Banker’s Rounding Explained (and Why Excel and Python Disagree)

Rule Statement

Banker’s rounding, also known as round half to even, is a rounding method where numbers exactly halfway between two possible rounded values are rounded to the nearest even number. This method minimizes cumulative rounding bias in large datasets and is preferred in scientific and financial calculations.

The basic rule is as follows:

If the digit to be rounded is exactly 5, and there are no subsequent digits or only zeros, round the preceding digit to the nearest even number.

For example, when rounding to the nearest integer:

• 2.5 rounds to 2 (because 2 is even)
• 3.5 rounds to 4 (because 4 is even)

This contrasts with the more familiar round half up method, which always rounds 0.5 up.

Banker’s rounding is codified in several standards, including ISO 80000-1:2009 (Clause 8.4) and ASTM E29-18 (Section 7.3.2). The Guide to the Expression of Uncertainty in Measurement (GUM) also recommends this approach to reduce systematic bias (see Clause F.2.2).

Worked Examples

Consider rounding these numbers to the nearest whole number using banker’s rounding:

1. 4.5: The digit before 5 is 4 (even), so 4.5 rounds to 4.
2. 5.5: The digit before 5 is 5 (odd), so 5.5 rounds to 6.
3. 2.35 (rounding to 1 decimal place): The second decimal digit is 5. The first decimal digit is 3 (odd), so 2.35 rounds to 2.4.
4. 2.25 (rounding to 1 decimal place): The second decimal digit is 5, the first decimal digit is 2 (even), so 2.25 rounds to 2.2.

Step-by-step for 2.35:

• Check the digit to round: second decimal place is 5.
• Look at the digit before 5: 3 (odd).
• Since it is odd, round up to 4.
• Result: 2.4.

Common Mistakes

Banker’s rounding is often misunderstood, leading to errors such as:

• Always rounding 0.5 up: Many users default to the “round half up” method, which can introduce upward bias.
• Ignoring trailing zeros: Not recognizing that trailing zeros after the 5 digit indicate a true halfway case.
• Applying banker’s rounding inconsistently: Using it only on some halves but not others leads to bias.
• Expecting software to behave uniformly: Different software implements rounding differently, confusing users (see Software Behavior Note).

Understanding these pitfalls helps ensure accuracy and consistency in your calculations.

Software Behavior Note

Banker’s rounding is implemented differently across common platforms, which can cause confusion:

Why the disagreement? Python’s built-in round() follows banker’s rounding, but some other rounding functions and languages default to round half up. Excel’s ROUND function also uses banker’s rounding, but its internal floating-point arithmetic can sometimes lead to unexpected results due to precision issues.

When precise control is needed, especially in financial and scientific applications, use libraries or functions that explicitly support banker’s rounding and verify behavior with tests.

Convention Comparison Table

The following table compares several common rounding methods to clarify how banker’s rounding differs:

Standards Citation

Banker’s rounding is documented and recommended in several international standards and guidelines:

• ISO 80000-1:2009 — Quantities and units, Clause 8.4: Defines round half to even as the preferred rounding method for numerical values.
• ASTM E29-18 — Standard Practice for Using Significant Digits in Test Data to Determine Conformance to Specifications, Section 7.3.2: Recommends round half to even to avoid bias.
• GUM (Guide to the Expression of Uncertainty in Measurement, 2008), Clause F.2.2: Advises round half to even when reporting measurement results to reduce systematic errors.
• NIST Special Publication 811, Appendix B: Discusses rounding rules consistent with banker’s rounding.

These sources establish banker’s rounding as the authoritative method for scientific and engineering rounding tasks, supporting its use here as the precision-and-rounding standard.

Counter-Examples

To emphasize the importance of correct application, consider these common errors:

1. Mistaking 2.50 for 2.51: Rounding 2.50 as if it were greater than halfway might lead to rounding up to 3 instead of down to 2.
2. Ignoring trailing digits: Rounding 3.5001 using banker’s rounding would round up to 4, but treating it as a pure half incorrectly rounds to 3.
3. Applying round half up to scientific data: Systematic upward bias accumulates over many calculations, skewing results.

These cases highlight how subtle misinterpretations can lead to significant discrepancies, especially in high-precision or large-scale datasets.

Related Rules and Articles

• Rounding Rules: An Overview
• Significant Figures: Counting, Rounding, and Reporting
• Round Half Up: The Default Rounding Method Explained
• Software Rounding Behavior: Python, Excel, JavaScript

For more detailed calculations and interactive examples, visit our Significant Figures Calculator.