What is Management?
For the purpose of this paper we need to establish a basic premise for management-by-prediction. First, from Deming, we get the simple but effective argument:
"Theory of knowledge helps us to understand that management in any form is prediction. The simplest plan - how may I go home tonight - requires prediction that my automobile will start and run." (Deming,1993:pp.101,2).
From this we may deduce that management is in many ways, and in the simplest of terms, planning and organising work. Therefore, it stands to reason that if we plan and organise 'work', it is implicit that we are predicting something will happen in the future as a result.
Wheeler, uses a similar premise for his argument on prediction:
"Since prediction is the essence of management, this ability to know what to expect when a process is behaving predictably is invaluable" (Wheeler,2000:24).
Here, Wheeler makes a link to process, as the source of information to help managers predict. We are told that control charts (now called process behaviour charts by Wheeler) represent the 'voice of the process'. Data, displayed on a chart indicates whether or not the process is in control. A process in control emits 'noise', but if it goes out of control, it emits a 'signal' with a data point outside the control limits. A process emitting noise cannot be used for prediction because it is probably out of control.
So while both Deming and Wheeler appeared to shift their thinking towards the notion of prediction, neither gave an in depth account on how to predict. For this we need to (re)turn to Shewhart's work first published in 1931. Here, Shewhart defines the notion of control:
"For our present purpose a phenomenon will be said to be in control if through the use of past experience, we can predict, at least within limits, how the phenomenon may be expected to behave in the future. Here it is understood that prediction within limits means that we can state, at least approximately, the probability that the observed phenomenon will fall within the given limits" (Shewhart, 1931:6) (emphasis in original).
Shewhart is therefore making a clear link between control and prediction at the beginning of his thesis. Space does not allow a full exposition of this here, but suffice it to say the notion of prediction is a dominant theme in this work.
What is Prediction?
A dictionary definition of predict is: "say or estimate that (a specified thing) will happen in the future or will be a consequence of something" with its root in Latin it comes from praedict meaning 'made known before hand (New OED,1998:1460). Prediction therefore represents a process, involving a judgement about what may or may not happen in the future.
To elaborate on this in more detail, we need to refer to Shewhart's later work, which began as a series of lectures given at Deming's request. These were then edited by Deming and published in 1939. Here we see the influence of the philosopher Lewis on Shewhart's (and Deming's) thinking. He developed histhe Theory of prediction around three components of knowledge: 1) evidence, 2) prediction, 3) degree of belief in the prediction based on the strength of the evidence.
Figure 1 Relationship between evidence, prediction and belief, adapted from (Shewhart, 1939,pp 85,86).
The unusual notion of belief is duly qualified in Postulate II :
"In what follows we need to keep clearly in mind that the statement that the quality of product is in a state of statistical control involves a prediction P which may or may not be true, and it involves the evidence E for believing in the prediction. The statement itself is a probable inference. I shall assume the basic.
Postulate II. The objective degree of rational belief pE' in an inference involving a prediction P based upon evidence E is not an intrinsic property like a truth but inheres in the inference through some relation of the prediction P to the evidence E" (Shewhart, 1939:42) (emphasis in original)
This is clearly an important feature of Shewhart's work, and one that we will try to unravel in this paper. To understand how figure 21 works, we must try to imagine the universe being in a state of flux, where knowledge of the universe is non-static (as opposed to static). Shewhart was irritated by engineers and scientists who were seeking "exact" or static knowledge about phenomena in the universe. What engineers should be concerned with is the effective validity of predictions.
To put Shewhart's epistemology into context, its roots can be traced back to Heracleities, who argued that one cannot stand in the same river twice. The idea that the universe is in a state of flux has been debated for centuries, using the notions of 'being' and 'becoming' which can be found in these pre-Socratic ideas (Guthrie:1971). Being, suggests that something exists, and takes an objective 'static' view of the world. Becoming, suggests that everything is in flux, or motion. To adhere to the latter, begs the question of 'how can we know anything, if everything is in flux?' Or more accurately what is ourthe Theory of knowledge, or epistemology. Needless to say, the early Greek philosophers gave this debate some considerable thought. This position can often lead to charges of solipsism or scepticism (Wilcox:2002).
