Not sure how to input replicate measurements to model a DoE

[#1]

Hello,

I hope someone can help me with the question below.

I have run a Custom Desing DoE with 3 factors (continuous variables) and 3 responses. The randomised desing calculated 16 runs. Because my data is a bit variable I have run this experiment 8 times.

What is the best way to input and model all the data available? Should I concatenate 8 tables of the 16 run Custom design and then run the model? Some of the runs whithin the replicates failed, how should I deal with these?

Thanks,

Nuria

[#2 · Accepted solution]

The analysis assumes that the each observation is a separate experimental run. If the outcome of one run is measured more than once, then the average measurement should be used. Otherwise, the measurement error serves as the experimental error. You would, in fact, have two levels or random effects.

In your case, you actually replicated each run 7 times, so each observation includes the necessary necessary random error for the modeling. I would enter all 8 runs for each treatment.

The failed runs should be omitted. Just leave the response as missing in these cases.

[#3]

Hi Mark,

Thanks for your reply. It is very useful.

In order to run the DoE model I created a table with the information for all the replicates, i.e. the 16 run structure repeated 8 times, and excluded the failed runs. What are the implications on the model calculations of doing this as opposed to leaving the responses for the failed runs blank?

Many thanks,

Nuria

[#4]

Exluding rows or leaving the Y value missing will have the same result.