Home › Forums › General Forums › Tools & Templates › Distribution-based Goal Statements
This topic contains 6 replies, has 4 voices, and was last updated by Chris Seider 5 days, 3 hours ago.
Hello All,
I currently need to develop a goal statement to essentially say that one company is “better” than another company at making low defect material. I have thousands of samples from one company and about 20 samples from the other. The mean of the data is not a strong value here because of how the data is distributed.
Better for them is all about the distribution of the defects. Its about the bulk of the data being lower.
Minitab’ distribution analysis tool seems to indicate that the company with lots of data follows a 3-parameter lognormal distribution. The other company also seems to fit, albeit not as well.
How can I make a S.M.A.R.T goal statement that will show the distribution has by and large lower specks at higher percentiles and doesn’t have any weird distributions like a bimodal distribution (Would be very bad) to through off the analysis?
Find a numerical measure of this “distribution” which I’m not sure I understand.
standard deviation and mean absolute deviation are measures to consider.
@chris,
Since the data isn’t normal, the mean doesn’t really seem to show much. Like Bimodal data with more data points to far left of the mean then the far right. I’m actually trying to compare different common grades from the companies. That’s why I really need to work out a way to accurately say one is better than the other. When the mean doesn’t seem to accurately describe the data much.
Thanks for the input.
I didn’t suggest the mean. See … https://www.khanacademy.org/math/cc-sixth-grade-math/cc-6th-data-statistics/cc-6-mad/v/mean-absolute-deviation
Are you benchmarking or just comparing the two companies? It makes a difference. If you’re benchmarking you either need more data, from more companies, or you need some independent evaluation that points to a particular company as a leader, suitable for use as a benchmark. I’d also insist that being better than a competitor should not be the goal. The goal should be to satisfy customer requirements (needs, specifications, and expectations) as fully as possible regardless of what competitors are doing.
Given that you are trying to use a 3 parameter lognormal, it appears that what you are saying is that the issue is more than a simple bean count of defects. In other words there are specific measurements connected a defect and/or defect types.
I’ve never used a 3 parameter lognormal but I have run analysis with 3 parameter Weibull for defects and in each case there were actual physical measurements associated with a defect.
For example fines (small particle diameters of the tail of the resin particle size distribution) are an issue with plastic resin powders. The distribution of the fines and the percentage of the fines between certain sizes can drastically impact resin performance. In the case of fines the “goals” statements were in terms of overall fines percent measures and percentages of distributions of ranges of fines sizes within the fines themselves. In other words there was no overarching single goal. Based on your post it sounds like this might be your situation.
@rbutler
I’ve seen that need for specific % within certain ranges for particle size distributions. Hopefully they know if tails or certain regions are important. I’m still a big believer that if one can’t quantify with 1 or 2 or … numerical metrics then it’s hard to quantify or call something a success.
Nicely done on your detailed ideas/answers.
© Copyright iSixSigma 2000-2017. User Agreement. Any reproduction or other use of content without the express written consent of iSixSigma is prohibited. More »