Suppose that you plan to run a \(2^{3}\) factorial experiment with 8 runs. You discover that your budget has been increased so that you can perform an additional 4 runs. Where would be the best places to run these additional 4 trials if you want to both get a model-independent estimate of error and reduce the variances of the model parameter estimates as much as possible? Restrict your choices to replicating some of the original 8 runs.