<b>Stopping a test the first day it hits p&lt;0.05 inflates your false-positive rate from 5% to ~26%.</b>

<b>Stopping a test the first day it hits p&lt;0.05 inflates your false-positive rate from 5% to ~26%.</b>
That's the cost of peeking — checking significance repeatedly and stopping at the first green light. Each look is another lottery ticket for a fluke.
— Fix one: fix the sample size in advance, look once at the end.
— Fix two: use a sequential method (mSPRT, group-sequential) built to allow continuous monitoring.
A p-value (chance the result is noise) only means 5% if you looked exactly once.
Read the number, not the story. [α inflation 5%→26%, ~5 looks]