Method
Adding up the errors
Every reading carries several small errors from different sources; an error budget lists them, sizes them and shows which one to fix first.
An error budget is a table of every separate way one reading can be wrong, with a size next to each, combined into a single honest uncertainty. Its value is not the total. Its value is that it shows which one or two lines dominate, so you spend effort there instead of on a better ruler.
Why a budget beats a single "accuracy" figure
People ask how accurate a measurement is as if the answer were one number. In practice a reading is the true value plus several unrelated disturbances: the instrument, the zero, the angle you look from, how hard you press, the state of the tissue, and the rounding you apply when you write it down. Each has its own size and its own sign, and each responds to a different fix.
Metrology has a formal habit for this. The Guide to the Expression of Uncertainty in Measurement (JCGM 100:2008, the GUM, published through the BIPM) tells you to identify each source, express it as a standard uncertainty, and combine the independent ones. You do not need the whole formalism here. You need the habit: list, size, combine, rank.
The definitions of precision and accuracy are covered in precision versus accuracy, and this piece does not repeat them. What it adds is the arithmetic that sits between "several small errors" and "one stated range".
The six lines for a length reading
A bone-pressed length reading has six sources worth a line each. The table gives the character of each, what supports the size, and which post goes deeper.
| Source | Direction | How big | Where it is covered |
|---|---|---|---|
| Instrument scale | Either way, tiny | Tiny on a decent rule; checkable in a minute | Checking your instrument |
| Zero offset | Reads short | A few millimetres if the scale starts inside the end | Ruler zero offset |
| Parallax and viewing angle | Either way | Small, but consistent for one person | Viewing angle |
| Pressure and landmark | Mostly reads long if pressed hard | Varies; set by whether the ruler stops on bone | How hard to press |
| State of the tissue | Either way | Largest for flaccid, smaller for erect | Temperature and flaccid readings |
| Rounding | Either way if done blind, upward if done by eye | Up to half the step you round to | Catching rounding bias |
For girth the list changes. Tape stretch, tape thickness, tension and skin movement take the place of zero offset and pressure. The same method applies and the table is rebuilt.
Notice that only two rows have a size you can state from first principles. The zero offset is whatever your ruler's scale starts at, which you can read off the instrument. Rounding is bounded by the step you choose. The rest are things you size by testing yourself, which is the point of the repeat readings later in this piece.
Systematic and random lines are different kinds of entry
Before combining anything, sort the lines into two piles.
Random errors scatter around the true value. Take enough readings and they average out. Parallax from a wobbling head, small differences in how you seat the ruler, and fine rounding when done blind are mostly in this pile.
Systematic errors push the same way every time. A ruler whose scale starts three millimetres in from the end is short by three millimetres on every reading, and averaging a hundred readings leaves it exactly as wrong. A tape that has stretched, a habit of pressing a little hard, a tendency to round up: these are all in this pile, and systematic versus random error explains why most self-measurement error belongs here.
This sorting decides what you do next. Random lines go into the combined uncertainty. Systematic lines you either correct, if you can size them, or you flag as a known offset that the repeat readings will not reveal.
Combining independent errors
If the errors are independent, they do not add in a straight line. They add in quadrature: square each one, add the squares, take the square root.
Take a deliberately illustrative case. These numbers are assumptions chosen to show the arithmetic, not measured values. Suppose the random lines come to 1 mm from seating and sighting, and 0.3 mm from reading a millimetre scale to the nearest mark. The combined random uncertainty is the square root of 1 plus 0.09, which is about 1.04 mm.
The second line barely moved the total. That is the main lesson of quadrature: a source a third the size of the biggest one contributes about a tenth as much, so it earns no attention. Only when two lines are comparable in size does the combination matter much.
