an instrumental variable does: isolates causal effect when you can't randomize

Why can't we always trust what we see?

Imagine you're trying to figure out if eating chocolate makes you happier, but you can't just give some people chocolate and not others. Some people might already prefer chocolate, which could affect their happiness.

We can't just compare happy chocolate lovers with non-lovers because their happiness might be linked to other things they do. An instrumental variable helps us figure out if chocolate really causes happiness by using something related to chocolate but not directly linked to happiness.

Example

If we find that people who live closer to chocolate factories (but don't eat chocolate) are happier, we might use the distance to the factory as an instrumental variable to see if chocolate really makes a difference.

Remember this

An instrumental variable helps us find the true effect of chocolate on happiness by using something related but not directly connected to happiness.

Text adapted from Wikipedia, licensed under CC BY-SA 4.0.

Related concepts

the back-door criterion identifies: sufficient adjustment sets for causal estimation

Can we trust studies without random experiments?

the condition number κ(A) measures: sensitivity of Ax=b to perturbations

How small changes affect big outcomes

Clinical trial

Why do scientists trust their experiments?

Bias vs variance: high bias = underfitting, high variance = overfitting

Can a perfect fit to past data predict future events?

log-loss / cross-entropy loss penalizes: confident wrong predictions more heavily

How wrong predictions hurt more than they're supposed to

Masking (behavior)

Can you not see what's right in front of you?

Swipe through 100 ML concepts daily

Open Pocket Polymath