MixSIAR
MixSIAR copied to clipboard
brianstock
mixtures as sources?
Adding a vignette with worked example of a situation such as arises in intraguild predation or cannibalism where the mixture may itself be a source would be useful at some point. My own view is that ideally one would separate out large body predators from small body prey if that made sense ecologically (e.g. gape limited predators). Alternatively one might include the mixture as a source, and add a prior that sets the corresponding proportion to be small.
Interesting. I've only thought briefly about this for cannibalism, not intraguild predation (assuming for a multi trophic level model?)... seems like mixing models are limited in this regard. How do you prevent the mixture-as-source from being estimated as 100%? Not sure how a prior gets around this, then you're not really letting the tracer data inform the proportions (would have to be a strong prior, right?). How is that different than running the mixing model without the mixture-as-source, say there's 10% cannibalism and then reduce the proportions to sum to 90%?
Open to writing a vignette for this case if there's a good published example.
i think the intraguild situation arises if you aggregate species into guilds, and treat them as fuctional groups within a food web… but the canibalism issue is the obvoius situation.
I guess the TDF means that the model cant just assume its 100% eating itself? but given the uncertainties, i reckon that is a solution that comes out very high.
Using the prior to constrain it i guess means that either there is an assumption that cannibalism is low (which it really shouuld be because its just not sustainable from a population dynamcis sense to rely on high rates of cannibalism). Idelaly i suppose you might have some idea about what the cannibalism proportion from other data, but if youre using isotopes its very likely you have no idea!
Youre idea to reduce the proportions to sum to 90% is interesting, but im always uneasy about leaving out a source entirely for the same reasons im against aggregating sources (especially distinct sources) as per the reasons in our reply in the MEPS case.
Ill keep my eyes open for a good example for a vignette. It often crops up at least once per mixing model course i teach in someone’s data, and each time we kind of go “hmmm good question… .well you could sort of try…."
--
Dr Andrew Jackson, PhD, FTCD Associate Professor Irish Research Council Laureate School of Natural Sciences, Department of Zoology Trinity College Dublin, the University of Dublin Dublin 2, Ireland.
+353 1 896 2728 | [email protected]mailto:[email protected] Twitter: @yodacomplexhttps://twitter.com/yodacomplex http://www.tcd.ie/Zoology/research/http://www.tcd.ie/Zoology/research/groups/jackson/groups/jackson/
Trinity College Dublin, the University of Dublin is ranked 1st in Ireland and in the top 100 world universities by the QS World University Rankings.
On 4 May 2018, at 15:33, Brian Stock <[email protected]mailto:[email protected]> wrote:
Interesting. I've only thought briefly about this for cannibalism, not intraguild predation (assuming for a multi trophic level model?)... seems like mixing models are limited in this regard. How do you prevent the mixture-as-source from being estimated as 100%? Not sure how a prior gets around this, then you're not really letting the tracer data inform the proportions (would have to be a strong prior, right?). How is that different than running the mixing model without the mixture-as-source, say there's 10% cannibalism and then reduce the proportions to sum to 90%?
Open to writing a vignette for this case if there's a good published example.
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/brianstock/MixSIAR/issues/136#issuecomment-386620571, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADK1sEaVyGymtUNk_5CZkf_rBF8nUyIgks5tvGa4gaJpZM4TyRYz.
Ah, didn't think of the TDF. Maybe it would be prevented from estimating mixture-as-source as 100% then, and the prior wouldn't have to be too strong... but with the "fully Bayesian" / source fitting behavior, the model may be able to push the mixture-as-source back on top of the mixture. Could set n=10000 to turn this off for the mixture-as-source.
Good TDF estimates would be especially important in this case. Do you know of any experiments where people have measured the TDF for cannibalism?
Leaving out a source is just the "ultimate strong prior" in a sense, right?
I think the cannibalism / mixing model is super interesting, and something that would be testable in a lab setting. I've only read the abstract, but this might be an example for prehistoric humans: https://onlinelibrary.wiley.com/doi/pdf/10.1002/oa.2269
Interesting - I'll check that out.
Also, cannibalism could be occurring, but adults on juveniles (or bigger on smaller in general), which likely have different isotopic signatures. Not uncommon in fish I think. Agree with Eric this would be interesting to test in a controlled setting!
some good ideas here for us to try.
to be honest i hadnt really thought of leaving out a source as being a very strict prior… i think i’d have to simulate it to convince myself!… there goers another few hours of productivity!
--
Dr Andrew Jackson, PhD, FTCD Associate Professor Irish Research Council Laureate School of Natural Sciences, Department of Zoology Trinity College Dublin, the University of Dublin Dublin 2, Ireland.
+353 1 896 2728 | [email protected]mailto:[email protected] Twitter: @yodacomplexhttps://twitter.com/yodacomplex http://www.tcd.ie/Zoology/research/http://www.tcd.ie/Zoology/research/groups/jackson/groups/jackson/
Trinity College Dublin, the University of Dublin is ranked 1st in Ireland and in the top 100 world universities by the QS World University Rankings.
On 4 May 2018, at 15:54, Brian Stock <[email protected]mailto:[email protected]> wrote:
Ah, didn't think of the TDF. Maybe it would be prevented from estimating mixture-as-source as 100% then, and the prior wouldn't have to be too strong... but with the "fully Bayesian" / source fitting behavior, the model may be able to push the mixture-as-source back on top of the mixture. Could set n=10000 to turn this off for the mixture-as-source.
Good TDF estimates would be especially important in this case. Do you know of any experiments where people have measured the TDF for cannibalism?
Leaving out a source is just the "ultimate strong prior" in a sense, right?
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/brianstock/MixSIAR/issues/136#issuecomment-386626878, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADK1sIbYpVuAPDBgNdJ2VbsE4Zg_J4EZks5tvGuXgaJpZM4TyRYz.
cool example eric.
feeding experiments would be great. It might also reveal some interesting (annoying) TDFs if the distance between your own and your food’s stoichiometry matters. Gammarus are known for cannibalism and would be easy to work with… perhaps an undergrad project sometime (too late for next year’s lot here though)
--
Dr Andrew Jackson, PhD, FTCD Associate Professor Irish Research Council Laureate School of Natural Sciences, Department of Zoology Trinity College Dublin, the University of Dublin Dublin 2, Ireland.
+353 1 896 2728 | [email protected]mailto:[email protected] Twitter: @yodacomplexhttps://twitter.com/yodacomplex http://www.tcd.ie/Zoology/research/http://www.tcd.ie/Zoology/research/groups/jackson/groups/jackson/
Trinity College Dublin, the University of Dublin is ranked 1st in Ireland and in the top 100 world universities by the QS World University Rankings.
On 4 May 2018, at 16:12, Brian Stock <[email protected]mailto:[email protected]> wrote:
Interesting - I'll check that out.
Also, cannibalism could be occurring, but adults on juveniles (or bigger on smaller in general), which likely have different isotopic signatures. Not uncommon in fish I think. Agree with Eric this would be interesting to test in a controlled setting!
— You are receiving this because you authored the thread. Reply to this email directly, view it on GitHubhttps://github.com/brianstock/MixSIAR/issues/136#issuecomment-386632344, or mute the threadhttps://github.com/notifications/unsubscribe-auth/ADK1sJRiSKAYGZx29sABjRX1PTDe_eFMks5tvG_TgaJpZM4TyRYz.