Hint to W. Edwards Deming (on the occasion of the Financial Crisis)

One year ago the Lehman Brothers Inc. went broke, an important milestone in the course of the ongoing Financial Crisis. Because there is a quest for solutions, I want to point to the teachings of a man where the often used saying “If they just had listened to him!” even after critical analysis really applies:

By the way: From Deming’s books “Out of the Crisis” and “The New Economics” still no German translation is published.

Punctuality of busses – an example from the city of Aachen

After I had to regularly travel certain routes per bus or train, I had to endure one delay or two. Because of my statistical background I came upon the idea to record the delays and to find out how punctual those means of transport really are. After a long time in mid-January I started again with it and now want to give an indication what you can make of it.

As an example I took the bus stop “Kuckelkorn” in the city of Aachen which is in my neighbourhood. There I recorded for 6 busses when they’ve departed. The busses are the following, driving each for the transport system of Aachen, the AVV:

1. Bus 35 to Breinig Entengasse in Stolberg (into the city). Departure: 6:26 pm.
2. Bus 55 to Vaalserquartier (out of the city). Departure: 6:28 pm.
3. Bus 3A to Uniklinik (into the city). Departure: 6:28 pm.
4. Bus 3B to Uniklinik (out of the city). Departure: 6:28 pm.
5. Bus 45 to Aachen-Brand (into the city). Departure: 6:36 pm.
6. Bus 45 to Uniklinik (out of the city). Departure: 6:36 pm.

I started the measurements on the 14th of January. Those went to the 24th of April, which provided – with the exception of Carnival Monday and Easter – 72 working days. On the 20th of March I could not be there, so I could measure on 71 days. For bus 35 on the 14th of January and bus 3A on the 28th of January there each has been a missing value.

At first the are some measures for a quick view. Because occasionally some big delays took place, robust measures are appropriate:

• The median: When you sort the data, then the value in the middle is the median. It is relatively invulnerable against extreme outliers.
• The lower and upper quartile: . The quartiles divide the sorted data into four parts of equal size. The first quartile (Q1) cuts off the lowest 25% of the data. The third quartile (Q3) cuts off the upper 25% of the data.
• The interquartile range (IQR): The IQR is the difference of the third and first quartiles and acts as a measure of dispersion. It is also called the midspread.

Delays at the bus stop “Kuckelkorn”
Bus	35	55	3A	3B	45	45
Departure	6:26 pm	6:28 pm	6:28 pm	6:28 pm	6:36 pm	6:36 pm
1. quartile	-0:28,0	1:52,0	0:28,0	0:12,0	-1:09,0	0:29,0
Median	-0:13,5	4:56,0	0:36,0	1:00,0	0:03,0	1:12,0
3. quartil	1:04,3	7:27,0	1:54,0	3:28,0	0:50,0	2:46,0
IQR	1:32,3	5:35,0	1:26,0	3:16,0	1:59,0	2:17,0

(times in minutes and seconds)
(Example: The delays of bus 55 had a median of 4 minutes and 56 seconds.)

Among the busses examined bus 55 stands out, which drives from the border to Belgium at Aachen-Lichtenbusch via Kornelimünster and the inner city and is departing later than the other busses. The dispersion of the departures is bigger too.

Besides the busses which has to pass through the city seem to depart later, which might be because of the advanced position on their routes. To prove this more busses had to be examined.

A further examination of the delays requires observing them over the time. The aforementioned measures are assuming that the circumstances remained the same. But weather, school holidays or similar thing may have an influence. This way you also can find out if some busses has been relatively late (or punctual!). This would indicate special causes or events, which have to be examined.

A possible advantage of recording delays is that one bus may turn out to be punctual. On the other hand I’ve experienced much more unpunctual busses. When I was working in Baesweiler, bus 51 coming from Aachen, for which I had to wait at the endpoint at the Reyplatz, has been generally late.

However, with the trains of the Deutsche Bahn I’ve experienced more. Out of the trains which I took at the weekend – departing in Bremen and supposed to arrive at 8:48 pm in Cologne – around autumn every fourth arrived more than 45 minutes late, effectively making me miss the connecting train to Aachen.

Complying to regulations is enough, isn’t it?

