A student called about the proper test to evaluate the differences in yields between eight different products. They told me nothing works.

The yield was at an electrical interconnection where each product had a different number per item. The number of opportunities varied to a level that could not be considered as equal or even close.

The hypothesis test roadmap was followed and it leads to a chi-square test or analysis of means. But they did not work.

ANOM could not be used due to the varying number of opportunities.

Upon questioning, the problem ended up being that the yields were too good as in 100% on a couple of the products. You can not make attribute comparisons without having defects. Otherwise there is no information to compare, so where do you go?

First, the chi-square test for tables (in Minitab) or Chi-square test for independence (from statistics books) has a data requirement before it is works well. This is that every cell in the table needs to have a predicted value of 5 or more. This usually means that each cell in the table needs to have a value of 5 or more. If you do not have that what can you do?

The standard answer is to combine categories with small counts until you have a single category with more than 5 counts, sort of like what a Pareto charting program will do with the last 5% or so of choices. This is difficult if there is not physical meaning to the grouping. It is also does not work if all the data has zeros. The real risk in low table counts is that the Ch-Square test will under estimate the true p-value (declare significance before it should). If the counts are less than 5 and the p-value is high, there is no real worry.

What can you do? There is no clear hypothesis test method, but a few can be misused.

How about doing a whole series of two proportion tests? If there are eight categories, that means that there would be 28 paired comparisons. [n*(n-1)/2] You can do this but you need to set the confidence level of each test at 99.82% so that the overall confidence is 95%. (.9982^28 = .95) This is the same concept that multiple comparisons are performed as part of ANOVA.

One possible way to compare the groups is to generate a confidence interval for each product and comparing them graphically. It would show differences, but would not be a hypothesis test and it would probably be adequate to make the proper business decisions.

A student recently asked about transforming time-between-event data so it can be used in an individuals chart.

This is a great question.

The answer is to use a square root transformation. It turns out that the time-between-failures typically has a poisson distribution if it is stable. In this case, the square root is the correct transformation. You can check this with simulated time data using a poisson random number generator. You will find the square root to be the right transform. when the mean exceeds 40 to 50, it is still benefiical but not a requirement.

It is great that you are using the ImR chart in the non traditional uses, most would say use a C or U chart which will end up being truly useless for infrequent events.

Frustrations in Enterprise Improvement
Responses to Webinar Questions
Smarter Solutions, Inc.
www.SmarterSolutions.com
512-918-0280
Forrest W. Breyfogle III
forrest@SmarterSolutions.com

View questions and answers, and download a pdf copy of the webinar questions and answers by simply using the password provided in an e-mail sent after the webinar. If you do not have the password and registered for this event, please contact info@smartersolutions.com to obtain password.

Read the rest of this entry »

As I talk with students they routinely ask about e-learning. After the earlier Blog on this topic, I decided to look deeper into this trend.

I found a number of positive views on e-learning as the future, but the sources usually ended up as providers of the e-learning tools and consulting. I did find one set of information that I would like to share. The articles came from Chief Learning Officer magazine. (www.clomedia.com).

The discussions over the e-learning topic were more targeted at an understanding of its benefits and drawbacks than a sales pitch. Probably my favorite was “Technologies to Support Leadership Development” by Kerry Patterson. In this article four rules of training success, of which the third is titled “Know When to Forgo the Flash … or Not”. In this there is a discussion that when courses are targeting behavior change or to break away from traditional methods, that instructional methods that emphasize the social connection and allow for direct feedback are usually the best. Why would that be true?

Think about it. To challenge your personal behavior or methods, let along going after the business behavior or methods is tough. It is always out of the students comfort zone. Think about an attempt to motivate that level of change with a 1-way e-learning presentation. I know that a student in this situation needs re-assurement and an ability to ask questions specific to their business or the frustration will be huge. I find that my students drive the training classes off the agenda to talk about how the ideas will be applicable in xxxxx. Where xxxxx is their business, organization, workgroup or what ever. These side discussions are what takes the Lean Six Sigma learning from the head to the heart.

Successful Lean Six Sigma students are the ones that find a connection between the tools and skills to their past experience. I can see the light come on when the connection is made. I hear things like “I wish I would have known that before” and “Now I know why xx worked and yy didn’t”. These revelations come after the side discussions, where the topic is made real. The tough part is that the same discussions do not reach everyone. Every student is different and needs different social support to allow the significant change to occur.

The last reason e-learning alone will always be tough for the Good Lean Six Sigma programs, the lack of experiential learning opportunities. Many of us use catapults in the learning, why? It is because they generate data with noise effects and clear predictive X’s. A simulation could do this too, but a simulation will not allow the mistakes, and people to people differences that the student will see as the catapult is used. Now they know the context of the data, and when the tools are applied during analysis the student gets to see how the tools represent changes and mistakes real time. If this was a MasterCard Ad, this would be the “Priceless” take away.

Can e-learning be part of a blended learning to support the classroom work, of course it can. But, where do you make the cut? I have a peer in a large company using Lean Six Sigma with employees all around the world. The decided on e-learning to supplement the classroom work and bought a set of Black Belt e-learning courses from a very successful consultancy. Started using them between training weeks in an attempt to reduce the classroom time. After a year, it was found that the e-learning increased training costs and was not allowing the classroom time to be reduced due to the need to question and discuss the topics. After the analysis, they stopped the blended black belt learning and have decided to only use e-learning when it is to pass information and tools, but not to change behavior.

If you have experiences or thoughts on this, please add a comment to this blog.

Many individuals consider these as two different sets of tools, or two different training courses.

Why is that, because it sells better or even sounds better? Who knows and what are the real differences?

Differences in the way a Process is viewed:
In manufacturing it is easy to see the work as a process. Building a process flow chart or a Value Stream map is straight forward. In the transactional areas, it is not so clear. Why? Read the rest of this entry »

If you evaluate enough process improvement consultancies, you will find there is no standard.

Get your Black Belt in a week! … via Web based classes at your own pace!

Get your certification through an extensive 4 week (or 5 week) program.

Lean Methods provide quicker improvements than Six Sigma.

Design of Experiments (also known as MVT) is all you need and provides greater business gains.

Six Sigma can be used to solve all business improvement needs!

Certify as a Black Belt by taking our test and attesting to skill proficiency!

Lean Six Sigma is the best of both systems! It is the new generation improvement system.

You need transactional (Lean) Six Sigma! You need Manufacturing (Lean) Six Sigma
…Lean Six Sigma in Healthcare!

Success requires Black Belts, Master Black Belts, Green Belts, champions, … and on and on…

Wow, how can you figure what is best for your company (or for yourself)? Read the rest of this entry »

Most all books say start with 95% confidence unless you know otherwise. When is otherwise? I remember reading examples of the medical and nuclear fields, well I was in the Nuclear Power field and we used 95% for everything except safety analysis, whish was at 99%.

So what is the right thing for the improvement practitioner?

Lets start with what a typical person would use in their qualitative judgement. Read the rest of this entry »

This is always a good topic to examine: E-learning or Classroom instruction.

Why not have both? What are the pros/cons?

It was paraphrased from a student comments, “…I’m a very disciplined person, self-starter I might add. Why should I take the time to book travel, hotel, and miscellaneous expenses when I can participate in threaded discussion, self-pace study, less time — greater perceived value (intrinsic), and not to mention I save on plane ticket, hotel, and expenses… why not elearn?” Read the rest of this entry »