Resources to support Regression Analysis in ArcGIS

[SteveCline]

Is anyone aware of any books or other publications to support using the regression analysis tools in ArcGIS?  I have completed the Regression Analysis Using ArcGIS module on the Virtual Campus and I am looking for more supporting documentation.

Thanks for your help.

[ChrisDonohue__GISP]

Let me add someone is very likely to know

Dan Patterson

Edit:  recent threads that may help

Questions about spatial dependence and Moran's I in Regression Analysis

Creating Feature Classes for Regression Analysis

Re: Regression kriging

and some more knowledgeable folks:

Steve Lynch

Eric Krause

Chris Donohue, GISP

[DanPatterson_Retired]

I presume that you don't mean the volumes of academic texts that are available.

A useful link to explore all things statistical is Cross Validated on Stack Exchange.

A general comment about using any statistical tool in any piece of software... the user needs to be aware that they are responsible to ensure that the underlying requirements of the 'test' are being met.  It is far too easy to obtain a result and the inconvenience of testing for such things as normalcy are often deemed inconsequential. 

[SteveCline]

Dan -

Actually I am interested in academic texts in this case so any suggestions which are specific to ESRI's Regression Tools would be appreciated.  I am doing work on a graduate thesis which relies on several regression tests, including spatial auto-correlation.  I understand the testing that I am doing from a statistical point of view but I want to ensure that the manner in which I am using the software is sound since I am using it as a tool for the first time.

Thanks for your advice.

Steve

[DanPatterson_Retired]

from the link Darren provided which contains one of my posts, there is link to

esri's spatial stats blog which contains useful resources done by esri...

Spatial Statistics Resources | ArcGIS Blog

there are a few other considerations in that link as well.

I hate to dump you into the pit, but there are esri publications and other resources here

http://esripress.esri.com/display/index.cfm?CFID=19557025&CFTOKEN=18749870

Then there are the classic texts, such as " Principles of Geographical Information Systems" by any incarnation of Burrough et al (Oxford press)

With regards to any particular tool, check at the very bottom of the tool and see if they have a reference list...having said this they tend to be a bit thin on that side of things, but sometimes, they are noted if there is any algorithmic suggestions.

Esri tools are ... shall we say... thin on non-parametric statistical tests, for example, regression, so one is obliged to ensure that the fundamental requirements for the test are met or the data can be transformed so that those requirements are met.

If you have some specifics and/or a description of what it is you are going to be looking at, I could discuss further from a data or analytical perspective since I have seen many projects over my teaching years and there may be insights that are unique from that perspective.

You should also be forewarned, that many problems that people have with using the tools lie in poor file management...so brush up on those before you begin any meaningful trial runs.  I would also suggest that you take a look a ArcGIS PRO since it gives you access to SciPy and there are some access points to R use within ArcGIS that will broaden your arsenal and analytic capabilities.

Keep us posted.

PS

my arsenal of academic texts are in my office and I can't think of anything offhand that is superior, names will bubble to the surface, if you come across any that you want opinions on, put them out and I will see if I can provide an 'opinion' but what suits me may not suit you

[SteveCline]

Dan -

Thank you for the resources.  That is what I need.

Here is a general overview of  my project:

The research question asks, "To what degree is the geographic distribution of charter schools in the United States associated with characteristics of either modernizers or traditionalists?"  The Modernizer-Traditionalist model is one developed by Paul Luebke at the University of North Carolina.  Essentially, modernizers tend to be younger, more mobile, more ethnically diverse, more educated and more likely to live in urban areas than traditionalists. 

I have been constructing the data using NCES public schools data through the US Department of Education and I have a county level file with both schools and students in charter schools as a proportion of all public schools as well as the change in that proportion over the last ten years.

For the social data I am primarily using the most recent ACS 5 year estimate at the county level. In addition, I have 2010 religion data from the Association of Religion Data Archives.  I have used SPSS before for this kind of work but thought that since I will be using ArcGIS for the visualizations it might work to use the statistical package with it as well.  Documentation on these tools is fairly thin though.

I have just recently begun using ArcGIS Pro so I may give that a try.  Any other suggestions are welcomed with regard to this specific problem.

Steve

[DanPatterson_Retired]

I will give this some more thought but I would suggest that you spend some time exploring the descriptive side of your data first...and extensively, both in terms of the attributes, their inter-relationships and their spatial pattern.  i cringe when I see the words correlation and regression spring forth in the early stages of work since I suspect you are not looking to see how changes in one variable produces changes in another variable...nor are you looking at how to predict values for intervening spaces in the observed patterns where no observations were made.

I would also suggest that your data is not likely "normal" which many...but not all.. would argue is a prerequesite to performing traditional parametric correlation or regression without wildly strange transformations to the data to yield the normal distribution in the variables.  You will also have to deal with the inter-correlation in the variables, since things rarely progress in isolation or in step with a single other variable.  The place to start would be to see if there are differences in the things like cross-tabulations ie chi-square testing which can be interpreted as tests of association/difference.  For just about every parametric test, there is a corresponding non-parametric test.  For example, Spearman's vs Pearson's tests.  These tests are largely absent in GIS software, but they can be integrated into the analysis through interfaces to SciPy, R and other packages.  Where Arc* will allow you to shine is the spatial stats, they have a good solid suite of tools if they are used wisely and used to guide what inferential tests in a non-spatial context you may want to examine.  The introduction of more temporal tools (space-time cube etc) should enable you to at least look at patterns in place over time, rather than concentrating on simple attribute associations.

It would be useful exercise on your part, to separate your particular topic and their variables out of the exercise and make a list of the types of analyses you should do, and in what sequence to analyse ANY problem.  Once you have a handle on that, then, and only then, can you really approach your topic in an unbiased fashion.

Keep us posted, and ask further questions as needed.  There is a very broad and strong skill set in this community that can assist you in your work.

[SteveCline]

Dan -

Thank you for the advice.  I have been working on the descriptives for a while now and I am just beginning to wade into the inferential side.  I had been told to use SPSS, which I have a fair amount of experience with, and then GeoDa, wIth which I have none, but when I came across the tutorial for Arc's regression tools it looked promising.  My intent here is to deepen my understanding of the tools in Arc so I can decide if they are the best fit for this work.  I will pull together some more specific information and post it here.

Steve

[DanPatterson_Retired]

a number of collegues and students use both, but at some time you are going to have to do some mapping and if you want to compare the inferential statistics side, there is no harm in doing some comparison.  Most people recommend what they were taught in grad school... it is called inertia...