topic Re: What exactly is "Location relationship" in reliability tool? (fit life by X) in Discussions

What exactly is "Location relationship" in reliability tool? (fit life by X)

LikelihoodPanda — Sat, 10 Jun 2023 23:57:34 GMT

Hello,

I'm using JMPpro for reliability data analysis.

I understand that "fit life by X" tool can be used for ALT data analysis and the "relationship" seems to show an accelerated test condition-lifetime relationship(Arrhenius, Linear etc.).

What is "Location" and "no effect"?

JMP help document says: Location means that μ is different for every level of X

Does this mean that it is used when there is an accelerated condition-life relationship (ALT data) but the specific relationship is unknown, or does it mean that it can be used even if the accelerated condition-life relationship is not existed? (not ALT data)

And for "no effect", I cant find any clue.

Thank you.

Re: What exactly is "Location relationship" in reliability tool? (fit life by X)

ih — Mon, 12 Dec 2022 22:19:56 GMT

From my limited testing just now, no effect appears to give the same coefficients as the same as not entering the x variable at all. I think using Location would be similar to using a 'by' column in most other platforms, but I am not sure how that relates to ALT data.

Re: What exactly is "Location relationship" in reliability tool? (fit life by X)

peng_liu — Tue, 13 Dec 2022 03:26:36 GMT

"No Effect", "Location", and "Location and Scale" are relationships between Life and X when X is discrete. While the remaining are relationships between Life and X when X is continuous.

If X is discrete, e.g. suppliers, model types, etc., three relationships fit models under three different hypotheses: (1) X has no effect on lives (2) X has effect on the location parameter of the distribution (3) X has effect on both location and scale parameters of the distribution. Usually, one should fit all three and select the most appropriate one.

If X is continuous, e.g. Arrhenius, the three "No Effect", "Location" and "Location and Scale" still play an important role in model checking, i.e. checking whether, e.g. Arrhenius, is an appropriate. If, e.g. Arrhenius is appropriate, then a test between "No Effect" and "Regression" should reject "No Effect", but a test between "Regression" and "Location" should NOT reject "Regression".

Since you are checking out documentation, please take a look at materials related to "Nested Model Tests" and see whether they make sense to you.

Nonetheless, the above are only necessary steps in model checking and selection. You should look at other outputs in the report. It is a fairly complex task to check model in this context, and Fit Life by X report has substantial amount of outputs, which you may want to get familiar with.