Hi everyone
I've got a delicius problem with a testmachine in our process.
We have four almost identical product tested in this machine and now I'm to check if these four is different from eachover.

I've checked the distribution of my reject (rework database per week) in PPM or dpmo. The correlation distriution for lognormal shows 0,992 and is the higest number of all distribution test.

How and wich method do I use to control if there is a difference between the products?
I've tried one sample T-test and no product stays around zero in CI 95 % and that should mean theres a difference or? I looked in Statguide and read to my big suprice "Only for normal distributed data" buhuu :mg:.

Help me please. What should I do?

Hi everyone
I've got a delicius problem with a testmachine in our process.
We have four almost identical product tested in this machine and now I'm to check if these four is different from eachover.

I've checked the distribution of my reject (rework database per week) in PPM or dpmo. The correlation distriution for lognormal shows 0,992 and is the higest number of all distribution test.

How and wich method do I use to control if there is a difference between the products?
I've tried one sample T-test and no product stays around zero in CI 95 % and that should mean theres a difference or? I looked in Statguide and read to my big suprice "Only for normal distributed data" buhuu :mg:.

Help me please. What should I do?
You mention ppm, so I assume your data are go-no go data. This would imply a p-chart (http://www.hanford.gov/safety/vpp/pchart.htm), where we can plot the percent defective for each of the 4 machines, and hopefully also trend versus time for each machine using p-chart.

However, as discovered in another discover of bent poles, it would be helpful to have the actual measurements, so that you could then run x-bar r, and indeed, t-tests would be much more meaningful.

Do you have any raw data you could share and let us play with?

You mention ppm, so I assume your data are go-no go data. This would imply a p-chart (http://www.hanford.gov/safety/vpp/pchart.htm), where we can plot the percent defective for each of the 4 machines, and hopefully also trend versus time for each machine using p-chart.

However, as discovered in another discover of bent poles, it would be helpful to have the actual measurements, so that you could then run x-bar r, and indeed, t-tests would be much more meaningful.

Do you have any raw data you could share and let us play with?

Hi Thanks for fast response.
I see what you meen and here is a file from the measurement instrument

In this file you can see i column A the results and in column B is the product seperated with a number.

:thanx:

Hi Thanks for fast response.
I see what you meen and here is a file from the measurement instrument

In this file you can see i column A the results and in column B is the product seperated with a number.

:thanx:

What does the number in Column A represent? Is that an actual dimension, or is it related to number of failures? If it is related to the number of failures, what was the number of trials?

A simple ANOVA gives that the 4 sets of numbers are different at a 2% significance level. But it would help to know more about the data, and if they are in time sequence.

What does the number in Column A represent? Is that an actual dimension, or is it related to number of failures? If it is related to the number of failures, what was the number of trials?

Hi again
It is the raw data in column A. It is a leakage test who measure flow mm^3/s. Our upper limit is 40 mm^3/s.
Every row in this xls book is one measurement from one unit ie 31 units from each product type (4 types).

Hope this would clear something out for you.
:bonk:

Before Steve has all the fun ;) , here are a few more observations. I popped open Minitab and found:

1) my ANOVA results were identical to Steve's (not big surprise there!)
2) Using I-MR charts shows that all machines are in control EXCEPT for the range chart for Machine 1
3) An analys of the data shows that not only are the means different, but the variations are also different (at greater than 99% confidence).

I tried putting the graphs into Excel so I could attach the file. We'll see if that works.


Tim F

Before Steve has all the fun ;) , here are a few more observations. I popped open Minitab and found:

1) my ANOVA results were identical to Steve's (not big surprise there!)
2) Using I-MR charts shows that all machines are in control EXCEPT for the range chart for Machine 1
3) An analys of the data shows that not only are the means different, but the variations are also different (at greater than 99% confidence).

I tried putting the graphs into Excel so I could attach the file. We'll see if that works.


Tim F

Now we talking.
The data is in time serie for each product (you call it machine 1 - 4) but they not taken after eachover ie machine 1 at day 1 machine 2 at day 4 etc.

Two questions:
1: P-level is approx. 2 %. That would mean there is no difference between the products/machine?
2: Dosent ANOVA analyse variance and not avearge?

:applause: for you both!

Now we talking.
The data is in time serie for each product (you call it machine 1 - 4) but they not taken after eachover ie machine 1 at day 1 machine 2 at day 4 etc.

Two questions:
1: P-level is approx. 2 %. That would mean there is no difference between the products/machine?
2: Dosent ANOVA analyse variance and not avearge?

:applause: for you both!
2% alpha would generally mean I reject the null hypothesis that the four machines are the same. Usually a 5% threshold is used in industry.

But more important is to look at the control charts. Machine 1 is not in control. And looking at the Upper Control Limits for each machine, it does appear that machines 1 and 2 are NOT capable of meeting the 40 specification, but machines 3 and 4 are capable.