FDE & prediction formula

Question

Hello,

I had a customer requesting how to save the prediction formula from FDE/B-spline which I did not find.

As instance in Fermentation / Ethanol, we can save data from "B-spline on Ethanol" which gives "B-spline saved data" datatable of 10,000 rows which has the Predicted Ethanol at the specific timepoints (the ones used to build the model) but there is no formula so we cannot interpolate for other timepoints.

Consequently through the "Function Summaries" I customized (save formula) and saved summaries. From the "B-spline model summaries" datatable I 1) unhide the column Time, 2) de-link Batch-ID, 3) add rows. On this new rows I populate "batch-ID" and timepoints and used the "Ethanol Conditional Prediction Formula" as prediction formula.

This solve the customer pain who wants to compare the effect of treatments on various animals at the same timepoints (while they were tested at various timepoints due to experimental errors).

Any thoughts on a simpler process to achieve the same result ?

PS: I also copy / paste the "Ethanol Conditional Prediction Formula" on another JMP table but yhe customer prefered the first path aforedmentionned.

jiancao · Answer

There might be some advantages to using a random coefficient as what the conditional prediction formula does than using a simple linear regression. Besides the r.c. model has already been fit :-). However, the conditional prediction formula cannot extrapolate since there are no basis function columns while a simple regression of y on time that Mark has suggested can.

A side comment--the conditional predictor formula should be saved to the B-spline Saved Data table since it is where a r.c. model is fit and that the predictions for are made.

(I briefly lost internet connection due to the Isaias that is bringing heavy rains and gusty winds while writing this post :)

Mark_Bailey · Answer

I think that this customer is not thinking 'functions.' They are merely using the interpolating functions to estimate the response at a particular time. They would be better to fit Y(time) versus X if that is how they think.