At a Glance

A question becomes checkable once its comparison and measure are set before the data.

1. Part 1"What killed them?" has four true answers, so it has to be split.
2. Part 2An estimand names who, exposed to what, which outcome, compared with whom.
3. Part 3Mendelian randomization is Snow's experiment, run at conception.
4. Part 4Each question needs two groups alike in everything but one.

Death, and other events partway through, change what a study can count.

1. Part 1Each death names one culprit and keeps quiet about its accomplices.
2. Part 2A heart attack ends the outcome instead of changing it, so the plan allows for it.
3. Part 3The women who died stay in the denominator, as a competing risk.
4. Part 4The competing risk grows into a multistate model that counts every fall.

Healthy years are made where a body's capacity meets what its surroundings ask of it.

1. Part 1Weak legs cost one woman nothing and the other her whole street.
2. Part 2A grab bar lets a body with less still live a full life.
3. Part 4At sixty-six, the surroundings' side is a rail on a stair.

An answer can be right and still late; an early reading leaves time to act.

1. Part 2A surrogate arrives sooner than the outcome it stands for.
2. Part 3Snow's count was right; the handle came off once the dying had slowed.
3. Part 4No one took a reading in her house, so the first one was the fall.

The gap closes when the sick years shrink, not when they only move later.

1. Part 1Compression squeezes the sick years into a short stretch at the very end.
2. Part 4Of three versions of her, the second is delay and the third is compression.