Music taste and intelligence

0 Quick Answer

There is a measured relationship between music preference and cognitive ability, it is small, and it is not the relationship people think it is. The best known study, Satoshi Kanazawa and Kaja Perina's Why More Intelligent Individuals Like Classical Music, published in the Journal of Behavioral Decision Making in 2012, did not find that intelligent people prefer classical music as such. It found a preference for instrumental over vocal music, with classical grouped alongside big band and easy listening on the instrumental side.

The size is the part that gets lost. In their American analysis of the 1993 General Social Survey, the standardized coefficient for verbal ability on instrumental music preference was .1245, while the coefficient for years of education in the same model was .2075. Education predicted the preference more strongly than measured ability did. In their British analysis, the whole model explained 6.19 percent of the variance in instrumental preference, and the coefficient for intelligence was .0990.

What people usually have in mind is a different thing altogether: a chart of average SAT scores by favorite band that circulated widely from 2008 onward. That chart tested nobody. It matched school level averages to school level popularity, and in it the entry Beethoven ranks first at 1371 while the entry Classical ranks 119th out of 133 at 978. That single contradiction is the clearest available demonstration of why aggregate data cannot answer an individual question, and it is covered in full below. Playing music is a separate topic with a separate evidence base, handled on the page on music training and cognitive transfer.

1 How Big These Effects Actually Are

Every finding on this page is small enough that it changes group averages and disappears at the level of a person. Putting the numbers side by side first makes the rest of the article readable, because the temptation with this topic is to treat a real effect as a large one.

The largest published difference in raw means comes from Kanazawa and Perina's illustrative figure. Among General Social Survey respondents in 1993, those who said they liked classical music very much averaged 107 on the survey's verbal measure, and those who disliked it very much averaged 93. That looks dramatic until you read the effect size the authors report alongside it: an eta squared of .0859, meaning classical music preference accounted for about 8.6 percent of the variance in the verbal score, in a bivariate comparison with nothing controlled.

Once controls enter, the relationship shrinks sharply. In their multiple regression predicting intelligence from preferences for all 18 genres simultaneously, with age, race, sex, education, family income, religion, marital status, and number of children in the model, the coefficient for classical music preference was no longer statistically significant. The coefficient that survived was for big band.

The British replication behaves the same way. In the 1986 follow up of the 1970 British Cohort Study, 661 sixteen year olds who said they usually listened to classical music averaged 107 against 100 for the 5,000 who did not, a difference of seven points that is highly significant on that sample size. In the regression model with controls, the standardized coefficient for verbal ability was .0990 and the model as a whole accounted for 6.19 percent of the variance. Teacher rated academic performance had a larger coefficient, at .1303.

An effect of that magnitude is real at the population level and useless as a signal about anybody. It is the same arithmetic that applies to every group difference in the outcomes literature: a few points of difference between group means sits inside distributions that overlap almost completely, so knowing which side of the difference somebody falls on tells you nearly nothing about where they sit.

2 Every Study That Measured Both Things

Four studies have measured music preference and cognitive ability in the same people, and one widely cited artifact measured neither. The table separates them, and each row carries the confound its own authors could not remove.

Read the last two columns together in every row. The pattern across the four real studies is consistent and modest: a preference for instrumental music tracks measured ability with a standardized coefficient somewhere between .10 and .14, and in every case at least one non ability variable in the same model carries a larger coefficient.

3 What the Classical Music Study Actually Found

The finding is about instrumental music, not about classical music, and the distinction changes the conclusion. Kanazawa and Perina were testing a specific evolutionary claim, not a claim about taste or sophistication. Their argument runs that music in its evolutionary origin was vocal, following Steven Mithen's account of a shared precursor to music and language, and that purely instrumental music is therefore evolutionarily novel. Their Savanna-IQ Interaction Hypothesis predicts that more intelligent people acquire evolutionarily novel preferences more readily and that general ability has no effect on evolutionarily familiar ones.

That is why they split 18 General Social Survey genres into three instrumental categories, big band, classical, and easy listening, and 15 vocal ones. Classical was in the instrumental group as a matter of classification, not because the hypothesis was about classical music. The authors acknowledge the split is approximate, note that some classical and some big band music is vocal, and say directly that classifying jazz as vocal made their test statistically conservative.

