Ridge Regression or centering of variables

DrPRW

I am performing a regression and structural multicollinearity is an issue I am trying to deal with. I can either center the variables or perform a ridge regression. (I am referencing this website: https://statisticsbyjim.com/regression/multicollinearity-in-regression-analysis/).

Two questions:

1. Is there a way to automatically center the variables in JASP?
2. How do I perform a ridge regression in JASP? I have looked and cannot figure out if it is included in the available analyses.

I am by no means fully functional in using statistics- I know enough to be dangerous. So, I respectfully ask that your help be phrased so that a lay person can understand (to the extent possible). Thank you in advance!

EJ

Hello DrPRW,

1. You can use the "compute new columns" functionality to center the variables. Maybe we ought to make this even easier. I am a little surprised that centering the variables would help actually. Johnny/Don?
2. Wrt the link you gave: Johnny/Don, can JASP compute the Variance Inflation Factor?
3. Johnny/Don, I don't see ridge regression in there -- if not, we ought to add it I think!

E.J.

JohnnyB

Hi @DrPRW,

Centering your variables can help in remedying multicollinearity in cases where you have multiple terms per variable such as square or interaction terms. Here, subtracting the means influences the interaction estimate, which will in turn also affect the estimates of the other regression weights. Unfortunately JASP cannot do this centering automatically yet, and you would have to do this by hand, using the compute columns functionality (see the blogpost here for instructions). In short, when you have the compute column interface, you click the variable you want to center, then you click the "-" sign, then in the list on the right you look for the button "mean(y)" and click it, then you click again the variable you want to center, and then click "compute column". Hopefully you don't have too many variables in your model ;-)

JASP can compute the VIF: under regression, you tick "multicollinearity diagnostics". This will add two columns to your "coefficients" table: tolerance and VIF, for each of your predictor variables.

Ridge regression (which I know only a little about) is a variable selection algorithm, aimed at minimizing the number of included predictors in your model, for instance by eliminating predictors that are highly correlated/collinear. JASP does not have this functionality yet (although the network module has some related algorithms like EBICglasso, but that's maybe a whole other story). I would advise against using such an automatic procedure, and rather look at which predictors are correlated, think about why, and then possible exclude certain predictors from your model to combat the multicollinearity manually.

Hopefully this clarifies your problem a little bit - if you have any further questions then please let me know!

Kind regards

Johnny

vandenman

Hi @DrPRW ,

It is possible to do ridge (and lasso) regression in JASP, but only within the machine learning module. As a consequence, the results are focused on prediction and not so much on statistical inference.

it is possible to obtain the regression coefficients by ticking `Regression coefficients` under Tables.

Best,

Don