I am trying to bake the best sugar cookies out there!
Currently, I have about 15 different types of sugars to choose from and I'd like to know which ones or combinations provide the best tasting results.
I have a specific amount that I'd like to put in each time (2 cups) and the oven will remain the same throughout each batch so the only variable is what sugar or combinations affect the result.
What is the best approach to this without cooking a bunch of batches?
My thought is to do an initial screening test where each sugar is tested individually at 100% but after that, what is the best way to capture synergies between combinations of sugars?
Thoughts would be greatly appreciated!
Sure, I'll "bite", but it may not be the answer you're looking for. Are you wanting to understand why cookies are best or just to create multiple tests and pick a winner?
The first question I have is what defines "best"? What is the response variable? Are there more than one response variables? Are the measurement systems capable? Are you using an ordinal scale? How do you account for individual bias? If you are using an ordinal scale, there are some recommendations I would have for you.
It seems to me that you have one factor, "sugar type" with 10 levels. You also indicate you are keeping oven (and other factors) the "same". This is both unrealistic and a poor strategy. There is both within oven and between runs of the oven variation which need to be counted for.
I am not a "baker", but my guess would be the following factors may also have some affect on best:
There may also be interactions between the above ingredients and the baking process. If we want to restrict our studies to experimentation, then I would start with a screening design. Lots of factors at 2-levels (pick extremes). Then iterate based on what is learned in the first experiment.
The good news is there are some excellent options to use split-plots. Making batches based on factorials of ingredients and splitting the batches to run factorials with baking conditions.
Ideally, I would like to know which sugar(s) to use that will provide the most sweet taste. I'm not really interested in measuring other variables for now.
My thoughts would be to use a ordinal rating system 1-5 (not sweet to very sweet) and the testers would be put through a blind taste test to eliminate bias that they might have towards some sugars. I'll have roughly 100 participants with the taste order randomized with each participant being provided a "standard" for what each value represents on the scale.
I know assuming the oven and mixing is always perfect is not the best of assumptions. It was more just to simplify the experiment rather than running numerous studies.
However, I do like your suggestion on the spilt-plot designs to incorporate some of these other factors.
I'll start with the screening design and then move to the split-plot.
Statman adds many more questions and layers to your experiment. If we step back to just the sugar idea and assume that you have figured out your measurement system for quantifying your results (if not then stop, figure out your measurement system, then you may return to designing your experiment) then what you have is a mixture experiment. If you only had 3 sugars then you could do say 8 runs where 3 runs are the 100%, 3 runs are the 50/50 blends, and then a run of the 33/33/33 blend. However, you have 15! That is a lot of sugars! The problem with just screening at 100% is that some of your sugars might be terrible as the primary sugar but excellent as a portion of the sugar. You would also probably want to limit the total number of sugars used in any one batch (say 4?). Since sugar cookies are typically refrigerated prior to baking you probably also want to consider a blocking structure such that you bake more than one batch of cookie at a time where the location of the cookie on the tray is also taken into consideration. The baking layout is a 2nd puzzle once you figure out the batches. So you are going to need a mixture design either using the mixture design platform or the custom design platform with constraints....beyond that lots to think about and consider. Looking forward to reading what some of the other DOE users have in mind for your experiment.
This is actually my current mental block.
Based from other forms and online, my thoughts right now are to just screen each sugar individual and categorize each one from a 1-5. Then "randomly" choose maybe 50 pairs of each at 50/50, and re-categorize from 1-5 again based on the pairs. After those 50, take those pairs of twos and pair them up (now 4). And from those 4, choose from the sweetest of the bunch
It's not the best design but rather just something quick and easy.