[Feature Request]: Predictions and prediction intervals for new observations in linear regressions

[sfn-mes]

Description

To many the purpose of a regression model is to be able to predict responses for new observations. For linear models most would expect to get these with prediction intervals.

Purpose

Make regression useful also for prediction

Use-case

Prediction of new observations

Is your feature request related to a problem?

No

Is your feature request related to a JASP module?

Regression

Describe the solution you would like

I would want to be able to feed new covariate values into a regression model and obtain point predictions as well as prediction intervals.

Describe alternatives that you have considered

No response

Additional context

First: I have notice a similar feature request (#51) dating back to 2017. It does not ask for prediction intervals (I think) and does not specifically want predictions for new covariate values (the current versions of JASP allows me to save fitted values - not what I am looking for). But maybe this request should just be merged with the old one.

The formulas needed are "easy": lm::predict(model, newdata, intervals='p'). The tricky part is really how to get "newdata" into JASP and the results out. In JMP this is solved by putting the prediction equation(s) into the data window/data table as a formula. The user can write new covariate values into the data table and the predictions are calculated in the data table (which works as a spread sheet). However, with JASP's "any change to the data changes all analyses instantly"-policy this is probably a dangerous solution.

I can think of a work-around using the console (if I was able to interact directly with M1 as an lm-object, I could just use lm::predict, but typing in a data.frame for newdata would be a hassle), but nothing that is user friendly.

[tomtomme]

@koenderks
Is this Something the machine learning Module provides?

[koenderks]

No

[TarandeepKang]

I think this is a duplicate of #51 & #2582? Which is to say that this is a popular feature with a (IMHO) natural use case and a long history of being requested. For example, in our own "introduction to R" MSc course this gets covered in the second lecture which introduces linear regression. This might go more quickly than others would suppose, but the difference being our students have already gone through the whole thing of basic statistics with SPSS. NB, relying here on my memory as as I was a teaching assistant in this course a couple of years ago.

Not trying to pressure or anything, just saying that I do think it would be useful, both for researchers, and for teachers/instructors. And it is obvious how busy you all are with both making the software, and with your actual full-time jobs!

Best,

Tarandeep

[tomtomme]

@sfn-mes
Do you agree, that this is a duplicate of
#51
together with
#2582

then we can close this one

[sfn-mes]

It is very much a duplicate of #2582

[tomtomme]

then lets close this one and discuss there. Please reopen if you disagree