Fifteen-unit rule for rounding numerical results

This page is a knowledge crystal of subtype method. The page identifier is Op_en5327
Moderator:Erkki Kuusisto (see all)
This page is a stub. You may improve it into a full page.
Upload data {{#opasnet_base_link:Op_en5327}}

Scope

When presenting the final results of a study, what is the proper way of rounding numerical results of the form "average ± probable uncertainty"?

Definition

Usually, from scientific studies (whether based on physical measurements or mathematical modelling), numerical results are obtained that consist of a an average value or a best estimate (e.g. 4.5678 meters) and a measure of the probable uncertainty (e.g. ± 0.2345 meters). However, either or both of these figures may be too precise (i.e. contain more digits than is justifiable or meaningful).

A systematic method for rounding those results to a justifiable and meaningful precision is needed, that will not convey a misconception of excessively (in)accurate results.

Result

The 15-unit rule says that:

• in the average value, the uncertainty of the least significant digit (LSD) must not exceed 15 units,

• the probable uncertainty must not exceed 15 units (where one unit pertains to the LSD of the average value)

• the probable uncertainty is always rounded upwards.

Please see the following examples:

• (1062 ± 41) meters is incorrect, because the LSD of the average value (2) is associated with an uncertainty of 41 units (and the uncertainty is expressed using a precision of 41 units)
• (1060 ± 50) meters is correct, because the LSD of the average value (6) is now associated with an uncertainty of 5 units only (and the uncertainty is expressed using a precision of 5 units)

- Note that the original uncertainty (41) has been rounded upwards (to 50).

• (0.8765 ± 0.0132) kg is incorrect, because the LSD of the average value (5) is associated with an uncertainty of 132 units (and the uncertainty is expressed using a precision of 132 units)
• (0.876 ± 0.014) kg is correct, because the LSD of the average value (6) is now associated with an uncertainty of 14 units only (and the uncertainty is expressed using a precision of 14 units)

- Note that the original uncertainty (0.0132) has been rounded upwards (to 0.014).