Understanding Variation 26 YEARS LATER IN 1931, Walter Shewhart published his landmark book Economic Control of Quality of Manufactured Product . He asserted that his theory and methods were an innovation to the science of management and wrote: “We are sold on the idea of applying scientific principles. However, a change is coming about in the principles themselves, and this change gives us a new concept of control.” 1 W. Edwards Deming supported this idea in a foreword he wrote for the 1986 republishing of Shewhart’s Statistical Method From the Viewpoint of Quality Control : “Another half-century may pass before the full spectrum of Dr. Shewhart’s contributions has been revealed in liberal education, science and industry.” 2 In 1990, QP published “Understanding Variation” by two of this article’s authors. 3 The article included examples of the economic and psychological losses associated with interpretations of data without a framework for un- derstanding variation. The economic losses included misguided changes to service delivery, investigations of trends where none existed and increased costs from increased variation. The psychological losses included blaming workers for what were actually faults of the system and experiencing anxi- ety from false hopes of improved operating conditions.
DATA ANALYSIS Correctly assessing variation is fundamental to sound decisions 26 YEARS by Thomas Nolan, Rocco J. Perla and Lloyd Provost LATER In 50 Words Or Less • Decisions made without knowledge of common and special causes often lead to increased variation, poor perfor- mance and misattrib- uted credit or blame. • Extending the ap- plication of Walter Shewhart’s approach to variation to datasets across a range of pub- licly available sources is an opportunity to improve decision mak- ing and learning from reported data. November 2016 • QP 29
Today, 26 years later, big data analytics, data-driven outcomes of the system. decision making, business intelligence and transparency • Special causes— Those causes that are not always in public data have created new opportunities for an un- part of a system (process or product) or do not affect derstanding of variation to guide decision making. The everyone, but arise because of specific circumstances. 5 U.S. Bureau of Labor Statistics, for example, is the prin- A process or system that has only common causes cipal federal agency responsible for measuring labor- affecting the measurement of interest is called a stable market activity, working conditions and pricing changes process. A stable process is one in which the cause sys- in the economy. Its mission statement emphasizes the tem for the measure of interest remains essentially con- collection, analysis and dissemination of “information stant over time. A stable process implies only that the to support public and private decision making. ”4 Without variation in outcomes is predictable within limits, not useful interpretation, however, this dissemination could that it has desirable or undesirable performance. actually degrade decision making. A process with outcomes affected by common and We want to extend the application of Shewhart’s special causes is called an unstable process for the mea- methods presented in our 1990 article to data sets across sure of interest, with the magnitude of the variation from a range of publicly available sources—sources that are one time period to the next being unpredictable. As spe- used by and the basis for the U.S. government and other cial causes are identified and appropriately acted on, the organizations to assess conditions and make decisions. process becomes stable. 6 Compared with data from individual organizations, the This theory of variation provides a basis for action to scale of these data sets underscores the importance of improve a system. A stable system requires a fundamen- understanding and applying Shewhart’s theory. tal change to affect its future performance (because it is stable), while an unstable system requires local action Shewhart’s theory of variation depending on the special cause. Shewhart’s theory of variation differentiated between In addition to providing the basic concepts of the the- common and special causes of variation in data: ory, Shewhart also introduced the control chart method • Common causes— Those causes that are inherent to determine whether variation in a process is due to in