Plato deconstructed various theories of knowledge in his dialogues, and at one stage Socrates was considering the Protagorean thesis of "man is the measure of all things" and the "Heracleitean state of flux" in discussion with his pupil, Theaetetus.
"Socrates: Their doctrine that 'being' (so-called) and 'becoming' are produced by motion, 'not-being' and perished by rest, is well supported by such proofs as these: the hot or fire, which generates and controls all other things, is by itself generated by movement and friction - both forms of change.... And so with the condition of the soul. The soul acquires knowledge and is kept going and improved by learning and practice, which are of the nature of movements" (Plato;1934:36-7).
Here, we may detect signs of continuous improvement in Plato's thinking, two and a half centuries ago. It is no surprise, therefore, that Shewhart's work was premised on the notion of becoming, and we can see how this led to him writing such phrases as; "Mass production viewed in this way constitutes a continuing and self corrective method for making the most efficient use of raw and fabricated material" (Shewhart,1939:45).
And referring to the same feature in the conclusion:
"In fact an economic standard of quality is not a written finality, but it is a dynamic process. It is not merely the imprisonment of the past in the form of a specification (step 1, fig 10,p45) but rather the unfolding of the future as revealed in the process of production (step II) and inspection (step III), and made available in the running quality report" (ibid:.119) (emphases added).
Given that Shewhart was writing for an engineering/scientific audience as well as the managers at the Bell Telephone company, he became adept at using many small figures and diagrams to make his point. Here, he tries to demonstrate the non-static notion of everything being in flux, and where the 'present' is the point for prediction based on our knowledge of the past.
(Figure 2, adapted from Shewhart, 1939:133)
Prediction therefore involves a fluid relationship between the past, the present and the future. If we take figure 2, we see how the past, present and future interact, to allow a process to be depicted, the present is ephemeral and represents a non-static universe. The present provides an opportunity for evidence to be gathered as data, from which we may predict the future. Our degree of belief in the prediction is relative to the quality of the evidence and our knowledge of the past. So how does this work in practice?
What is Management by Prediction?
Shewhart had a riddle to solve which provides a clue to how prediction works in practice.. "Knowing begins and ends in experience; but it does not end in the experience in which it begins" (C.I.Lewis quoted in Shewhart,1939:80). What does this riddle mean?
A key to understanding the riddle is in the premise that knowledge is non-static (becoming) (Shewhart,1939:104). While the universe is perceived as being in a state of flux, we also need the ability to measure certain phenomenon in-order to make sense of the universe. The universe is non-static - it is in movement - so when we measure x, it is a snapshot in time. The data point x is treated as evidence of how a phenomenon is behaving, from which we predict how this may behave in the near future. Measuring is a scientific investigation but it is not to establish "facts" in a deterministic or positivistic sense, this static view of the universe and knowledge is rejected. This is the domain of the proponents of 'being'.
"I take it that the object of a scientific investigation is to organise past experience and so to direct the acquisition of new experience that it will be possible to make valid predictions on the outcome of any proposed experiment that is capable of being carried out, and to make predictions in less time than it would take to carry out the proposed experiment" (Shewhart,1939:105) (emphasis in original).
The aim should be to predict. Engineers/managers are not trying to establish facts, for static knowledge to be presented in some tabular form. Therefore, evidence, prediction and degree of belief in the prediction, are all variables in the process of predicting. The correlation between the three components of knowledge fluctuates, so the:
"results of an experiment should be presented in a way to contribute most readily to the development of the knowing process" (Shewhart,1939:105) (emphasis in original).
I will now try to describe this process using Shewhart's theories.
