Blog / Standard Error of Measurement: Why a Reliability Coefficient Is Not a Person’s Score Error
Standard Error of Measurement: Why a Reliability Coefficient Is Not a Person’s Score Error
A reliability coefficient has no score unit; SEM does. That difference matters before interpreting any individual result.
October 5, 2026 · 6 min read · By Cepingku Editorial Team
Ready to start?
Start the Big Five Personality Test →A reliability coefficient of .84 is sometimes turned into a personal statement: ‘this person’s score has 16 percent error’ or ‘this individual result is precise.’ Neither statement follows. Reliability first concerns how consistently observed scores preserve people’s relative ordering within a specified group and for a specified source of measurement error. ETS explains that a reliability coefficient is affected by the spread of scores in the group. Change the group’s score dispersion and the coefficient can change even when the measurement procedure itself has not changed. That makes the coefficient a property of a particular score use and population context, not a unit-labelled error attached to one respondent.
The standard error of measurement, usually called SEM, answers a different question. It expresses the typical size of measurement error in the same units as the observed score. If a total is reported in raw points, SEM is in raw points. If a legitimate scoring transformation reports a 0–100 display score, the corresponding SEM must also be expressed in that display-score unit. ETS describes SEM as a typical measurement error stated in raw-score units. A coefficient such as .84 has no point unit at all, so it cannot by itself tell a reader whether the relevant error is four points, eight display points, or something else.
| Measure | Question it answers |
|---|---|
| Reliability coefficient r | How consistently scores preserve relative ordering within the specified group; it is not an individual’s percentage error. |
| SEM | The typical measurement error in the stated score unit; it is not an absolute individual decision boundary. |
A fictional calculation needs shared conditions
Use this deliberately fictional teaching example only. Suppose a questionnaire has a total-score standard deviation of 10 points in one group, a reliability coefficient of .84, and an SEM referring to the same group and the same error source. Under those shared conditions, the relationship may be written as r = 1 − (SEM/SD)^2, or rearranged as SEM = SD × √(1−r). The arithmetic is SEM = 10 × √(1−.84) = 10 × √.16 = 4 points. ETS discusses this relationship between reliability, SEM, and score standard deviation when they refer to the same population and error source.
The result is not ‘16 percent error,’ and it is not an estimate for this site’s scores. The coefficient .84 is unitless; the four points inherit the unit of the fictional total score. A different group can have a different standard deviation and a different reliability coefficient because its scores are more concentrated or more spread out. You therefore cannot lift a coefficient from another manual, language version, or sample and use it to announce one respondent’s personal error range.
Changing the score scale changes the SEM unit
Continue with the fictional four-point SEM. If a 0–50 total is transformed linearly to a 0–100 display, each raw point becomes two display points. The four-point SEM becomes eight display points. Reporting only r = .84 cannot reveal which of those units is in use. The same issue arises when a score is converted to a mean, a percentage scale, or a new total with a different item count: state the scale, the transformation, and the SEM in the resulting unit. A familiar coefficient is not a substitute for that information.
This is also why a high reliability coefficient alone does not establish that a small difference between two scores is meaningful. Interpretation depends on the instrument version, scoring rule, population, score distribution, and error source. Reliability is not another name for validity. It does not, by itself, establish norms, validity, or measurement precision for this site’s assessment.
Do not turn ±1 SEM into a personal guarantee
Writing a score with ‘±1 SEM’ can sometimes communicate typical uncertainty under a specified model. It is not an absolute cut line, and it does not mean a person’s true score must lie inside that interval. The meaning of any interval depends on the error model, intended use, and probability statement. Without those conditions, two numbers drawn around a score can imply more certainty than the evidence supports.
Before interpreting an individual report, ask three concrete questions. Which instrument version and population produced the coefficient? What is the numerical SEM and its score unit? Do the coefficient and SEM refer to the same scoring method and error source? If those details are unavailable, the appropriate conclusion is that the score’s measurement error cannot be estimated from the coefficient alone. The Big Five Personality Test can help organise responses across five dimensions. It should not be read as if this fictional calculation, another instrument’s reliability coefficient, or another instrument’s SEM supplied a personal error estimate, norms, or validity evidence for this assessment.
Want to see where you land?
Take the Big Five Personality Test and get your results in minutes.
Take the Big Five Personality Test →