The result matched the prediction in both countries. Instrumental preference rose with verbal ability, significantly and modestly. Vocal preference did not. In the American data the coefficient for vocal music was .0954 and not significant; in the British data it was .0176 and not significant. The asymmetry, not the size, is what the paper was arguing for.

Two things should temper how anyone uses this. The first is that the ability measure in the American analysis is a 10 item synonym test. The authors defend it, citing Miner's 1957 review of 36 studies finding a median correlation of .83 between vocabulary and general ability, and Wolfle's 1980 report of a correlation of .71 between that specific survey measure and the Army General Classification Test. It is still a verbal measure standing in for a general one, which matters given that verbal knowledge is precisely the domain most sensitive to education. The distinction between accumulated knowledge and reasoning on unfamiliar material is set out on the fluid versus crystallized page.

The second is that the theoretical framework is contested and does not have to be accepted for the descriptive result to stand. Whether or not instrumental music is evolutionarily novel, the correlations are what they are. The hypothesis is an interpretation laid over them, and the authors themselves raise a rival explanation in the discussion, that people may cultivate certain preferences as a costly signal once the association is widely known.

4 The Table Inside the Paper That Undercuts the Headline

The most useful result in the classical music paper is the one that contradicts the title, and the authors published it themselves. Anticipating the objection that their instrumental and vocal split was really a proxy for musical complexity, Kanazawa and Perina regressed intelligence on preferences for all 18 genres at once, so that each coefficient describes a genre while holding the other 17 constant.

With no controls, classical preference predicted 1.84 points per step on the five point scale, significant at p below .01. Big band predicted 2.12 points, a larger coefficient. Gospel predicted negative 2.46 and rap negative 2.13. When the demographic controls were added, in the model with 517 respondents, the coefficient for classical fell to 1.01 and lost significance, while big band stayed significant at 1.76 and oldies became significant at 1.41.

The authors draw the obvious conclusion and print it. It would be difficult, they write, to argue that big band is more cognitively complex than classical, or that oldies, reggae, or Broadway musicals are more complex than opera. They note that Rentfrow and Gosling classified blues, jazz, classical, and folk as structurally complex, and that in the controlled model none of the four correlated significantly with intelligence, with jazz and folk running slightly negative.

They also flag a methodological trap worth carrying forward. Preferences for all 18 genres load positively on a single factor, because some people like music and some people do not, and those who like it tend to like many kinds. Preference for classical was significantly positively correlated with preference for 12 of the 18 genres, including bluegrass and reggae. Any per genre analysis therefore has to partial out general enthusiasm for music, and once it does, the individual genre coefficients become much harder to interpret as statements about that genre.

5 The Croatian Replication With a Nonverbal Measure

The most useful check on the classical music result is a study that replaced the vocabulary test with a nonverbal reasoning test and got a similar answer. Elena Račevska and Meri Tadinac of the University of Zagreb published Intelligence, Music Preferences, and Uses of Music From the Perspective of Evolutionary Psychology in Evolutionary Behavioral Sciences in 2019, volume 13, pages 101 to 110.

They tested 467 Croatian high school students with the Nonverbal Sequence Test, a Scale of Music Preferences, and a Uses of Music questionnaire. That third instrument matters, and it is the paper's main contribution: it distinguishes listening to music for cognitive reasons, such as attending to a performer's technique, from listening for emotional reasons or as background to other activities.

In a forward stepwise regression on 373 cases, intelligence predicted preference for instrumental music with a standardized coefficient of .134, p equal to .003, and the model as a whole explained 33.3 percent of the variance. Intelligence was the smallest of the five significant predictors. Preference for vocal-instrumental music came first at .326, cognitive use of music second at .290, duration of formal music education third at .175, and mother's level of education fourth at .163. Intelligence did not predict preference for vocal-instrumental music once the other variables were in.

Two of the null results in that paper are as informative as the significant one. Higher ability students were no more likely to use music cognitively, with a correlation of .21 that failed to reach significance on 454 cases, and no more or less likely to use it emotionally or as background. And the parental education variables did their own work: students whose parents had more education were more likely to have had structured music education and to have pursued it longer, which is a direct route from social position to the instrumental preference the study is trying to explain.

