🕸️Lecture 13 - Confounding and what regression cannot tell you

1. A 🔤 Confounder is connected to both the thing you are blaming and the thing you are measuring. Leave it out and its effect is credited to whatever is standing nearest.
2. A regression slope is a pattern in these rows. Turning it into because needs an argument the software cannot make.
3. Adjusting helps only for what you thought to record. The dangerous confounders are the ones not in the spreadsheet.
4. Added 19 November: I did not reread this page before fitting the project model. See the last box.

Where this came back

• 📌 Statistics II - Project - What predicts vessel delay at Greywater
• 🎓 Lecture 18 - Causal questions with observational data
• 🎓 Lecture 9 - A-B tests and their traps
• Port Authority open vessel-delay data

• guest essay went up this morning. spent the first ten minutes not listening
• confounder = a third thing that moves both the cause you like and the effect you see
• ice cream and drowning. everyone knows this one and everyone still does it
• regression will give you a slope for anything. it has no idea what causes what
• adding the confounder to the model can flip the sign of the slope
• you can only adjust for what you measured
• assignment 3 set tonight, due in two weeks