nav[aria-label="Primary Navigation"] { padding: 0; & ul { list-style: none; width: 100%; display: flex; flex-direction: row; justify-content: start; align-items: start; gap: 30px; padding: 0; & li { margin: 0; } & ul li { list-style: none; } } }

Extend linear regression line in JMP version 13

RobRobeyns

Hi,

I have a question regarding the Fit Line functionality in version 13.

In version 10, I was able to make graphs with a linear regression line that extended beyond the data points (see figure). However in version 13 the linear regression line stops at the last data point. Is there a setting in version 13 to achieve the extension of the linear regression beyond the last data point?

Thanks!

Accepted answers

jules

I agree with Jim's concerns about extrapolating beyond the range of data observed. However, if you wish to do this you can add in a new row to the table with a value beyond the range of data for just your X variable, and JMP will then extend the line to that point. Since there is no Y value associated with that row this will not change the regression results.

All comments

txnelson

You can use either the annotation tools, or in JSL use the Add Graphics capability to add to the fit line, however, since it is a regression, which is based upon the parameters within the data, and not a forcasting tool, having the fit line project beyond the known data points would be statictically improper. So what you are seeing is a correction of an error in JMP 10.

jules

RobRobeyns

Great! Thanks for the feedback!

cfranklin333

Adding a 0 on x didn't work. Had to use Annotation tools with fit where intercept is forced to zero where it is impossible to have any other intercept value other than (0,0).

Dan_Obermiller

There are, obviously, times where you want to force the intercept to 0, but a caution: this changes the interpretation of many of the regression statistics that you are looking at. Essentially, by forcing the intercept to be 0, you are saying that the mean is 0 (think of centered/scaled data). This can be especially problematic if you are extrapolating beyond the range of your data.

Another simple way to think of this: think of a plot of height versus weight. You can fit a line to that data. Should that line be forced to go through the (0,0) point? If you have no height, you would not have a weight so it makes sense. However, the linear relationship likely would not hold from the lowest height datapoint to 0. Forcing it to be linear across the entire range can cause problems.

cfranklin333

Also need to extrapolate for stability / shelf-life extension based on 95% CI of trend. Unable to get Degradation to provide similar to existing visual plots.

Mark_Bailey

You can use Bivariate > Fit Special command to select a No Intercept trend analysis.

This has been one of my fustrations with the JMP program, as new rollouts have made the program less flexible and useable. If alternative work-arounds are being offered for functions that use to be available in past versions, then why was that functionality removed in the first place?

For large and complicated datasets the work-around is not feasible.

Kevin_Anderson

When first-generation food processors cut off fingers, responsible manufacturers added interlocks to the covers.

When Taylor Series approximations were improperly used to extrapolate beyond the range of the data, responsible manufacturers disabled that functionality.

I think Jim and Dan are right. Just say no to the workarounds. If you simply must extrapolate, fit a deterministic (usually nonlinear) model.

nocaltim

Regarding this issue of extending best-fit lines to the limits of the (say) X graph range limits, I would like suggest a scenario where it is very useful to have these best-fit lines extend beyond the data. In my work, EVERY SINGLE DAY I do an analysis where two variables should have perfect 1:1 correlation (slope of 1, intercept of zero). And this 1:1 correlation should extend across range that far exceeds my data limits. The actual X vs Y never a that true 1:1 correlation -- and that's very useful to know. But to help grasp this, I like to explicitly add the 1:1 line to the graph. And I want this 1:1 line to extend far beyond the data limits so that when I also do a best-fit on the actual data, even though the best-fit line is limited to the range of available data, I can use the 1:1 line as "tilted floor" of comparison. I fully understand why best-fit data shouldn't be extrapolated (too far) beyond the available data, but why can't I include an extended 1:1 line??? (Yes, I can add it as a graphical object, but that's just JMP *creating* work for me rather than JMP simlifying my work.) Comments?

quantum

Mark, I have a need to make the plots extrapolate to a population not existing in the data set to provide predictive models of what we will see from known variation in time. Constraining slope wrecks the statistical validity. I had developed this method in a previous revision of JMP and this will force an entire team of several hundred engineers to move away from JMP as a platform. Do we have a better fix than this?

Chris_Kirchberg

A great place to request (or add back) certain functionality would be in the new JMP Wish List:

https://community.jmp.com/t5/JMP-Wish-List/idb-p/jmp-wish-list

Developers are looking at this list for upcoming versions and this sounds like a great thing to add to that list (if it is not already there).