egc wrote:Tapirus wrote:Hi,
When I run any Robust Design Model (RDM) in RMark with only two primary occasions, the R crashes. Jeff (principal author of RMark package) gently told me that the RDM runs only with three or more primary occasions.
Unfortunately, I have a bunch of places where I wanna estimate the population size (using RDM) before and after a determined impact (turtles before and after the building of hydroelectric ponds). My design is BACI (before-after, control impact); therefore I have only two primary occasions for every place. Has anyone some suggestion for me overcome this RMark limitation?
Thanks.
Two issues:
1\ this might be an RMark question, and should probably have been posted there. I'll leave it here for now.
2\ So, Jeff Laake, who created RMark, and knows more than just a little bit about this (and related) subjects than most people, made a suggestion, that you're now hoping others will help you avoid? Seriously? Your problem is limited data. Period. There is no trick to get around that, and it isn't particularly a limitation of RMark. With only 2 primary samples, there is precious little you can do that will be of much interest. if all you want to do is look at abundance before and after, then simply run closed abundance estimates for each primary period separately, as a standalone analysis, and then do whatever you think appropriate with those estimates.
1 - Jeff is really a great scientist, and gently gave me some advisement. Despite his great skills, he is very polite and respectful. My experiences have demonstrated that education and respect inform a lot about the quality of the scientist.
2 – I put this post at this section because Jeff suggested me, by email, to post this exact question on this exact section. Sorry, if I misunderstood it, Jeff. You could exclude it, if it is necessary.
3 – I consider the impossibility of running RDM with two occasions a limitation in RMark, because the original idea behind the RDM was to estimate both the abundance within primaries occasions and survivorship among them. This is the classic design, and it is necessary just two occasions to explicitly estimate these parameters (N, phi and p).
4 – Analytical advances in RDM included two further parameters, named GammaPrime and GammaDoublePrime, to deal with movements “in” and “out” the sampling area. Although the inclusion of these two parameters (an wonderful new parameterization) opened new opportunities in model fitting, the importance of them could be small (if not null) for lots of organisms, questions and sampling designs. Actually, I guess that it is still possible to run the classic RDM in Mark. Conceptually, it does not make sense eliminate the possibility to fit the classic model. Because this I consider the absence of the classic RDM as a limitation, since the classic model could be the most adequate for lots of biological questions.
5 – I completely agree that 2 population snapshots usually are few information to take big conclusions. However, the science walks with good ideas, strong theory, appealing hypotheses, not only with big data. I have a bunch of pairs of occasions under a controlled field experiment. This kind of natural experiments are rare to see and hard to get in ecology/conservation. Therefore I believe that my data can provide an interesting knowledge.
6 – I am particularly interested in estimating explicitly the survivorship (not Nt+1/Nt). So, two closed estimates will not fit my needs. Anyway, thanks for the suggestion.
7 – I am scientist, and consequently, I am not constrained by the opinion of others. Poor is that scientist that declines of an idea because some “authorities” said that this is “impossible” . “Really, do you do this?” “Really, do you work only in ideas that someone tell you?” "Seriously, do you do only what the authority does?” Sorry, I am not this kind of investigator.
8 – I spent this afternoon thinking in alternatives to overcome what I considered a limitation. One possibility would be: Split my first primary occasion in two occasions (the first will be composed by only two secondary occasions, or only one if you have few secondary occasions), and to fix the survivorship parameter between occasions 1and2 equal 1. Now I have 3 occasions (the RMark runs
), and as I am not interested in the survivorship between 1and2 (artificial, all individuals must be there yet), I can estimate the survivorship between 2and3 (my initial goal).
Ironically, the best models were those which I fixed both GammaPrime and GammaDoublePrime equal 0. That means that these parameters are completely useless for my organism/design, that means that the classic model still has its place in population ecology science, that means that the absence of the classic RDM in RMark is a limitation of this amazing package.
9 – I have thought about other possibilities, and I could share it if someone desire.