Hello, I ran a DSD with 3 continuous and 2 categorical factors and am getting pretty weak significances in my model. Only parameter with a p-value below 0.05 is the Intercept. Next best is 0.12 When I uncheck the checkbox for 'interactions obey strong heredity' I got at least one p-value below 0.05. It's an interaction with 1 main effect estimate and 1 effect that is not listed in stage 1. * Would you suggest it to be a good idea to uncheck the boxes in my case? What am I loosing by doing it? As I can see in the 'prediction profiler' I am loosing curvature. Is there more? * Is there some documentation about this? In the jmp documentation I found the checkboxes (page 282 in The Fit Definitive Screening Platform), but there is no proper explanation.. @martindemel

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; } } }

Fit of DSD: influence of heredity checkboxes

JL_LZH

Hello,

I ran a DSD with 3 continuous and 2 categorical factors and am getting pretty weak significances in my model. Only parameter with a p-value below 0.05 is the Intercept. Next best is 0.12

When I uncheck the checkbox for 'interactions obey strong heredity' I got at least one p-value below 0.05. It's an interaction with 1 main effect estimate and 1 effect that is not listed in stage 1.

Would you suggest it to be a good idea to uncheck the boxes in my case? What am I loosing by doing it? As I can see in the 'prediction profiler' I am loosing curvature. Is there more?
Is there some documentation about this? In the jmp documentation I found the checkboxes (page 282 in The Fit Definitive Screening Platform), but there is no proper explanation..

@martindemel

Accepted answers

Mark_Bailey

It is difficult to answer a question about a specific number of runs without more details, but if you can afford to run more than thirty runs (including the 18 that you already have) it would go a long way towards settling the decision about the important effects and towards predicting the response very well.

All comments

Mark_Bailey

A DSD for five factors is a very small screening experiment. Did you include extra runs?

A DSD for five factors would require large effects to be significant. Perhaps four or five times the standard deviation of the response. How large is the SD compared to the mean response?

Heredity is a useful principle that often holds but not always. Strong heredity holds less often. So the check box gives you the option to restrict the effects that are tested or not, in the hope that you can find more effects.

The prediction profiler is based on the selected model. You did not find a significant non-linear term so the profiler would not show any curvature.

JL_LZH

Hello Mark,

thanks for the information.

My DSD was 5 factors and I it contained 18 runs. As I was restricted to one day for my experiments I did not include extra runs.

With the heredity checked the smallest p-value that I get is 1.1209 and its a main effect. But the estimate is 1.1858 and StdErr is 0.134 (gives a t Ratio of 1.3861).

Without the heredity the smallest p-value that I get is 0.0006 and its an interaction. The estimate is 0.64 and StdErr is 0.1095 (gives a t Ratio of 5.8461).

From the model and also from playing around with the graph builder I anticipate some effects that lead to a change of my factor ranges. So I am going to augment my design (you already helped me on that topic in another discussion) which would give me the opportunity to include some extra runs. I want to add one level to one of the categorical factors and enlarge the range for one of the continuous ones.

Would you suggest a certain number of extra runs? JMP suggests 28 runs including the 18 initial runs.. I'm thinking about boosting it to 35, do you think that will make it?

Mark_Bailey

JL_LZH

Ok Mark, I'll try. Thanks a lot!