Is it possible to conduct a correlation analysis over time?

[#1]

Hi all,

I had like to conduct - if possible - a correlation analysis on Jmp using three different parameters. Usually I conduct correlation analyses using the Fit Y by X-platform pasting the two traits (two columns) I want to correlate in the "Y" (response) screen respectively in the "X" (factor) screen of the platform - And read off the R squared value that it shows in the output file. However, this time I had like to correlate two traits over time (five different time points to be specific). And I dont see how to go about to conduct that analysis...

Or is it as simple as that I make a column for my time data and paste this information in the "By"-section of the Fit Y by X-platform? (But how do Jmp know which data belongs to which time point?). My question is therefore: Can anybody help me to construct such an analysis? If possible I had like to receive an R2-value for the correlation of the two traits over time. Currently, I only know how to correlate between the traits at each time point respectively.

Thank you in advance for help!

If you need more information to help me please ask for that.

Best regards,

Jim 

[#2]

Another approach is to use some form of multivariate SPC, ala in the Quality and Process -> Control Charts -> Multivariate Control Chart sub platform. This format is extensible to any number of variables and helps control risk. With only 5 points in time, your data set might be a little skimpy wrt to rational groups of observations...but it's a start. There are other useful analysis options such as the change point detector for determining when a change has occurred in the multivariate structure in the data set. Here is the reference within JMP Help (version 12):Multivariate Control Charts

[#3]

Thank you very much. I havent had to time to make this test yet, but will make it in a near future. I hope to be back to update on this question then.

Best regards,

Jim

[#4 · Accepted solution]

I don't think correlation between two variables at a single point in time has any meaning, but you could look at the correlation between two variables during some interval of time, and compare that to some other interval of time. I think that would just be a matter of doing a Fit Y by X with your time intervals selected  by excluding the data outside of your time interval of interest. I don't have any experience with JMP scripting but I imagine you could write a script to automate selecting the different time intervals and repeating the analysis for each. Say you had a day's worth of temperature data from two points in a process, with readings every minute. You'd do a Fit Y by X for hour 1, then another for hour 2, and so on. Then you'd see how they correlate over time. I don't think five data points is enough data to get a meaningful correlation in this way.

[#5]

I would use Multiple Linear Regression, setting time, and your X variables as the Effects and your Response variable as the Y.  Using the Fit Model platform, it will provide you with an R2 of how well time and your second X variable can predict your Response variable.