a system (process or product) over time, affect common or special causes. The Shewhart control chart everyone working in the system and affect all consists of three lines and points plotted on a graph. Shewhart control chart example for an economic measurement / FIGURE 1 Gross domestic product percent change (basis in 2009 U.S. dollars) Measure 12 10 UCL 8 7.1 6 4.6 4 2 2.0 1.7 1.2 0 LCL -2 -3.0 -4 -6 -8 -10 1996 Q1 Q3 1997 Q1 Q3 1998 Q1 Q3 1999 Q1 Q3 2000 Q1 Q3 2001 Q1 Q3 2002 Q1 Q3 2003 Q1 Q3 2004 Q1 Q3 2005 Q1 Q3 2006 Q1 Q3 2007 Q1 Q3 2008 Q1 Q3 2009 Q1 Q3 2010 Q1 Q3 2011 Q1 Q3 2012 Q1 Q3 2013 Q1 Q3 2014 Q1 Q3 2015 Q1 Q3 2016 Q1 Q3 2017 Q1 UCL = upper control limit LCL = lower control limit 30 QP • www.qualityprogress.com
DATA ANALYSIS Nursing facility residents with one or While there are numerous books describing how to construct Shewhart’s charts, we will focus on more falls with major injury / FIGURE 2 the broader method and its modern applications. 7 20% Shewhart control chart method 15% Figure 1 shows an example of a Shewhart control 10% chart for a popular federal economic measure- ment—the quarterly change in gross domestic 5% product (GDP). This measurement is usually pre- sented in published reports and on the U.S. Depart- 0% 2005 Q1 Q2 Q3 Q4 2006 Q1 Q2 Q3 Q4 2007 Q1 Q2 Q3 Q4 2008 Q1 Q2 Q3 Q4 2009 Q1 Q2 Q3 Q4 2010 Q1 Q2 Q3 ment of Commerce’s website. 8 Business media re- port reactions when the quarterly value is released or revised. For example, the Wall Street Journal Source: U.S. Department of Health and Human Services, “Health System Measurement Project,” https://healthmeasures.aspe.hhs.gov. used the headline “U.S. GDP Grew a Disappointing 1.2% in Second Quarter” for an article that offered Nursing facility residents with one this summary of 2016’s second quarter: Declining business investment is hobbling an al- or more falls with major injury ready sluggish U.S. expansion, raising concerns (Shewhart chart) / FIGURE 3 about the economy’s durability as the presiden- tial campaign heads into its final stretch. Gross 15.4% UCL domestic product, the broadest measure of goods 15.2% and services produced across the U.S., grew at a 15.0% seasonally and inflation adjusted annual rate of 14.8% LCL just 1.2% in the second quarter, the Commerce De- 14.6% 14.4% partment said Friday, well below the pace econo- 14.2% mists expected. 9 2005 Q1 Q2 Q3 Q4 2006 Q1 Q2 Q3 Q4 2007 Q1 Q2 Q3 Q4 2008 Q1 Q2 Q3 Q4 2009 Q1 Q2 Q3 Q4 2010 Q1 Q2 Q3 Each quarter is treated as a special event. For example, the Wall Street Journal recently pub- lished these headlines for three sequential quar- UCL = upper control limit ters: LCL = lower control limit 1. “U.S. Economy Shows Signs of Gearing Up”— reporting on 2013’s fourth quarter in which economic implications for global markets as well as the there was a 3.5% increase in GDP. 10 United States. Applying the Shewhart chart method can 2. “U.S. Economy Shrinks by Most in Five Years”— minimize these losses. reporting on 2014’s first quarter in which there was a The method’s five key components are: 2.1% decrease in GDP. 11 1. A selection of a measurement and statistic to be 3. “Growth Rebound Stokes Fed Debate”— report- plotted. The choice of measurement will give differ- ing on 2014’s second quarter in which there was a 4% ent insights about a process or system. In the GDP ex- increase in GDP. 12 ample, the key statistic reported was the change (per- These reports clearly suggest big, quarter-to-quarter centage difference from the previous quarter) in GDP. swings in our economy as if they confer actionable in- 2. A method of data collection from the process formation. The Shewhart chart in Figure 1, however, in- or system—observation, measurement and dicates a stable system for the previous five years. The sampling procedures. These methods provide an economic losses associated with the misinterpreted vari- operational definition for the measurement, and ation in quarter-to-quarter data include the consequenc- information in the Shewhart chart always will be es of actions taken by individuals and institutions based conditional on how data are collected and a mea- on nonexistent trends such as potentially raising or surement is obtained. The U.S. Department of Com- lowering the U.S. interest rate, which carries profound merce’s website offers an extensive explanation November 2016 • QP 31
Recommend
More recommend