During my study I once encountered the following situation:

Within informatics (my minor field of study) I had to buy a lecture note for a basic lecture. The lecture note shop was only open at noon. When I turned up there, there were about 100 students standing in a line and wanting to buy that note too. When the opening hours came to their end, the staff closed the door, selling the last notes of the day to the students which were still in the shop and sent them out through a back door.

And now to the big question: The staff has complied to the regulations which said that the lecture note shop had certain opening hours. But has – regarding the fact that many students were waiting in vain and had to show op the next day and wait again – the selling of the notes maybe been poorly conducted? Or in more general terms: Is it enough to comply to regulations when these regulation are potentially not optimal? For the staff of the shop there just one thing was certain: After the opening hour no script could be sold.

Others suffer more! or: How much improvement do you need?

One thing is sure: The German railways (Deutsche Bahn) are having a serious image problem. Around 2002 I’ve read in a newspaper the results of a survey according to which in Europe only the Italians and the Dutch had a worse opinion about their railways (the Dutch railways seem to have improved somewhat since then). You can concede to the German railways that they are doing much and that they do some of it well. However, you can’t help feeling that some regularly occurring problems (e.g. the same delays on a route) are accepted without being noticed, analysed, let alone reduced. The fact that the Deutsche Bahn is often acting self-centered in customer relations (new price system in 2002, abolition of buying tickets en route etc.) doesn’t really improve matters.

On the German website bahn-spass.de special particular stories and adventures as well as other interesting things about the German railways are compiled. One post tells the reader that surprisingly there also are problems with the railways in countries such as India, Switzerland, Argentinia and England. In the post the question is raised if in the light of these problems the Germans are just complaining on a high level. Considering that they would be glad in Africa when they are having something like the German railways, this question is most obviously understandable. But you should consider that in those countries they wouldn’t have reported these problems if they haven’t got the impression that something went wrong.

Generally some contemporaries are having strange conceptions about quality. During my study the copy machines at the University of Dortmund has been replaced by new ones. The operator of the machines had then anticipated the current trend of total profit maximisation and reduced the maintenance costs. This lead to not every copy machine working and that the remaining ones were not properly working. After on the bulletin boards notes were showing up asking the students to report the copying problems, the operator took the psychologically adept to remove the notes. Before a revolt has risen, the contract went to another operator. After a report in the campus newspaper InDOpendent a reader could not understand the anger, because problems are belonging to “modern technology”. Obviously he missed the point that a technical device is supposed to fulfill a task and that a modern version of that device is distinguished by more options and/or a better fulfillment of its task.

Generally some people seem to consider the reporting of a problem worse than the actual problem itself. One example is a post on Bahn-Spass.de, where the write complained about her bus in Berlin, which didn’t show up. Neither did the four following busses of the line. Because the bus had a 10-minute interval, this meant a delay of about 50 minutes. The outdoor temperature of 4 °C didn’t actually make things better. Nonetheless in a comment this report has been marked with the terms “lack of patience” and “hypocrisy” (“fehlender Geduld” and “Heuchelei”). As motive has been mentioned that in Berlin more than 1.300 busses of the BVG 300.000 kilometers of public transport are managed which are sometimes hampered by a demonstration, a roadblock or similar things. To this the following has to be noted:

• A bus is supposed to depart at the time noted in the timetable.
• The more the actual time of departure deviates from the timetable, the worse. The “damage” grows exponentially with the deviation (this means the additional damage added grows with each minute).
• The post suggests that the delays (resp. cancellations – the German railways don’t count cancellations as delays) are based on a special cause, thus were not a common event on that line. These special causes you have to look up and to consider how you can alleviate the consequences.
• The Kano model for the classification of customer’s demands shows that there are basic demands which the customers regard as absolutely necessary, so that they would be especially dissatisfied when those are not fulfilled. The cancellation of a bus can be considered a non-fulfillment of a basic demand.

When people are criticised for expecting that a product and a service works as announced, this is often tied with the remark that this expectational attitude is common only in Germany. Barring that we’ve just seen that expectations are part of the human nature, recognising and anticipating problems as well as the comparing the results with ideal values are necessary conditions for the improvement of the present conditions. Because in Germany traditionally great importance is attached on following orders, you might explain that criticism of customer’s demands with them being regarded as some kind of resistance. This kind of customer bashing is what you can consider as especially German (although these criticism occurs in other countries too – for instance in the Netherlands).

