d-separation - directional separation, some variables on a directed graph are independent of others (no path connecting them, independent)
Shows the connection between variables matters greatly when designing experiment
Gives example of adding moisture as a variable, leads to fungus appearing to help the growth of plants, even though it had no effect (next section talks about this)


Directional Separation and Independence

- Fungus is a post-treatment variable
- Shows there is a flaw in our experiment because soil has an effect on the growth through reducing fungus. Better to omit the post-treatment variable fungus 

Affects of a Post-Treatment Variable

The experiment testing the height difference is done
Shows the average growth to be about 40%
Fungus proves to negatively affect growth

Experiment Difficulties due to PTV

Omitted Variable Bias - missing variables that could affect the experiment
Included Variable Bias - including variables that could affect the experiment
Post-treatment Bias - example of included variable bias, pg 171 example

University of Connecticut

Bias in Multiple Regression

Collider Bias - when you condition on a collider, it creates statistical - but not necessarily causal associations among its causes. 
Relationship between trustworthiness, worthiness and selection

Learn Before

Related

Learn After