How do I test for unequal variances in mixed model?I want to test if the measurement (intermediate) precision from two different labs are different. My data usually consist of the same sample being measured k times per day, across n days in each of the two labs. I analyze the data using Fit Model with lab as fixed effect, and Day as random effect. I would like to compare the total variation (within and between days) between the two labs - something similar to test for unequal variances in the Fit Y by X platform. Any ideas about how to do this in JMP?
As an example I use the '2 Factors Nested' sample data set. Treating Operators Sally and Willy as they were two different labs I wanted to compare, and treating Part as random, I get the following Fit Model output for each Operator:
Sally measures with a total variation standard deviation of 0.43, and Willy measures with a total standard deviation of 0.58. I would like to test, if these are statistically different?
Advanced Statistical ModelingBasic Data Analysis and ModelingQuality and Process EngineeringRe: How do I test for unequal variances in mixed model?I might be oversimplifying this, but would you get what you need by treating this as a measurement system analysis? Specifically, the Bias Comparison and Test-Retest Error options in the EMP personality tests if nuisance factors are measured differently. The picture below is from the 2 Factors Crossed data set, since you mentioned that each lab (Operator) is measuring the same part multiple times.
Re: How do I test for unequal variances in mixed model?just a quick add-on. the MSA is a variance only model... a mixed model except without the fixed effect.
Re: How do I test for unequal variances in mixed model?N_Sally = 12, N_Willy = 12, F(0.05, N_Sally-1, N_Willy-1) = F(0.05,11,11) = 0.355
U = SQRT(0.58)/SQRT(0.43) * SQRT(1/0.355) = 1.95 less than 2 (USP <1010>)， No significant different.
SQRT: Square Root
Refence: USP-NF 〈1010〉 Analytical Data—Interpretation and Treatment
