The response surface equation obtained from observational data can
serve as a useful prediction equation as long as care is taken to ensure
that the points for which predictions are to be made are valid points in the
sampled population. This requires that the values of the independent variables
for the prediction points must be in the sample space. It is easy, for
example, when one variable at a time is being changed, to create prediction
points that are outside the sample space. Predictions for these points can
be very much in error.