A topic in the Open Knowledge Graph — a free, open map of 15,290 topics and the order to learn them in.
Norms transform raw scores into interpretable reference points by comparing an individual's performance to representative samples. Common norm types include percentiles (rank-based), standard scores (z-scores, T-scores, IQ scores), and grade/age equivalents. Selecting appropriate norm groups and understanding their limitations—including when populations have changed—is critical for valid score interpretation.
Examine published norm tables from major tests (WISC, SAT, MMPI-2) and practice converting raw scores. Understand the relationship between percentile ranks and standard score distributions. Compare how the same raw score can be interpreted very differently depending on which normative sample is used.
Your prerequisites in statistics and the normal distribution give you the mathematical tools — z-scores, means, standard deviations, the bell curve — but those tools only produce numbers. What transforms a raw score on a test into something meaningful is norming: the process of situating that score within a reference distribution collected from a real, representative sample of people. A child who answers 34 out of 50 vocabulary items correctly tells you nothing useful until you know how many vocabulary items typical 8-year-olds answer correctly. Norms provide that reference.
The most intuitive norm-referenced score is the percentile rank: the percentage of the normative sample that scored at or below a given raw score. A child at the 72nd percentile performed better than 72% of the comparison group. Percentile ranks are easy to explain to parents and teachers, which is why they are ubiquitous in educational testing. But they have a statistical trap rooted in the normal distribution: because scores cluster near the mean, small raw-score differences in the middle of the distribution translate into large percentile jumps, while the same raw-score difference near the tails moves the percentile barely at all. A student going from the 45th to the 55th percentile may have gained very few raw points; a student going from the 95th to the 99th percentile may have gained many more.
Standard scores solve this by using the normal distribution's equal-interval property. A z-score expresses how many standard deviations above or below the mean a raw score falls. Most applied contexts use linear transformations of z-scores that avoid negative numbers and decimals: T-scores (mean = 50, SD = 10), IQ scores (mean = 100, SD = 15), and SAT scores (historically mean = 500, SD = 100). These are all the same mathematical operation — they differ only in their chosen scale. Because they are interval-scaled, a 10-point difference in IQ has the same meaning anywhere on the scale.
Choosing the right normative sample is as important as the math. Norms are only valid for populations that resemble the norming sample. If you administer a test normed on American adults to recent immigrants, or use norms from the 1990s on today's children, the interpretive reference frame breaks down. The Flynn effect — the well-documented finding that average IQ scores have risen roughly 3 points per decade across the 20th century — illustrates the consequence: outdated norms make current populations look artificially elevated. For high-stakes decisions (educational placement, disability determination), the currency and demographic representativeness of the normative sample are not technical footnotes; they are the foundation on which the entire interpretation rests.