The current model fits a polynomial term for detectability, following van Strien et al 2010. I am concerned this is numerically unstable, possibly because different combinations of parameter values have quite very similar fits. Yesterday (1/3/19) I wrote a simulation to create data for sparta. I found that sparta was not particularly effective in estimating the parameter values for a small but rich dataset (few sites but many visits per site per year). I wonder whether an alternative formulation would be better. Specifically, I envisage the detection phenology being fitted as a normal distribution. The beta1 parameter would estimate the mean detection date and beta2 the standard deviation.
To implement this we’d probably have to code up the density function:
f(x) = 1/(√(2 π) σ) e^-((x - μ)^2/(2 σ^2))
where μ is the mean of the distribution and σ the standard deviation. In R:
The f_x line would be a modifier on the basic detection model.
@JackHHatfield91 does this make sense?
I suspect this would be easy to implement in BUGS: the difficulty would be in editing getObsModel to allow ensure backwards compatibility.
The current model fits a polynomial term for detectability, following van Strien et al 2010. I am concerned this is numerically unstable, possibly because different combinations of parameter values have quite very similar fits. Yesterday (1/3/19) I wrote a simulation to create data for sparta. I found that sparta was not particularly effective in estimating the parameter values for a small but rich dataset (few sites but many visits per site per year). I wonder whether an alternative formulation would be better. Specifically, I envisage the detection phenology being fitted as a normal distribution. The
beta1
parameter would estimate the mean detection date andbeta2
the standard deviation. To implement this we’d probably have to code up the density function:f(x) = 1/(√(2 π) σ) e^-((x - μ)^2/(2 σ^2))
where μ is the mean of the distribution and σ the standard deviation. In R:which is testable as follows:
In BUGS:
The
f_x
line would be a modifier on the basic detection model.@JackHHatfield91 does this make sense? I suspect this would be easy to implement in BUGS: the difficulty would be in editing
getObsModel
to allow ensure backwards compatibility.