-
Notifications
You must be signed in to change notification settings - Fork 14
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Shared contributions in ANOVA #120
Comments
Hi Weverton, these are good questions - as coauthor of the Leibold paper, so let me try to clarify:
|
I just wanted to ask if the above Anova function is bugged or If I have a bad installation. I initially thought I just had a very bad fitting model and came looking here for solutions, but I've just tried replicating the above code verbatim and my results are very different as shown below. Nagelkerke values almost looks normalised?
Unsure if McFadden is affected by this also but results look more reasonable.
Session info below.
|
Hi, After some further investigation, I think this function is not working as intended in later packages. A working example below of differences between v1.0.4 and v1.0.5. I have been able to replicate this issue on Windows, Fedora and in the rocker/verse docker
Results from sjSDM 1.0.4
Results from sjSDM v1.0.5, the list names are different in this version to get the shared R2 values
|
Hi @dansmi-hub, |
Hi Maximilian,
I have a question regarding the shared contributions in ANOVA. If I understand correctly, after reviewing the get_shared_anova function, the shared contributions are calculated in the following manner:
However, in this recent paper, the authors mentioned that they also followed the approach of Leibold et al. (2022), but they calculated the shared R2 differently by employing a Type III ANOVA. They explained that:
“The fractional variance explained by each model component and interaction is determined using a Type III variance partitioning approach (Peres-Neto et al. 2006), working backwards from the most complex model to identify single effects and overlap terms (Fig. 1b). Then, to more efficiently summarise these seven quantities, we again follow the approach of Leibold et al. and group the sub-partitions of the total variance explained into three partitions of interest. The sub-partitions overlapping with codistribution (Venn segments ‘f', ‘g' and ‘e' in Fig. 1b) are assigned to the spatial and environmental partitions, respectively, because the flexible latent variables are likely to efficiently capture residual variance in the combined models. The variance explained by the combination of environmental and spatial predictors (Venn segments ‘g' and ‘d') is split 50:50 between them.”
They used the following formula to compute the shared contributions:
I ran a simulation to see the difference between the approaches, and it appears that the method used in this paper in Ecography gives greater contribution to spatial factor and less contribution to environmental and biotic factors:
The sjSDM results make more sense to me, but I'd like to know if I'm understanding the analysis correctly or if I'm missing something. Is there a conceptual distinction between the two methods? If so, what is the "best" method?
Thanks!
The text was updated successfully, but these errors were encountered: