Using Statistics in the Real World
Home page
About Us

“If you want to retain those who are present, be loyal to those who are absent.”

-- Dr. Stephen Covey, The Seven Habits of Highly Effective People

“Learning cannot be disassociated from action.”

-- Peter Senge, The Fifth Discipline

“The most important measures are both unknown and unknowable.”

-- W. Edwards Deming, Out of the Crisis


Using Statistics in the Real World by Kevin McManus

First published in Industrial Engineer magazine November 2004

My son is now in his final year of course work for his Industrial Engineering degree. While I am proud of his accomplishments and can tell that he has learned a lot during his first three years in college, I am somewhat concerned about the types of concepts he is, and is not, being exposed to. Take statistics for example. He is now taking his third statistics course, and yet I do not think he has spent much time in these classes talking about the basics of variation, six sigma levels of quality, or statistical process control. Should I be concerned? It may all depend on the type of company that he goes to work for.

I have probably worked for more different companies than I should have during my 23 plus years in business, and I have yet to work in a place where anyone, engineers included, used the types of advanced statistical concepts that I was taught in school or that my son has been learning. Don't get me wrong - I think these are valuable concepts and tools, but I really wonder if the workplace in general appreciates them and if simpler tools really might make a bigger difference. On the other hand, I may have not worked in the types of companies where advanced techniques were needed. Systems are system right? Doesn't statistical analysis help us better understand system performance and predict future system outcomes, no matter what is product is being made or what service is being provided?

Dr. Deming was a statistician, but if you look at the bulk of his more popular published works, you won't find a lot of discussion about the different types of distributions or the application of probabilistic theory. You will however find a lot of book pages being devoted to common cause and special cause variation, and the roles which human beings should play in helping to reduce process variation, and in turn, create more dependable, predictable, and effective work systems. Perhaps that is where the distinction should be made. If a system is highly mechanistic and the degree of required process precision is high, then advanced statistics is needed. If, on the other hand, system performance is much more human dependent, basic statistical theory is sufficient to help you significantly save time and reduce costs.

That has been my experience. I have used control charts and histograms in a variety of settings to help understand process behavior and to help reduce process variation. On the other hand, I have rarely used regression analysis, chi squared analysis, or z tests to analyze the work processes that I was responsible for helping to improve. In fact, in most of the cases where I tried to use the tools I was taught in school, I was chastised by my bosses for being too theoretical and complex. In those cases, I really did not take their criticism to heart however, because they did not have the same math background that I had, and in turn, just really did not understand what a difference such tools could make … or did they?

In order to perform a sound statistical analysis, you have to be able to isolate the system that you are attempting to analyze. When you are working with a system that has a high level of human interface, and in turn is significantly influenced by human attitudes and behavior, you are dealing with a different system each day. Additionally, as we attempt to become more flexible in terms of meeting different customer needs, we find that we are using the same equipment to make or provide different products or services to these customers each day. With a highly human work system, you are inducing variation into its performance simply by letting the people go home each night. Because peoples' attitudes change from day to day, the systems they affect do as well. If you are trying to investigate system performance for a particular product or service, and you only make that product or provide that service two or three days a week, you will battling a lot of variation that comes from the mood shifts and personal lives of the people that are involved.

I learned this the hard way as I attempted to take the variation out of a simple production system. As I watched the control chart and histogram results vary from day to day, I learned a lot about the need to really try to keep different systems separate from each other in my data samples. While this may sound like common sense, it is really a much more daunting task if you think about it. What is meant by the term ‘different systems?' If a system changes from day to day as the attitudes and behaviors of its human operators change, does that make the system we study on Wednesday is different than the system we analyze on Thursday, even if the product being made is the same? If this is even somewhat true, what are the implications of trying to use statistical tools to analyze things like accident rates or major equipment failures, which often occur only a few times each month (or less)?

Most people aren't statisticians. In fact, most people don't even like math. These are the same people however that we count on for consistency in order to perform a valid statistical analysis, and who scrutinize our engineering studies with a fairly sharp pencil. Statistical tools are needed to properly understand, analyze, and predict system performance, but which tools are the most applicable for regular use and best meet the needs of our internal customers? Which tools should we be teaching our future engineers how to use, and which ones should we as aging engineers be attempting to stay up to speed on the application of?

I think it all depends on the type of process you are expected to use these tools to study. If you work with very machine-based systems, then you will benefit from knowing how to use the plethora of statistical tools found in today's Excel spreadsheet software or in other statistics-focused packages. If you are like many of us however, you will probably find it challenging enough to use the most basic tools Deming and Shewart wrote about, because the high degree of human interface results in a different set of study parameters every time the sun comes up. What types of systems are you trying to improve each day? What experiences have you had in trying to use advance statistics on highly human systems? Are we teaching the right things to our future engineers and in our company's statistical process control and six sigma classes?

The current use of statistics in this political season shows us that few people care about using statistics properly and that even fewer understand what distinguishes a valid analysis from one that is taken from a biased sample size or based on other forms of flawed methodology. If our internal and external customers don't care about the fancy statistical tools, should we? Who knows, I may feel differently about all of this when I wake up tomorrow.

Would You Like to Learn More?

Click on one of the following links to learn even more about Great Systems! and the types of systems improvements I can help you make:

Great Systems! home page
More articles on performance improvement
Systems Change: The Key to Getting Better Results
Do You Need Great Systems!
Types of Systems I Can Help You Improve


“The only thing I know is that I do not know it all.” -- Socrates

Copyright © 2015, Great Systems!
Last Revised - March 1, 2015
Contact me at: