Short Answer
In scientific and engineering practice, the decision to truncate or round a numerical value is not a matter of convenience—it is a matter of correctness. Truncation and rounding are two fundamentally different operations that affect the accuracy, bias, and uncertainty of reported results. This article provides a definitive reference on when cutting digits is appropriate, grounded in international standards and metrological principles.
Rule Statement
Truncation (also called chopping) removes all digits beyond a specified position without any adjustment to the remaining digits. For example, truncating 3.14159 to three decimal places yields 3.141, regardless of the value of the fourth decimal digit.
Rounding replaces a number with a nearby value that has a shorter representation. The most common rounding rule is round half up, where a discarded digit of 5 or greater causes the last retained digit to increase by 1. However, other conventions exist (see the Convention Comparison Table below).
The general rule for measurement and calculation: round when you need to minimize error and maintain statistical properties; truncate only when the discarded portion is inherently irrelevant or when the application explicitly requires it (e.g., integer division in computing, or when representing a lower bound). Truncation always introduces a systematic negative bias (for positive numbers) and is never appropriate for reporting measured values unless specified by a standard.
Worked Examples
Example 1: Rounding to Three Significant Figures
Value: 0.0045678
- Identify the first three significant digits: 4, 5, 6 (the leading zeros are not significant).
- Look at the next digit: 7 (which is ≥5).
- Increase the last retained digit (6) by 1 → 7.
- Result: 0.00457 (three significant figures).
Example 2: Truncation to Three Decimal Places
Value: 12.345678
- Keep digits up to the third decimal place: 12.345.
- Discard the rest (678) without any adjustment.
- Result: 12.345 (truncated).
Note the difference: rounding to three decimals would give 12.346 because the next digit is 6.
Counter-Examples
Common error: truncating a measurement to meet a precision requirement. Suppose a balance reads 2.3456 g, and you need to report to the nearest milligram (0.001 g). Truncating gives 2.345 g, but rounding gives 2.346 g. The rounded value is closer to the true reading (2.3456 vs 2.3450 vs 2.3460). Truncation introduces a systematic error that can accumulate in subsequent calculations.
Another error: rounding intermediate results. In multi-step calculations, you should retain extra digits until the final step. Rounding early (e.g., rounding 3.14159 to 3.14 before multiplying) can cause significant error propagation. This is known as premature rounding.
Truncation in a negative context: For negative numbers, truncation moves toward zero (e.g., -2.999 truncated to two decimals gives -2.99), which is not the same as flooring. This can cause unexpected bias if not recognized.
Convention Comparison Table
| Method | Rule for discarded digit | Example (2 decimals) | Bias |
|---|---|---|---|
| Truncation | Ignore all discarded digits | 3.149 → 3.14 | Systematic negative (for positive numbers) |
| Round half up | If discarded digit ≥5, increment last kept digit | 3.145 → 3.15 | Slight positive bias for random data |
| Round half down | If discarded digit ≥6, increment; if =5, leave | 3.145 → 3.14 | Slight negative bias |
| Round half to even (Banker’s) | If discarded digit =5 and last kept digit is odd, increment; if even, leave | 3.145 → 3.14 (since 4 is even); 3.135 → 3.14 (since 3 is odd) | Unbiased for random data |
| Round half away from zero | If discarded digit ≥5, increment the absolute value | -3.145 → -3.15 | Unbiased in sign |
Standards Citation
Several standards explicitly define when truncation is acceptable and when rounding is required:
- ASTM E29-13 (Standard Practice for Using Significant Digits in Test Data to Determine Conformance with Specifications) – Section 6.2 states that rounding shall be performed in accordance with the “round half up” method unless otherwise specified. It explicitly discourages truncation because it can cause false acceptance or rejection in conformance testing.
- ISO 80000-1:2009 (Quantities and units – Part 1: General) – Annex B provides rules for rounding, recommending “round half to even” for statistical applications to avoid bias. Truncation is not recommended except for specific technical reasons.
- JCGM 100:2008 (GUM) – Section 7.2.6 advises that when reporting measurement uncertainty, the numerical value of the uncertainty should be rounded to two significant digits, and the measurement result should be rounded to the same decimal place. It does not permit truncation.
