search

salbeke / rKIN

Estimate Niche space using Kernel Density Estimates
4 stars 0 forks source link

Kernel issue #17

Open AylenM opened 1 year ago

AylenM commented 1 year ago

Hi! I'm trying NMDS scores in rKIN (following a paper idea) and I have found 95% and 75% kernel levels area in group B (not the other one) being exactly the same: -Method Group ConfInt ShapeArea 1 Kernel A 50 0.02000903 2 Kernel A 75 0.05167157 3 Kernel A 95 0.07948332 11 Kernel B 50 0.00819259 21 Kernel B 75 0.03357224 31 Kernel B 95 0.03357224 (Also, Why do the rows always have this numbering?)

Trying ellipses method resulted ok: Method Group ConfInt ShapeArea 1 Ellipse A 50 0.01877666 2 Ellipse A 75 0.03755332 3 Ellipse A 95 0.08115138 4 Ellipse B 50 0.01742192 5 Ellipse B 75 0.03484385 6 Ellipse B 95 0.07529630

For my understanding, my kernel results are incorrect and maybe the NMDS scores (with a lot of negative and near-to-zero values) are affecting the model performance.

Any help/suggestion/advice will be welcome! Thank you in advance!

salbeke commented 1 year ago

Hi Aylen, I don't think the function is broken, but probably more of a response to your small sample sizes (if I am recalling correctly). In reality for a 2D Kernel, we would have a minimum of 19 samples, but I believe you are well below that number. Thus there may just not be enough samples to provide a measurable difference between the two confidence intervals. Tough to tell without actually having your data though.