Data collection for prediction
Shewhart argues that is not possible to treat measurements as facts so. A true value of phenomena x does not exist. The process of measuring x is governed by the same rules that we have outlined above. It is subject to variation. Therefore knowledge of x is only probable. So evidence from measuring x is interpreted by Shewhart using an adaptation of Lewis' riddle: "knowledge provided by such measurements begins in these measurements and ends in measurements, but it does not end in the measurement in which it begins: such knowledge can only be probable" (op-cit).
The point of measuring x is to provide a summary of x in order to make a prediction. The degree of belief in the prediction, is the same as that of the original data used to make the summary. He argues that what may be an effective summary for a prediction one day, may not be effective at a later date. Why? Because of the non-static nature of the universe. Therefore, given the same data tomorrow, as today, our prediction may differ because of new theories that have emerged in the light of experience. With this in mind, the criteria for measuring are important to ensure the process and procedures are themselves in a state of statistical control.
The critical part of the process of prediction is in taking great care in the collection of data, which will be used as evidence. We have to distinguish between data collected under controlled conditions and data that are not. If the data collection process is in statistical control then we can make reasonably accurate predictions from this data. This then relates to the degree of belief we may have in the prediction. Conversely, if the data collection process is not in statistical control, then we cannot make accurate predictions and our degree of belief will be far less.
Shewhart makes the distinction between precision and accuracy of data collection. Using a classic book of its time by Goodwin (1908), he explains the importance of this distinction and how it is often lost. The distinction is clearly important, for if one is to make predictions based on available evidence, then one needs to be sure that the data were reliable.
"Careful writers inthe Theory of errors, of course, have always insisted that accuracy involves in some way or other the difference between what is observed and what is true, whereas precision involves the concept of reproducibility of what is observed" (Shewhart, 1939: 124).
Goodwin provides his own interpretation:
"By the precision or precision measure of a result, ... will be always understood the best numerical measure of its reliability which can be obtained after all sources of error have been eliminated or corrected for. ... By the accuracy of the result, should, strictly speaking, be understood the degree of concordance between it and the true value of the quantity measured. Since, however, the latter is usually unknown, it is seldom that we can obtain a numerical value of the absolute accuracy of a measurement... The terms 'accuracy' and 'precision' are often carelessly used indiscriminately" (Goodwin,1908:8).
Taking this advice, Shewhart demonstrates in far more detail than can be shown here how this would work in practice. But, as was his tendency, he simplified the process with the aid of diagrams and models. Here, he explains how the measurement process works.
Figure 3, adapted from (Shewhart, 1939:89)
Where X is the data being gathered, H is the person collecting the data, and C is the conditions. The concomitance of the three signs in the figure represents the original data being collected, to be used in various forms of prediction. Each time this process takes place it creates data for a control chart.
This process is best explained by Shewhart, as this is clearly an important feature of histhe Theory.
"In passing from the original data on the left… to the predictions on the right, the interpreter takes three steps, involving the introduction of assumptions and interpretive constructs; he adds something to the original data… It is necessary now for us to note more carefully than heretofore how knowledge differs from original data and predictions. Knowledge as has been stated begins in data and ends in other data. It starts with original data and makes predictions about data not yet taken, involving, at the same time, something more - it involves a certain degree of rational belief in a prediction based upon evidence derived from the original data: this relationship between prediction and evidence is of great importance from the view point of the presentation of the results of measurement as knowledge" (Shewhart,1939:101-103)
Analysing and interpreting data for prediction.
When the data has been collected it has to be analysed and interpreted in a predictive process. Shewhart'sthe Theory of prediction draws on the use of the past to interpret the present in-order to predict the future. He uses Figure 2 above to explain the dynamic nature of systems thinking, and how statistical methods fit into that process.
To try to explain the dynamic nature of this model he refers to CI Lewis' paradoxical riddle quoted at the beginning of this section. This riddle encapsulates the fluid nature of systems thinking, where phenomena are in a state of flux. Shewhart adapts the riddle to suit his audience:
"I shall assume that knowledge begins and ends in experimental data but it does not end in the data in which it begins" (Shewhart, 1939:85).
