VALUE
IMPROVEMENT
LEADERS
TOPIC #12
543 words + 2 activities | 1 hr, 13 min (2 to read, 11 to watch video, 1 hour data analysis)
SPECIAL CAUSE VS. COMMON CAUSE
PRINCIPLE
All datasets contain noise and signal. You can’t rely on your lying eyes to tell the difference; a test is required.

TOOLS
•  Inferential statistics
•  Control charts

APPLICATION
2.  After gathering and cleaning your data, select an appropriate method to filter for signal
The Ancient Low Cost of False Positives

Value improvement work is threatened by both false positives and false negatives. Humans are notoriously bad at recognizing randomness. We want patterns. There’s a reasonable (and fun) hypothesis suggesting that humans won our place atop the food chain because of our ability to recognize patterns and the high rate of false positives. That is, seeing pattern where there is none. 

Imagine two Pleistocene hunters, Grak and Krurk, headed back to their caves with fresh kills. 

Grak is prone to anxiety about the dangers of the savannah. Every snap, crackle, and pop causes Grak alarm followed by swift defensive turns, spear raised, ready to fight. All these false positives cost Grak little more than stress. 

Krurk is more relaxed and understands that 99% of nature’s noises pose little risk. As he walks, Krurk spends time thinking about how best to prepare the fresh meat and impress the target of his affections. Perhaps a lime-based marinade would woo. Krurk was more prone to false negatives and he died without passing on his chill DNA because he assumed that rustling behind him was the wind, not a stalking cheetah. 

Eons later, the cheetahs have culled most of the Krurks from the proto-human herd, leaving the Graks to pass on their DNA; thousands of generations later we’re wired to spot everyday objects when gazing at clouds and potato chips . Value leaders needn’t worry about cheetahs but they better worry about false patterns in data that should only be attributed to randomness. 

That’s two emails in a row involving the Pleistocene. I think I see a pattern. 

Click here to see the caveman name generator I used.
Special Cause Variation vs. Common Cause Variation

Thanks to Grak and Krurk, we must question intuition. Again, data and analysis tools will save us from ourselves. 

When we measure the output of any process, there is always randomness that results from hundreds of minute interactions. These are impractical if not impossible to track. Colloquially this phenomenon is known as the butterfly effect. We call it “noise” and noise is the result of common cause variation. There’s not much you can do about common cause variation, and 99% of the time, it’s detrimental to try. At best you just waste resources. At worst, you screw up a process that’s working. 

Process outputs can drift and/or shift in measurable ways due to changes of inputs. We call this “signal,” which is the result of special cause variation. Signal warrants investigation to identify the special cause of the variation you are seeing and perhaps make a corrective change. Or maybe you purposefully implemented a process change and you are now watching for, hoping for, signal to confirm your change was successful. 

Every dataset contains noise; some datasets contain signal. Value improvement leaders have, at their disposal, tools to differentiate them. 

Watch the "Special Cause vs Common Cause Variation" video (11:07) explaining the most common methods that provide bright line separation between signal and noise. 
ACTIVITIES


2. After gathering and cleaning your data, select an appropriate method to filter for signal.
LINKS

Quickly locate all course videos, slides, and previous emails here .
LEARN  |  CONNECT  |  EXPLORE  |  ABOUT
Accelerate | University of Utah | healthsciences.utah.edu/accelerate
Questions? Email:  kim.mahoney@hsc.utah.edu