Dear users and developers,
I have a general question concerning the output data that is obtained from the InVEST Recreation and Tourism Model. When looking at one pixel of AOI, would the usage of similar predictors (i.e rivers, wetlands and water bodies) have an impact on the results acquired? Is there a risk of double-counting an area that has layers of similar predictors?
Your thoughts are highly appreciated!
Liyana
Hi Liyana, could you be more specific about the results you are concerned about? Is it pud_results.shp, regression_coefficients.txt, or scenario_results.shp? And which column from within the table?
Hi Dave,
mainly the result of the estimate in the regression_coefficient.txt for each predictor. For example, in my study area, I have three separate predictors for recreation: wetlands, lagoons, and rivers. However one of the value in the wetlands’ attribute table already includes areas with lagoons and rivers. So I am wondering if that could influence the result in the reg. coeffs, and that there might be double-counting. I hope I am able to clarify my question. Here I attach an example of my result with 10 different predictors, with three similar predictors: wetlands, lagoons, rivers. regression_coefficients.txt (851 Bytes)
Appreciate your input in the matter!
Thanks,
Liyana
Thanks for the clarification. You are correct to be concerned about two different predictors representing some of the same information. The more general form of your question is something like “what are the consequences of correlated predictors in a linear regression?”. That’s a big question, this resource looks like a decent entry point: https://stats.stackexchange.com/questions/86269/what-is-the-effect-of-having-correlated-predictors-in-a-multiple-regression-mode
To give some very general advice, as much as possible you probably want each predictor to represent only one “thing”. That way the regression coefficient (aka estimate) can be interpreted as the effect of that “thing”. The “thing” could be very specific or quite general. For example, you might modify your wetlands to remove the features that are already represented by other predictors, so you have a specific wetlands predictor. Or you might merge predictors into a single one so that you have a general “wetlands & lagoons” predictor. That’s a good option if you suspect that wetlands and lagoons are functionally similar when it comes to influencing visitation rates.
Hi Dave,
thanks a lot for your input. It makes sense to ensure that all predictors should represent one specific thing that is not correlated.
Liyana