The 100 most visited websites of Nizhny Novgorod

v is visitors from the region over 31 days. The class sets the uncertainty, and the uncertainty sets the significant digits: d = ⌊−lg ε⌋ + 1.

A direct measurement

The counter keeps a regional tally. Three significant digits.

B conversion between measures

v̂ = ĝ·u, u is the total audience by Top.Mail.Ru; lg (v/u) ~ N(μ, σ²) over sites with both counters: n = 5, μ̂ = 0.44, σ̂ = 0.34, σ_LOO = 0.38; ĝ = 10^μ̂ = 2.7 (95% CI 1.0 … 7.2); PI₆₈: v̂ ×÷ 2.6. One significant digit.

C upper bound

The ranking threshold θ = 0.20 gives v < θ u_RU — a consequence of the selection rule, not an assumption.

D estimate from rank

lg v̂ = β₀ + β₁ lg r, r is the place in the list; OLS; σ_LOO is the leave-one-out error in lg, ρ_S the rank correlation of r and v.
Similarweb: β₀ = 9.05, β₁ = −1.13; n = 79, R² = 0.73, σ_LOO = 0.40, ρ_S = −0.73.
Tranco: β₀ = 12.66, β₁ = −1.74; n = 23, R² = 0.14, σ_LOO = 1.58, ρ_S = −0.39.
Majestic: β₀ = 5.68, β₁ = −0.40; n = 21, R² = 0.01, σ_LOO = 1.61, ρ_S = −0.20.
v̂ only where σ_LOO ≤ 0.70; v̂ ≤ med(v), the median of the measured rows of the hundred; sections of national sites and domains that host other sites lie outside the domain of definition.
Δk = k(v̂_−i) − k(v_i), k is the place in the hundred: med |Δk| = 28, max |Δk| = 70.

The ranking includes Nizhny Novgorod sites: those with a local owner, a local newsroom or an office where the site itself is made. A city section of a national or global service — jobs, maps, directories, catalogues, classifieds — is not ranked, even if the service has a branch in the city; otherwise search engines would qualify too. A traffic counter on such a section says nothing about belonging to the city: it reflects how the service keeps statistics, and the largest national services keep no city counters at all. Regional editions of national media with a local newsroom stay in the ranking.