DOE: Replicate Tolerance

Question

A customer has in addition to JMP also MODDE. There they have something called "replicate tolerance", which allows Modde to detect when runs are seen as replicates:

„Replicate tolerance - The replicate tolerance [0 .. 0.20] is the number used when MODDE decides whether experiments can be considered to be replicates or not. Default is 0.1, that is, 10% of half the range for each factor. “

Is there something similar in JMP? Can it be set?

I know that in JMP with JMP 10 the individual run replication has been added, but I have not heard yet "replicate tolerance" when talking about DOE in JMP. Thanks for enlighten me :)

martindemel · Accepted Answer

Hi Phil,

I've taken a deeper look into the Modde knowledge base and documentation now (as I first believed I just missed that wording and everbody else would know).

Modde seems to replicate only the whole design. In addition it has an analysis for replicates in an replicate plot:

The plot shows the variation in results for all experiments in order to provide a quick overview of raw data. The values of the responses (green and blue points) are plotted vs. experimental runs displaying the variation in the response for replicated experiments. The ideal outcome is that the variability of the repeated experiments is much less than the overall variability. Experiments deviating significantly from others should be checked. It is not recommended to exclude experiments. Failed experiments can be excluded from the investigation by using Exclude once verified with a repeated
experiment.
Replicated experiments: MODDE checks the rows of all the factors (both included and excluded) in the worksheet for replicates. Rows in the worksheet  with the same factor values plus or minus the replicate tolerance are considered replicates. The default Replicate tolerance is 0.1 (10%). You can change the Replicate tolerance in MODDE options on the File tab. MODDE shows replicated experiments as blue (default) points on the same replicate index.
 They use it to e.g. identify outliers when exxamining the Replicates plot (caused by e.g. incorrect data entries, incorrectly carried-out experiments, deviating result is correct and experiment does not fit the model).

This can be done easily with JMP as well by using the color coding and any of the exploratory tool we have.

So thanks Phil for asking back :)

PS: To be more concrete on the analysis:

ANOVA, Lack of Fit, and summary of fit are changed if replicates are detected

* Model validity might be low in very good models due too high sensititvity in the test or extremely good replicates

* Reproducability (variation of replicates compared to overall variability) will be checked as well

* ANOVA:

If there are replicated observations, the residual sum of squares is further partitioned into Pure Error and Lack of Fit. A goodness of fit test is performed by comparing the MS (mean square) lack of fit to the MS (mean square) pure error. * LACK OF FIT

SD-pe (Pure error): Shows the variation due to the replicated experiments (observations) adjusted for degrees of freedom and in the same units as Y. This is the square root of MS (mean square) pure error.
 
May this is of help for future questions about Modde and the replicate tolerance :)