Rounding is a small but important component of correct and effective data presentation. Here’s a quick tutorial on how to round values in your manuscript submission.
The Round-to-Even Method
Radiology follows the AMA Manual of Style (section 19.4.2) in using the round-to-even method, which is slightly different from the traditional rounding you might have learned in school. In traditional rounding, you round down if the digit one place to the right of the “last significant digit” (ie, the place value you’re rounding to) is 0–4, and you round up if the digit one place to the right is 5–9. There is, however, a slight systematic bias towards rounding up with this approach, which the round-to-even approach avoids (see this SAS blog post for a deeper dive).
The round-to-even method is identical to traditional rounding if the digit one place to the right of the last significant digit is 0–4 or 6–9. Where round-to-even differs from traditional rounding is when the digit one place to the right of the last significant digit is a 5. If it is a 5 followed by any other non-zero digits, then you round up, just like in traditional rounding (eg, 8.6503 rounds to 8.7). But if there is only a 5 to the right of the last significant digit, then you round down if the last significant digit is even, and up if the last significant digit is odd. Here is a set of examples involving rounding to the tenths place:
If the last significant digit is even, round down:
8.05 → 8.0
8.25 → 8.2
8.45 → 8.4
8.65 → 8.6
8.85 → 8.8
If the last significant digit is odd, round up:
8.15 → 8.2
8.35 → 8.4
8.55 → 8.6
8.75 → 8.8
8.95 → 9.0
Which Place Value to Round To
If you are wondering which place value is most appropriate to round to in any given instance, the AMA Manual of Style (sections 4.1.8, 19.4.1, and 19.4.2) and the Radiology Scientific Style Guide provide the following guidance:
Measured values should not be more precise than the limits of the instrument used for the measurement.
Calculated values (eg, means, SDs) can be expressed to one significant digit beyond the accuracy of the instrument (eg, if values were measured to the nearest millimeter, means and SDs can be expressed to the nearest tenth of a millimeter).
Odds ratios, risk ratios, and hazard ratios should be provided to the nearest tenth or, if warranted, to the nearest hundredth.
Area under the receiver operating characteristic curve, intraclass correlation coefficient, and κ values should be presented to the nearest hundredth.
Apparent diffusion coefficient values can be provided to the nearest thousandth.
Pearson and Spearman correlation coefficients should generally be presented to the nearest hundredth, though they can be presented to the nearest thousandth for sample sizes of 100 or greater.
Percentages should have a number of significant digits equal to or less than the number of digits in the denominator of the associated proportion. For example, the percentage for 67 of 150 can be presented as 44.7% or 45%, but not 44.67%. For a single-digit denominator, give the percentage to two significant digits (eg, 5 of 9 is presented as 56%).
P values of .01 or greater should be expressed to two digits, except where rounding to two digits would make the P value appear nonsignificant (eg, .047 rounded to .05). P values less than .01 should be expressed to three digits, with P values less than .001 presented as P < .001. (When the threshold for statistical significance is much lower than P < .05, as is the case in genome-wide association studies and in some methods of adjusting for multiple comparisons, P values can be expressed with more digits.) P values that would round to 1.00 if rounded to the nearest hundredth should be presented as P > .99.
Final tips to round like a pro: Present comparable data to the same decimal place. This makes it easier for readers to compare values and, as a bonus, keeps your tables looking sharp. And, finally, check that rounding is consistent between the text, tables, and figures.