The best DAWGS
Statistical model as golem is a great metaphor because it forces you to think from the model's perspective. Similar to when we
debug, the first step is to realize that the unwanted behaviors might not be a bug but exactly what you are asking for.
Note. By clicking on the headers, you'll be sent to the corresponding location in the lectures.
Context & discrimination
For instance, in the Berkeley admission problem, we get that women might choose departments that are more competitive (think maths vs. psychology at graduate level), and maybe there is also a gender bias:
The gender bias is confounded by department choice. Or, more controversially
Varying confound
It is similar to context & discrimination, but here we have a collider. For instance, in the children education problem, parents influence children outcome. But grand parents and neighborhod (a proxy for socioeconomic status), also influence children outcome AND parents. Similarly, one could see the following model of becoming a National Academy of Science, given citation patterns, gender, and hidden quality of the works: