Different results depending on who's "first" in i/o

[geof]

Hi everybody. I'm a relative newbie to m/rc and an utter newbie with MARK/RMark. I can sense that my question is naive, but I can't make sense of things despite poring over the early chapters & 14 & appendix C, and quick reviews of Seber and Borchers et al.

I have a simple closed population with two independent nearly simultaneous observers and an independent matching mechanism to identify recaptures. My question is: why do I get very different answers depending on which observer I call the "first", whereas with (e.g.) a Petersen estimator N=n1*n2/m2 I get the same answer regardless of who corresponds to n1?

Here are counts of capture histories for a toy example.
10 35
01 45
11 14
and results (via RMark)
g=c(rep("11",14),rep("10",35),rep("01",45))
try2=data.frame(ch=g) ; try2$ch=as.character(try2$ch)
mark(try2,model="Closed")
<...>
Real Parameter p
1 2
0.1200178 0.1200178
Real Parameter c
2
0.2857144
Real Parameter N
[,1]
416.1087

And now reverse the roles of the observers.
10 45
01 35
11 14
with results
g=c(rep("11",14),rep("10",45),rep("01",35))
try2=data.frame(ch=g) ; try2$ch=as.character(try2$ch)
mark(try2,model="Closed")
<...>
Real Parameter p
1 2
0.433398 0.433398
Real Parameter c
2
0.2372881
Real Parameter N
[,1]
137.9453

As I tilt things further in the direction of example 1, phat and Nhat quickly become absurd (e.g. 11=14,10=30,01=50 gives Nhat=96268).

Whatever I'm misunderstanding also applies to the Huggins model. Thanks to anyone who can set me straight.

Geof

[jlaake]

Look at your parameter estimates and you'll see that you are not fitting the Petersen estimator. What you fitted was model Mb and Petersen is Mt. Mb will be order dependent because the recapture probability (c) differs from the capture probability (p) and the recapture probability is that those seen on the first occasion are re-seen on the second occasion. The default model for RMark is p==~1 and c=~1. The Petersen estimator is p=list(formula=~1,share=TRUE). See edwards.eberhardt example in the help file. If you are going to use RMark you should also look at any example data sets and read the workshop notes as well.

If you are analyzing independent observer sighting data you should look at DISTANCE which can do tht type of analysis with the mrds R package that comes with DISTANCE.

PS. This was really an RMark question and should have been in that forum. I'm more likely to reply if RMark is also in the subject line.

regards --jeff

[egc]

PS. This was really an RMark question and should have been in that forum. I'm more likely to reply if RMark is also in the subject line.

regards --jeff

I'll move this topic to the RMark subforum as a result.