[Basta-users] post-fledging survival - recapture probability etc.
Fernando Colchero
colchero at imada.sdu.dk
Thu Sep 17 14:49:43 CEST 2015
Dear Dries,
It’s still going to take a while to implement that, unfortunately, we don’t have the man power to do it, although it is in our priority list…
By looking at your traces, I think you should run longer the model, particularly for the b1.FRAGNG. Still, your CIs will be somewhat large, but that should help reducing them. What happens is that, when there’s a lot of autocorrelation, the SE’s end up being inflated.
Best,
Fernando
Fernando Colchero
Assistant Professor
Department of Mathematics and Computer Science
Max-Planck Odense Center on the Biodemography of Aging
Tlf. +45 65 50 23 24
Email colchero at imada.sdu.dk <mailto:colchero at imada.sdu.dk>
Web www.sdu.dk/staff/colchero <http://www.sdu.dk/staff/colchero>
Pers. web www.colchero.com <http://www.sdu.dk/staff/colchero>
Adr. Campusvej 55, 5230, Odense, Dk
University of Southern Denmark
> On 17 Sep 2015, at 14:44, Dries Van de Loock <dries.vandeloock at ugent.be> wrote:
>
> Dear Fernando,
>
> thanks for your rapid and clarifying reply.
>
> About the resighting covariates. How realistic are the prospects of implementing this feature in the near future ?
>
> About the large CI. According to the model output, all parameters should have converged. I have added the traces for all the parameters to the dropbox folder <https://www.dropbox.com/sh/kk1wztsftfljbw5/AAC4OHKT-pNXepioaR4qf78oa?dl=0>.
>
> Best wishes,
>
> Dries
>
>
> 2015-09-17 10:46 GMT+02:00 Fernando Colchero <colchero at imada.sdu.dk <mailto:colchero at imada.sdu.dk>>:
> Dear Dries,
>
> Thanks for sending this very detailed explanation. I’ll try to answer to each of your points below:
>
>> (i) Recapture probability is around 10 %. We however did not check for presence/absence of the bird every day after fledging. Usually on the 5th/6th day (range 3-9). This thus underestimates recapture probability dramatically. We noted on the BaSTA website that future versions would include covariates on recapture and recovery probabilities. We could for example add "search effort" as covariate, stating the specific days when we actually looked for that individual. Is there a possibility of trying this out already ?
>
> Fernando: Unfortunately we have not yet been able to implement this.
>
>> (ii) Our confidence intervals on the mortality rates just after fledging are big. To that extent that it obscures the real shape of the mortality function when added to the graphs. See PlotFancyBaSTA with CI and regular plot without CI <https://www.dropbox.com/sh/kk1wztsftfljbw5/AAC4OHKT-pNXepioaR4qf78oa?dl=0>. Because we only searched for the juveniles some days after fledging, the first resightings are only done on day 3 and later. Is it possible that the particular sampling strategy is on the origin of this big CI's ?
>
>
> Fernando: By looking at your plots, particularly the PlotFancy, I suspect that not all your parameters had converged… Did you have a chance at checking the traces?.
>
>> (iii) Based on the lowest DIC from running all basic models (no covariates), we selected the WEIBULL model with a Makeham shape. Adding the Makeham constant avoids creating immortal individuals when having a declining mortality shape (i.e. when 0 < b0 < 1) by setting a "base mortality". How do we interpret this value when having other shapes (b0 <= 1) ? Why is this in these cases not (partially) embedded within the scale parameter (b1) ?
>
> Fernando: Actually embedding the third parameter (c) into b1 does not solve the problem of starting mortality at 0, inherent to the Weibull model, while the intended effect of the Makeham constant is to shift up or down the mortality. If it’s embebed in b1, then you are only changing mortality proportionally.
>
>> (iv) The serial autocorrelation of most coefficient estimates appears to be rather high (> 0.4). Would you recommend readjusting the thinning argument or other model settings ? Or does this not really effect our estimates and outcome ?
>
>
> Fernando: Yes, I would recommend running the model for more iterations with a large thinning argument. If you run it in parallel then you can run more chains for less iterations but with higher thinning, that’ll be even better and it would save you quite a bit of time.
>
>> NB : we noticed a discrepancy between the default values in the package versus the ones listed in the BaSTA vignette (niter = 11000, burnin = 1001 and thinning = 20 vs. niter = 50000, burnin = 5001 and thinning = 50). Which settings to you recommend as default values ?
>
>
>
> Fernando: Thanks for spotting this! We’ll check the settings.
