nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Difference in shelf life estimates when row-excluding data versus using local data filter?

Ressel

I am working with JMP 14 and am relatively heavy user of the JMP stability module. I found behavior which I don't understand. For one test I found that the shelf-life estimate calculated by JMP varies with the method used for excluding the outlier. The methods used are:

A) Excluding the outlier via row-exclusion and performing the stability analysis.

Excluding the outlier directly in the stability module using a local data filter.

Is this a user error? If so, how can I avoid it?

Accepted answers

Dan_Obermiller

I think the confusion is how shelf life is defined. I believe all three curves are establishing your shelf life, not just the red curve. A quick way to check is to scroll down in your report and look at the DF listed under the model summary. When you exclude one row by using the row exclusion, your DF will likely be 26 for the Simple Linear Path. When you exclude three rows by using the local data filter, you will see the DF will be 24 for the Simple Linear Path.

JMP is using ALL of the data to estimate the degradation curve. You use more data which provides better estimates.

So, the result does NOT depend exclusively on the behavior of the red curve.

All comments

Dan_Obermiller

I think your filter actually removed two observations. Looking at your green curve on the first graph, there are four points after the peak. With the local filter graph I only see three. Plus, your filter is set at 32, that last green point looks to be around 35 or so. I'm not sure you have the same sets of data for these analyses.

Ressel

Dan, I am not sure I understand you correctly. The data is spread out over the following time points in months: 0, 1, 2, 3, 6, 9, 12, 18, 24, 30, 36. So the only data removed via the local data filter is for 36 months (which corresponds with your observation of 3 and 4 points after the peak in the grean curve). The underlying data set is the same, but as I explained above the result is dependent on how the "outlier" or last datum at 36 months is excluded.

the local data filter sumarily excludes all 36 month data (3 points). Row exclusion only excludes the 36 month result for the red curve. Since the red curve is also defining the shelf-life, I do not understand why the different methods of data removal should give different results. The result should exclusively depend on the behaviour of the red curve, correct?

Dan_Obermiller

JMP is using ALL of the data to estimate the degradation curve. You use more data which provides better estimates.

So, the result does NOT depend exclusively on the behavior of the red curve.

Mark_Bailey

The default settings for the modes is also different between the two types of filters. Only the Select mode is enabled by default in the Data Filter. The Include mode is also enabled by default in the Local Data Filter. If you did not enable the Include mode in the Data Filter, then the platform will not recalculate results.

Ressel

Now I am entirely confused... Since I already accepted Dan's solution (Thank you very much!), I am not demanding further explanation. But if you could spare a few minutes to explain to me what you mean, I'd be even more in love with JMP.

Mark_Bailey

I assume that by "you" you mean me! If so, read on. If not, have a great weekend!

To see what I mean, simply launch the Degradation platform with the Stability Test. Add a Local Data Filter eith your filter column. Now go to the Rows menu and select Data Filter with the same filter column. Now compare the active modes. They are not the same by default.

You might know this or you might have over-looked it. I just wanted you to be aware because the behavior of the platform will not be the same if different modes are in effect.