BIBD design for MarketingHi Team,
I am planning to show a BIBD design for a Marketing professor. I found this example.
Factor A: Placement of promotional displays (A1: Front of store, A2: Middle of store, A3: Back of store).
Factor B: Type of promotion (B1: BOGO, B2: Discount, B3: Free gift).
Factor C: Timing of promotion (C1: Morning, C2: Afternoon, C3: Evening).
Resulting Design to be created to run across 5 stores
How can i create this using BIBD? (Or is there a way to recreate this using Custom Design?
thanks in advance
Re: BIBD design for MarketingMurali, hi!
I don't think the example above is really a BIBD, at least as we define it in our software.
According to our documentation: "The BIBD design is constructed such that each treatment and each pair of treatments occurs together in blocks an equal number of times." How does this concept of "Treatments" map onto the example you described? Another section of the documentation explains: "Treatment size can refer to the number of levels of a single factor or combinations of levels of two or more factors." Your example has 3 categorical factors with 3 levels each; 3^3=27 "Treatments". In other words, the combination of [A1, B1, C1] could be assigned to treatment level 1, and [A2, B1, C1] could be called treatment level 2, etc.
Adopting this framework, the example above is a design with 27 treatments, 5 blocks (stores), and 2 runs per block. With 5 stores * 2 runs per block = 10 total runs, it would not be possible to examine all 27 combinations of the three factors. So it is not possible for JMP to generate a BIBD for this scenario. If you try to generate a similar design using the BIBD tool in JMP, you'll find that with 27 treatments, the lowest number of permissible runs per block is 13:
You can produce something similar to the example using the custom designer, by specifying random blocks:
Cheers, Jordan
Re: BIBD design for MarketingWe're only capable of generating BIBDs with a single treatment. Here's a link to a paper looking at ways to create BIBDs with more than one treatment, what they're calling multi-part BIBDs. Given that the paper was published in 2019, If this research is novel, it's newer than our implementation of BIBDs. Briefly skimming though the construction techniques has me hopeful that this might be something that can be done in JSL.
BTW, you'd need at least 18 runs to estimate the main effects and 2 factor interactions for the experiment.
Re: BIBD design for Marketing@Jordan_Hiller and @DonMcCormack : thanks a lot for your inputs. Yes it helps me a lot.
Meanwhile, I will add a request for multi-part BIBD (citing the link which you provided) to internal wish list.
Thanks and have a good day
Murali