>
> Best,
>
> Fernando
>
> Fernando Colchero
> Assistant Professor
> Department of Mathematics and Computer Science
> Max-Planck Odense Center on the Biodemography of Aging
>
> Tlf. +45 65 50 23 24 <tel:%2B45%2065%2050%2023%2024>
> Email colchero at imada.sdu.dk <mailto:colchero at imada.sdu.dk>
> Web www.sdu.dk/staff/colchero <http://www.sdu.dk/staff/colchero>
> Pers. web www.colchero.com <http://www.sdu.dk/staff/colchero>
> Adr. Campusvej 55, 5230, Odense, Dk
>
> University of Southern Denmark
>
>> On 16 Sep 2015, at 15:13, Dries Van de Loock <dries.vandeloock at ugent.be <mailto:dries.vandeloock at ugent.be>> wrote:
>>
>> Dear BaSTA community,
>>
>> I'm involved in the same project as Diederik Strubbe on post-fledging survival of a tropical passerine (see archive <http://lists.r-forge.r-project.org/pipermail/basta-users/2015-August/thread.html>).
>>
>> Nests were monitored until fledging and juveniles were relocated every 5 to 6 days after fledging.
>>
>> Our input matrix is formatted accordingly : the birthyear is considered the point of fledging and is set to "1" for all individuals. The resighting matrix starts at day 1 and is "0" for all individuals on that day. Day 2 in the matrix is thus the first day after fledging. See screenshot <https://www.dropbox.com/sh/kk1wztsftfljbw5/AAC4OHKT-pNXepioaR4qf78oa?dl=0> from RStudio.
>>
>>
>> When running the models, we bumped into following questions :
>>
>> (i) Recapture probability is around 10 %. We however did not check for presence/absence of the bird every day after fledging. Usually on the 5th/6th day (range 3-9). This thus underestimates recapture probability dramatically. We noted on the BaSTA website that future versions would include covariates on recapture and recovery probabilities. We could for example add "search effort" as covariate, stating the specific days when we actually looked for that individual. Is there a possibility of trying this out already ?
>>
>> (ii) Our confidence intervals on the mortality rates just after fledging are big. To that extent that it obscures the real shape of the mortality function when added to the graphs. See PlotFancyBaSTA with CI and regular plot without CI <https://www.dropbox.com/sh/kk1wztsftfljbw5/AAC4OHKT-pNXepioaR4qf78oa?dl=0>. Because we only searched for the juveniles some days after fledging, the first resightings are only done on day 3 and later. Is it possible that the particular sampling strategy is on the origin of this big CI's ?
>>
>> (iii) Based on the lowest DIC from running all basic models (no covariates), we selected the WEIBULL model with a Makeham shape. Adding the Makeham constant avoids creating immortal individuals when having a declining mortality shape (i.e. when 0 < b0 < 1) by setting a "base mortality". How do we interpret this value when having other shapes (b0 <= 1) ? Why is this in these cases not (partially) embedded within the scale parameter (b1) ?
>>
>> (iv) The serial autocorrelation of most coefficient estimates appears to be rather high (> 0.4). Would you recommend readjusting the thinning argument or other model settings ? Or does this not really effect our estimates and outcome ?
>>
>> NB : we noticed a discrepancy between the default values in the package versus the ones listed in the BaSTA vignette (niter = 11000, burnin = 1001 and thinning = 20 vs. niter = 50000, burnin = 5001 and thinning = 50). Which settings to you recommend as default values ?
>>
>>
>> We performed all analysis using BaSTA 1.9.4
>>
>>
>> Your feedback and suggestions are greatly appreciated,
>>
>> Best wishes,
>>
>> Dries
>>
>> -----
>> Dries Van de Loock
>> PhD student
>> Terrestrial Ecology Unit
>> Department of Biology
>> Ghent University
>> KL Ledeganckstraat 35
>> B-9000 Ghent, Belgium
>> Phone: +32 (0)9 265 50 39 <tel:%2B32%20%280%299%20265%2050%2039>
>> http://www.ecology.ugent.be/terec/ <http://www.ecology.ugent.be/terec/>_______________________________________________
>> Basta-users mailing list
>> Basta-users at lists.r-forge.r-project.org <mailto:Basta-users at lists.r-forge.r-project.org>
>> https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/basta-users <https://lists.r-forge.r-project.org/cgi-bin/mailman/listinfo/basta-users>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.r-forge.r-project.org/pipermail/basta-users/attachments/20150917/dba304c7/attachment-0001.html>
More information about the Basta-users
mailing list