Employee Performance and The Drunkard's Walk

 

I recently picked up a new book called "The Drunkard's Walk.  How Randomness Rules Our Lives." 

I am a big fan of randomness - or more accurately - how we mistake randomness for causation or how we ignore the effects of it in our decision making process.  Some of my favorite all-time books are about randomness - The Black Swan, Fooled by Randomness, Sway, Predictably Irrational.  All of which I've posted on before.

While all those books are fascinating reads (I recommend all managers read all of them) "The Drunkard's Walk" had an anecdote at the beginning that really brought home what I think is one of the chief causes of poorly planned incentive and reward programs (and possibly compensation strategies as well.)

Fooled by Proximity

The author, Leonard Mlodinow, recounts a story about flight instructors for the Israeli air force.  At a lecture where a speaker was highlighting the benefits of rewards over punishment for modifying the behavior of the instructors' students, the flight instructors unanimously disagreed.  They said their experience was that if they praised a student for well-executed maneuvers, the next time they did worse.  But if they yelled and screamed at them when they did poorly they almost always did better the next time.

Boy - if I had nickel for every time a client told me a similar thing I'd be retired in Bimini throwing back shots with Gary Hart.

The instructor, who was using evidence from animal studies, couldn't understand the disconnect.  Why would the studies he's seen say rewards work better but the experience from the instructors indicate the opposite.  While the yelling preceded the performance improvement it did not cause it.  We like to to think that because two things are closely related they are causal.    Think of it this way.  When you get up in the morning the sun comes up.  Did our rising cause the sun to come up?  Not likely.

The answer is something more subtle.  The answer was found in something called "regression toward the mean."

Normal Punctuated by the Abnormal

Regression toward the mean simply means that in any series of random events an extraordinary event is most likely going to be followed by an event closer to the normal or ordinary type of event.  In other words, when a pilot had an extraordinarily good landing, it would most likely be followed by a landing closer to their normal performance level - a worse landing than previous.  Same with an extraordinarily bad landing - it would be followed by one closer to their normal level - a better one.

So the instructors, looking back on their experience concluded that when someone had a great landing and they praised them - their next landing was worse.  Yet when someone had a bad landing and they yelled - their next landing was better.  What else would they to conclude.  Praise doesn't work, yelling does.  (Now do you see why this stuff is so fascinating!)

Application Time!

We've all heard about "management by exception" - it is a process by which you only concentrate on the things that fall outside expected ranges.  A very smart way to manage a machine that we want to produce specific parts with specific tolerances (can anyone say Six Sigma?) 

But this doesn't translate as well to people - yet we do this a lot in business. 

Think about your own employees, typically, we focus on those that are below the standard - usually with performance plans and threats of firing.  For those above the norm we reward.  Yet we always have this itch in the back of our brains that when I threaten I get better results than when I praise.  And now you know why.

Don't look at performance as a point in time but as a series of points - look for the trends and reward based on improvement in the trend. I'm not saying ignore the high points - you want to recognize that behavior because you want more of it - but don't worry if the next event is closer to the average - that's going to happen.  But over time, you will have a better result by rewarding ongoing improvement than you will by punishing specific instances of low performance.