What inspired this wish list request?
The Fit Y by X platform cannot analyze data from a Randomized Incomplete Blocks Design (RICBD), i.e. if the blocks are unbalanced, i.e. 1 plot (or more) is missing in 1 block (or more), and therefore in JMP 18, the Fit Y by X automatically invoked SLS in Fit Model platform, and that is what users needed!
Unfortunately, this very important and very useful feature was removed in JMP 19,and JMP invokes a mixed model in Fit Model platform instead, in order to allow for a valid test for Unequal Variances (see Detailed rationale section).
What is the improvement you would like to see?
Simply return the way this issue was handled in JMP 18. Or as an alternative (less-preferred) option, allow an option (e.g. in Oneway Preferences Menu) for JMP to handle the Randomized Incomplete Blocks Design case with Fit Model - Fit SLS (the same way it is handled in JMP 18; i.e. with no Unequal Variances test available).
Why is this idea important?
Many researchers using Fit Oneway are not familiar with Mixed Modeling approaches and may not understand that JMP is treating block as a random factor (not a fixed factor), with an unequal variance structure (repeated on the X factor). Researchers, Scientits, and Engineers using Fit Oneway often prefer and are most familiar with the simplest (least technically or statistically complex) model, and the equal variance assumption can often be appropriate for their experimental situation.
Allowing a default work-flow to Fit the SLS regression model for RICBD in Fit Oneway gives these users the most flexibility without leaving the platform, since now as of JMP 19, to fit the arguably simpler model, they have to go to Analyze > Fit Model and specify the SLS regression model there. This is particularly true for JMP standard users who don't have options to generate LS Means tests in the Fit Mixed report outline produced by Fit Oneway (since Fit Mixed is a JMP Pro only feature).
Detailed rationale for why is this idea important?
In Quantitative Genetics, mixed models are used to analyze data from pedigree populations. In this type of research, the main objective is to estimate the variance components between families of progeny obtained by matings between many random fathers (sires) and mothers (dams), and use these variance components to calculate estimates of heritability. These heritability estimates are the sole objective of this type of research, whereas the fixed effects in the model (e.g. hatch when the studied population consist of several hatches, or rearing house when there are several of them) are added only to remove these non-genetic effects from the data.
Additionally, the objective in many standard experiments in teaching and industry experimental practice is to compare the means of alternatives (call 'Treatments' in other jargon, and 'Levels" in JMP) of fixed-effect factors (Varieties, Bedding materials, Teaching method, etc.). Thus, the focus is on the LSMeans of these 'Levels' and the significance of the differences between them.
The objective of using RCBD, typically with only 4 to 6 blocks, is only to remove the variation between these blocks from the Error variance, in order to reduce the Standard Errors of the estimated means of the fixed-effect 'Levels'.
But because the objective of such experiments is also to explicitly compare between the LSMeans of the factors' 'Levels', the SLS report of Fit Model offers a practical advantage over the Fit Mixed report in JMP standard. Because it offers 6 different multiple-comparison tests. Using one of these tests is a must in this scenario, and therefore Fit Model's SLS is needed as the default for analyzing data from RCBD experiments when one (or few) data points ('plots') are missing.
... View more
