CDF and core and peripheral use

[Thomas Collins FAL]

Hi all,

A question regarding the relationship between the CDF in regards to core and peripheral use.

I want to use CDF to acknowledge the probability of use for a given location. So, where in a group of groups distribution is an event (there are many groups living in close proximity on a peninsula)? I’m getting flak from my supervisor for using CDF in this ’raw form’. I had thought that it was acceptable to take the CDF values of locations and plot them for different reproductive groups asking whether there were differences among those groups.  

What the supervisor wants appears to be categories of use – so the core and the periphery. To me, I have shown these areas with the CDF value – it shows the 50% and the 95% quite clearly, and I will surely discuss core- and peripheral-use in my write up (a masters project). But, in my head, it isn’t that they either occur in the core or the periphery, it’s just that there are differences. 

Christen, you discuss core home range in “A new kernel density estimator for…” (P. 578). I understand from this that core (50%) use is a good benchmark because of its relationship to the median. I have also read Powell who encourages the most "parsimonious simplification" and to think of the biological reasoning behind using thresholds. In my case, I am asking if there is a difference between those reproductive groups? I think it worth remembering that groups are separated somewhat arbitrarily, right? Groups still interact with regularity.   

My questions:

Is my thinking about this all wrong? 

could it be dogmatic to use core/periphery?

Am I doing something crazy by discussing CDF in a bounded but continuous way? 

i suppose my main question is, can anyone please point me toward literature that discusses these themes?

Very best and many thanks,

Tom

 

[Christen Fleming]

Hi Tom,

If I understand your inquiry, I would say that the general advice of the statistics literature is to avoid categorizing/discretizing continuous variables if you can help it:

• https://onlinelibrary.wiley.com/doi/abs/10.1002/sim.2331
• https://bmcmedresmethodol.biomedcentral.com/articles/10.1186/1471-2288-12-21
• http://www.ajnr.org/content/32/3/437.short
• https://onlinelibrary.wiley.com/doi/abs/10.1002/9781118445112.stat04857
• https://search.proquest.com/openview/96804e1fb78526aee66f9ab7ebb2c888/1
• https://journals.sagepub.com/doi/abs/10.5127/jep.008310
  

When reporting a predicted area of use (which has to summarize an entire distribution), I think it's a little more understandable to make a choice.

Best,

Chris

[Thomas Collins FAL]

Thanks for that, Christen. 

best, 

T

[Patrick Green]

Hi Tom and Chris,

I had a follow-up to this thread that I'm hoping you can help me with. I'm planning to use the same data that Tom used to calculate a metric of "territory value"--the idea being that an event (here, a fight) that occurs closer to the focal group's territory center, as opposed to the rival group's territory center, should be won more often by the focal group. I was thinking that I could calculate the CDF values for each group relative to the location of the event and simply subtract them: focal CDF - rival CDF. In this case, more negative values would represent a fight closer to the focal group's center (e.g., focal group core = 0 CDF, rival group periphery = 1 CDF, relative score = -1).

Is it feasible to use the CDF in this way: as a continuous variable that can be manipulated and included as a predictor in a mixed model? I just want to make sure there isn't any statistical or other reason for avoiding using this metric in this way.

Happy to clarify if this question is confusing, and thanks very much for the help!

Patrick

[Christen Fleming]

Hi Patrick,

I would think that density (PDF) dependence would make more sense for these kinds of interactions. Also, I don't know what other predictors you are considering, but having attraction to your own group's density would interact strongly with any predictor that explains site fidelity.

You might also look at the literature on MHRA scent marks for ideas.

Best,

Chris