@statistical-rethinking
11/02/23 08:55:20
- Knowing the cause, means being able to predict the consequences of intervention.
- An example is shown of a panda, that some netweork identifies as such with a confidence, even a tiny amount of noise completely distorts the computers prediction. Also, even some filtering, that to us shows an image of, still, clearly a panda is a completely different prediction by the machine of a gibbon.
- The lesson here is that the model doesn’t have any relation between objects, it’s using statistics to imitate but has no guarantee of correctness.
- Statistical models contain no information about causation, they don’t know if it’s the trees that causes the wind or the wind that causes the trees.
- Another model is needed for causation relationships(?).
- Didn’t get the rest of the causal salad topic.
Two Moms example: Pairs of moms and daughters along with a range of variables on them like family size, birth order… we’d like to use the data to estimate the causal effect of mom’s family size on daughters.
Causal Design
- Make a causal model to design data collection and statistical model.
- Fundamental component of causal model: Function that determines how some variables are influenced by others.
- DAG’s don’t show interactions.
- Need to maybe practice constructing some of these. The whole worlds a causal system, but if we can focus on subsets of causal relationships.
- The causal map allows you to instill what you know about the variables. In the example, the birth order of the daughter might affect their family size, but birth order of the daughter is effected by the family size of the mother, so you put an arrow in for this relationship. “You can say things about the data that are not in the data themselves”
- Details the idea of forks, pipes and colliders, not too sure if I understand what the z line is.
- Interesting how association changes as you learn information. Would have to review this.
- Colliders pretty interesting.