22
  efficiency also influences investigative decisions.
• 2:27
  We don't passively process information.
• 2:29
  Instead, we actively interpret it based on our understanding of
• 2:33
  the world. These shortcuts help us focus on
• 2:36
  perceived important areas and ignore or reinterpret other
• 2:40
  things. As skills increase, so does
• 2:43
  reliance on these efficient, less consciously accessible
• 2:46
  cognitive processes. Personal factors such as an
• 2:50
  individual's disposition and beliefs can also play a role.
• 2:55
  For instance, a high need foreclosure can lead
• 2:58
  investigators to quicker conclusions with less in depth
• 3:01
  processing. Let's look at how the
• 3:04
  environmental, cultural, and experiential factors influence
• 3:07
  what investigators construct. So investigators background and
• 3:11
  organisational environmental factors contribute to bias.
• 3:16
  How professionals develop initial hypotheses can be shaped
• 3:19
  by their education. For example, engineers might
• 3:22
  favour linear systemic explanations for safety
• 3:25
  incidents where site managers might consider more very
• 3:28
  contributing factors. Time pressure and fatigue so
• 3:33
  both act as constraints where fatigue can lead to greater
• 3:36
  reliance on intuitive rather than analytical cognitive
• 3:39
  processing. Cultural and structural
• 3:41
  pressures within an organisation can also bias judgement.
• 3:45
  For instance, an emotionally charged event could promote more
• 3:48
  cognitive shortcuts, or sadness might encourage a more thorough
• 3:52
  understanding of evidence. Job status and allegiance can
• 3:55
  also play a role, so an investigator's job status can
• 3:58
  correlate with their explanations for events.
• 4:01
  Higher status safety specialists might offer more worker centric
• 4:05
  explanations, while those with lower status might focus on
• 4:08
  management or design factors. Personal allegiance to the
• 4:12
  company can also be a source of bias.
• 4:15
  Another factor is base rate factors, which we covered in the
• 4:18
  Last Podcast. So while experience generally
• 4:21
  supports good judgement, you can also direct decisions based on
• 4:25
  investigators experiences of the event likelihood rather than
• 4:29
  solely on the current information.
• 4:32
  One study found that investigators knowledge about a
• 4:34
  worker's or the equipment's unsafe history significantly
• 4:39
  biases their attributions of the event 'cause this means an
• 4:43
  investigator's own past experience influenced their
• 4:45
  perception of how an event occurred, potentially leading to
• 4:49
  base rate fallacies. Irrelevant contextual
• 4:52
  information also plays a role. For instance, information not
• 4:56
  directly relevant to the task can still influence judgments.
• 5:00
  Information accompanied by an image, even if non prohibitive,
• 5:05
  can make the information seem more truthful and the source
• 5:07
  more credible. Even if it's not directly
• 5:10
  related. The order of the information
• 5:12
  plays a role. The information encountered
• 5:14
  earlier in an investigation tends to be weighed more heavily
• 5:18
  and also irrelevant opinions. Opinions such as the driver was
• 5:22
  careless, while future projections like this will
• 5:25
  happen again can bias appraisals even though they cannot be
• 5:29
  verified as facts. So how does evidence get
• 5:34
  interpreted? The specific information
• 5:36
  available can also be biassing. Resources such as images,
• 5:40
  documents, and tools can influence judgements, suggesting
• 5:44
  that the resources consulted may actually shape an investigator's
• 5:47
  cognition more than the actual evidence itself.
• 5:52
  Investigators may not follow a linear process.
• 5:55
  Instead, they might consult reference materials before or
• 5:59
  during evidence evaluation, allowing these materials to
• 6:02
  influence their interpretation. An example is a cause analysis
• 6:06
  chart with a checklist of substandard conditions and
• 6:09
  actions. This could influence
• 6:11
  investigators to fit evidence into predefined categories.
• 6:15
  One study showed that experts using such a chart sought less
• 6:19
  information about how the event occurred compared to a control
• 6:22
  group. And just think about it, ICAM,
• 6:25
  HVACS, almost everything has some sort of predefined
• 6:29
  categories. Perception also plays a role, so
• 6:32
  certain types of data can be more biassing than others.
• 6:35
  For instance, while a tyre print might not be biassing, an
• 6:39
  ancient disrepaired piece of industrial equipment might be,
• 6:42
  even if irrelevant to the event's cause.
