When reviewing breath alcohol test results, it is easy to conclude that a device is accurate when the same number appears on the screen several times in a row. However, a breathalyzer is a measuring instrument that takes breath as a sample and displays a reading. A repeated number only shows how closely the results cluster, not how close they are to the reference value. When safety managers or quality and technical staff evaluate the reliability of results, they need to keep these two questions apart.
Here, accuracy means how closely a measured value agrees with the true value of what is being measured. Repeatability, which people tend to lump together under the word "reproducibility," is precision concerning how consistently results come out under set repetition conditions. So the same number may repeat because the device matches the reference value well, but it may also repeat because the device displays readings that are stably offset from the reference.
What does the same number tell you?
If repeated readings are identical or very close, you can first observe that the scatter of results in that measurement process was small. This is the starting point for evaluating repeatability. However, because of the number of display digits, small differences in the actual internal readings may show up as the same number, so you must not conclude from identical screen values alone that all the readings were exactly the same.
Conversely, a run of identical values does not guarantee accuracy. Without a reference value, you can only confirm that readings agree with one another, and you have nothing to compare them against to see whether they match the reference. Readings can cluster tightly while being consistently higher or lower than the reference.
Accuracy asks how far you are from the reference value
The International Vocabulary of Metrology (VIM) describes measurement accuracy as the closeness of agreement between a measured value and the true value of the measurand. It also states that accuracy is not a quantity given as a single number in itself, and that a measurement is said to be more accurate when the measurement error is smaller. So when you see a phrase like "the accuracy is X," you also need to read which reference value, unit, and test conditions the comparison was made against.
To compare breathalyzer results with a reference, the source of the reference value also matters. You can calculate the difference from the readings only when a reference value is given, and if the reference value itself carries uncertainty, the comparison is affected too. The tolerance in a published specification and the result-scatter criterion set in an internal repeat test may serve different purposes, so it is better not to combine them and interpret them as the same test item.
Repeatability is a set of results under fixed conditions
In the VIM, repeatability is measurement precision under a set of repeatability conditions. The key point here is not that results were obtained several times, but what was kept the same between measurements. If the conditions you want to compare change, such as the same device, the same test procedure, and the same sample conditions, it becomes hard to explain differences in results by device repeatability alone.
For example, even if the results from several people blowing in turn are similar, the data include the variable of different people's breath. These data can be used to look at the distribution of results on site, but it is hard to call them a single repeat test under identical conditions. You need to record cases where the person changed separately from cases where the same person repeated the test in order to separate the causes of scatter in the results.
| Question to separate | Data to record with it | What you can interpret |
|---|---|---|
| Does it match the reference value? | Reference value, unit, source of the reference value, individual readings | The difference between the reference value and the readings |
| Is it similar when repeated? | Test order, individual readings, conditions under which it was repeated | Scatter of results |
| Why did the results change? | Changed device, test procedure, sample conditions | The relationship between changes in conditions and changes in results |
Two conclusions from a hypothetical case
As a hypothetical example, suppose a sample with a stated reference value was tested several times with the same device and the same procedure. If all the individual results are close to one another but skewed to one side of the reference value, the fact you can observe is that "the scatter of results is small." The interpretation that follows is that repeatability can be examined, but whether the results agree with the reference must be evaluated separately.
In this case, a possible action is not to record only the average, but to set out each result, the unit applied, the source of the reference value, the test order, and the conditions kept the same. If you look only at the average without the individual values, you cannot tell whether a particular result differed from the others or whether all values were skewed in the same direction. The completion standard is a state in which the data are complete enough to explain separately the distribution of repeated results and the difference from the reference value.
The opposite case is also possible. If the average looks close to the reference value but the individual results are widely spread, you cannot say each test was stable just because one average is close to the reference. In this case, you must also record which results were excluded from the calculation and why, so that the summary value does not hide what was actually observed.
Split data requests according to the question
When you request product data or test results, rather than asking only "Does it give the same value every time?", it is useful to ask separately whether the supplier can provide the individual results obtained from a reference sample and the test conditions. A request aimed at accuracy needs the reference value and the comparison method. A request aimed at repeatability needs the conditions kept the same in each repetition and the scatter of results, more than the number of repetitions.
Leave unconfirmed conditions as items for further review rather than interpreting them as already verified. In the end, a repeated number is one piece of what you need to judge reliability. Only by looking together at the distance from the reference value, the distribution of individual results, and the conditions fixed or changed between repetitions can you tell whether that number supports accuracy or shows only repeatability.