Coming back to the question whether complaints about the German railways are justified regarding the problems in other countries, you can best point out the Deming circle, which is a thinking model for continuous improvement. The opportunities for this you can find everytime and everywhere. Therefore the railways should look for reasonable improvements always and everywhere too.

Statistician bashing 2

While attending an exercise to the lecture in economic policy - with which I did have my problems - the tutor made a remark about the topic “statistics in economy”. Basically, he remarked that statistics are being used for such interesting subject as the per-capita-consumption of carrots and that Germany is having a mid-table position in this discipline. Obviously he doesn’t regard that application of statistics as that important.

I’m not an expert on that subject, but I can imagine that every person who takes part in the production, trade or sale of carrots, can use informations about their consumption well. The prediction of demand (like prediction in itself) is in fact an important part of management, and this should work with statistics much better than without statistics.

Like I said: The tutor who did make this remark, did this in the context of a lecture on economy.

The IT-company, a statistics-free zone?

“There is no substitute for knowledge.” (W. Edwards Deming)

After the publication of his book “Out of the Crisis”, W. Edwards Deming worked at a theoretical instrument, With which you are able to judge the status quo and to understand the necessary changes. The result is the System of Profound Knowledge. It consists of four elements:

1. Apprehension for a system: A system consists of two or more elements and fulfils a purpose which the elements on their own are not able to do. One example is a orchestra consistig of different musicians. To perform well depends less on the individual talents but on how they are playing together. When this is not possible – because every musician just wants to look good – it goes to the disadvantage of the orchestra. A company also forms with its suppliers, customers, employees, shareholders etc. a system, where you can call the cooperation some sort of economy. Like in an orchestra the attemps to optimise individual parts without regard for the system may result in its destruction. Because of this Deming proposed as a goal of a system that everybody should win on the long term. In this context he described the conventional proceedings of a company as the attempt to get a bigger piece of the cake. Instead they should attempt to make a bigger caken.
2. Knowledge about variation: In the statistical sense variation is the term for the diffusion of values. This diffusion can be noticed among persons, products, results, services etc.. Knowledge about variation is important for the following reasons:
   • For production it is important that the results remain consistent. The more you deviate from a optimal value, the worse is that. A train is f.i. expected to arrive at the same time.
   • The results of a process are having in themselves variation. When the variation is supposed to be reduced, the process has to be chaged. The aforementioned train can be expected to normally have the same amount of driving times. It is therefore not advisable to treat each drive as a special event.
   • An unusual result is indicating an unusual event. When a train which usually is 5 minutes late arrives 50 minutes past its schedule, which usually does not happen, then there must be a special cause which ca be detected.
   • When a desired goal lies outside the capability of a process, then it cannot be reached. When a train basically arrives 15 minutes late, you cannot expect it to be on time.
   • To know what a process is able to achieve, it has to be stable, which means that it has to be free from special events. The more the delays of the train are within a certain frame, the easier you can examine ad improve the train .
3. Theory of knowledge: To explain the world and the consequences of a deed, you need a theory. Management too has to know what’s gonna happen when implementing a measure. Thus management also is the Art of Prediction. A theory is also important to assess examples and experiences, because examples cannot prove a theory but at most disprove them. On a plane the angles of a triangle add up to 180 degrees. On a sphere this theory would not work. Finally the nature of theory makes clear that everything which is being observed and measured, doesn’t possess a “true” value, because the value depends on the method of the observation or the measurement.
4. Psychologie: Because humans are involved in work, you have to realise that cooperation is a deciding factor in a gathering. As well you have to take into account the fact that there are differences between people. This shows f.i. in the manner of learning (some learn by reading, some by pictures etc.). Especially important is the difference between intrinsic and extrinsic motivation. Intrinsic motivation is motivation out of respect for oneself and for others. Extrinsic motivation is the attempt to motivate by reward and punishment. The problem with extrinsic motivation is that it goes at the expense of intrinsic motivation, because reward and punishment just motivate to get the reward and to avoid punishment, with all consequences. The American educationist Alfie Kohn points in his articles out that this way grades and even praise can be considered harmful.

