Non-Normal Data F test and T test

[bobdoering]

I certainly wanted you to be comfortable with the idea that what you are testing is truly "good enough" to prove bad versus good in your product. For ours, as an example, it simply is not adequate.

But...it still poses the amount of error in the tensile testing process and whether the sources of error not related to the actual failure mode are statistically insignificant an not masking the condition you are trying to make a decision on.

If it is black and white good versus bad for your product than you may be fine....and blessed...

But..is it true?

 

[Miner]

Hi Bev,

Thanks for your response and guidance.....attached is my excel file. The first page is the data as taken, the second page has the one outlier data point I removed and replaced with a Median value data point 32.00. The process I am examining has very limited data history, settings etc.

Thanks again...

Regards,
Marty

Marty,

You forgot to attach the data file.

 

[optomist1]

Hi Bev Take II,

Thanks for your response and guidance.....attached is my excel file. The first page is the data as taken, the second page has the one outlier data point I removed and replaced with a Median value data point 32.00. The process I am examining has very limited data history, settings etc.

Thanks again...

Regards,
Marty

 

[Bev D]

sorry for the delay - I was out of the office last week.

I have attached my analysis.
first I always just plot the data and look at it. (best statistical analysis tool you have)

I ran a simple t-test (assuming unequal variances - JMP will automatically assess and perfrom the correct test). I also plotted the confidence intervals on the means (they are the two green diamond shapes wihtthe points of ht ediamonds representing the confidence limits.) I also performed the analysis with and without the extreme value. When you look at the charts you will see that this is much ado about nothing. The statistical analysis isn't really helpful at all. the plots of the data however, are very telling.

The automatic crimping tool has a lower mean, BUT the spread of the individual data points is so large that that very small difference in the mean hardly matters at all.

 

[optomist1]

Hi Bev,

No hurry, thanks for your assistance.

This site is fantastic...and thank you for your consistent input and guidance.

Thank you for the analysis...well put much ado about nothing......I agree with your assessment....and especially as a first step simply plotting the data.

I guess I could re-run the test with a larger sample size, and tighter confidence intervals, but I don't have the extra time. For me current purposes this may be sufficient

Regards,
Marty

 

[optomist1]

Hi Bev,

I was sitting here thinking about my last response.....I don't think it made sense. If I chose to re-run with larger samples....this would or should result in narrower CIs and I would be more likely to detect differences, which is not a concern here. If I ran the test for mean differences and had not detected a difference then it might make sense.

Make sense??

Regards,
Marty

 

[Miner]

I think the key issue here is not the statistics, but what you will do with the results.

Let me summarize a few key points from the previous posts as well as my own analysis:

• There is a statistical difference between both sets of data regardless of the test used (e.g., t-test, 1-way ANOVA, Mann-Whitney, Mood's Median, etc.) or whether the outlier is included or not.
• Is this difference of any practical importance? That is, will it make you reject the automatic process?
• Despite the fact that the manual process has a higher median, the outlier may be an indicator that the manual process is susceptible to sporadic crimping defects.
• Despite the fact that the automatic process has a lower median, you may be able to make adjustments to increase this median.
• The outlier may be an indication of testing issues, such as slippage.

I think you need to focus on the cause of the outlier. If it is a product of the test, address the test. If it is a product of the manual process, it may be additional support for the automated process. You may also want to consider a DOE on the automated process to see whether the median of this process may be increased.

 

[optomist1]

Hi Miner,

Thank you for the thorough and thoughtful response;

During the interim I looked into the "testing procedure", your last bullet is right on. The testing device is susceptable to slippage due to improper loading of the sample.

This has been addressed by a visual and written clarification to the test procedure regarding proper loading and orientation of the test specimen.

Regarding DOE, I am considering this for a future effort.

Thanks again...

Regards,
Marty

 

[Bev D]

Optomist1: To address your question concerning large sample sizes and tighter confidence intervals and to expand on what Miner said -

• Yes the confidence intervals will be tighter.
• Yes in general larger sample sizes and therefore tighter confidence intervals will 'detect' smaller differences
• BUT every test you could possibly run (and we must remember that just because we can doesn't mean we should) already indicates a small difference exists. so why run more samples?
• The automatic process has a lower 'average' pull strength with or without the extreme value. given teh nature of pull strength my assumption is that you are looking for a process with a HIGHER pull strength than the current manual process? If so no amount of re-sampling will change the fact that the automatic process is worse (not much but still...)
• The mean is not of any real help here. The real clue to improving this process lies in the variation of the individual values.

 

[optomist1]

Hi Bev,

Thanks for the follow up...I agree with you and Miner......for this particular situation, investigating the process variation further, will not be cost effective. With a specification tensile strength minimum of 13lbs, and in both cases the process means are +2X spec and both mins are approx 2x spec, doing so will not be worth while.

Thank you.....

Regards,
Marty