The terms accuracy and precision are often confused or misunderstood. But every scientist, forensic or otherwise, should understand what they mean. In simple terms, ‘accuracy’ relates to how closely the value comes to the real score or true value (being ‘on target’). ‘Precision’, on the other hand, relates to the consistency of the value in repeated testing. Any given test, statistic or process may produce results that are one or the other, both or neither.
I personally find that images help when discussing these concepts. Shown below are a few ‘classic’ images that demonstrate the distinction between the terms.
First, the ‘target’ analogy…
Here the true value is the centre of the target. The results of the process/method are shown by the red dots.
Now, a somewhat more statistical example using ‘distributions’…
Here the true value is shown by the red arrow and line. The results of the process/method are represented by the bell-shaped distributions.
The above images, using either targets or ‘distributions’, are functionally equivalent. Consider the second image using data distributions. The data in the distributions might be scores from almost any type of test procedure, process, or method. If a process is to be considered ‘valid’ (another concept discussed more fully elsewhere) it should ideally be both accurate and precise. In other words, the mean value of the results should ‘equal’ the true value (shown by the red arrow and line) and, at the same time, the scores should not display ‘excessive’ variation.
This is, of course, neither feasible nor realistic. From a statistical and experimentation point-of-view, the situation wherein the mean of values is exactly equal to the ‘true’ value with no variation is so unlikely and unexpected that, should it ever occur, the procedure itself would have to be considered highly suspect and flawed.
An observant reader will be asking herself two obvious questions. First, how closely does the mean test value have to match the true value in order to be considered ‘equal’? Second, what amount of variation is considered acceptable or, alternatively, ‘excessive’? Unfortunately, complete answers to those questions, both of which are critical to issues relating to ‘validation’ (and ‘competency testing’), depend upon several factors. And that would require a discussion beyond the scope of this posting.
‘Validation’ of any procedure consists of testing it to determine the limits of its accuracy and precision, under appropriate conditions. In my opinion, a procedure or process can never be completely and absolutely validated; only validated to a certain, determinable degree. More about that in another post.
Understandably, the terms ‘accuracy’ and ‘precision’ come up a lot when talking about science, scientific research and forensic science. These graphics may be simple, but they are very effective when describing, visualizing and/or understanding these terms.