• 6:45
  Investigators with prior knowledge of involved
• 6:47
  individuals may also be biassed. Research indicates workplace
• 6:51
  investigators tend to attribute a disproportionately greater
• 6:55
  amount of responsibility for an event to the workers action than
• 6:59
  the evidence actually supports. This is influenced by cognitive
• 7:02
  tendencies to perceive others actions as intentional and
• 7:06
  overemphasise dispositional qualities like they're lazy or
• 7:09
  inattentive, while underestimating contextual
• 7:12
  factors. The severity of an event also
• 7:15
  biases the data search, the interpretation and judgments.
• 7:18
  More severe events can prompt investigators to perceive worker
• 7:22
  decisions as more careless. Hindsight bias is also a factor
• 7:26
  where investigators are from an event.
• 7:29
  They believe it was more predictable and preventable than
• 7:32
  it truly was beforehand. This has been called delusional
• 7:35
  clarity in other research paper. Also covers some other sort of
• 7:39
  threats in the investigation process.
• 7:41
  Confirmation bias. This is where an initial
• 7:44
  hypothesis can create tunnel vision, causing investigators to
• 7:47
  seek evidence that supports their initial theory and
• 7:50
  minimise or ignore contradictory information.
• 7:54
  Studies show investigators may stop collecting information once
• 7:57
  they find the causes that they confidently endorse.
• 8:01
  They find what they're expecting to find.
• 8:03
  Another one's called coherence based reasoning.
• 8:06
  Instead of updating their hypothesis based on new
• 8:08
  information, investigators may unconsciously reinterpret
• 8:12
  previously evaluate information to fit a new coherent narrative.
• 8:17
  A bias cascade biassed opinion at one point in the
• 8:20
  investigation can systematically bias subsequent judgments and
• 8:25
  bias snowball effect. When a bias investigator shares
• 8:28
  information, it can lead to others forming seemingly
• 8:32
  independent corroborating opinions that in reality
• 8:35
  originate from a single bias source.
• 8:38
  The paper covers some techniques and some solutions around like
• 8:41
  debiasing, noting that the evidence for debiasing is pretty
• 8:45
  patchy. The first one's blinding,
• 8:49
  minimising irrelevant or contaminating information.
• 8:51
  For investigators. This technique is considered
• 8:54
  effective, but requires careful consideration of when
• 8:57
  information is truly necessary versus merely interesting.
• 9:01
  Linear sequential, unmasking extended.
• 9:04
  This technique controls the flow of information, presenting less
• 9:07
  biassing information 1st and more potentially biassing
• 9:11
  information later. This tries to provide more of a
• 9:14
  linear investigation process. This technique's actually been
• 9:17
  recommended elsewhere too. Documenting alternative
• 9:20
  hypotheses. For situations where
• 9:23
  investigators prior knowledge or affiliation could lead to bias,
• 9:27
  documenting alternative hypothesis can be useful
• 9:30
  independent assessment and verification.
• 9:33
  Having individuals without personal connections to those
• 9:36
  involved and as separate from the organisation as possible can
• 9:40
  also help ensure some level of impartiality, as practically
• 9:43
  it's really difficult for internal investigators to remain
• 9:47
  truly objective, if that's even really a thing that's possible.
• 9:51
  Information saliency. So providing up to
• 10:08
  And the last one, noise reduction through standardised
• 10:10
  approaches using more standardised practises with
• 10:14
  demonstrable reliability between and within investigators can be
• 10:18
  beneficial. It's noted that most workplace
• 10:20
  investigation techniques lack empirical testing, with very few
• 10:24
  exceptions. In conclusion, the paper
• 10:27
  emphasises that knowledge about cognitive bias is not
• 10:30
  necessarily intuitive, but can actually require explicit
• 10:33
  teaching. People tend to believe their
• 10:36
  perceptions is accurate, making bias often undetectable.
• 10:41
  While investigators are susceptible to bias, certain
• 10:44
  techniques can help service these biases and potentially
• 10:47
  reduce the noise in their judgments.
• 10:50
  That's it on Safe as I'm Ben Hutchinson.
• 10:53
  Please help share, rate, and review, and check out Safety
• 10:56
  insights.org for more research. Finally, feel free to support
• 11:01
  Safe As by shouting a coffee link in the show notes.