Measurement variation
- Concept
Measuring the same thing twice almost never gives the same result: the person changes, the observer changes and the device changes.
What it is
Every clinical measurement varies for several reasons: the person's biological variation from one moment to the next (within-subject), the same observer repeating a reading (intra-observer), different observers (inter-observer) and the instrument. Knowing these sources helps decide how many measurements to take, how to standardise them and when a change between two visits is real. This page covers the sources of variation and how to reduce them; the difference between validity and reliability has its own page.
Worked example
- A patient's systolic pressure: the standard deviation between readings in the same person is 6 mmHg.
- With a single reading, 95% of values fall within ± 1.96 × 6 ≈ ± 12 mmHg of the usual value.
- With the mean of 3 readings: 6 / √3 ≈ 3.5 mmHg; interval ± 1.96 × 3.5 ≈ ± 7 mmHg.
- With 4 readings: 6 / √4 = 3 mmHg; interval ± 5.9 mmHg.
- A reading of 146 mmHg with a usual value of 136 fits within the variation of a single measurement.
- Reading: averaging several readings cuts the noise nearly in half; deciding on one reading can misclassify.
Sources of variation and how to reduce them
| Source | Example | How to reduce it |
|---|---|---|
| Within-subject (biological) | Blood pressure changing during the day | Several measurements, same time and conditions |
| Intra-observer | Same radiologist, different reading another day | Written criteria, reading blind to the previous one |
| Inter-observer | Two doctors hear different sounds | Joint training, operational definitions, measure kappa or ICC |
| Instrument | An uncalibrated blood pressure monitor | Regular calibration against a standard |
Markers
- Intra-observer
- Inter-observer
- Kappa
- ICC
In practice
- Before diagnosing on a single value (blood pressure, glucose, spirometry), confirm it with repeated measurements.
- A change between two visits is real only if it exceeds the expected measurement variation.
- In studies, random error in the exposure usually weakens the association toward the null.
Common pitfalls
- Believing that averaging fixes an uncalibrated device: it only reduces random error.
- Comparing studies that measured with different protocols.
- Not reporting the reliability (kappa, ICC) of the main measurements.
Test yourself
Think of the answer, then tap the question to see it.
Within-subject SD of 6 mmHg. What is the standard error of the mean of 4 readings?
6 / √4 = 3 mmHg.
Which type of error does averaging reduce, and which does it not?
It reduces random error; it does not fix systematic error (bias), which needs calibration.
How is inter-observer agreement measured for a categorical variable?
With the kappa statistic, which corrects for agreement expected by chance.
Connections
Educational content on epidemiological and statistical methods; the examples use made-up numbers to teach the calculation. Not a clinical or legal guide: current Colombian rules are in the official sources.
Sources: 2.1.1.4. Variability (NIST/SEMATECH e-Handbook of Statistical Methods, 2012); 2.1.1.3. Bias and Accuracy (NIST/SEMATECH e-Handbook of Statistical Methods, 2012); 2.4. Gauge R & R studies (NIST/SEMATECH e-Handbook of Statistical Methods, 2012).
Spotted a mistake?We are always improving for the community. Tell us what is wrong and our team will review it.Report a mistake
Representative model, for educational purposes only.