topic Re: Prediction Interval for New Observations Not in the Data Set in Discussions

Prediction Interval for New Observations Not in the Data Set

jlrouquette — Fri, 04 May 2018 21:27:32 GMT

I’ve built a linear regression model and been asked to create a tool in excel, using the prediction expression, for a team of stakeholders to use. The will apply the tool to new observations not seen in the original dataset. I’d like to include a prediction interval but I’m not sure how to use the JMP outputs to give a prediction interval on data not already seen in the model.

Re: Prediction Interval for New Observations Not in the Data Set

dale_lehman — Sat, 05 May 2018 18:02:55 GMT

The red arrow on the regression output will permit you to save the predicted values and either a mean or individual confidence interval - which to use depends on whether the predictions will be for individual observations or an average of many observations with those characteristics (note that the difference between the two is likely to be very large, so be careful deciding which is appropriate). However, it is also important to note that these predictions and intervals only apply to new observations drawn from the same population. Usually, the new observations are from a later time (such as the present or future) than from the past (where the data came from). So, there is always the question of whether the sample you based your analysis on (even if it was the population at a point in time) is still representative of the population about which the new predictions are to be made. There is really nothing to be done about making predictions about a changed population unless you can specify the ways in which it may have changed.

Re: Prediction Interval for New Observations Not in the Data Set

jlrouquette — Mon, 07 May 2018 17:09:27 GMT

Perfect. Thanks, Dale! I hadn't thought about noting that we can only apply the data to observations from the same population.