aec 2013 big data pg&e
TRANSCRIPT
Preliminary Learnings: Big Data Office Ergonomics Field Study
Ron GoodmanRemedy Interactive
Arnold NeustaetterPacific Gas & Electric
Main Ergonomics Goals
Reduce injuries and optimize injury prevention
• Step 1: Identify the greatest risks
• Step 2: Mitigate those risks
Typical Methods to Identify Risk
Typical Methods to Identify Risk
In-PersonAssessment
Typical Methods to Identify Risk
In-PersonAssessment
Remote Assessment
Typical Methods to Identify Risk
In-PersonAssessment
Self Assessment
Remote Assessment
Typical Methods to Identify Risk
In-PersonAssessment
Self Assessment
Remote Assessment
Combination of Methods
Typical Methods to Identify Risk
In-PersonAssessment
Self Assessment
Remote Assessment
Combination of Methods
EqualsEmployee Risk Status
“Before the computer age, progress in science was achieved mainly by: gathering empirical data and crafting [hypotheses to explain] our observations...
This theory-observation-refine (TOR) cycle has provided many of our most profound insights into how the universe works.
It has not worked so well, however, for developing our understanding of complex systems.”
Christoph Adami, Professor of Microbiology &Molecular Genetics, Michigan State
Are there Better Ways to Predict Risk?
PG&E Study
• Step 1: Collected as much risk factor data as practical, using an epidemiological study model, with premise that we don’t know which factors influence risk or why
• Step 2: Using predictive analysis tools (a la Netflix) to consider each factor separately and in combination with others to see where factor(s) predict risk
• Step 3: Using these results to create an algorithm that accurately predicts risk of discomfort and time-to-onset of discomfort
What We Learned
• Factors with predictive value aren’t necessarily intuitive
• We can use predictive analysis to quantitatively guide the degree to which an ergonomics program should consider different factors
Understanding Optimal Risk Factors
Initial Findings – Example 1
• Disc = Discomfort• OR = Odds Ratio
Key Take-away:
An employee’s perception of how often they take breaks is a significant predictor of risk of injury
Initial Findings – Example 2
• Disc = Discomfort• OR = Odds Ratio
Key Take-aways:
• Not all questions are valuable risk predictors (surprisingly, this one wasn’t)
• Since this is a survey question, this doesn’t mean that external devices aren’t important (could be the question, or inaccurate reporting)
Initial Findings – Unexpected Predictors
What question would you imagine results in this discomfort distribution?
• Disc = Discomfort• OR = Optimal
Risk
Key Take-away:
• It’s important to consider all factors without bias as to which will be the strongest risk predictors
Future Use of Predictive Data• Shorten assessments to focus on
questions with significant predictive value
• When possible, use automatically collected data to predict risk (timeliness, easier on employee)
• Focus on interventions that are shown to reduce discomfort incidence
• Look at multiple factors (e.g. notebooks + exposure hours)
• Rely on software solutions to automatically take measures to reduce detected risks
Our Preliminary Learnings
Collect as much ergonomic data as possible before making any assumptions about what factors cause risk
What your data reveals may surprise you!
Questions?
Thank you!
Arnold Neustaetter
Ron Goodman