Accuracy vs. Precision

The words accuracy and precision are tossed around a lot in the CAD, mapping and engineering fields, and they are often thought of as having the same meaning. It is important to know the difference, and it’s even more important to verify whether data you are getting from someone else is accurate, precise, or both, and to what degree, then compare that with your needs.

Accuracy refers to how close something is to the truth. In the case of map data, this means how closely map features reflect real-world features.  Locations of points, lengths and bearings of lines and areas of polygons are all represented on maps, and how closely these scaled entities match the features they represent is expressed as map accuracy.

Precision refers to the degree of detail that can be observed and compared. When a surveyor measures a distance several times, it might be observed to be a few tenths of a foot different each time, depending on the number of transects shot, how plumb the rod was held each time, etc. How close each of the observed measurements were to each other is expressed as precision. The smaller the variation between the measurements, the more precise the collective information is.

It is possible to have very precise information that is also inaccurate.  If a survey rodman has a tendency to hold the rod at a slight angle, for instance, and does so about the same amount each time, or if the surveyor’s instrument is incorrectly calibrated, then several redundant error checking shots could result in very similar, and therefore precise measurements, all of which are inaccurate.

A similar problem can occur in digitizing.  If you calibrate your tablet based upon erroneous measurements, and then digitize some line work, all of those decimal places of precision that your tablet and AutoCAD are capable of are meaningless. The geometry will be inaccurate, no matter how painstakingly careful you are in positioning your crosshairs for each pick point.

Another way to see the differences between accuracy and precision is in weather forecasting. If a meteorologist predicts the temperature will reach a high of 67.5, and the actual high is 69.2, it was an unnecessarily precise forecast (who can feel a tenth of a degree ?) and acceptably accurate (within 2%). You wore a heavy shirt and got a little warm – close enough. However, if the prediction had been for a high between 62 and 78, it would have been more accurate (69.2 is closer to 70, the mid-point of that range, than it is to 67.5), but the lack of precision would have left you wondering whether to wear a sweater or a T-shirt!

Decimal places imply precision that may not exist. Stating a value as “8.5” implies it is known to within one tenth of a unit. Calling it “eight and a half” implies it is known to within one half of a unit.  Saying “somewhere between 8 and 9” implies it is only known to within one whole unit.  Keep this in mind when someone assures you that AutoCAD maintains a precision of 16 decimal places, since few data capturing techniques are capable of benefiting from it.

Being aware that a high degree of precision does not necessarily indicate accuracy, and vice versa is very important. It is always a good idea to ask questions about how data was collected, what other outside sources were used and what the original purpose of the information was, before relying upon it.

Obviously, the best data is both accurate and precise, but generally the cost increases proportionally for each. When in doubt, ask, and if only the word “precise” is used to describe it, make sure you know how accurate it is.

It is often better to be approximately correct than precisely wrong!

Leave a Comment