The System of Profound Knowledge is the result of the interaction of these four elements.

To grasp and to implement the System of Profound Knowledge, you don’t need to be a system analyst or a statistician or a philosopher or a psychologist. The System shows how you can judge an organisation and how it ideally looks like. From this the necessary steps and tools for transformation are coming forth. When the System of Profound Knowledge is learned on a voluntary base, the knowledge necessary for the transformation is readily learned.

At the same time the System of Profound Knowledge helps to establish or to comprehend the methods of quality management such as the ”7 basic quality tools (Q7)” to Six Sigma. Just with “fads” it is important to comprehend the requirements for their successful implementation as well as their strengths and weaknesses. Deming mentioned in this context that in an attempt to learn the “Japanese secret”, excursions has been made. During these they learned for instance that quality circles are common over there. The lack of a theory lead to quality circles being implemented at home without grasping the requirements or their meaning. When they didn’t deliver the expected results or lead into problems, the eyes were set on the current “Method of the Day”.

In a conventional company the principles of the four elements are followed more or less, even without knowledge of their interaction. This knowledge may be missing for an effective implementation in quality management, the base is there so far. In the IT world things are the same except for one thing: In the common IT company variation is not being regarded. As it turned out in a discussion, Deming has attributed a special importance to variation and confirmed the impression of the writer that 70 to 80 % of his lectures are based on it.

To make this clear: In the IT sector they do apply statistical methods. Internet usage statistics are just an example of many. The problem with this examples is that they are applied in external departments like marketing or that – when they are used internally – the existence of some common variation is ignored. An example is performance appraisal where the employees are ranked and the position or the performance are only ascribed to the individual. But Deming has found out that performance depends to 94 % to the system and to only 6 % to the individual. Statistical methods with the objective to optimise the IT company in a context of quality management seem to be applied sparsely.

I just can make assumptions about the reasons. One reason can be that the IT world regards itself with its programming work as a logically acting system where each departure has a special cause (a view which according to Deming leads to uncalculatable damage). A connected reason might be that the IT sector regards itself as a pioneer who has to invent everything which it needs themself. From other sectors the Not-Invented-Here syndrome is known, where group think lead to not looking at developments in other groups.

But there is hope. On the one hand in the IT sector too there are people who are grasping the importance of the knowledge of variation for programming. One is David J. Anderson, who applies the Theory of Constraints for the improvement of project management in software developement and who for this purpose is taking variation into account. In his blog he delivers examples for common ad special causes in the IT sector. These you can analyse with statictical methods.

On the other hand a small company doesn’t need an elaborate system such as Six Sigma, whose implementation comes along with big efforts and may thus lead to more damage than benefits. When you start to solve existing problems with simple methods like the seven basic quality tools while taking into account the System of Profound Knowledge, you can communicate the effectiveness of quality management and later implement more advanced methods like the new seven quality tools. This way you can work your way from the solution of local problem to the improvement of the whole system (where local problems should be regarded in the context of the system).

When somebody knows about examples of the potential or actual application of statistical methods in an IT company, I would gladly hear about it.

• ZDNet.be: “In our own small IT world we remain after all the years still in an US mentality from the fifties, regarding quality.” – The Belgian consultant Peter Hinssen is waiting for a W. Edwards Deming for the IT sector. (in Dutch)

Take a look at the unvarnished data!

The internet service Alexa compiles usage statistics for particular websites. Using the Alexa toolbar page views, reach and rank are estimated. There are discussions whether the estimates are correct, what should not be surprising because the results are determined by the method of estimation and can be accordingly optimised. On the other hand, each day you get new results.

At Alexa you can look at the trend of the usage statistic of the popular websites. For instance you can look how the usage of the website of Apple has developed over the time. When you look at the graph, you can see at the bottom right a control. When you move the mouse over it, the tooltip “Graph Smoothing” pops up. This control stands by default on the highest level. The curve you get to see at first is therefore strongly smoothed.

When you’re smoothing a curve, you expect to see trends more clearly. This is nice, but you should be trying to push the control to the left and see the original data. This way you can see how strong the data is fluctuating day by day. These fluctuations are showing up even without a special occurrence or a change in the mean value. In statistics they’re talking about two kinds of errors:

• Concluding that something special has occurred when it hasn’t.
• Concluding that nothing special has occurred when it has.

