Blog

Using Analytics to Minimize Bias and Luck: Cross Your Data, Not Your Fingers

Using Analytics to Minimize Bias and Luck: Cross Your Data, Not Your Fingers

Humans are cognitively wired for pattern seeking. Unfortunately, these “patterns” are laden with our judgments, experiences, biases, and expectations, such that we struggle to differentiate between real patterns and random events. This bias is even more prevalent when we are fed information one instance at a time (as in the case of workplace incidents).

To ensure we are identifying true patterns and effectively using safety data for prediction requires overcoming two common tendencies: (a) focusing on single incidents in isolation and (b) analyzing only the incident data, independent of other sources of data in the organization. Many users of safety data have not considered connecting the dots beyond injury reports or studying many instances all at once. Sometimes, there are just too many organizational barriers in the way or the resources are unavailable.

For most safety professionals, incident investigations are the primary resource for diagnosing problems and eliminating exposure. However, investigations provide information limited by the occurrence of an incident, with more severe incidents providing more information. This practice leaves most safety professionals looking at the outcomes and extrapolating factors from there. It is like explaining how rotten apples fell out of a tree by only looking at the rotten apples, and ignoring the tree. This practice creates distortion and focuses more attention on immediate causes (e.g., the employee put his hands in the equipment), rather than the larger contributing causes like equipment speed, on-the-job experience, employee fatigue, or overtime. Furthermore, the population who did not get hurt never enters the equation. For example, given only records of injured employees, one may conclude that newer employees appear to have more incidents. However, if the entire population is made up of newer employees, there may be no significance to that fact.

In our experience helping clients analyze their own data for precursors and other important factors, we have identified the following best practices that can make your safety data yield more fruitful insight:

  • Expand more detailed incident investigations to include severe potentials. Base investigations on risk potential rather than on actual outcomes alone.
  • Factor in the population. Given a span of time for a cross section of population, one can study whether some groupings have higher rates of incidents than other groupings (accounting for the actual distribution of employees and not just those that happened to appear in incidents).
  • Create connections to data upstream of safety outcomes. Factors that are likely to predict incidents include: employee overtime or fatigue, management of change, pre-job inspections, maintenance completions, weather, and employee experience. These data are likely to reside in other databases maintained separately from safety logs. Survey data others in the organization collect. When something sounds interesting or possibly related, capture a sample to explore.
  • Do not require perfection. Most data are imperfect, disjointed, or limited in some way, but can still allow exploration and provide information. What you find might surprise you. If you find something, then collecting more high-quality data could be a good investment of resources.

In the end, finding patterns in the data helps us eliminate risk exposure. When we find out what the incident precursors and related factors are, we need to eliminate them or modify the exposures so they are no longer precursors or factors of risk. Then we will need to start the learning process all over—perhaps with new measures of safety, since we will have no incidents left to predict.

Keys to Successfully Employing Leading Indicators
Motivating Employees with Success Feedback

Related Posts

 

Comments

No comments made yet. Be the first to submit a comment
Already Registered? Login Here
Guest
Monday, 14 October 2019