How to Recognize When Special Causes Exist,
A Guide to Statistical Process Control
© Ends of the Earth Learning Group 1998
Linda Turner and Ron Turner
TABLE OF CONTENTS
Chapter One: The Logic Of Statistical Process Control (SPC) Charts When sales fall or error rates rise, how bad do things have to get before you decide to investigate for root causes? Intuitively most of us realize that if something "special" has occurred, then we may need to "fix" things. The dilemma we all face is that all work processes produce some normal downturns in results even though the work force was doing everything the same as on "good" days. How do you keep yourself from "fixing" something that isn't really broken, but rather is simply experiencing "normal" variation? How do you calculate what is "normal"? Includes the Eight Rules for Identifying Specialness, and guidelines for deciding which kind of SPC Chart is appropriate.
Chapter Two: Proportion Charts (P-Charts and np-Charts). If you know that on average you make 6% errors, then how high would the error rate have to rise before you would say, "Let's investigate. Something special is happening." Proportion charts track the percent times that things are correct or incorrect. Examples: % questions missed by a student on a test, % phone calls answered in three rings, % random audits that showed two or fewer errors, % paperwork that was error-free, % of sales contacts that resulted in a sale. For tips on how to interpret SPC Charts and how to determine subgroup sizes, see Appendix A: Tips for using any SPC Chart For those of you who want a non-technical explanation as to why SPC Charts work, there is a short description in Appendix B: Why does SPC work? of the reasoning underlying SPC Charts.
Chapter Three: Unit/Count Charts (U-Charts and C-Charts) When it is possible to make more than one error per task, then you can't think in terms of proportions, but rather need to think in terms of average errors. For instance, a front desk clerk in a hotel can probably make an almost infinite number of "goofs" per customer. Unit/Count Charts track the average number of errors. Examples: size of backlogs and queues at different times of the day, number of phone calls per day, number of complaints per 1000 customers, number of contacts generated per sales-person, number of "children's meals" sold per week at a restaurant, number of items purchased per customer, number of "dings" in a car body freshly painted, number of errors (such as dents, punctures, mislabeling, etc.) per 1000 cans of soup.
Chapter Four: Averages Charts (X-bar charts) and Range Charts (R-Charts). These charts are used for finding the averages of data that can have fractional values. For non-manufacturing environments, the most common use of these charts is for measuring how long something takes to accomplish. Example: You might track how long clinicians took for patients who had been scheduled for fifteen minutes. The "day" would be your subgroup. The Averages Chart would track the daily average over time. In this case, that daily average might be 14.2 minutes. The Range Chart tracks the range between high and low for that day. If you had a high of 30 minutes and a low of 10 minutes, then the range would be 20 minutes (30-10.)
Both Proportion Charts and Unit/Count Charts always have integer data without fractions. (You either goofed three times or two times, but you can't goof two and a half times.) Averages and Range Charts can have fractional values like 14.5 minutes. Other examples: length of time to complete a task, length of time it takes a customer to enter a store and leave, length of time it takes a customer to call back with a complaint. You can also use Averages and Range Charts for measuring physical data that can have fractional values. Examples: average weight of a grocery bag, average height of a customer (perhaps in an airplane or other cramped space), average temperature of freezers, etc.