Hi Donatien, the range of ar values is not very informative in itself, it depends on many confounding factors, the effects of which are hard to dissect. What is informative (because theoretical models -such as Rousset 2000- established its relationship with interesting population demographic parameters that can generally be inferred) is how fast ar increases with the logarithm of distance (in two dimensions). The inverse of the regression coefficient of ar on log-distance (over a particular range of distances) is related with (and allows estimating) the product of the effective population density and the effective dispersal parameter (see Rousset 2000 for details), assuming migration-drift equilibrium. It is this slope (or its inverse) that should better summarize the figure that you mention, and allow comparison with results from other studies. There is software available to infer these demographic parameters from genetic isolation by distance patterns, based on refinements of this theoretical expectation (such as Migraine http://kimura.univ-montp2.fr/~rousset/Migraine.htm, or Spagedi http://ebe.ulb.ac.be/ebe/SPAGeDi.html). Hope it helps.

Hi Juan José,

Your answer helps me in the sense that I did not know if I could interpret the scatter plot besides the inverse of the regression coefficient.

On other hand, I read in Rousset 2000 that we should not add pairwise individuals with a geographical distance smaller than σ in the linear model. So I used SPAGeDI to calculate σ but I have some trouble on the way I implemented it because I do not have the effective density population D. Unfortunately, I am currently in my third year of PhD and I spent 2 years to collect DNA samples according to the best coverage possible on the field, which is difficult to combine with a CMR.

So I used as proxy of density the number of adults by squared distance units (here, D = 280 adults / 70 000 000 m² = 4e-6 / m²) in order to compute σ and make the linear model without individuals distant from ≤ σ. If I do that, I will compute a maximal σ since more D increase and more σ decrease. So this would allow me to remove individuals which are too close each other to improve the linear models, but I think this is not really a good way to do this. So is that acceptable? And if not, how remove individuals geographically too close without compute σ?

In any case, thanks for your answer and your time

Best regard

Hi Donatien,

It's common practice to use different values of the unknown effective density (De), in order to estimate σ, such as D, D/2, D/5, D/10, where D is the estimated census density. That way you can check whether your σ estimates are sensitive to the assumed De, and if yes provide a range of σ estimates for a reasonable range of assumed De values.

I cannot think of other way to estimate σ from your data, nor of a criteria (other than an estimate of σ) for defining "nearby" individual pairs to be removed before inferring Nb from the IBD pattern.

You may want to provide a range of Nb estimates, based on the range of σ estimates (in turn obtained from the range of assumed De values).

Hope it helps.

Hi Juan José,

Thank you very much for your help. I believe i follow your recommendations and I will propose a range of D in order to compute σ, Nb and several linear models based on.

Thank you again

Best regards