The authors name their own limitation. The sample was large and diverse on income and parental education but homogeneous on age and educational level, all being high school students at one point in time, and they note that music preferences change with age, which bounds how far the result generalizes.

6 The Structure Underneath Genre Labels

Genre is a marketing category, and the research that made this field tractable started by replacing it with dimensions. Peter Rentfrow and Samuel Gosling published The Do Re Mi's of Everyday Life in the Journal of Personality and Social Psychology in 2003, volume 84, pages 1236 to 1256, across six studies covering more than 3,500 individuals.

Their analyses converged on four preference dimensions rather than a list of genres: Reflective and Complex, Intense and Rebellious, Upbeat and Conventional, and Energetic and Rhythmic. The abstract states that preferences on these dimensions related to personality dimensions such as openness, to self views such as political orientation, and to cognitive abilities such as verbal IQ. The paper is behind a publisher paywall, so this page cites its structure and its stated correlates rather than reproducing coefficients it cannot verify.

Two things follow from the dimensional approach. The first is that a genre label maps onto a dimension imperfectly and differently in different populations. Račevska and Tadinac factor analyzed their own Croatian preference scale against the Rentfrow and Gosling structure and did not recover a clean four factor solution. They obtained five factors, reflective, popular, conservative, intense, and sophisticated, with the equivalents of Reflective and Complex splitting across two of them and several styles loading on more than one factor.

The second is that the same team later revised the structure. Rentfrow, Goldberg, and Levitin published a five factor model of musical preferences in the Journal of Personality and Social Psychology in 2011, volume 100, pages 1139 to 1157, and a replication and extension of that model in Music Perception in 2012, volume 30, pages 161 to 185. Any table that assigns an average score to a genre is therefore working with a unit of analysis that the field itself moved away from twice.

For a reader, the practical upshot is that questions phrased as which genre indicates intelligence cannot be answered on their own terms. The variable that carries the association is a dimension of preference, the dimension is not the genre, and the association with ability is small even when it is measured properly.

7 Openness Is Doing Most of the Work

The trait that predicts music preference most consistently is not intelligence, and treating it as intelligence is the most common mistake in this area. Rentfrow and Gosling reported openness to experience among the personality correlates of their preference dimensions, and openness has remained the strongest and most reproducible personality predictor of musical taste in the work that followed, including Schäfer and Mehlhorn's meta-analysis in Personality and Individual Differences in 2017.

Openness and measured ability overlap, which is why the two get conflated. They are not the same construct and they behave differently. Openness is a disposition toward novelty, aesthetic engagement, and unfamiliar ideas. Measured ability is performance on cognitive tasks against an age reference group. A person can be high on one and average on the other, and the correlation between them is far from unity. The page on IQ and personality sets out where the two constructs meet and where they separate.

Once openness is in the picture, several findings in this area look less mysterious. A disposition to seek out unfamiliar aesthetic experience would predict exactly the pattern Kanazawa and Perina observed, since instrumental music is less immediately accessible than vocal music for most listeners, and it would do so without any claim about ancestral environments. It would also predict Račevska and Tadinac's strongest non ability variable, cognitive use of music, which is essentially a measure of attending to musical structure for its own sake.

The same trait is central to the small and much argued relationship between measured ability and creative output, which is covered with its effect sizes on the creativity page. The pattern to notice is that openness, not ability, is the variable that keeps appearing when taste is the outcome. A page that reports a small ability correlation and never mentions the trait that outperforms it is telling half the story.

8 The Chart Everybody Cites, and the Ecological Fallacy Behind It

The single most circulated artifact in this area tested nobody, and its own author said so. Virgil Griffith's Music That Makes You Dumb, first archived in December 2008, plots an average SAT score against each of 133 favorite music entries. It has been reproduced in newspapers, magazines, and countless posts ever since.

The method is stated plainly on the page. Someone collected the ten most frequent favorite music entries at each college from that college's Facebook network statistics page. The average admission test score for each college came from the College Board. Each favorite music entry then received the average of the school level scores where it appeared. Griffith describes the output as a correlation between musical tastes and dumbitude, adds that he is aware correlation does not equal causation, and writes that the results are hilarity incarnate regardless of causality.

