I just conducted my first DSD study and just wanted understand some of the inner workings behind it that I couldn't find/understand from the JMP manual here. * I want to understand if there are any assumptions regarding the response data that need to be met prior to analysis by the "Fit Definitive Screening" script. I am asking this because if you want to run a ANOVA test, you need to ensure that the data is normal, so just wondering if there are any similar assumptions that need to be met before analyzing the data. * Why does the "Fit Definitive Screening" script display a Main Effect Residual Plot and not simply a Main Effect plot (as is common in two-factor designs)? Furthermore, as mentioned in the link above, I am not able to grasp why "A factor with residuals that differ across levels indicates an important main effect"? * Why does the script not display an interaction plot? Is this something to do with the structure of the DSD?

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; } } }

Are there any assumptions regarding the response data that need to be met before analyzing DSD?

AsymptoticRules

I just conducted my first DSD study and just wanted understand some of the inner workings behind it that I couldn't find/understand from the JMP manual here.

I want to understand if there are any assumptions regarding the response data that need to be met prior to analysis by the "Fit Definitive Screening" script. I am asking this because if you want to run a ANOVA test, you need to ensure that the data is normal, so just wondering if there are any similar assumptions that need to be met before analyzing the data.
Why does the "Fit Definitive Screening" script display a Main Effect Residual Plot and not simply a Main Effect plot (as is common in two-factor designs)? Furthermore, as mentioned in the link above, I am not able to grasp why "A factor with residuals that differ across levels indicates an important main effect"?
Why does the script not display an interaction plot? Is this something to do with the structure of the DSD?

Accepted answers

statman

Some clarifications:

1. There is no assumption of normality of response variables from a DOE nor is there an assumption of normality to use ANOVA. In fact, ANOVA is fairly robust to distributional abnormalities. The assumption for quantitative analysis is Normally and Independently distributed residuals with a mean of 0 and a constant variance (NID(o,variance)).

2. Residual plots are quite useful for identifying model issues. The Prediction Profiler is what you are looking for for main effects plots. If residuals are quite different around one level than another, this can indicate the factor has an effect.

3. You must make sure there are interactions in your model to get interaction plots (Stage 2).

All comments

statman

Some clarifications:

3. You must make sure there are interactions in your model to get interaction plots (Stage 2).

AsymptoticRules

On point #2, could you explain this statement: "If residuals are quite different around one level than another, this can indicate the factor has an effect?" I am not able to grasp this. Does this mean that the residual in this graph are of the line, i.e. model, created for the particular response vs. factor?

Mark_Bailey

It means that the factor affects the variance of the response, regardless of how it affects the mean of the response. This outcome is a violation of the least squares model which assumes the variance is independent of the response level. The variance is estimated as a constant value for all X.