I am planning to build a predictive model for land productivity using arcGIS 10.1. and logistic regression. In my model, the Independent Variables would be, among others, distance from rivers, distance from spring-line, distance from the coastline.

I am wondering if it is sounder/useful/essential to use the same value classification for all the three aforementioned rasters, or if it would not pose any problem in analytical perspective to have rasters with different classification.

1 Answer 1


Regression does not require all of the variables be on a common scale in order to be compared to each other. Whether you reclass them before running the analysis, or the analysis tool you choose lets you do so within the tool is up to you and the tool choice. You would of course want to keep a copy of any original data.

I might suggest reviewing the ArcGIS help files on regression analysis starting here. There is also a tutorial linked in the help file.

Some additional resources:

Since I also note you specifically mention a logistic regression, you may want to also take a look at this discussion regarding that method not being available in ArcGIS without use of R.

Disclaimer: I'm not an expert on regression and have only done a few simple analysis with it. What I know is from a one day lecture and reading resources linked above and similar.

  • Thank you for the answer and for the link. I would use either R or other stat tools. That would not pose a problem.
    – NewAtGis
    Jul 14, 2014 at 9:35
  • 2
    It is not correct that "all variables have to be on a common scale." Logistic regression estimates the values at one raster based on linear combinations of those in other ("explanatory") rasters: the coefficients in that linear combination are sufficiently flexible that no initial rescaling is needed. More importantly, the model will suffer by classifying the explanatory raster in the first place. Better methods exist, including non-linear re-expressions of the values and splines. This is extensively discussed on Cross Validated.
    – whuber
    Jul 14, 2014 at 16:03
  • @whuber Thanks for the correction. I'm afraid math/stats is not one of my strong areas, and it appears I've misread my notes. Can this answer be salvaged through editing, or would it best be deleted? I don't want to lose the information you've provided, but removing my first paragraph turns it into a (too long) comment and not an answer. Editing to say 'no, see whuber's comment' (or quoting it) doesn't seem right to me, and I lack the knowledge to provide what I feel would be a good answer.
    – Chris W
    Jul 14, 2014 at 18:12
  • Chris, I think you can easily retain all the information you have researched and kindly made available through this comment: just delete the first sentence. (Alternatively, if you're comfortable with this, consider negating it: regression does not require the variables to be on a common scale.)
    – whuber
    Jul 14, 2014 at 18:57
  • 2
    @NewAtGis That is correct, and part of why I mention keeping a copy of any original data if you do use reclassification in my answer - what if you decide to change your classes to improve/alter the outcome of the analysis? You can play with the visualization all you want, because it doesn't alter the data.
    – Chris W
    Jul 16, 2014 at 18:42

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.