Standard Error of Measurement — Definition, Formula & Example
The standard error of measurement (SEM) is a number that tells you how much a person's test score would typically bounce around if they could take the same test many times. A small SEM means the test gives consistent scores. A large SEM means any single score is less trustworthy.
In classical test theory, each observed score is modeled as , where is the true score and is random error with mean 0. The standard error of measurement is the standard deviation of . It is estimated by , where is the standard deviation of observed scores and is the test's reliability coefficient.
Key Formula
Where:
- = Standard error of measurement, in the same units as the test scores
- = Standard deviation of the observed test scores for the group
- = Reliability coefficient of the test, a number from 0 to 1
Worked Example
Problem: A reading test has a score standard deviation of 15 points and a reliability coefficient of 0.96. A student scores 112. Find the SEM, then build an approximate 95% band for the student's true score using ±2 SEM.
Step 1: Subtract the reliability from 1 and take the square root.
Step 2: Multiply by the standard deviation of scores.
Step 3: Go 2 SEMs below and above the observed score.
Answer: The SEM is 3 points. The student's true score most likely lies between 106 and 118, with about 95% confidence. This assumes measurement errors are roughly normal and the same size across score levels.
Why It Matters
Schools use the SEM to decide whether a student just below a cutoff, such as a gifted-program or placement score, truly differs from one just above it. Score reports for tests like the SAT and state exams often show a score range alongside the reported score for this reason.
Common Mistakes
Mistake: Treating the SEM like the standard error of the mean and expecting it to shrink when more students take the test.
Correction: The standard error of the mean, , measures how precise a sample average is. The SEM measures how precise one person's score is, and it depends on the test's reliability, not on sample size. On a fixed score scale, lowering the SEM takes a more reliable test.