Read the structure of that method carefully. No individual in the dataset ever took an intelligence test. No individual's music preference was ever paired with that same individual's test score. Both variables are school level averages, and the unit of analysis throughout is the institution. The chart is a statement about the composition of student bodies.

The underlying table makes the problem visible without any statistical argument. Beethoven ranks first at 1371, computed from 12 schools with a standard error of 27.7. The entry Classical ranks 119th out of 133 at 978, computed from 47 schools. Jazz ranks 127th at 946, from 82 schools. Gospel ranks 131st at 925, from 115 schools, and Lil Wayne ranks last at 889, from 42 schools. The mean across the 1,352 schools in the dataset is 1071.

The same music appears near the top and near the bottom depending on how the label was typed. Whatever separates Beethoven at 1371 from Classical at 978 is not a property of the music, because there is no music that is in one and not the other. It is a property of which twelve schools happened to list one label and which forty seven listed the other. Anyone using this chart as evidence about listeners is reading institutional composition and calling it taste.

That mistake has a name. Inferring an individual relationship from grouped data is a named statistical error, and it is the error that produces the popular version of this topic. William Robinson set it out in Ecological Correlations and the Behavior of Individuals, American Sociological Review, 1950, volume 15, pages 351 to 357, showing that a correlation computed on group aggregates can differ in magnitude and even in direction from the same correlation computed on the individuals inside those groups.

The reason is that aggregation throws away within group variation, which is usually most of the variation. When you average a school, you discard every difference between the students in it, and what remains is a comparison between schools. Any relationship you then find is a relationship between school characteristics. It may exist at the individual level too, but the aggregate analysis cannot tell you whether it does or how large it is.

Applied here, the chain of reasoning that produces the popular claim has a break in it at every link. Students at more selective institutions have higher average admission test scores, which is true by construction since selectivity is defined partly by those scores. Students at more selective institutions list somewhat different favorite music, which reflects the demographic composition of those institutions. Neither step licenses the conclusion that a person who likes a given artist has a given ability level.

The distortion runs in both directions and cannot be signed in advance. A genre popular at a handful of small institutions with unusual student bodies will get an extreme average from a tiny number of schools, which is exactly what happened with Beethoven's 12 schools and its standard error of 27.7. A genre popular almost everywhere will regress toward the overall mean regardless of who listens to it, which is what happened to Rap, computed from 510 schools with a standard error of 3.8.

This is the same class of error that produces confident claims from every other kind of aggregate table, including national rankings and occupational averages. The page on occupational averages carries the same warning for job categories, and the page on birth month works through another case where a small population level pattern gets read as a personal one.

9 Heavy Metal and What the Swedish Cohort Found

The best evidence on metal fans measures an outcome rather than an ability, and the distinction is the whole of its interpretation. Martin Hällsten, Christofer Edling, and Jens Rydgren published School's Out Forever? Heavy Metal Preferences and Higher Education in PLOS ONE in 2019, volume 14, issue 3, article e0213716.

They followed 2,232 Swedish adolescents born in 1990, of native Swedish, Iranian, and Yugoslav background, collecting favorite bands through open ended questions and linking them to subsequent enrollment in tertiary education at ages 22 and 23. The controls were unusually thorough for this literature: prior school grades, parental socioeconomic status, personality, personal network, and neighborhood characteristics. Net of all of that, metal fans showed a transition rate into higher education roughly seven percentage points lower, which the authors describe as about 13 percent in relative terms.

What that finding is not is a measurement of ability. Grades entered the model as a control, so the effect is on top of measured school performance rather than a proxy for it. The authors also report no difference in career aspirations between fans and non fans, which points away from an explanation based on lower ambition. A preference associated with a different educational path is a sociological result, and it is compatible with identical distributions of measured ability in both groups.

The contrast with the popular narrative is worth holding. There is a widely repeated claim that metal fans are unusually able, usually traced to work on gifted adolescents using music to manage social pressure, and a widely repeated claim that they are unusually not. In the controlled preference research neither claim survives cleanly. In Kanazawa and Perina's per genre model, heavy metal preference carried a coefficient of negative .11 points with full controls, entirely non significant. In Račevska and Tadinac's five factor solution, the equivalents of the intense and rebellious styles formed their own factor and the cognitive use of music correlated positively with preference for it.

