What is survivorship bias in equipment fleet sizing?

Two old, rusted tractors sitting idle and abandoned in a snowy field.

Survivorship bias in fleet sizing is what happens when your installed-base number only counts the machines that are still around to be counted. Dealer surveys, registration rolls, warranty records: all of them skew toward units that are active, visible, and still generating a paper trail. Machines that got scrapped, cannibalized for parts, or sold off-lease into a market with no reporting requirement just drop out of the dataset. Nobody files a form when a 2009 wheel loader goes to the shredder.

The result is a fleet count that looks bigger and younger than reality. If you're using that number as the denominator for a market-share calculation, you're dividing by a figure that's wrong in one direction only, a direction that always flatters your position: fewer competitor units, a healthier-looking replacement cycle, a fleet age distribution skewed younger than the real fleet.

How it gets into a regional estimate

Most installed-base estimates start from some combination of registration data, dealer reporting, and historical sell-through numbers run forward with an assumed useful life. Each of those sources has its own survivorship problem.

Registration and titling records only capture machines still registered. A unit that's been parked behind a shop for three years, waiting for a part or a buyer, often stays on the books. A unit that got exported, parted out, or quietly retired usually doesn't get a corresponding deregistration. So the active count overstates the working fleet, and it does so more in older age bands, because that's where attrition is heaviest and least likely to be reported.

Dealer survey estimates have a mirror-image version of the same issue. Dealers report what they see come through service bays and parts counters. A machine that's stopped visiting the dealer network, because it's dead, because it moved to an independent shop, or because the owner switched brands, disappears from that picture even though it might still be running somewhere. You end up over-weighting the fleet toward units still loyal to the channel doing the counting.

Sell-through-plus-useful-life models dodge some of this but introduce their own flavor: they assume a retirement curve instead of observing one. If the assumed useful life is too long, which is common because nobody wants to write down their own machine's expected lifespan, the model keeps units "in the fleet" for years after they've left it for good.

Why decommissioned machines matter more than they seem to

The instinct is to treat decommissioned units as noise, a rounding error at the tail end of the age distribution. In a mature equipment category they're not noise. Depending on duty cycle and application, a meaningful share of any 15-plus-year-old cohort is already gone, scrapped or exported, by the time most estimates get built. If your model doesn't subtract that attrition explicitly, you're not just missing a few stragglers. You're carrying an entire generation of machines that exist on paper and nowhere else.

This matters most when the number feeds a market-share denominator. Market share is your install count divided by somebody's estimate of the total fleet. If that total is inflated by machines that no longer exist, every OEM's share looks smaller than its real share, uniformly, which means the comparison between OEMs is still technically consistent but the absolute figure you're reporting upward is fiction. Boards and planning committees tend to notice when "the market" keeps growing but nobody's unit sales explain where the growth came from.

What a defensible count requires

A fleet estimate that accounts for survivorship has to start from something that doesn't care whether a machine still files paperwork: physical presence on a job site, in a yard, at a quarry, wherever the equipment sits today. That's a count of what's there now, not what was registered five years ago and never un-registered.

Installed Fleet Estimation scales that kind of site-level machine count up to a regional installed base, with coverage and occlusion correction built in so a parking pattern or a roof overhang doesn't inflate or deflate the number. If your current denominator came from a dealer survey or a registration roll and you've started to wonder how much of it is ghost fleet, it's worth requesting access and comparing the two.

Request early access

← Back to the blog