Very informative video with a refreshing amount of humour (which is rare to find in the world of statistics). Never had so much fun watchin this kind of videos, congrats haha! I would love to see more videos on GLMMs cause its cool but also complicated af
Absolutely amazing! thanks so much! Another request, what to do with data that has **a lot** of zeros? Please keep posting videos like this; you are indeed amazing at explaining what is going on and what we should be looking for! Thanks!
This was an AMAZING video!!! I was told to do GLMM's for my undergraduate thesis, but I had no idea how to decide when something should be a random effect versus a fixed effect. Should I be leaving things as random effects (i.e. individual) even if it isn't really explaining anything? I'm wondering if it changes the accuracy of the model by including them!
Hmmmm, usually I would keep them in there. Always make sure that your model is passing DHARMa checks. If it is (with the random effect included in your model structure). You can then mention that ID are not explaining much variation, probably because your treatments are doin' the job! Overall, it is tends to be a good idea to be like -- "hey! this is a potential source of variation; maybe one mouse (or frog or penguin) was acting like A FOOL." -- then you see it's not-- that's good to be able to say!
Can you do a video explaining what family of curve to choose? Here you have picked gaussian each time. I am struggling with this step in my own glmm's. Thanks for all your help so far, this was an interesting example!
Commenting for others reference - I believe the family is determined by the type of response variable. For continuous data, you'll want to use gaussian, count is well fit for the poisson, and binomial is best for presence/absence data.
Very specific but should you have expertise in setting up MaxEnt with SWD files to run several projections out of the same model...videos would be highly appreciated 🙃
Hey, thank you for your videos. Very informative. I have a question. My summary matrix is composed of species (around 18 thousand species), coordinates, species body sizes, and environmental variables, and those species occupy different oceanic basins. The models I am trying to build have size as the response variable (the question is related to what explains body size variation across the world`s oceans). The limitation of my database is that each species has a unique size measure. That is, the same species appears in multiple oceanic basins with the same size. In that case, I was told that I should use species as a random effect. Do you think that this is the best way to go? Any suggestions?
I mean that is definitely better! But man oh man, try recapturing bank voles in the middle of an enormous forest over a summer. They are tiny little jerks!!!
Hi, thanks for the explanation of the GLMM analysis. I have a question: Do you applied DHARMA::simulateResiduals after created the last model? Thanks :D
Very informative video with a refreshing amount of humour (which is rare to find in the world of statistics). Never had so much fun watchin this kind of videos, congrats haha! I would love to see more videos on GLMMs cause its cool but also complicated af
Great explanation, you have a talent for explaining complex things well !
The thumbnail tho!!!! LOL why did you have to make me feel old?! LOL!
antarctic research student here! Super helpful and interesting video!!
super helpfuil! Would love a video showing how to compare models (and esp mixed effect models) for best fit!
Absolutely amazing! thanks so much! Another request, what to do with data that has **a lot** of zeros? Please keep posting videos like this; you are indeed amazing at explaining what is going on and what we should be looking for! Thanks!
Thanks a lot! This video was very helpful!!!
This was an AMAZING video!!! I was told to do GLMM's for my undergraduate thesis, but I had no idea how to decide when something should be a random effect versus a fixed effect. Should I be leaving things as random effects (i.e. individual) even if it isn't really explaining anything? I'm wondering if it changes the accuracy of the model by including them!
Hmmmm, usually I would keep them in there. Always make sure that your model is passing DHARMa checks. If it is (with the random effect included in your model structure). You can then mention that ID are not explaining much variation, probably because your treatments are doin' the job! Overall, it is tends to be a good idea to be like -- "hey! this is a potential source of variation; maybe one mouse (or frog or penguin) was acting like A FOOL." -- then you see it's not-- that's good to be able to say!
Can you do a video explaining what family of curve to choose? Here you have picked gaussian each time. I am struggling with this step in my own glmm's. Thanks for all your help so far, this was an interesting example!
Commenting for others reference - I believe the family is determined by the type of response variable. For continuous data, you'll want to use gaussian, count is well fit for the poisson, and binomial is best for presence/absence data.
I recognise that mixed effect model schematic on the thumbnail from Silk, Harrison & Hodgson 2020 PeerJ
Yes! Beautifully written by some great researchers ;) I linked the paper in my GitHub, but added it now to the video description as well!
Very specific but should you have expertise in setting up MaxEnt with SWD files to run several projections out of the same model...videos would be highly appreciated 🙃
Hey, thank you for your videos. Very informative. I have a question. My summary matrix is composed of species (around 18 thousand species), coordinates, species body sizes, and environmental variables, and those species occupy different oceanic basins. The models I am trying to build have size as the response variable (the question is related to what explains body size variation across the world`s oceans). The limitation of my database is that each species has a unique size measure. That is, the same species appears in multiple oceanic basins with the same size. In that case, I was told that I should use species as a random effect. Do you think that this is the best way to go? Any suggestions?
No no no... I say to have at least 30 in each group.... If the data is sparse, it can cause problems...
I mean that is definitely better! But man oh man, try recapturing bank voles in the middle of an enormous forest over a summer. They are tiny little jerks!!!
Hi, thanks for the explanation of the GLMM analysis. I have a question: Do you applied DHARMA::simulateResiduals after created the last model?
Thanks :D