Serrogate data

[vnigam]

I have two sets of failure data of unequal samle size. The two sets of parts tested at different times, are very similar in design and were subjected to the same test. I want to combine them in my analysis so my sample size is high. I tend to think I should do a hypothesis test to check if the samples have similar mean and variance. Does weibull++ have that module in it? Or do you have any advise for this analysis?
Thanks

[Arai.M]

Hello vnigam,
You can analyze the 2 data sets separetly and then do hypothesis testing on the whole population via contour plots. For more information on contour plots see http://www.weibull.com/hotwire/issue19/relbasics19.htm. If there is a particular measure of interest you can use that for your test. You can just use the bounds around the particular measure that you obtain from the QCP.
Regards,
Arai

[vnigam]

Thanks.

Is this analysis option not available in the version 6? When I go to QCP, i have just the fisher matrix method available. Please advise.

VN

[Arai.M]

It is available in 6, both likelihood ratio and fisher matrix are available. The underlying assumption in the contour plots is the likelyhood ratio but you use 2 edgrees of freedom since you are comparing both parameters (as opposed to 1 metric in the QCP). In the QCP the method will be tied to the bounds method you have chosen to analyze your data.
Hope this helps.
Regards,
Arai

[vnigam]

Thanks.

I just did an analysis of two hypothetical data sets and I got the multiple plots as attached. I want to know if this tells me that the two datasets are similar so I could pool the data set together ?

I could not set analysis for likelihood ratio method. I had just the Fisher Matrix method available.

regards.

[Arai.M]

The conclusion is that at a 90% CL you cannot tell the difference between these data sets. Hence at that confidence level (you have 10% chance of being wrong) and backed up by your engineering knowledge that you expect those units to be the same (same design, same test) then you could pool them together.
You should be able to obtain likelihood ratio bounds by changing the bounds from the "set analysis" tab on the control panel and changing the confidence interval method to Use Likelihood ratio.

[vnigam]

I got it!

To summarize, if there is an over lap in the data and from engineering perspective it is similar, it can be pooled in for futher analysis?

Do I have to look for high degree of over lap or any degree of overlap with good engineering judgement, we can decide to pool the data or not?

Thanks a lot!

[Arai.M]

Any degree of overlap gives you that conclusion (at that confidence level). If you increase the confidence level, the degree of overlap will decrease. Conceivably you could increase the confidence level until you find no overlap. That threshold would be the critical CL at which you can say they are statitically different.
Engineering judgement always has to be used along with statistical tools. If you have reasons to believe something is different, then you could expore that.

[vnigam]

Thanks a lot.