I published this a few years back in our local ASQ newsletter. Actually I have since found out the "Lies ****ed Lies and Statistics" quote is attributable to Disraeli, not Mark Twain (though he used it with attribution to its source).
late entry
Interesting - I see the editor edited me automatically. So I will continue the story - I did write (and officially cleared by the US Government) a paper called Lies ****ed Lies and Statistics" this year. They held an authors breakfast, and since I made the database of the authors and papers, I made sure the MC read off that title. . .
__________________
Steve Prevette
"A Passionate Statistician", ASQ CQE, Fluor Government Group
The opinion stated above does not necessarily reflect that of my employer.
Dear Steve: Excellent article. You have highlighted three problems very nicely. Now, I have a question about the fourth. If we take any 25 numbers, such as what you have, can we necessarily say that they follow the "normal" distribution. Quite coincidentally, I was looking at an exactly similar problem and I came to the conclusion that the numbers do not represent a normal distribution. I did not "randomize" them using a random number generator, but did do some other types of "randomizing." Again, thanks for a lovely and thought provoking article.
Steve,
Great little article. It is very easy to understand and gets across a very big picture of the problems with charting and the way data is projected.
I must admit that I have abused graphs to get a point across or to win an argument. My problem has always been that people usually see the negative of the graph, as your 'LIE' graphs have shown, and resort to the blame game, as stated.
It is sooo easy to increase or decrease the data (add a Good month/period or drop off a bad month/period etc) to suit the point you wish to make. I find this especially true when charting 'Lag' Indicators such as LTIs or Corrective Actions. If I want to push a point, good ro bad, I can paint a suitable picture by showing a graph with 'Selected' data that shows a marked increase/decrease over a given period. It has come in handy from time to time. The worrying thing is that no-one has picked it up....ever.
We us the 'Control Chart' for tracking our Product Specification Results with a UCL and LCL set. We then only report on those that breach these limits.
Greg B
__________________ - Greg B - 'Vision without action is a daydream. Action without vision is a nightmare'
Last edited by Greg B; 29th July 2004 at 06:25 PM.
Dear Steve: Excellent article. You have highlighted three problems very nicely. Now, I have a question about the fourth. If we take any 25 numbers, such as what you have, can we necessarily say that they follow the "normal" distribution. Quite coincidentally, I was looking at an exactly similar problem and I came to the conclusion that the numbers do not represent a normal distribution. I did not "randomize" them using a random number generator, but did do some other types of "randomizing." Again, thanks for a lovely and thought provoking article.
Charmed
You are very correct - they are not normal. I simply used a uniform distribution. Anticipating the next question - Dr. Shewhart tested control charting with not only the normal distribution, but also a uniform distribution, and also a triangular distribution. Control charting is very robust to non-normal distributions, and does not (versus how most people teach SPC) rely on the normal distribution. The true basis, as stated by Dr. Shewhart is the Tchebychev Inequality. Perhaps that discussion would make a good follow on article, though I must admit Dr. Wheeler has done a lot of work already in that area.
OOPS, rewind, rewind. As I looked at the file again, I did shift the uniform data to a normal distribution. If you double click on the charts, you can find the underlying excel file (which is why the file is somewhat large, lately I have been trying to be sure that only the chart object comes into the word file). You can see my random number generator routine.
__________________
Steve Prevette
"A Passionate Statistician", ASQ CQE, Fluor Government Group
The opinion stated above does not necessarily reflect that of my employer.
Last edited by Steve Prevette; 29th July 2004 at 02:55 PM.
I like it, Steve. Good job. Do we get to read "Lies ****ed Lies and Statistics"?
I gotta pace myself - can't give away all my secrets at once
__________________
Steve Prevette
"A Passionate Statistician", ASQ CQE, Fluor Government Group
The opinion stated above does not necessarily reflect that of my employer.
Steve, whether you win $150 or not. I LOVE your articles!!
I think the best part of this article is the very end. where after several conflicting interpretations of the data - you reveal that the data has no information at all.
I just finished reading your May 2004 article, "Tyrrany of Targets" - An argument against Management by Objective. Another lovely article. I notice that you have used the same control chart example in both articles. Do you have the actual values that were used to prepare the control chart? If it is not a bother, can you post the numbers?
***********
Never mind Steve, I read your response to another question and was able to get the number by double clicking on the Figures. I rounded off each value to two-decimal places and calculated the average and the standard deviations. I was also able to get the same least squares regression equation that you got, in spite of the rounding-off two decimal places.
Charmed
Last edited by Charmed; 31st July 2004 at 10:02 AM.
Reason: Add notes to Steve
Liars Figure and Figures Lie
OOPS, rewind, rewind. As I looked at the file again, I did shift the uniform data to a normal distribution. If you double click on the charts, you can find the underlying excel file (which is why the file is somewhat large, lately I have been trying to be sure that only the chart object comes into the word file). You can see my random number generator routine.
Linear Mode
