Latest / Safe As Podcast / E59: Predictors of the use of higher-order risk controls

Transcript

E59: Predictors of the use of higher-order risk controls · Safe As Podcast

• 0:07
  What factors predict the use of higher order risk controls?
• 0:10
  Let's jump in and find out. Good everyone.
• 0:14
  I'm Ben Hutchinson and this is Safe as a thrifty analysis of
• 0:17
  safety and risk research. visitsafetyinsights.org for more
• 0:21
  research. Today's study is from Ooze Ekal
• 0:26
  ET al, 2024 titled Predicting Serious Injury and Fatality
• 0:31
  Exposure using Machine Learning and Construction Projects,
• 0:35
  published in a Journal of Construction Engineering and
• 0:38
  Management. For background, Over a long
• 0:41
  time, safety professionals have largely relied on historical
• 0:45
  incident data, which is really a reactive approach.
• 0:49
  So traditionally then, safety prediction models typically use
• 0:53
  information linked to past incidents to forecast future
• 0:56
  incidents. However, relying on failure data
• 1:01
  means that these models, according to paper, cannot
• 1:04
  distinguish effectively between success and failure without well
• 1:08
  informed comparison. Additionally, recordable
• 1:12
  incidents are so statistically rare that the resulting data is
• 1:16
  often too sparse for high statistical power.
• 1:19
  It's very difficult to make sense of it, at least over short
• 1:22
  term timeframes. So instead this research focused
• 1:26
  on predicting success and exposure conditions.
• 1:29
  A core part of this study was use of HACKER, which was used to
• 1:33
  capture both its success and the exposure conditions.
• 1:37
  HACKER stands for High Energy Control Assessment and it's said
• 1:41
  to align with the modern safety view that safety is the
• 1:44
  continuous presence of defences and controls against safety risk
• 1:48
  instead of the mere absence of injuries.
• 1:51
  HACKER is an observation based measure.
• 1:54
  Exposure is defined as a condition in which workers are
• 1:56
  exposed to a high energy hazard without a corresponding direct
• 2:00
  control. Success, conversely, is when all
• 2:04
  high energy hazards observed have adequate direct controls.
• 2:09
  This allows analysts to measure safety performance frequently
• 2:12
  without waiting for an incident. They also applied a machine
• 2:15
  learning approach in this study, but I've really skipped a lot of
• 2:17
  that. If you want to find out more
• 2:20
  about Mecca, I've covered in a previous POD, and I highly
• 2:22
  recommend checking out the Construction Safety Research
• 2:25
  Alliance's research around it. So what did they find?
• 2:30
  Overall, the studies show that data sets with both success and
• 2:33
  failure information yield more reliable and meaningful
• 2:37
  predictions than data sets with failure alone.
• 2:40
  So, perhaps not surprisingly, observing real work and the
• 2:44
  controls that are actually used yields more insights than just
• 2:47
  responding to incidents. Critically, they discovered that
• 2:51
  safety cannot be accurately predicted by looking only at the
• 2:55
  frontline crew. Crew level features alone,
• 2:58
  though important, didn't yield a high performing prediction
• 3:00
  model. And the models were highly
• 3:03
  predictive only when business and project features were added.
• 3:07
  Probably no surprise to most of my assistant's friends.
• 3:10
  So what were the most influential factors?
• 3:13
  So they looked at the crew, project and business levels and
• 3:16
  which factors were most predictive of whether the crews
• 3:20
  were using direct controls. So at the crew level, the single
• 3:25
  most influential factor centred on team's stability and
• 3:27
  supervision, the authors noted. In all models, the time that the
• 3:32
  crew has worked together and supervisor experience with the
• 3:36
  crew are highly predictive and influential.
• 3:40
  This means smaller crews that have worked together longer and
• 3:44
  with a consistent supervisor tend to perform better with the
• 3:47
  use of direct controls. Other crew factors were the
• 3:51
  number of frontline workers the supervisor is supervising on the
• 3:54
  day of the assessment and the total number of high energy
• 3:57
  hazard classes observed around the crew. 2 at the project
• 4:02
  level. Project factors related to scale
• 4:05
  were also key. Specifically, the total number
• 4:08
  of workers, total number of worker hours, average crew size
• 4:13
  and size of the project were all significant in the Hacker
• 4:16
  predictions. This suggests that how a project
• 4:19
  is scaled and organised has a quantifiable impact on the
• 4:23
  exposure risk, again via the use of direct controls.
• 4:28
  As before, crew experience was a another major predictor, as was
• 4:32
  their supervisor. Interestingly, at the project
• 4:36
  level, the number of consecutive days worked without a day off
• 4:39
  was also correlated, and three at the business level.
• 4:45
  Organisational investment matters The impact of providing
• 4:49
  leadership training to the frontline supervisors, which is
• 4:51
  a business feature, was ranked second in overall importance,
• 4:55
  again after crew experience. Furthermore, while lagging
• 4:59
  indicators are often criticised, the study found that Triffa has
• 5:04
  a place in SIFT exposure Predictions specifically noted
• 5:08
  that the previous year's TRIFFA is predictive of future
• 5:11
  exposure, but it was located pretty far down the predictive
• 5:15
  ranking. So what could we make of the
• 5:17
  findings? For one, they argued that the
• 5:21
  findings had overlapped with other SIFT predictors.
• 5:24
  Yet despite this overlap, their predictive model uncovered new
• 5:28
  predictors such as equipment repair funds, leadership
• 5:32
  training for frontline supervisors, importance given to
• 5:35
  tracking observations and the total number of high energy
• 5:38
  hazards in the workspace and of production responsibilities of
• 5:41
  the supervisor. They also highlight the
• 5:45
  importance of more broader project and business level
• 5:47
  features like the pre qualification orientation
• 5:50
  programmes and how these higher level factors influence their
• 5:55
  crew level. They demonstrate the value of
• 5:58
  incorporating safety success along with exposure, focusing on
• 6:02
  the presence of direct control. Remember, this wasn't about
• 6:05
  injuries and fatalities, it was assessing whether the crews in
• 6:10
  response to a high energy hazard are using a direct control.
• 6:14
  And they show the value in shifting from tracking very
• 6:17
  statistically rare and random outcome measures like injuries
• 6:22
  to more frequent success or exposure conditions during
• 6:25
  observations of work. Now of course they note several
• 6:28
  limitations. 1 is the sample size and it was limited to 28
• 6:33
  construction companies in the US and Canada.
• 6:35
  And while further research is needed, it does provide pretty
• 6:39
  interesting and decisful research around some of the
• 6:41
  predictors for adoption of more effective, higher order direct
• 6:46
  controls. That's it on Safe.
• 6:48
  As I'm Ben Hutchinson, please help share, rate and review and
• 6:53
  checkout Safety insights.org for more research.
• 6:57
  Finally, feel free to support Safe AS by shouting a coffee
• 7:00
  link in the show notes.