Lecture 01 - The Golem of Prague
Rose / Thorn
Rose: I LOVE DAGS. Really interesting to think about the non-uniqueness of null models in ecology.
Thorn: the difference between regression/intervention – have i always just done regression?
Third Edition
- peach boxes instead of blue boxes
Causal Inference
- statistical models require scientific (causal) models
- correlation is a very limited measure of association
- association can occur without correlation
- causation requires intervention - it is not just the behaviour without intervention
- causal prediction = prediction of the consequences of an intervention (implications of changing one variable on another variable)
- knowing the cause of an action allows you to create predictions
- what happens if I do this?
- causal imputation = knowing the cause of an action allows you to reconstruct possible outcomes (i.e., what if I had done something else?)
- Even for description, causal models are required
DAGs
- abstract causal models: includes names of variables and their causal relationships
- tells you the consequences of an intervention
- tells you what you can decide/ask without additional assumptions
- facilitates you asking scientific questions
- each causal query requires a different model
Golems
- statistical models = golems
- often not possible to design and outline a null hypothesis that is meaningful to reject in observational science
- what is a null ecological community?
- think of good example/explanation for no null ecology/previous two slides
- takeaway is that null hypothesis does not give you cause/process behind outcome
- what is your null? is it unique?