ModelsBefore_doing_any_models_we__.html

<h3>Models</h3><p>Before doing any models, we need to do different things:</p><ol><li><p></p><div>Expanded cumulative yield plots showing the start and end of the experiment, the rainfall, the yield standard error, etc…<br></div><p></p></li><li><p>Plot all variables against each other to understand correlations and establish if any are collinear and thus not mixable into the same model<br></p></li></ol><div></div><p>From the perspective of the least effort, we should do simple models first and if they are inconclusive, move up to more complex and complete ones.</p><div></div><p>Different approaches</p><ol><li><p>test proximal effects on yield of: temperature, tree size, herbivory, pollinators, soil, rain. Every effect that is not significant is removed, and the ones that are significant are replaced by their causative effect(s) and their interactions, if any. For example: if herbivory has an impact, we should replace it with herbivores. Then if herbivores have an impact, they should be replaced with the bird, bat, and insect predator data. Thus, we would be moving up the chain (or out of the onion) and trying to prove effects from more and more indirect and distant variables (that are more and more interesting and relevant to ecologists and practitioners). <br></p></li><li><p>SEM approach. That’s the point we always get to with Lisa. Is it feasible or not? Lisa thinks might not be possible due to low number of replicates (replicates at exclosure level, not at tree level)<br></p></li><li><p>Test simple relationships with models we can handle:<br></p></li><ol><li><p>yield ~ (insectivorous birds + insectivorous bats) * predatory ants<br></p></li><li><p>yield ~ ((insectivorous birds + insectivorous bats) * predatory ants * herbivory)^2<br></p></li><li><p>herbivory ~ (insectivorous birds + insectivorous bats) * predatory ants<br></p></li><li><p>etc…<br></p></li></ol><li><p>Start with full model, stepwise selection with AIC. Or: dredge with all non-correlated explanatory variables<br></p></li><li>Alternative packages: mboost?<br></li></ol><div></div>