Loading…
Reading the draws and working out what chance predicts.
LotteryOracle · F-45
Compare the modeled preference for 1–31 in the earlier and later halves of the latest complete records, and below versus at or above their median jackpot.
Popularity changes how much a win is shared, never whether it happens. These are estimates from prize tables, not records of individual choices.
Powerball · 5/59 + 1/35 era · Jan 2012 – Oct 2015 · main · 388 Draws · Canonical archive; verification varies
2012-01-18 — 2013-11-23 · 194 Draws
At least 200 complete draws are needed in this group.
2013-11-27 — 2015-10-03 · 194 Draws
At least 200 complete draws are needed in this group.
2012-02-15 — 2015-10-03 · 169 Draws
At least 200 complete draws are needed in this group.
2012-01-18 — 2015-09-30 · 219 Draws
153 training draws · 66 later validation draws
Four fixed comparisons; intervals are approximate and widened for multiplicity. They omit model error and serial dependence. Jackpot associations are not causal effects.
Median advertised jackpot, in the game’s currency: 80,000,000
crowd-timeline-birthday-v1
The latest eligible records are split into two chronological halves and two groups separated by the median advertised jackpot. Every group needs at least 200 complete draws. A single 1–31 coefficient is fitted with the existing crowd model, using 70% for fitting and later draws for validation. The displayed intervals reuse quasi-Poisson errors on the log multiplier and widen them for four estimates. They remain approximate: model error, dependence and changing player composition are not captured. A bounded, unidentified or unvalidated estimate is withheld.
interval = exp(log(multiplier) ± Φ⁻¹(1 − 0.05/8) × SE/multiplier)
Compare the modeled preference for 1–31 in the earlier and later halves of the latest complete records, and below versus at or above their median jackpot.
Popularity changes how much a win is shared, never whether it happens. These are estimates from prize tables, not records of individual choices.