- NIST Technical Note 1297 – Section 7.4 emphasizes that rounding should be performed only at the final step of a calculation, and that truncation is not acceptable for reporting.
Common Mistakes
- Using truncation when rounding is required – especially in scientific reports, engineering tolerances, or any context where accuracy matters.
- Rounding intermediate values – always carry extra digits through calculations and round only the final result.
- Ignoring the context of negative numbers – truncation and rounding behave differently for negative values; always consider the sign.
- Mixing rounding conventions – e.g., using half-up for one value and half-even for another in the same dataset, which introduces inconsistency.
- Assuming truncation is the same as floor or ceiling – truncation always moves toward zero, not toward negative infinity.
Practice Problems
Test your understanding. Answers are provided below.
- Truncate 9.8765 to three decimal places.
- Round 9.8765 to three decimal places using the half-up rule.
- Round 0.0004567 to two significant figures.
- Which operation (truncation or rounding) would you use to report a measured length of 12.345 cm to the nearest 0.01 cm? Why?
Answers: 1) 9.876; 2) 9.877; 3) 0.00046 (since the third sig fig is 6, and the next digit is 7, round up); 4) Rounding, because truncation would introduce a systematic error and does not reflect the true proximity.
Software Behavior Note
Different programming languages and spreadsheet applications implement truncation and rounding in distinct ways:
- Excel –
TRUNC()truncates,ROUND()uses round half up (away from zero for positive numbers).ROUNDDOWN()truncates toward zero,ROUNDUP()rounds away from zero. - Python –
math.trunc()truncates toward zero;round()uses banker’s rounding (round half to even) for floats, but for integers it rounds half to even as well. This can surprise users expecting half-up. - JavaScript –
Math.trunc()truncates;Math.round()rounds half up (toward positive infinity for negative numbers? Actually it rounds half up, but for -2.5 it returns -2, which is half-up toward positive infinity). - MATLAB –
fix()truncates toward zero;round()rounds half away from zero (default).
Always verify the default rounding behavior in your software, especially when working with financial or scientific data.
Quick Reference Table
| Scenario | Recommended Operation | Rationale |
|---|---|---|
| Reporting a measured value | Round | Preserves accuracy and minimizes bias. |
| Conformance testing (ASTM E29) | Round half up | Standard explicitly requires rounding, not truncation. |
| Statistical analysis | Round half to even | Unbiased for random data. |
| Integer division in programming | Truncate | Language-defined operation for integer types. |
| Representing a lower bound | Truncate | Ensures the value is not overestimated. |
| Intermediate calculations | Do not round | Carry extra digits to avoid error propagation. |
Related Rules
- Sig Figs in Scientific Notation
- Banker’s Rounding
- Double Rounding Error
- Rounding vs Significant Figures
Sources & Further Reading
- ASTM E29-13, Standard Practice for Using Significant Digits in Test Data to Determine Conformance with Specifications.
- ISO 80000-1:2009, Quantities and units – Part 1: General.
- JCGM 100:2008, Evaluation of Measurement Data – Guide to the Expression of Uncertainty in Measurement (GUM).
- NIST Technical Note 1297, Guidelines for Evaluating and Expressing the Uncertainty of NIST Measurement Results.
Reviewer Box
This article was reviewed by a metrologist with 15 years of experience in calibration and uncertainty analysis. The content aligns with current standards and best practices. Suggestions for improvement are welcome.
Changelog
- v1.0 – Initial release (2024-03-01)
- v1.1 – Added software behavior section (2024-06-15)
FAQ
Is truncation ever acceptable in scientific reporting?
Only if a standard or specification explicitly requires it. For example, some engineering tolerances specify truncation to ensure a conservative estimate. In most cases, rounding is the correct approach.
What is the difference between truncation and rounding down?
Truncation always moves toward zero. Rounding down (floor) moves toward negative infinity. For positive numbers they are the same, but for negative numbers they differ: truncating -2.9 gives -2, while rounding down gives -3.
How does double rounding error occur?
Double rounding happens when you round a number in two steps, e.g., first to three decimals, then to two. This can produce a different result than rounding directly to two decimals. Always round in one step.
Leave a Reply