Double Rounding: Why Rounding Intermediate Results Ruins Precision

Double rounding is the quiet precision killer in scientific and engineering calculations. It occurs when a value is rounded to an intermediate precision and then rounded again to the final reported precision. The result can differ from rounding the original value once. This reference article explains the rule, standards, examples, and pitfalls. For a quick calculation, use our significant figures calculator; for the underlying rules, explore our rounding rules and measurement uncertainty guides.

Rule Statement

The governing rule is simple: round only once, at the end of the calculation. Carry at least one or two guard digits beyond the final significant figures or decimal places. Do not round intermediate results unless a standard or specification explicitly requires it.

ISO 80000-1:2009, clause 7.3, specifies that rounding should be performed only once to avoid the accumulation of rounding errors. ASTM E29-13a, Section 6, provides a consistent rounding method for test data. JCGM 100:2008 (GUM), clause 7.2.6, warns against excessive digits when reporting estimates and uncertainties.

When an intermediate rounding is unavoidable—for example, when a measurement instrument displays a limited number of digits—document the intermediate precision and apply a consistent tie-breaking convention. The most common conventions are round-half-up and round-half-even (banker’s rounding). See Round Half Even and Rounding Methods.

Worked Examples

Example 1: 2.4445 to 3 significant figures

Direct rounding: The number 2.4445 has five significant figures. To round to three significant figures, inspect the fourth digit (4). Because 4 is less than 5, the retained digits stay: 2.44.

Double rounding (round-half-up): First round 2.4445 to four significant figures. The fifth digit is 5, so round up: 2.445. Then round 2.445 to three significant figures. The fourth digit is 5, so round up: 2.45. The double-rounded result differs by 0.01.

Double rounding (round-half-even): First round 2.4445 to four significant figures. The digit to drop is 5, and the preceding digit is 4 (even), so it stays: 2.444. Then round to three significant figures: 2.44. Here the tie-breaking convention masks the error, but double rounding remains risky.

Example 2: 0.125 to 2 decimal places

Direct: 0.125 rounded to two decimal places using round-half-up is 0.13.

Double: Round first to one decimal place: 0.125 → 0.1 (since the second decimal is 2). Then round 0.1 to two decimal places: 0.10. The result is off by 0.03. This example shows that rounding to a precision lower than the final target destroys information before the final step.

Example 3: Uncertainty reporting

Suppose a measurement gives mean y = 10.2345 and standard uncertainty u(y) = 0.0123. GUM clause 7.2.6 recommends reporting the uncertainty to two significant figures: 0.012. The mean should then be rounded to the same decimal place: 10.234. If you first round the mean to 10.235 and then to 10.23, you may introduce double-rounding error. Always round the uncertainty first, then the estimate, using guard digits in the intermediate calculation.

Counter-Examples

Double rounding is often mistaken for harmless tidying. These counter-examples show where it fails.

• Summing rounded addends: Compute 1.005 + 1.005 + 1.005. Direct sum = 3.015, which rounds to 3.02 at two decimal places. If you round each addend first: 1.01 + 1.01 + 1.01 = 3.03. The difference is 0.01.
• Chained unit conversions: Convert 1.2345 m to cm (123.45 cm) and then to mm (1234.5 mm). If you round to 123 cm and then to 1230 mm, you lose 4.5 mm. Carry full precision through the chain.
• Percentage of a rounded base: 0.445% of 1000 = 4.45, which rounds to 4.5. If you first round 0.445% to 0.45%, you get 4.5. Here the result matches, but if the base were 999, the error would appear. Test with multiple values before assuming safety.

Convention Comparison Table

Standards Citation

Several international standards directly address rounding and double rounding.

• ASTM E29-13a, Standard Practice for Using Significant Digits in Test Data to Determine Conformance with Specifications, Section 6. It specifies a consistent rounding method and warns against intermediate rounding in conformance decisions.
• ISO 80000-1:2009, Quantities and units — Part 1: General, clause 7.3. It states that rounding should be performed only once and gives the round-half-even rule for exact ties.
• JCGM 100:2008 (GUM), Evaluation of measurement data — Guide to the expression of uncertainty in measurement, clause 7.2.6. It advises against excessive digits and recommends rounding the uncertainty before the estimate.
• NIST Special Publication 811, Guide for the Use of the International System of Units (SI), section 9.4. It provides practical rounding guidance for SI values.

"The numerical values of the estimate y and its standard uncertainty u(y) or expanded uncertainty U should not be given with an excessive number of digits." — JCGM 100:2008, 7.2.6.

Common Mistakes

• Rounding at every step. This is the classic double-rounding error. Keep intermediate values in full precision or with guard digits.
• Mixing tie-breaking rules. Using round-half-up for one step and round-half-even for another can create inconsistent results.
• Trusting displayed values. A calculator may show 2.445 but store 2.4445. Re-entering the displayed value introduces double rounding.
• Rounding uncertainty after the mean. GUM recommends rounding the uncertainty first, then the estimate to the same decimal place.
• Confusing significant figures with decimal places. These are different concepts. See Rounding vs Significant Figures.

Practice Problems

1. Round 3.14159 to three significant figures directly. Then round it to four significant figures first, and then to three. Compare the results. Answer: Direct 3.14; double: 3.142 → 3.14. No difference here, but try 3.1455.
2. Compute 1.005 + 1.005 + 1.005. Round each addend to two decimal places, sum, and compare with the direct sum rounded to two decimal places. Answer: Double 3.03; direct 3.02.
3. Mean = 12.3456, u = 0.0789. Report according to GUM 7.2.6. What happens if you round the mean to two decimal places first? Answer: u → 0.079; mean → 12.346. If mean first: 12.35 → 12.35, which may misalign with uncertainty.

Software Behavior Note

Software often uses binary floating-point (IEEE 754), where decimal fractions such as 0.1 and 0.125 are not exact. This can cause double rounding even when you round only once in decimal. For example, Python’s round() uses round-half-even, Excel’s ROUND() uses round-half-away-from-zero, MATLAB rounds half-away-from-zero, and JavaScript’s Math.round() rounds half-up toward +∞. Chaining these functions or re-entering displayed values can produce double-rounding errors. Use decimal libraries, carry guard digits, or use our significant figures calculator for controlled rounding. See Excel, Python, and JavaScript floating-point.

Discipline Note

Different fields have different conventions. Chemistry and biology often emphasize significant figures for lab measurements. Physics and engineering use uncertainty propagation and standards such as ASTM E29 for conformance. Astronomy may carry many digits but still rounds final results. Finance may use banker’s rounding to avoid systematic bias. Always follow the governing standard for your discipline. For more, see Chemistry, Engineering, and Measurement Uncertainty.