Understanding accuracy and precision

Determining the position of an object is a statistical task. It is based on measurements and observations of the environment around and signals from neighbouring devices – which are also affected by the environment. All of the measurements are subject to noise and uncertainty. When the measurements are combined to compute an estimate of the position, the result is statistical. This means that any position we compute has a probability and error margin associated with it.

It is common practice to describe a positioning system as ‘accurate to 2.5 metres’ (or some other number according to the system and technology used). So what does this actually mean?

What it does not mean is that every single location it ever generates will be within 2.5 metres of the ‘actual position’! So this raises several important questions:

1. What proportion of measurements are within 2.5 metres?

2. How do we know when a measurement is worse than the specified value?

3. Under what conditions would we expect this performance to be achieved?

4. What is the ‘actual position’ with which the system is being compared?

5. Are the errors repeatable? In other words will the same error occur if the test position is revisited at a later time?