Report 04 — Delays and the memorylessness of chance

The 'overdue number' under the statistical microscope

For one number, at each new draw the probability of appearing remains 1/18, regardless of how long it has been absent.

P(G = g) = (17/18)g × (1/18),   E[G] = 17
Delay distributions are compatible with the geometric law on every wheel: none of the 11 tests is significant after Holm. The expected number of non-hit draws between two appearances is 17; all observed means are extremely close to this value.

Wheel-by-wheel comparison

WheelIntervalsMeanMedianMaximum observed gapp rawp Holm
Bari5367016.96712201 (n. 21)0.835901.00000
Cagliari3629516.96512210 (n. 76)0.448551.00000
Firenze5450016.97512198 (n. 82)0.884471.00000
Genova3665516.95712184 (n. 38)0.616031.00000
Milano5453016.97412181 (n. 30)0.935391.00000
Napoli5451016.97312179 (n. 79)0.445091.00000
Nazionale1734016.90512257 (n. 53)0.814661.00000
Palermo5441016.97112185 (n. 22)0.174651.00000
Roma5453016.97212201 (n. 8)0.992631.00000
Torino5453016.97412181 (n. 11)0.784951.00000
Venezia5452516.97712191 (n. 67)0.489101.00000

Empirical vs theoretical survival

The theoretical probability that an interval contains at least g non-hit draws is (17/18)^g. The table uses all complete inter-arrival intervals in the archive.

gEmpiricalTheoreticalΔ
01.0000001.000000+0.000000
50.7519330.751419+0.000514
100.5647380.564630+0.000108
200.3186120.318807-0.000195
300.1795790.180008-0.000429
400.1012000.101638-0.000438
600.0317470.032403-0.000656
800.0101200.010330-0.000210
1000.0031570.003293-0.000136
1200.0009510.001050-0.000098
1600.0001010.000107-0.000006
2000.0000110.000011+0.000001
A long delay is rare, but rare does not mean “due”. The geometric distribution is memoryless: after 5, 50 or 150 misses, the probability at the next draw is still 1/18.
Note: the delay test pools complete inter-arrival intervals for the 90 numbers; its p-values are a secondary asymptotic check, because intervals for different numbers are not perfectly independent within the same five-number draw. The main conclusion does not rely on this test alone.
LottoStatistics 1.0.1 — reproducible descriptive/inferential analysis; not a prediction system.