I am intrested in what statistical methods you would use to compare same items on different tools and sites? Purpose of the study is to define a specification limit. Repeated measurements are not planned as the repeatability of those tools are quite good (Showed by GRR). Still there are deviations within this tool group related to drift and calibration standards.
I was thinking to compare them by "Matched Pairs" with each other. Is there a possibility to include all tools as well as the Grand Average, USL, LSL in one Matched Pair chart? I have seen it before but I am not sure how to get there using JMP.
Thank you for you feedback.
I suggest reading Help > Books > Quality and Productivity chapters for Measurement System Analysis and Variability Chart.
My initial thought was to compare the means by ANOVA combined with Tukey HSD post-hoc test. To verify the results i wanted to use Box-Whisker plot if several tools match with each other with an given specification limtit. It turned out, the residuals of each group (n=24; tools=10) are non normal distributed. I have read in the literature that for one-way anova the normal distribution for the residuals of a sample does not necessarly have to be fulfilled, if other assumptions (indenpendency, homogenity of variances) are not violated. Tests that variances are equal are passed by all provided tests in JMP. This was expected as same items were measured at different but same tools.
Graphical anaylsis show that there is a (Mean/Median) shift between the groups, while the data points show evidence of symetrie.
I would actually like to stick to the paramteric test approach and not use non-parametric tests because it is easier to understand third parties.
My alternative would look as the following:
- Nonparametric Comparisons For Each Pair Using Wilcoxon Method
I already concudted the alternative. Using this approach does not really fit my graphical analysis, instead it tells me that most pairs a significant different.
Additional I already tried to tranform my data to log values, but this does not change anything as the groups itself show evidence of within 3 data clusters.
Thanky you for your feedback
You could also use the bootstrap feature in JMP Pro to compare your sample statistic to the empirical sampling statistic.
Have a look here:
In addition - This is not an exact answer to your question but perhaps worth looking at:
Also worth reading the discussion.
There are no labels assigned to this post.