Rounding has a clean closed form. If you round readings to the nearest step of size s and the true values fall anywhere within a step, the standard uncertainty from rounding alone is s divided by the square root of 12, about 0.29 s. Rounding to the nearest half centimetre therefore contributes about 1.4 mm by itself, which is more than the two lines above combined. Rounding to the nearest millimetre contributes about 0.3 mm. The cheapest improvement in many budgets is simply recording to a finer step, provided the instrument and technique can support it, as rounding to the nearest millimetre discusses.
Do not combine systematic lines this way. A known offset is corrected or stated separately. Folding a 3 mm zero error into a quadrature sum hides the fact that you could simply have subtracted it.
Finding the line that dominates
A budget is only worth building if it changes what you do. So rank the lines after you size them.
Start with the ones you can bound cheaply: read the scale's zero, check the scale against a reference, decide your rounding step. Then size the rest by looking at your own repeat readings. Three readings in a session expose the within-session spread, which carries seating, sighting and pressure together. Readings on separate days add the state of the tissue. Within-session versus between-session variation shows how to tell those two apart, and how many times to measure gives the practical minimum.
The usual result is unglamorous. The biggest lines are pressure and state, which are habits and conditions, not hardware. The instrument line is nearly always the smallest, which is why you do not need to buy anything to get a good budget. A person who spends money on a calliper before fixing a three-millimetre zero offset has optimised the wrong row.
A worked pass through a budget
Put the pieces together in order, again with assumed numbers. A person checks their rule against a reference and finds the scale correct, so that line is zero to within the marking width. They read the zero and find the scale starts 2 mm in from the end, a systematic offset, so they add 2 mm to every reading and drop that line from the random sum. Their three readings in a session scatter with a standard deviation of about 1.5 mm, which carries seating, sighting and pressure together. Their readings across separate days scatter a little more, say 2.5 mm, which is the within-session scatter plus state.
Two things follow. The state contribution is the difference in quadrature: the square root of 6.25 minus 2.25, which is 2 mm. That makes state the larger single line, bigger than all the technique lines put together, so the next improvement is a more consistent pre-condition, not a better tape. Second, rounding to the nearest half centimetre would add about 1.4 mm in quadrature, taking the total from 2.5 to nearly 2.9 mm, so recording to the millimetre is the cheapest gain in the whole budget.
None of this needs software. It needs a pencil, the log and the habit of asking which line is biggest.
From budget to a stated reading
Once you have the combined random uncertainty and a list of any known offsets, you can write the reading properly. The convention is a central value from your repeats, the method line, and a range. Stating an uncertainty from three to five readings covers the arithmetic, and write the method next to the number covers the wording that makes it comparable to anyone else's.
The range you state should usually be wider than the instrument's resolution suggests. A ruler marked in millimetres does not give a millimetre of certainty. It gives a millimetre of display, and the budget says how much of that display is real.
It also helps to compare your total against the published spread. Veale et al. (2015) pooled studies in which a health professional took the reading under a standard procedure, and even that data carries a floor of measurement error, as measurement error inside the studies describes. If your own budget comes out comfortably larger than the difference you care about, the measurement cannot settle the question, whatever the digits say.
What the budget cannot capture
A budget covers errors you thought of. It cannot cover the ones you did not, such as measuring a different state than you meant to, or measuring a different landmark on the second visit. Those are mistakes of definition rather than noise, and the defence is a written routine, which is what the one-page measurement card is for.
It also says nothing about what the number means once you have it. A tight, well-bounded length is still just a length. Placing it against a population is a different step, with its own uncertainty from the reference data. If you want a judgement of a photograph instead of a distance, that is another kind of output entirely: Rate Cock produces a score from an image, which carries no instrument error budget because nothing is read off a scale, and its own scoring behaviour is described by an AI scoring service's accuracy notes. If you are curious how people price that sort of service, Cock Rate's pay hub lays out the commercial side, and a numeric score from a photo pool shows what the output looks like, none of which belongs in a table of ruler errors.
The part of the budget that is yours to improve is the part on the table above: zero, sighting, pressure, state, rounding. Size each, combine the random ones in quadrature, correct the systematic ones, and state the range your own repeats support.