I am designing a model for a processing facility. This facility runs certain processing steps on products. Depending on certain details of the product/step, these steps are grouped together into step groups. Basically I need to create some kind of aggregate model number for these step groups to use for predictive analysis. I've chosen trimmed mean, as mean and median tended to under or overestimate the predictive model.
My question is, how can I best determine trim values for these trimmed mean calculations, using statistical methods such a mean, median, mode, n, SD, skewness, ect.? The histograms of the step groups tend to skew right, so I'm guessing the trim would be more heavily favor the right side (like bottom 5%/top 10% for example).
Any and all information/articles about this topic would be helpful. Thanks