The herd that was not counted

Counting animals in a forest is harder than it sounds, and the honest methods all begin by admitting that most of the animals will not be seen.

The naive approach is to walk a line and count. This produces a number that depends on how far you can see, how quiet you are, how dense the vegetation is and what mood the animals are in, and comparing two such numbers from different years tells you mostly about the observers. The method is not useless, but what it measures is encounters, not population.

Mark-recapture does better by using the overlap. Catch some animals, mark them, release them, and catch another sample later. If a tenth of the second sample carries a mark, the first sample was roughly a tenth of the population. The arithmetic is simple and the assumptions are strong: marks must not fall off, marked animals must mix back in, and being caught once must not change the chance of being caught again. Each assumption, when it fails, pushes the estimate in a known direction, which is why a careful report states which way its number is likely to be wrong.

Camera traps have changed the practice more than any statistical advance. A grid of cameras records individuals that can be told apart by markings, over months, without anyone present. The limiting factor is no longer detection but identification, and software now does most of it.

The constant across all three methods is that the published figure is an estimate with a stated interval. A single number, confidently given, is the clearest sign that somebody has counted encounters and called it a population.