detfct not working on models with adjustment terms

[Greg Forcey]

I'm attempting to predict detection probabilities at various distances from my data and the models from Distance. I know the predict function does not do this, but I saw in some posts from 2017 and 2018 where it was suggested to use detfct to accomplish this:

detfct(seq(0,250,50), model$ddf$ds$aux$ddfobj)

What I've noticed is that this only works if the model has no adjustment terms. if there are any adjustment terms, detfct just returns a blank numeric vector.

Is there any way to use models with adjustment terms to predict the detection probabilities at defined distances? With my data, models with adjustment terms are often a better fit than models without them. If not, then is the next best solution to use the adjustment = NULL when fitting the models with the ds function?

Thank you,

~Greg

[Eric Rexstad]

Greg

I've had a look at detfct​ and it has had some revisions in the last couple of years. I agree it does not produce what you want in its current state.  Give me a couple of days to poke around with it and get back to you.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Greg Forcey <gregf...@gmail.com>
Sent: 10 January 2023 20:56
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] detfct not working on models with adjustment terms

 

--
You received this message because you are subscribed to the Google Groups "distance-sampling" group.
To unsubscribe from this group and stop receiving emails from it, send an email to distance-sampl...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/distance-sampling/015e35ca-56cd-4ccc-b6df-a93fe0d286b1n%40googlegroups.com.

[Eric Rexstad]

Greg

I've got a partial solution for you.  The detfct​ function gets used in the iterative model fitting process, so it is a workhorse. It expects to work with the actual data, hence I've not made it behave when I send a set of perpendicular distances that are not equivalent to the observed distances.

However, if I send the function a vector of distances from my data, I get success, regardless of the presence of adjustment terms.  The secret in the presence of adjustment terms is to also include the width=​ argument to detfct​.  Here are examples using the ducknest data shipped with the Distance package.

library(Distance)

data("ducknest")

unif <- ds(ducknest, key="unif")

hr <- ds(ducknest, key="hr")

hn <- ds(ducknest, key="hn")

hnadj <- ds(ducknest, key="hn", adjustment = "cos", nadj = 1)

hradj <- ds(ducknest, key="hr", adjustment = "cos", nadj = 1)

unif.detfct <- detfct(ducknest$distance, unif$ddf$ds$aux$ddfobj, width=2.4)

hr.detfct <- detfct(ducknest$distance, hr$ddf$ds$aux$ddfobj, width=2.4)

hradj.detfct <- detfct(ducknest$distance, hradj$ddf$ds$aux$ddfobj, width=2.4)

hn.detfct <- detfct(ducknest$distance, hn$ddf$ds$aux$ddfobj, width=2.4)

hnadj.detfct <- detfct(ducknest$distance, hnadj$ddf$ds$aux$ddfobj, width=2.4)

plot(ducknest$distance, unif.detfct, ylim=c(0.65,1), col="blue", pch=20,

     xlab="Distance", ylab="Detection probability")

points(ducknest$distance, hr.detfct, col="green", pch=20)

points(ducknest$distance, hn.detfct, col="salmon", pch=20)

points(ducknest$distance, hradj.detfct, col="black", pch=20)

points(ducknest$distance, hnadj.detfct, col="red", pch=20)

legend("topright", legend=c("unif", "hr", "hn", "hradj", "hnadj"),

       col = c("blue", "green", "salmon", "black", "red"), pch=20, cex=1.2)

Hope that's useful for you.

---

From: distance...@googlegroups.com <distance...@googlegroups.com> on behalf of Greg Forcey <gregf...@gmail.com>
Sent: 10 January 2023 20:56
To: distance-sampling <distance...@googlegroups.com>
Subject: [distance-sampling] detfct not working on models with adjustment terms

 

--

[Greg Forcey]

Eric, this seems to work. Thank you very much.