While Figure 2 is an effective display of the process is does not convey the cognitive or mental process of prediction, which was a major part of Lewis' thesis and the role of 'mind'. Therefore, we have to consider how the interpretive and predictive process takes place. In Shewhart'sthe Theory, engineers and statisticians played a pivotal role in this process. Engineers were in charge of designing products and processes as well as being trained in statistical methods. Take this following passage as an example.
"The operation of control is in this sense a dynamic process involving a chain of actions, whereas the criterion of control is simply a tool used in this process. The successful quality engineer, like the successful research worker, is not a pure reason machine but instead is a biological unit reacting to and acting upon an ever changing environment" (Shewhart, 1939:38) (emphasis added).
The concept of an engineer acting as a 'biological unit' is similar to Lewis' use of 'mind' in histhe Theory of pragmatism. We can therefore deduce from this that 'mind' is part of the interpretive and predictive process. Interestingly, Shewhart rarely mentions the human element of work in his writing, but this remarkable short passage is very illuminating.
In the same vein, he was explicit on the use ofthe Theory in the analytic and interpretive process. The influence of Lewis, can be found in Shewhart's postulate, that there is no knowledge withoutthe Theory. Data do not interpret themselves, so each data point requires the use ofthe Theory to provide knowledge. It is at this point, that we detect more pragmatism in Shewhart's work. For Lewis and other pragmatists, theories are inherently practical, so what is useful today, may not be useful tomorrow in the light of new evidence or different circumstances. Hence engineers should use theories purely on their practical application in interpreting the here and now. The importance of this concept is that time is of the essence for the engineer on the shop-floor. The pragmatic use and application of theories in interpreting the present is therefore a crucial part of thisthe Theory.
Presenting data for prediction
One of Shewhart's most important inventions was the control chart. Different types of control charts can be used in many ways to present data (Wheeler:2000). Control charts are the means for displaying the data that have been collected in a way that illustrates time and variation in a process. The horizontal line on the chart is the vector, and the vertical line the scalar. Data is plotted on the chart and a mean or median calculated and drawn along the chart in the appropriate place. Three sigma control limits are computed to find the upper and lower control limits of the process. As the data is collected and displayed on the chart the engineer is given a visual perspective of what is going on in the process (the voice of the process). This is where the interpretation and prediction takes place. A suitably trained person will be able to interpret any variation in the process, and follow the rules and guidelines laid down by Shewhart.
With remarkable insight, Shewhart developed the notions of 'chance' (noise) and 'assignable' (signal) causes of variation to help us to interpret control charts. Put simply, a chance cause is a sign of random variation in a process. Assignable causes are where an external influence has created non-random variation in a process and the data point falls outside the control limits. Assignable causes are traced and removed in-order to bring the process back into control and predictability. Only in the absence of assignable causes can one predict the future performance of a process. Needless to say, this is an ongoing process, so the ability to predict is continuously reassessed, as data is collected and displayed on the chart.
Again, we find remarkable detail, to ensure the rigour of data collection, analysis and interpretation in his work. Here Shewhart details the criterion for collecting and presenting data. One should note the important point on the use of text, so that data can be put into context. Data without annotation may be easily misinterpreted.
"There are at least the following four characteristics of original data to be considered in presentation:
- Numbers representing the numerical values of the measurements
- Text describing the condition under which each measurement was made, including a description of the operation of measurement.
- Human element of observer H
- Order in which the numbers were taken" (Shewhart,1939:89).
So when considering data, we should be able to ascertain whether it was the same person measuring (the observer), whether the conditions were similar, and the order in which the data was gathered. The benefit of this method is that it allows engineers and managers to see the state of the processes in their organisation and know that the data they are being shown is reliable. Following Shewhart's rules, there is no point in tampering with processes displaying random variation. These processes are 'behaving normally' and tampering will probably lead to a state of non-random variation.
