The Difference A Few Data Points Make

Written by Dale Lehman, Associate Director, Center for Business Analytics, Loras College

In the news this morning, another grim and tragic story, killings of police officers are up this year – a 52% increase for the first six months of 2016 compared with 2015.  That got my attention, especially with all of the news this year about shootings of, and by, police officers.  I have no insight regarding the causes or remedies for these tragedies.  But I do have something to say about the use of such data.

While a 52% increase sounds alarming (and is), it must be placed into some kind of context.  How much variation is typical?  Is this size change unusual?  This is not an exercise in big data, but something that only requires a little more data to provide meaningful context.  According the FBI statistics, here is the number of “Law Enforcement Officers Feloniously Killed” for the first six months of each year, from 2006 through 2016:

The sharp uptick in 2016 is readily apparent, but so is the fact that there was a similar increase in 2014, as well as a prolonged increase between 2008 and 2011.  This does not reduce the importance of recognizing and addressing the issue.  But the addition of a few more data points does provide a sense that 2016 is terrible, but not terribly unusual.

I bring this to your attention for two reasons.  First, the media wants attention-getting stories – providing sufficient context to understand what is happening is secondary.  As citizens, we should demand better.  Second, the need to add a few data points is more important than commonly recognized.  Many businesses track their success (revenues, sales, profits, etc.) quarter to quarter, year to year, or year to date compared with last year.  This comparison of two data points is misleading and wasteful.  If sales are down, bonuses are not paid, people feel bad, and the business may scramble to determine what’s wrong and how to correct it.  If sales are up, the bonuses are paid and people celebrate.  However, in both cases, the behavior may be based solely on “noise.”

Randomness is a fact of life, and appreciating randomness requires more than two data points.  Big data is not required – only enough data to detect how much variability is “typical” and to identify true trends in the data.  You owe it to yourself and your organization to view recent experience in the context of how much background variation is typical.

Chasing noise is wasteful and potentially detracts from asking the right questions.  In the case of the killing of police officers, it is not why 2016 is so much worse than 2015, but why 2007, 2011, 2014, and 2016 were so much worse than 2008, 2012, and 2015.  Diagnosing the answers to this question may be more productive than focusing on the change from 2015 to 2016.


Dale Lehman