Because you usually can’t prevent both errors, you have to determine when you can talk about unusual values. To do this, you have to know how strong the values are usually fluctuating. Applied to the graph by Alexa, this means that you need the unvarnished curve to see, where there are days with unusual visits or whe the usage did really change.

When you thus apply statistical methods designed for original time series data on smoothed values, you will come to wrong conclusions.

Statistician bashing 1 – How to act on statistical fallacies

The problem with the application of statistics is that there may be fallacies. A concrete example came from the long-time minister-president of the state of Bavaria, Franz Josef Strauß:

“Two men are sitting in a tavern. One is eating a knuckle of pork, the other drinks two litres of beer. Statistically viewed that’s one litre of beer and half a knuckle for each – but one is overeaten, and the other is plastered.”

According to the manner how they react on such an example, people can be divided into two groups:

1. One group regards this example as a proof that statistics leads to wrong conclusions, is unnecessary and is only practised by out-of-touch crackpots.
2. The other group concludes from the example that the average is sometimes not able to describe everything, which means that you have to take this into account and therefore you have to try to further refine the conclusions won by statistics.

You may think yourself about which reaction will help us on the long term. By the way, the result can be transferred to other situations, where there are problems or open questions regarding approaches, methods and results.

How successful is “Bild am Sonntag”?

At Bild.de, the website of the big German tabloid “Bild”, the success story has been published that more and more readers are picking up the Sunday edition “Bild am Sonntag” at the newsstand. The story has been sparked by the recent “Media-Analyse” (media usage report), according to which the paper had 11,25 million readers, which were 640.000 readers more than the year before. The base has been the half-yearly survey of more tha 38.000 people about their reading customs at the request of the Arbeitsgemeinschaft Media-Analyse (Working Committee Media Analytics). The time of the survey was March to September 2007.

To this story there is a report at BILDblog.de, a website which keeps adding important details to the news published by “Bild” and its offsprings. In the report about the success story they mention the actually sold circulation of the “Bild am Sonntag” which is analysed and published by the IVW. Since 1998 the circulation went down by 750.000 copies, which is corresponding to a decrease by 30 percent. This clearly shows that the assumtion circulated by Bild.de that the readership is an indicator for the sales is obviously wrong. When a theory turns out to be wrong, it obviously should be changed.

That more and more readers are picking up the “Bild am Sonntag” at the newsstand, is thus not correct. Is at least the report correct that more and more readers are picking up the “Bild am Sonntag”?

The survey showed that the number of people who read “Bild am Sonntag” increased by 640.000 in a year. When you’re looking at the results of the surveys conducted over 5 years, it turns out that just since 1 year there were more and more readers of the “BamS”, because only in the last two surveys there has been ayearly increase. When you’re looking at the last 5 years, you would come to the conslusion that readership just remained the same. The 11,25 millionen readers of the last report were a peak but the value is not unusual. A trend couldn’t be determined too. The increase of the last year didn’t indicate one but can be explained by random movements.

With the available information you can assume that the readership of the “Bild am Sonntag” has been the same for the last years, the number of people who picked up the newspaper at the newsstand has decreased on the other hand. There only remains the question the news report at Bild.de has indicated something different.

Mapped crime

A further example for the power of the internet is the Misdaadkaart (Crime map). On this website the Dutch police reports are analysed and the cases are put on a map, using the places and streets retrieved in the reports by robots. The visitor can look for particular crimes and refine the results by place or postal code. Considering that the reporting is consistent at least in a particular city, you can get an overview about the safety in certain neighbourhoods. Via the mapped crime you also can look up the according report.

The success of the website encouraged the owners to use this principe on further projects. Besides a map for local neuws in the Netherlands a German version of the Misdaadkaart has been made. On the Krimikarte German police reports are analysed so that the visitor can look for cases in a particular area in Germany. At the time of the original post this map has been in the beta stadium so that it happened that crimes in Aachen where located at the police headquarters. The visitor can help improve the site by reading the according police report and reporting the right addresses. This is one of the examples on the web where you are not just a consumer but can actually participate in the making of a service.