Isotope variance without repeated, colocated samples

[Keenan Ganz]

Hello,

I am working on an isoscape project with timber samples, and after a few months of working with IsoriX I was very happy to discover this google group exists :)

I understand that IsoriX can estimate how the "background" variance of an isotope changes among locations. However, in my work I am finding that timber samples seem to have a fairly constant background isotope variance. For instance, we get an insignificant Levene test among disparate sites where we have duplicate samples. This is interesting because most of our data are single samples spaced relatively far apart, so it is difficult to explicitly estimate how isotopes vary over space.

Let's say I assume that isotope variance does not change as a function of location in my data. I could estimate variance in all locations as the nugget of a semivariogram. Then, my question: how do I incorporate a constant variance in an IsoriX assignment test? I can think of two possible ways:

1. Specify the resid.model argument in fitme to use this constant.
2. Ignore variance estimation with IsoriX altogether and use this variance in the assignment test.

I am leaning towards the latter for the sake of simplicity, but I welcome your input.

Best,

-Keenan

[Alexandre Courtiol]

Hi Keenan,

Welcome to the IsoriX micro-community.

Considering a constant variance in the resdidual model should be doable without having to "mess around".

I did not try it recently but I expect that setting all the inputs for disp_model_fix and disp_model_rand in isofit() to FALSE should imply a constant variance.

That should be easy to check by looking at the corresponding model output or by plotting the relevant isoscapes.

Let me know if that does not work, and I would have a closer look.

Best,

Alexandre Courtiol, www.datazoogang.de

--
You received this message because you are subscribed to the Google Groups "IsoriX" group.
To unsubscribe from this group and stop receiving emails from it, send an email to isorix+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/isorix/555863b2-e9ef-453f-b963-2e14d6ba2163n%40googlegroups.com.

[Keenan Ganz]

Hi Alexandre,

Thanks for your input. I was able to get a constant residual variance, and it matches what I expect. I did have to cheat a little by setting n=2 for all the input data so that isofit would actually consider the constant variance column.

A follow-up question. I notice that mean_respVar = mean_predVar + mean_residVar. To me this is saying that the total variance in the isotope of interest is the sum of uncertainty in prediction and residual uncertainty, which makes sense. However, in 3.7.1 of the isoscapes book it says that only mean_predVar is used for assignment. Is there a reason why assignment does not consider residual variance? To me, residual variance is coming from whatever fractionation process affected the input data, so our assignments should reflect that uncertainty.

Appreciate your input,

-Keenan

[Alexandre Courtiol]

Hi Keenan,

Am I correct to assume you are not performing any calibration? (i.e. your isoscape is made on wood, and you assign wood)

If so, then the residual variance is already accounted for during the assignment.

Indeed, in that case the respVar is used, which is the sum predVar + residVar.

You can see that in the code, here: https://github.com/courtiol/IsoriX/blob/35fb6a7b0be9d3140a6f235e433afe03bdd1db85/IsoriX/R/isofind.R#L311

++

Alex

To view this discussion visit https://groups.google.com/d/msgid/isorix/b2d40ffc-24f9-44f3-bb8b-db97ddf8d3ffn%40googlegroups.com.

--

Alexandre Courtiol, www.datazoogang.de

[Keenan Ganz]

Yes, there is no calibration in my workflow. Thank you for pointing out that detail in isofind and for your prompt responses to this group. Now I'm confident that I am using the correct variance in the assignment test. I feel like I've learned more about spaMMs in the past day than in the prior month :)

-Keenan

[Alexandre Courtiol]

You are very welcome Keenan.

++