Considering standard deviations in to DOE

[#1]

I made a DOE with custom design of 6 runs. The 2 factors are continous and response Y is in %. When i have the results for Y. the standard deviations is quite huge. So, if i just input the mean value, I believe that it will not do justice to the design predictions. So what can i do to add these standard deviations to be considered by the model to get a better predictions.

[#2]

Could you post an example?

It sounds like you have repeated measures of each of the 6 runs?

Unless you want to analyze the data using a mixed model in JMP Pro, a good alternative is to use the mean of the repeated measures and the log-transformed standard deviation as Y's.

The mean of the replicates takes advantage of the central limits theorem to give you a better estimate from a noisy system, and the Stddev lets you see if there is a portion of the design space that has less noise.

[#3]

In addition to @Byron_JMP's suggestion, you can also use the LogLinear Variance platform that is available through Analyze > Fit Model.

Were the factor ranges wide enough to elicit large effects?

[#4]

Not enough information/context to provide specific advice.  What exactly is the response variable?  % what?  For example, % yield is not a very useful response for understanding specific failure mechanisms (there are many things that could cause yield issues).  As Byron suggests, how are you calculating standard deviation?  Is this within treatment variation?  If so, the standard deviation is quantifying the effect of the noise changing within treatment.  You might certainly want to know if the model effects can impact this.  Also as Byron suggests, you will be able to model 2 responses: The mean (with increased precision as this will reduce the within treatment noise) and the standard deviation (or whatever transform you chose).