DOE: add historic data as block additional blocking variable or simply as categorical factor?A customer asked:
They want to do a DOE with a small chromo column (1.6cm diameter). They have already some historic data from a Lab-scale chromo column (5cm diameter). Can we include these results in the DOE to get better model results? Especially how to tell the design that the software understands that there have been used different scales? Using blocks or just as categorical factor? Even better would be to get confirmed that the chromo column does not have any effect. If it has, what it should tell us what offset we can use for scaling. (important for our scale down model qualification
(I exclude the option to use a blocking variable in the design itself, because we cannot run additional 5cm experiments)
My suggestion would be to use an additional blocking factor for scaling size 5cm and 1.6cm, therefore the model needs to be adjusted with a random effect for the blocking factor. If we use just the categorical scaling size factor, we may not get information about the desired scaling offset, as I just have the different groups but no information about the size effect itself.
Do you have any experiences on that or opinions on what's best to approach this? Could it be an issue (adding bias) when they use way more historic data than the small size experimental data.
Thanks!
Re: DOE: add historic data as block additional blocking variable or simply as categorical factor?I have augmented the original data when it was externally validated.
Re: DOE: add historic data as block additional blocking variable or simply as categorical factor?Thanks Mark,
I thought about this too, but was not sure how I should then treat the model to understand if there is a differnece between the scaling size. I would have to add a column to the data table to add the scaling size (5cm for the existing data, 1.6 for the augmented part). Would the best be to simply use it as categorical?
Re: DOE: add historic data as block additional blocking variable or simply as categorical factor?Hi Martin,
You certainly could have Scale as a categorical factor. But it depends what the domain experts think:
- How do they expect the system to behave at the two different scales?
- What are their goals for the experiment?
It would also be valid to have Scale as a continuous factor.
And it would be valid to have interactions between Scale and the other factor effects, if they think that the factors might have different effects at the different scales, which seems plausible to me.
I would encourage the domain experts to consider the different model possibilities and understand what they mean for how they will enable them to understand and predict the behaviour of the system.
Phil
