- I want to use Plackett Burman Design(PBD) for screening the factors.
- One categorical factor is type of polymer (Type 1, Type 2 and Type 3).
- In design expert software while selecting categorical factor, i can add only two level (-1 and +1) but there is no option to accommodate third level(0).
- For Numeric factor software takes middle level automatically. eg. I have given value for -1 and +1, middle level (0) was taken automatically.
- Please suggest me alternative screening design or any other way to solve this problem.
Solution i imagine:
If i consider categorical factor as Numeric factor than i can evaluate 3 level (Type 1, type 2 and Type 3 polymer) but i am not sure whether it will affect statistical algorithms followed for categoric and numeric factor. :-(
Is it valid?
Have you read Plackett and Burman's original paper?
What is it about the polymer you think might contribute to the problem? Some material property? A range of properties? I'd suggest that using three different complex materials might not be good parameters for a screening test.
Is this a numerical or physical experiment? If numerical, break the polymers up into their defining properties and vary each as individual parameters, at two levels. If it's a physical experiment, you could just pick the two most different polymers.
Remember that the point of a screening test is to work out what is worth looking into in more detail.
There are many other modelling techniques you could use, like taguchi.
The equivalent of Plackett & Burmann designs for qualitative factors are latin square designs. You can study up to 4 factors having each 3 levels in 9 experiments (add also at least 2 ot 3 repetitions of one experiment).
If some factors have only 2 levels, you can consider, for example, level 1 = level 2, and verify, when analysing results by ANOVA that these 2 levels are not different. You can study up to 5 factors with 4 levels in 16 experiments, ...
Hope this can help you,
Pierre Chagnon