Methodology
Global Cone Census, fourth edition. Field period 1990–2026. Released under no licence anyone has checked.
We combine aerial image classification, manufacturer output reconciliation and volunteer roadside transects to estimate the number of traffic cones in service worldwide. The three methods disagree by 7.8 billion, which is most of this document. After coverage weighting and attrition correction we report a central estimate of 10.0 billion units, ±0.9 billion at 95% confidence.
What counts as a cone
A unit is counted if it is conical or frustoconical, taller than 300 mm, weighted so that it stands unaided, and intended to redirect traffic of some kind. Colour is not a criterion: we count the blue cones used at airports and the green ones used by a small number of Nordic municipalities.
A cone remains in service until it can no longer stand on its own. This threshold is permissive and it does real work — requiring an intact base would remove an estimated 1.4 billion units from the total. We keep the permissive version because a leaning cone still functions, and because volunteers found the intact-base judgement impossible to make consistently in the rain.
The estimator
N̂ = Σ (wᵢ · Eᵢ) + S − D
Eᵢ is the estimate from method i and wᵢ its coverage weight; S is the stored-stock correction for cones sitting in depots and yards; D is the double-count deduction. The weights are 0.31 for orthophoto, 0.44 for manufacturer reconciliation and 0.25 for transect sampling.
Why the methods disagree
Orthophoto counting undercounts because a cone viewed from directly above presents a circle roughly 360 mm across — close to the resolution floor of most public imagery, and identical in plan to a great many other objects. It also cannot see indoors, and roughly a fifth of the world's cones are indoors at any moment.
Manufacturer reconciliation overcounts unless attrition is modelled well, and attrition cannot be modelled well, because the destruction of a cone is not an event anyone records. We fitted a survival curve to 4,100 individually tagged cones tracked since 2011 and extrapolated. The curve has a long tail. Cones are remarkably hard to kill.
Transect sampling overcounts globally because volunteers walk where volunteers are, and volunteers are in cities, and cities have cones. We call this cone-seeking bias and correct for it by road-density stratification, which helps but does not solve it.
Limitations
- The stored-stock correction rests on 61 depot inventories, all of them from countries that answer emails.
- We have no reliable figure for cones in private ownership, and the true number of cones in gardens may be very large.
- One classifier run in 2023 counted an orange hedge in Lisbon 1,100 times before anyone noticed. Those counts were removed. We mention it because it is the kind of thing that usually goes unmentioned.