As the title suggests, what is it about the nature of GAMs that make them so popular in ecology? Is it their ability to model latent variables? If so, why wouldn't they be more universally popular, since everything is affected by more than meets the eye.
TL;DR: GAMs are useful models when specific functional relationships are not hypothesized.
Ecology as a science (and like many other sciences, particularly population sciences) often has hypotheses for its statistical models which do not take specific functional forms. For example, causal relationships may be specified qualitatively in terms of direction of effect (e.g., ‘species $A$ causes a decrease in species $B$ in some trophic system,’ but perhaps not ‘$A$ causes some specified $f(B,\Theta)$’ in the way that we often encounter in physics or chemistry). GAMs are an appealing kind of model in such circumstances, because they are agnostic to the functional form relating $A$ and $B$.
2$\begingroup$ Is this not also true of gradient boosted trees, random forests, and neural networks? GAMs have the interesting property that the functional form can be determined separately for each input feature (see e.g. here), rather than the model trying to find a single "joint" functional form. $\endgroup$ Dec 1, 2022 at 17:58
1$\begingroup$ @shadowtalker Re: "interesting property", I think this is just the additivity assumption in GAM? (And, for that matter all linear additive models, e.g., multiple linear regression, multiple GLM, etc.) $\endgroup$– AlexisDec 1, 2022 at 18:31
2$\begingroup$ Yes, that's a consequence of additivity, but your answer doesn't mention additivity at all! So it left me wondering what, if anything, is particularly appealing about an additive model as opposed to any other model. Is it a matter of interpreting the model? Is additivity an appealing property in ecology research for some domain-specific reason? Etc. $\endgroup$ Dec 1, 2022 at 18:50