The Myth of the Lucky Streak and the Projected Confidence in Punishment

The movie Moneyball is an excellent example of how human perception is deceptive.  The idea that a player is “hot” or has a “hot hand” or is on a “winning streak” is more about perception than reality.  A great deal of our perception about players being “hot” or on a winning streak is evidence of availability and representative bias.  A great deal of our commitment to players and teams is emotional.  We remember the signature performance much more than the failures, we recall the star achievement because it was celebrated with more noise than the quiet failure.  We can learn a great deal about the hot hand fallacy from studies in gambling and attributed success to winning against the odds. 

Kahneman and scholars have shown clearly that there is no such thing as a lucky streak over time.  Every player has their average skill level and they perform to that level.  Sometimes they go up and other times down, but on average they maintain results around their mean score.  Don Bradman’s average was 99.94, his top score was 334 and he scores 7 ducks (no score).  The next best batsman of all time according to the Bible of cricket - Wisden’s, is Sachin Tendulkar.  Tendulkar’s average is 65.25, his top score is 248 and 14 ducks.  Bradman is not remembered for his last duck which robbed him of making the immortal average of 100 but he is remembered for his best performance of 5 centuries in the Ashes series 1936-1939.  It should be remembered too that in that series Bradman didn’t score hundreds in 5 of his innings in the series.  Bradman was dropped after his first test appearance with scores of 18 and 1.  Why all these statistics?  Kahneman explains that over time there is no lucky streak but simply “regression to the mean”.

“Regression to the mean” was first discovered by Sir Francis Galton 200 year ago.  Moneyball is based on a true story which shows that talent scouts and managers pick players more on emotion than on averages.  That even professional decision making is constrained by a range of cognitive biases.  The movie is about “sabermetrics” and the failed career of Billy Bean despite all predictions by talent scouts.

Gilovich (How we Know What Isn’t so: the Fallibility of Human Reason in Everyday Life  1991) demonstrates the fallibility of the belief in the hot hand or losing streak.  In research into patterns of scoring for the Philadelphia 76ers he compares the evidence of scoring clusters with the probability of scores around each player’s average. The evidence shows that scoring clusters soon disappear when time is extended beyond the recall of the moment. For example, the following pattern OXXXOXXXOXOOOXOOXXOO was recalled by 62% of survey participants in Gilovich’s research as a winning streak.  This may be so if one only looks at the first eight shots, if we look at the last eight shots it is the opposite.

 

Why does all this matter?  Why bother about “regression to the mean” in a discussion about the psychology of risk?

When it comes to perceptions about the effectiveness of punishment the law of regression to the mean is most important.  When safety professionals and managers look at safety statistics what are they looking for?  Do they notice the immediate changes in behaviour following an audit and inspection and then attribute success to that activity?  Do they believe in the effectiveness of punishment over reward based on their perception of changed behaviour following punitive action?  Would anything change from the average if nothing was done?  Are their indeed average scores for injuries and incidents for certain projects, complexity and staff size?  If scores vary from that average, what do people attribute this change to?

The research shows that humans attribute far too much to the power of punishment to change behaviour and remember far less about how people responded to reward.  Indeed, organisations don’t even keep comparative data like this, they only record LTIs etc and remember punishment effectiveness.  The truth is, claims to the effectiveness about punishment are attributed.  Claims to improvements in safety performance can just as easily be attributed to changes in reporting and definitions of injury and lost time!

Even more amazingly, regulators don’t even keep data on time of day and frequency of incidents as if such data is not relevant to incident causation.  This further goes to show that humans tend to want to confirm their own theories than be challenged by data which may disconfirm their theory.  How can you be challenged by data you don’t collect?  Why would the regulator want to collect such data when their reason and purpose is fixated on blame, catching people out, litigation and punishment?
The harmonisation of the workplace safety laws has substantially increased penalties for breaches of the Act.  It will be interesting to see how these new punishments are reported and what is attributed to them.