1,000 days of
puff counts.
Every morning: open spreadsheet, log yesterday’s count, reset counter. A small ritual that turned into a clear window into habit, nicotine dependency, and what it actually takes to stop.
All 963 days
Every recorded day as a dot. The purple line is a 30-day rolling average. The whole story in one view.
The shape of the data
Means hide a lot. The distributions before and after the quit attempt tell a richer story — and explain why you need more data than you’d think to detect any pattern reliably.
The After distribution is not just lower — it’s tighter. Standard deviation dropped from 149 to 103. More predictable behaviour, less day-to-day chaos.
How many days to reliably detect a difference?
Given the variance in this data, detecting a real signal from noise requires more observations than most people expect. Based on a standard power analysis (α=0.05, power=80%):
This is why the post-quit dataset (n=215) is only just becoming powerful enough to detect moderate effects — and why single-day observations are always misleading.
What actually drives the count?
Four candidate variables: where I was (location), what day it was, where I was in my cycle, and whether the kids were around. A decision tree fitted separately for Before and After the quit attempt — because the drivers changed completely.
Before the quit attempt, location explained 81% of what the model could find. After, day of week took over at 71%. The model also explains more variance After (R²=0.21 vs 0.08) — lower, more stable usage is simply more predictable.
Weekly patterns
Each panel: all days as ghost dots, highlighted day on top with a dashed mean line. Sat and Sun are combined — no statistically meaningful difference between them (p=0.22 before, p=0.31 after).
Before the quit attempt, weekends were actually the heaviest days — free time, fewer natural breaks (p=0.051 vs weekdays).
Where I was
Location is the strongest single predictor in the Before era. WFH consistently means more puffing — proximity, no social cues, fewer natural breaks.
Note: 2024 office data from desk booking records — may under-report actual office days. 2025–26 data complete.
Hormonal cycle
36 cycles tracked, phases calculated from period start dates. Ovulation is consistently the lowest phase — statistically significant before the quit attempt (p=0.025 vs Luteal). After, the signal fades as lower overall usage reduces detectable variance.
Hypothesis: nicotine and hormonal appetite suppression interact. With nicotine effectively gone, the signal disappears. More data needed to be certain.
What actually worked
Not willpower. Methodical nicotine reduction, one step at a time, watching the puff count for any compensatory increase. There wasn’t one.
Total nicotine consumed over 996 days: approximately 29 grams. Each step-down halved nicotine intake without any meaningful increase in puff count. The physical dependency was already gone before the behavioural habit was broken.