This message is available on the new blog from this link
I had decided to return to the fundamentals of the LSS but I am definitely passionate about the struggle being waged " Indicators Famous "and" Indicators of the shadows. " On side we have the couch potatoes and illusionists Average, Min and Max and the other we have the needy and robust median, quartiles, standard deviation. So far it seems very difficult to decide ... Then, the ring, two laws of distributions come do the show ...! ;-)
To my right, built like an ox, down from its 20-80 we have: Mr. Pareto ...
my left lying like a log, precision random: Mr. Benford ...
DDDDDDDRRRRRRRRrrrrrrrrrrriiiiiiiinnnnnnnnnnngggggggggg!
Well ... then I think you're conditioned enough to get into the thick of it. It was a few years, I was required to establish a computing cost of logistics products stored on a platform. So I had masses of apportionment for each item based on key distribution (also called work units). You agree, for logistics such as storage or shipping it was appropriate to take into consideration the volume of parcels. Also most of our calculation reallocated charges based on volume (volume stored, the volume received, shipped volume ... etc). So I made a pretty matrix in excel, with beautiful macro to update the zany and after a few days of work I before my eyes production costs by product ... ... Oh miracle!
Then came the time of the analysis ... And there: problem! I found myself with items that generated staggering costs ... A little research, I finally realize that no measures were inconsistent products ... :-( I called
the logistics manager and asked him to check the products that I had identified ... He returned with the good things ... So I redid the crunching and such ... zinzin articles dating back to surface with errors ...!
The problem is that it was impossible to ask to remeasure all 6000 products stored on platform! So I had me a nice dilemma: how to be sure of the reliability of my data without having to measure all of my products and redo endless iterations on taking new measures .. .!?
My first instinct was to turn to the Pareto law. As everyone knows Pareto highlighted a distribution law that is "arch-used" in business. This law says for example that 80% of my sales are generated by 20% of my clients. I decided to leave the 20% of my items that generate 80% of my volume. Also I leave the 80% of my items in stock representing 20% of my value Global stock. So if I meet my two files I end up with items that have the highest volumes with the lowest values so I can identify aberrations! In one detail that I almost had assumed (somewhat debatable ;-) my bulkiest items were necessarily the most expensive. And actually I could identify new aberrations ...
So: I relaunched the zany ... and always the same ... He left with errors Articles ... But I had another problem: the errors were smaller in magnitude because I corrected the 20% of my article that generated 80% of my erroneous volumes (Pareto is sacred!;-P).
After a few days the light brightened. If the law of Pareto distribution was limited by my problem: when I was going to use another distribution law, the latter much less known, but much more efficient for the problem I had to solve: Benford's law .
For those who do not know: Benford's law refers to a random observation of the phenomenon that we reserve nature. In any digital distribution (the list price of a supermarket, the list size of all trees in the territory, the list of net wages inhabitants of a country ... etc.) if we take the first digit (different from 0) and we account for these figures: while we have consistently said distribution:
not strange ...!? Whatever you take as distribution, if you count the number of 1 in first position, 2 in first place ... etc, you will always get this%. Even stranger is that there are more than 1 2 of 2 than 3 ... etc.. But it's like this ... To the extent that the distribution is random it would look like this.
I therefore decided to compare my distribution with that of Benford and here's what I got:
The result was obvious! I had too much of 5 and not enough three. So I made an extraction of all my items with a 5 in the first volume of my character and I review the list. I quickly see that all items of a family (those shoes) were the same size. The ingratitude of the task had led the people in charge of measures to copy-paste. They are justified by the fact that all the shoe boxes were the same size. In a proportion, perhaps, but you can not take the Benford's law, by default, so ...! The shoeboxes are not the same size when it comes to dress shoes, fishing boots or booties for children ... etc..
After measuring each of the references to the family shoe found in the warehouse, I redid an analysis of my new layout and here is what it gave:
This time I seemed reliable file ... I remoulins the zany and obtained success with the calculation of my costs back as expected ... ;-)
For leisure I LSS projects in a post to come, you talk about the tool that tests the reliability of data for registration thereof (GR & R) but often project starts with data already now registered and in such cases it is necessary to verify consistency of the data before providing any analysis ...!
Bravo to you Mr. Benford ... You won this battle hands down! ;-P
0 comments:
Post a Comment