Measurement Errors and Why They Matter

Everything Has Uncertainty

No physical measurement is perfectly exact. Even the most sophisticated instruments have finite precision, and even the most careful technique introduces some variation. Understanding measurement error — where it comes from, how it propagates, and how to quantify it — is essential in science, engineering, medicine, cooking, and any domain where numbers matter.

Percentage

Two Fundamental Types of Error

Random Error

Random error causes repeated measurements to scatter around the true value. On some measurements you get a slightly high result; on others, slightly low. The key properties:

• Unpredictable in direction for any single measurement
• Reducible by averaging many measurements (random errors average toward zero)
• Caused by: electrical noise in instruments, subtle changes in environmental conditions, human reaction time variation, thermal fluctuations

Example: Weighing a 100g reference weight 10 times on a lab scale: results are 99.97, 100.02, 99.98, 100.01, 99.99, 100.03, 99.97, 100.00, 100.02, 100.01 grams. The scatter is random error; the average (100.00) is close to the true value.

Systematic Error

Systematic error consistently biases measurements in one direction. It cannot be reduced by averaging — it must be identified and corrected.

• Predictable in direction (always too high or always too low)
• Caused by: miscalibrated instruments, incorrect technique, environmental bias, using the wrong model
• Often the more dangerous type because it goes undetected

Examples: - A bathroom scale that always reads 2 kg too high (calibration offset) - Measuring blood pressure in a stressful clinical setting (white coat hypertension) - Self-reported dietary intake consistently underestimates calories by ~20% - Thermometers placed in direct sunlight reading higher than ambient air temperature

Error Percentage

Accuracy vs. Precision: A Critical Distinction

These terms are often used interchangeably in everyday language but have distinct technical meanings:

Accuracy: How close a measurement is to the true (or accepted) value. An accurate measurement is correct on average.

Precision: How consistent repeated measurements are with each other. A precise measurement shows little variation, regardless of whether it is correct.

The classic illustration uses a dartboard: - Accurate and precise: Darts clustered in the center - Precise but not accurate: Darts tightly clustered but far from the center (consistent systematic error) - Accurate but not precise: Darts scattered around the center (averaged correctly, but each shot varies) - Neither accurate nor precise: Darts scattered far from the center

In practice, precision is easier to measure than accuracy (because you can repeat measurements), but accuracy is what ultimately matters. A drug dosage measured with poor accuracy but high precision could be consistently wrong in a dangerous direction.

Quantifying Error: Absolute and Relative

Absolute error is the simple difference between the measured and true values:

Absolute error = |Measured value − True value|

Relative (or percentage) error expresses this as a fraction of the true value:

Percentage error = (|Measured − True| / |True|) × 100%

Example: A blood glucose meter reads 118 mg/dL when the lab result is 112 mg/dL. - Absolute error = |118 − 112| = 6 mg/dL - Percentage error = (6/112) × 100% = 5.4%

FDA standards for home glucose meters require that 95% of readings fall within 15% of the laboratory value — so this meter is within specification.

Error Propagation: How Errors Combine

When multiple measurements combine in a calculation, their errors propagate. The rules depend on the mathematical operation:

Addition/Subtraction: Absolute errors add (approximately): - If A = 5.0 ± 0.1 and B = 3.0 ± 0.2 - A + B = 8.0 ± 0.3 (worst case) - A − B = 2.0 ± 0.3

Multiplication/Division: Relative errors add: - If A = 5.0 ± 2% and B = 3.0 ± 5% - A × B = 15.0 ± 7% (worst case: errors add) - A / B = 1.67 ± 7%

This is why long chains of calculations in engineering require careful error budgeting — each multiplication step can add relative errors.

Real-World Consequences of Measurement Error

Medicine — insulin dosing: Insulin dosing depends on blood glucose measurements. A systematic error of 20% in a glucose meter for a diabetic patient taking 10 units of insulin could mean they receive 8 or 12 units — with serious health consequences.

Engineering — material strength: Structural engineering uses safety factors precisely to account for measurement uncertainty in material properties. A steel beam rated at 400 MPa tensile strength might actually be used at 267 MPa (two-thirds of rated), providing a factor of 1.5 safety margin against measurement uncertainty, material variability, and unexpected loads.

Climate science — temperature records: The global average temperature anomaly (deviation from a baseline) is measured to within ±0.05°C at the 95% confidence level. When scientists report a record warm year, they are accounting for this uncertainty in their confidence intervals.

Navigation — GPS accuracy: Standard GPS is accurate to ±3–5 meters horizontally under good conditions. Differential GPS corrections reduce this to ±10 cm. For surveying property boundaries, this difference is legally significant.

Reducing Measurement Error in Practice

Increase sample size to reduce random error: the standard error of the mean decreases as 1/√n, so 4× more measurements halves the uncertainty.

Calibrate instruments regularly against known standards to detect and correct systematic errors.

Use controls and blind measurements to detect systematic bias from technique or expectation.

Report uncertainty explicitly: instead of "the temperature is 98.6°F," write "98.6 ± 0.3°F" to communicate both the measurement and its uncertainty.

Choose appropriate instruments: using a kitchen scale (±1g) to weigh a 5mg pharmaceutical dose is not appropriate, regardless of how carefully you read it.

Measurement error is not a sign of inadequate science — it is an honest acknowledgment that reality is harder to observe than we might wish. Learning to quantify, communicate, and reduce measurement uncertainty is a mark of scientific maturity.

Percentage