A difference is two counts you took on purpose. "Usually" is one count, plus a story about other people.
The Guardian writeup of daraxonrasib, a pill for pancreatic cancer, puts both numbers on one line. A year after the trial patients started it, 76.6% were alive.
The rest of that line is the other kind of number. Usually, 40 to 45% of patients given chemotherapy alone survive a year.
"Were alive" is this group. "Usually" is how the disease has gone for other groups.
The paragraph never describes a second set of patients kept on chemo, beside these patients, for this trial. The subtraction is a feeling the headline trusts you to finish.
A remembered rate can be real and still be the wrong object to subtract. Other years, other hospitals, other rules for who got through the door.
Both numbers can stand. The gap between them is not automatically the pill.
It is the gap between a cohort and a scrapbook, and hope can stay.
The label "measured effect" does not fit.
The body of the piece already taps the brake. Doctors say the pill cannot yet be called a cure, and they do not know how long it holds the disease down.
Longer studies are the stated next step. Delete those two limits and the percents look finished.
Leave the limits in and you still have an open question about time. The strong reading is what you get by pretending that question was answered.
You have shipped this bug with cleaner charts.
The new path is "faster than last quarter." The rewrite "beats what launch week felt like."
Errors are "down from the bad one people still quote." You had one group, and a number you already believed sitting in the control column.
A control is dull on purpose. Same days, same door, the other build still on.
If you cannot point at those requests, you do not have a lift. You have a before-picture and a wish.
None of this says the pill did nothing.
Keep the hope, and drop the fake precision.
- Don't subtract a historical rate from a new cohort and brief the remainder as the effect.
- Don't let the title finish arithmetic the paragraph never ran.
- Don't delete "not a cure" and "we don't know how long" because they slow the hook.
- Don't treat the follow-up as optional after you admitted duration is unknown.
- Don't grade your change against a memory and type "experiment" in the doc.
If the other arm is "usually," you counted one group. You did not measure a difference.
This post was generated automatically from public Hacker News data.
