Compare new data set to previously generated model?Hi all,
Question regarding how to compare a data set to the output of a previously generated response surface model. I have a response surface model that I developed based on a primary data set ("Data set A") with a response variable and three input variables. The input variables are environmental conditions - temperature, humidity, and light level, across specific ranges (e.g. 10-30C, 10-90%RH). I also have a second data set ("Data set B") that has the same response variable, obtained at a single temperature, RH, and light level. I want to test whether the response surface model's estimate of the response variable for the conditions used to generate Data set B is statistically different from the actual mean value of Data Set B. What would be the best way to do this? I thought about doing a one-sample t-test to compare Data Set B to the predicted value from the model, but that does not take into account the error associated with the model estimate.
Thanks for any suggestions.
Quality and Process EngineeringRe: Compare new data set to previously generated model?Perhaps you can attach dataset B to the dataset A and mark it as validation. Then you can use the model workflow to get the comparison result.Re: Compare new data set to previously generated model?The Fit Least Squares platform is aimed at the mean. The model is for the mean response. The Prediction Profiler provided a confidence interval for the mean:
You must save the interval for individual observations. I suggest that you save the formula, so that you may add new rows with the new data and the formulas will evaluate the conditions and give you the interval for comparison:
Re: Compare new data set to previously generated model?@Mark_BaileyThanks. Is there a way to view the prediction intervals from the model output window?
Re: Compare new data set to previously generated model?You have a single observation to test the null hypothesis, so you can't use a t-test.
You might instead use a prediction interval to decide if the new observation is unexpected.
