When to use replicate
Replicates are used to estimate pure error and parameter estimates more precisely. Adding replicates to a design allows you to reduce the variance of predictions and parameter estimates in your model.
You can use this method if there is a lot of noise in your response, or if you need to achieve a certain level of accuracy in your predictions. You can also add replicates in a second series of experiments (if you realize that the variance of your prediction is too high, for example). In this case, simply increase your design (click on DoE ->Augment design, then specify your factors and response(s), and finally click on "Replicate" for your choice of design increase). JMP will then ask you how many times each run should be repeated, allowing you to specify, for example, 4.
By the way, it's best to enter the replication condition directly in the design generation, to ensure that replicas are randomly distributed throughout the design/experimen
How many replicates?
There are no "hard and fast rules" for the number of replications, as this may depend on the subject, the equipment used, the precision required, etc... Please note that you have two options depending on the DoE platform you're using (classic or custom design):
- You can specify replicates, i.e. the number of times all experiments will be performed independently in the DoE,
- You can specify repeats, i.e. the independent repetition of an experiment in the design.
How to analyse your design with replicates?
If you have actually replicated series, you should include all individual replicates in the analysis, not the mean. The mean square error estimates the random variation between series. The mean squared error provides the appropriate denominator in the ANOVA F-test, and the mean squared error provides the appropriate standard error for parameter estimation t-tests.
On the other hand, if you have simply repeated the response measurement for a single series, it makes sense to use the mean if you believe that the variation between series is purely related to the measurement (repeatability).
