[R] oldlogspline probabilities

[David Winsemius]

On May 17, 2012, at 3:16 AM, Beutel, Terry S wrote:

> I using oldlogspline (from logspline package) to model data  
> distributions, and having a problem.
>
> My data are search area sizes. They are based on circular search  
> radii from random points to the nearest edge of the nearest grass  
> tussock. Search area sizes are distributed from 0 (the random point  
> intercepts a tussock) and upwards (as points are further from any  
> tussocks). The density of all my distributions (using doldlogspline)  
> are highest at size=0, and decline as search area increases.
>
> I am most interested in the probability of a value in the  
> distribution being equal to 0 (ie probability of a direct hit on a  
> tussock). I know I can just use the proportion of actual hits, but  
> am curious to compare this to an estimate from a density estimation.  
> Unfortunately when using poldlogspline, this probability is always=0  
> (simulated data example is below). How can this be, given that the  
> density is highest at area=0?
>
>> simdat<-c(rep(0,8),rexp(92))
>> myspline<-oldlogspline(simdat,lbound=0)
>> poldlogspline(fit=myspline, q=0)
> [1] 0
>
> Any help to work out the probability of an area value in my  
> distribution = 0 would be appreciated
>
Probabilities often are zero over a zero-length interval , even if  
their densities are positive. Surely this tussock has a finite width?

 > logspline::poldlogspline(fit=myspline, q=0.01)
[1] 0.04138533

-- 
David Winsemius, MD
West Hartford, CT