Shewhart spent considerable time in working out how best to present data. The control chart was a significant invention in that it provided the means to present complex information in a form that was easy to interpret and analyse. Significantly, they represent the fluid nature of the becoming while capturing the here and now with a data point. Control charts are a unique invention and draw on several other theories, not least semiotics andthe Theory of signs. Shewhart read widely, so for example he used the works of Morris (1938) Foundations ofthe Theory of Signs to help explain his ideas. In a similar vein he refers to Dwiggins (1928) Layout in Advertising - a classic in its own right - to show the importance of presenting data so that it can be easily understood, and more importantly, not misleading the interpreter.
So while the control chart was an important invention and tool in histhe Theory, he had to take great care not to distort the meaning of the original data. If the meaning of the original data was lost or distorted, then any predictions made from that data could be invalid. Hence we are given two rules for the presentation of data.
"Rule 1. Original data should not be presented in a way that will preserve the evidence in the original data for all the predictions assumed to be useful...
Rule 2. Any summary of a distribution of numbers in terms of symmetric functions should not give an objective degree of belief in any one of the inferences or predictions to be made therefrom that would cause human action significantly different from what this action would be if the original distribution had been taken as a basis for evidence" (Shewhart, 1939: 88, 92).
Wheeler (2000) summarised these rules to make them more user friendly.
" (1) Data should always be presented in such a way that preserves the evidence in the data for all the predictions that might be made from these data... (2) Whenever an average, range or histogram is used to summarise data, the summary should not mislead the user into taking action that the user would not take if the data were presented in time series" (pp:12-13).
While these two rules refer to the use of data, he has also to consider the use of words and language in general. Clearly, while exercising such rigour with the use of data he requires a similar approach to the meaning of words. Hence we are given an additional rule called the Criteria of Meaning. Shewhart refers to a book by Chase (1933) called the "Tyranny of Words", which also appears to draw on semiotics and linguistics for its main themes. Chase (1933:66) uses a model shown below to illustrate how words may function.
Figure 4, adapted from Chase 1938:66
Chase makes the point that his model has only a dotted line on the base. This is because there is no direct link between the word and object in histhe Theory. The object, say a cat, is the referent, which is 'called up' by our senses, which in turn 'call up' a word from our memory to give to this type of object. Hence, there is no direct link between word and object. Chase is at pains to show how confusing language can be, from which Shewhart takes his cue for a "criterion of meaning".
"Obviously all scientific predictions must have definite meanings, and we shall accordingly choose the following.
Criterion of Meaning: Every sentence in order to have definite scientific meaning must be practically or at least theoretically verifiable as either true or false upon the basis of experimental measurements either practically or theoretically obtainable by carrying out a definite and previously specified operation in the future. The meaning of such a sentence is the method of its verification" (Shewhart, 1939:94).
Clearly important in Shewhart's work, the rules provide the rigour for the use and presentation of data, to preserve the original meaning and context from the measuring process.
Summary
In this paper I have tried to summaries an immensely complexthe Theory developed over 70 years ago. Most people trying to read Shewhart's work are overwhelmed by the difficult statisticalthe Theory and mathematics involved. However, my approach has been to try to interpret his work as a practicalthe Theory, which I have called management by prediction. To do this I have had to interpret Shewhart's work from the basis of reading C.I.Lewis' treatise of conceptual pragmatism, which he read 14 times (Deming:1991). By taking this approach I have been able to shed new light on his work and hopefully make it more accessible to practising managers.
With modern software packages (e.g. WinChart), there is no need to be particularly adept at statistics to be able to operate this system. However, what will be challenging to anyone following the precepts laid out in this paper is (a) the degree of rigour (precision and accuracy) required, and (b) the need to actively predict from the ongoing process of interpreting data.
The most exciting aspect to emerge from this paper is the dynamic nature of the process of running an organisation in this way. Using statistical methods, as they were originally intended by Shewhart, will provide managers with a pro-active methodology for managing an organisation. Managers could be judged on their ability to accurately predict the random variation in the key business performance processes. They would be able to predict if the organisation will meet its objectives, by measuring the key processes underpinning them.
