How to read the histogram

A histogram of measured lengths is one of the more honest ways to show this kind of data, and one of the easiest to make misleading without anyone intending to. The same underlying numbers can be drawn tall and narrow, or short and wide, or spiky, or smooth, depending on choices that have nothing to do with the measurements themselves.

What a bar actually represents

Each bar covers a range of lengths - the bin - and its height is how many measurements fell inside that range. A histogram with 0.5 cm bins and one with 2 cm bins, built from the identical dataset, look like different distributions. The wider bin smooths over real texture in the data; the narrower bin can make ordinary sampling noise look like a meaningful bump or dip.

There is no single correct bin width. There is a reasonable range for a given sample size, and a chart with unusually wide or narrow bins for its stated sample is worth a second look at why.

Counts versus density

Some histograms plot the raw count in each bin. Others plot density, scaled so the total area sums to one, which lets you compare two datasets of different sizes on the same axes.

Reading a count chart as if it were a density chart, or the reverse, produces a wrong impression of how concentrated the distribution is. Check the y-axis label before drawing any conclusion from the shape - it is the single most commonly skipped step in reading one of these correctly.

The truncated axis

An x-axis that starts well above zero, or stops just past the tails of the data, exaggerates how spread out or how narrow a distribution looks. This is not unique to size data - it is a general charting habit - but it matters more than usual here, because the true spread in this distribution is already narrow, and a chart that clips or stretches the axis can make that narrow spread look wide, or an already-narrow spread look narrower still.

Measured histograms look different from self-reported ones

Beyond the charting choices, the underlying texture differs by data source. A self-reported dataset piles up sharply at whole and half units, producing visible spikes at round numbers that a clinically measured dataset, read to the nearest millimetre, does not show in the same way. A histogram with tall, narrow spikes at 5, 5.5 and 6 inches is showing you rounding behaviour as much as it is showing you anatomy, and that spiking pattern is itself a useful diagnostic for spotting a self-reported dataset before the article even says so.

A clinically measured histogram, like the pool behind Veale et al. (2015), tends toward a smoother, more continuous shape for exactly the opposite reason - a ruler read carefully has no reason to prefer round numbers. One caution: the smooth curves in that paper are not histograms of raw readings at all - its abstract describes "simulation of 20,000 observations from the normal distribution" from each pooled mean and SD, so their shape is an assumption drawn, not data binned.

Reading one without being led

Check the bin width against the sample size, check whether the axis is count or density, check whether the x-axis starts at zero or is cropped to the data, and look for suspicious spikes at round numbers before trusting the overall shape. None of that requires statistical training - it requires reading the axes before reading the bars, which is the step almost everyone skips.

A histogram is a good way to see a distribution's real shape, including whether it resembles the normal curve the published nomograms assume. It is a poor way to compare yourself to a single number, the way a subjective score from a service like Rate Cock or a rating hub's public distribution of scores invites - those are judgements plotted as data, and this is data plotted as a shape, and the two do not answer the same question even when they share an axis. A written verdict from a human reviewer is further still from a histogram bar, and an AI-estimated figure is not a measurement at all - an image model has no scale information to bin in the first place.