Since I did my engineering major in Instrumentation and Control, I have been reading all about measurements. Though my work has never been directly related to using or designing measurement devices, my work has always been, now that I think about it, related to measurements.
At my first job I was writing configuration software for power quality measurement devices. Post ISB I have been working on search, web and marketing analytics. All related to measurement. Which brings me to this post on the the suitability of accuracy vs. precision in the field of measurement.
Before proceeding further lets see what each means though in regular usage we are wont to use them interchangeably. Wikipedia has the following to say about accuracy and precision.
In the fields of science, engineering, industry and statistics, the accuracy of a measurement system is the degree of closeness of measurements of a quantity to its actual (true) value. The precision of a measurement system, also called reproducibility or repeatability, is the degree to which repeated measurements under unchanged conditions show the same results.
Since we generally use each of these terms as synonyms we are always striving for accuracy.
First we must agree that there can be no measurement without error. All measurement systems physical, on the web or data is prone to systemic errors. Whether due to imperfect calibration, external influences or imperfect methods of observation. We simply must live with error.
So how do we get around this omnipresent error?
Lets think of the true value of a variable as a point in space. And the measured value is a region around/some distance from this point. The distance and size of this region from the true value depends on the measurement error in the system. The distance (from the true value to the mean of the measured value) represents the accuracy of the system while the size is the precision of the system. So if we focus on accuracy the distance is lower, but the size could be big. Which essentially means that knowing one measured value, it is very difficult to predict the next value even though that may be pretty close to the true value.
On the other hand, for a high precision system, even if the measured value is very far away from the true value, it is much easier to predict the next measured value. In other words, accurate systems have lower errors which are more random in nature, while precise systems (could) have higher errors which are less random. Of course a system could be either accurate or precise or both or none.
As you see while striving for accuracy can reduce the absolute error, it increases the randomness of the error value as well. While it is just the opposite when striving for precision. The decreased randomness also renders itself more favorably towards predictive ability of the system.
Hence, analytics systems should really be chasing precision and not accuracy. Knowing that a system always has some definite error is surely better than knowing that a system has a small but relatively indefinite error.
All said, the pursuit for an accurate AND precise system continues.
Some more on the topic by Avinash Kaushik.