Perhaps the most challenging aspect to comprehend in thisthe Theory is to accept the paradigm shift from a predominantly 'positivistic universe' (being) to the fluid nature of systems thinking. For many managers, engineers, scientists and technicians running modern organisations, there is still a strong tendency to treat data as facts and rely too heavily on month by month comparisons of management information (Wheeler:2000). Carefully developed training programmes would be required to make this change happen, but I would argue that the focus should not be on teaching managers statistical techniques, but more to do with the philosophy of prediction.
Author details
Dr Mark Wilcox
Centre for Business Performance
Cranfield School of Management
Cranfield University
Bedford MD43 0AL
mark.wilcox@cranfield.ac.uk
Mark Wilcox is an active teacher, researcher and consultant in quality management and performance measurement. He is Director of Research with the Deming Special Interest Group in the Institute of Quality Assurance. His current research interests are in being able to predict performance in organisations and promoting the seminal work of Dr Walter Shewhart.
References:
Blankenship, B. and Petersen, P.B., (1999) W.Edwards Deming's Mentor and Others Who Made an Impact on His Views During the 1920s and 1930s. Journal of Management History. Vol 5, No 8, pp 454-467. MCB University Press.
Bird, G. (1986) William James, Routledge Kegan Paul, London.
Chase, S., (1938) The Tyranny of Words. Methuen and Co. Ltd. London
Chung-Ying, Cheng., (1969) Pierce's and Lewis's Theories of Induction, Martinus Nijhoff: The Hague.
Deming, W.E., (1986) Out of The Crisis, MIT Press, Cambridge, Massachusetts.
Deming, W.E. (1991) "A Tribute to Walter A. Shewhart On the Occasion of the 100th Anniversary of His Birth" SPC INK Newsletter Winter 1991. SPC Press, Inc.
Deming, W.E., (1993) (2nd edition) The New Economics, MIT Press, Cambridge, Massachusetts.
Dwiggins, W.A., (1928-48) Revised edition. Layout in Advertising. Harper and Brothers. USA.
Goodwin, H.M., (1908/1920 edition) Elements of The Precision of Measurements and Graphical Methods. McGraw-Hill Book Company, New York.
Guthrie, W.K.C., (1971) A History of Greek Philosophy. Volume 1: The Earlier Presocratics and the Pythagoreans, Cambridge University Press, Cambridge
Harrison-Barbet, A., (2001) (2nd edition) Mastering Philosophy, Palgrave, Basingstoke.
James, W., (1913) Talks to Teachers on Psychology: and to students on some of life's ideals. Longmans Green and Co, London.
James, W., (1917/1935) Selected Papers on Philosophy, Dent and sons, London
Lewis, C.I., (1929/1956) Mind and The World Order: Outline of athe Theory of Knowledge. Dover publications: New York.
Morris, C.W., (1938) Foundations forthe Theory of Signs. International Encyclopaedia of Unified Science. Volume 1. Number 2. The University of Chicago Press, Chicago, Illinois.
Peterson, G.T. (1999) Unpublished paper. Mind and the World Order: Chapter Summaries.
Plato, (1934)( Edited by Cornford, F.M) Plato'sthe Theory of Knowledge: Theaetetus and the Sophist translated with a running commentary. Routledge ∓ Keegan Paul, London.
Shewhart, W.A., (1931/1980) Economic Control of Quality of Manufactured Product ASQC: Milwaukee.
Shewhart, W.A., Deming, W.E. (ed) (1939/1986) Statistical Method from The Viewpoint of Quality Control. Dover Publications: New York.
Wheeler, D (2000) Understanding Variation SPC Press, Knoxville
Wilcox, M. (2002) Whither The Pragmatism? Proceedings of the 8th International Deming Scholars Seminar, Fordham University, New York.
Wilcox, M, and Bourne, M (2002) Performance and Prediction. Performance Measurement Association Conference. Boston 17th -19th July, 2002.