Not even sure how/where to start....some type of regression needed.

#1
Hello there, I am a chemist needing to do some type of a regression I think.

I have a recipe with 6 components (the amount of each component used gives rise to 6 independent variables). The resultant product of this recipe is evaluated with a single numerical value (1 dependent variable). I have prepared and evaluated this recipe 11 times. So I have 11 data sets, each comprising the 6 independent and 1 dependent variable.

I would like to try to pull out of this data, if any correlations exist between amounts of recipe components and the evaluation.

To use a cooking based example: I have 6 ingredients and rate the final taste on a scale of 1 -10. Perhaps the taste is only good (rated highly) when I used slightly more sugar, but ALSO less salt then the recipe called for.

Can anyone recommend ANY starting place for what type of analysis is appropriate for this situation? Any help would be greatly appreciated.

Cheers!
 

Miner

TS Contributor
#2
Technically, this calls for an ordinal logistic regression. You may be able to get an understanding of what is important using a multiple regression (6 IV plus some interactions). This biggest issue will be the small sample size of 11. This will severely reduce the power of the analysis to detect significant IVs.

A better approach would have been to design an experiment (possible a mixture design) with sufficient replicates before running the experiments.
 

Miner

TS Contributor
#3
Another approach that comes to mind since you mentioned you were a chemist is EVOP (Evolutionary Operations) where you make a series of small changes and follow the path of steepest ascent. I have only seen this used with 2-3 factors, so I don't know if it can be used for 6. The downside of this approach is that you do not end up with a model, This is purely an optimization methodology.