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; } } }

Variance Components confidence intervals differ, why?

martindemel

Hi,

if I have a data table with three columns:

measurement
mean shifted version of this measurement
grouping/level column like e.g. 3 operators

e.g.

When I do a variability analysis with only main effects model and look at the "within" variation I would expect same variance component and same confidence limits. But the analysis does show different numbers (attached the data used).

I wonder why? I looked in the documentation on variance components, I also thought about % of total might have influence but couldn't explain the difference. I think I miss something here.

Thanks,

Martin

VCA.jmp

Accepted answers

martindemel

Technical Support replied after some deep investigation:

Fit Model - REML is showing the appropriate results (specifically, the standard errors/confidence intervals) for this data (which does seem to be an edge case). The issue was related to a difference in coding surrounding the singularity criterion. The output for the two responses should indeed match, like they do in REML

The development team expects the issue in the Variability/Gauge platform to be resolved in JMP version 19

All comments

Dan_Obermiller

Your third level for Data 2 is a mean increase of 20 while the level 2 for Data 2 is an increase of 10. That changes the overall variability which explains the different results. Change the values of level 3, Data 2 to be 12.1, 12.2, and 12.3 and the results will match what you expect.

martindemel

Hi Dan,

if I got you right, if the mean change is different between the levels the confidence intervals within the level variance component will change?
I made some additional columns to the data. One reflecting your proposal, which creates the same results, one with different mean shifts but on a smaller level with the same mean, and one on a smaller level but with a overall mean shift but different ones for each level.

The analysis is for all data the same but the data 2. Again just looking at the Within variation.

When I follow your argumentation the shift in data 5 is then just too small to reflect any changes in the confidence limits.

VCA.jmp

martindemel

if I have 9.1, 9.2, 9.3, 11.7, 11.8, 11.9, 14.1, 14.2, 14.3 the confidence intervals are the same as data 1. If I have 8.1, 8.2, 8.3, 11.7, 11.8, 11.9, 15.1, 15.2, 15.3 the confidence intervals start changing for within. That would mean that there is a kind of threshold when the confidence limits start changing.

However I still do not understand completely why the confidence intervals change. the %of total is clear when it changes, but I thought the within variation stays the same and the confidence limits for that are based on the within variation only and not the overall variation when I check the main effects only.

martindemel

Technical Support replied after some deep investigation:

The development team expects the issue in the Variability/Gauge platform to be resolved in JMP version 19