Is AUC provided in OLS Regression output?

[SeanMcNulty]

Hello, I'm attempting to compare the results of two predictive models: one using ArcGIS's OLS Regression tool, and another using Maxent SDM.

From what I can tell, Maxent SDM only provides the area under the curve (AUC) statistic. However, ArcGIS doesn't seem to provide that. So I have a few questions a result.

1) Does ArcGIS actually provide the AUC value from the OLS Regression output and I just don't know what it looks like?

2) Is there another variable that is comparable to AUC located on the OLS regression output?

3) Is there another tool I should be using to develop the model in ArcGIS that will give me AUC as an output?

Thanks!

[DanPatterson_Retired]

Are you referring to where in the interpretation output that might be found?

Interpreting OLS results—ArcGIS Pro | ArcGIS for Desktop

[SeanMcNulty]

I just searched for AUC and area under the curve on that page but didn't find anything. Is AUC not provided as an OLS regression statistic?

[DanPatterson_Retired]

I was just wondering if you were looking for a more common signficance representation.  Area under the curve ... in my field ... relates distributions and/or but not limited to... p-value interpretation of significance tests... but i was more curious why your comparative software only puts out one indicator to interpret results.  esri has gone overboard in their descriptors IMO, but everything is well covered and visually displayed.  I thought the display might twig an association between the two

[SeanMcNulty]

Oh, I don't have a software that specifically compares the two. But at the end of my research paper I want to be able to draw some sort of conclusion like "the AUC for the Maxent model is higher than the AUC for the ArcGIS OLS regression model and therefore the Maxent model predicted species distribution with a higher accuracy."

But Maxent ONLY provides AUC as a measurement of model accuracy, and although ESRI provides a ton of descriptors for the OLS regression (as you pointed out), AUC doesn't seem to be one of them. So I was wondering if there's an alternative way to compare them.

The only other alternative I can think of is trying to compare the number of correct predictions given a sample. Maxent will provide one AUC for the model developed using the training data and another AUC for sample data. But I don't know if I can do something like that with the regression analysis in ArcGIS as well.

[DanPatterson_Retired]

hmmm ... might I advise going against going down the "this one is better than that one path" .  I have supervised and been involved in too many theses/research projects to know that that approach rarely go well.

I will boldly suggest some areas of discussion (in no direct order) that I am sure your academic adviser has covered or would have covered...

• I would objectively go through a comparison of the results and what they show. 
• Does one environment provide a more comprehensive analysis than the other? 
• What results are produced by both toolsets. 
• Don't assume that the results you obtained using one toolset in your area will be applicable to the same degree in another area
• What are "correct" results (assuming ground truthing was involved). What defines 'correct' if it is a qualitative property (and even sometimes this is an issue with quantitative measures)
• You can summarize the results independently and say what the tool/procedure/whatever did and produced without doing a direct comparison
• You could go with the strengths and weakness approach (one weakness you identified is the lack of appropriate statistical summary in maxent)

Good luck... the only good thesis, is a done thesis ...

[SeanMcNulty]

Thank you for your input! This is more of a research-oriented final class project than a thesis (though we had to come up with a topic on our own) and my statistical background is sorely lacking. I will keep your advice in mind when I write the results section of my paper.