That last detail deserves emphasis because it runs against the stereotype. The measure of attending to music for its structure and craft correlated positively with intense music preference, not negatively. Whatever separates metal fans in the Swedish enrollment data, it is not a lack of engagement with music as a thing to be listened to closely.

10 The Confound Nobody Removed

In every study on this page, education or social position carried a coefficient at least as large as ability, and none of them could disentangle the two. This is the reason the honest summary of the field is that the effects are small and mostly explained by things other than ability.

The pattern is consistent across designs. In the American General Social Survey model, education's coefficient was .2075 against .1245 for verbal ability. In the British cohort model, teacher rated academic performance was .1303 against .0990. In the Croatian model, duration of formal music education was .175 and mother's education .163, both above intelligence at .134. Three independent samples, three countries, three different ability measures, and the same ordering.

Race, class, and educational routes are also braided into how genres get categorized in the first place. Kanazawa and Perina's own per genre table shows gospel with the largest negative coefficient in the uncontrolled model, at negative 2.46, larger in magnitude than rap. Gospel's audience in the United States is defined by religious tradition and community rather than by musical complexity, and the authors say as much when they note that gospel is closely linked to religious ritual. A coefficient like that is carrying demographic information, and any reading of it as a statement about music is a misreading.

There is a further route from class to preference that the Croatian study documents directly. Parents with more education were more likely to enroll their children in structured music education, at correlations of .133 for mothers and .121 for fathers, and their children pursued it longer. Formal music education then predicted instrumental preference at .175 in the same model. Family position produces exposure, exposure produces preference, and preference then correlates with ability because family position also correlates with ability.

The attainment side of that chain has its own page. The education level page carries the measured association between completed education and test scores, which is where any residual explanation of these preference findings has to be settled.

11 What None of This Says About You

The individual version of this question cannot be answered from a genre, and the honest response is to measure the thing directly. An effect size of .10 to .14 in a preference model means that if you sorted people by their music and then measured them, the distributions would overlap almost entirely. There is no genre whose listeners are separable from any other genre's listeners on ability.

Guessing does not close the gap either. Self estimates of ability correlate around .3 with measured ability, loose enough that a personal hunch spans most of the usable range, which is worked through with the pooled figures on the self estimation page. Adding a taste based cue to a hunch adds nothing, because the cue carries less information than the hunch does. The same logic applies to every other proxy in this family, including the humor preferences examined on the humor page, which is the closest sibling to this one in the way its evidence behaves.

What can be measured is a profile. ACIS is a self-administered online assessment for adults aged 16 to 90, running 20 subtests across six CHC domains and reporting a Full Scale IQ with six primary indices, sold as a one time payment in three tiers: Quick at 15 dollars, Optimized at 30 dollars, and Full Scale at 50 dollars. For most readers arriving from a question like this one, the Quick tier is the appropriate entry point. It runs six subtests, Similarities, Vocabulary, Matrix Reasoning, Figure Weights, Digit Span, and Alphanumeric Sequencing, in about 45 minutes, and returns verbal comprehension, fluid reasoning, and working memory.

The published reliability figures for those composites are an omega of .9745 for the verbal comprehension index with a standard error of measurement of 2.40 points, .9727 for fluid reasoning with a standard error of 2.48, and .9247 for working memory with a standard error of 4.12. The derivations are in the technical manual, and the domain structure is explained on the CHC model page. The ACIS reference frame is documented in the technical manual.

ACIS is not a clinical instrument, is not diagnostic, and is not appropriate for hiring decisions, accommodation requests, or high IQ society admission. What it produces is a structured estimate of a profile with its measurement error attached, which is a different object from a genre and a much more useful one. What the resulting figure means against the general population is on the percentile chart.

12 What the Evidence Actually Supports

Three claims survive the sourcing, and everything else in popular circulation on this topic does not.

First, preference for instrumental music is associated with measured ability, in three independent samples across three countries, with standardized coefficients between .0990 and .134. The association is small, it is not specific to classical music, and in every one of those models at least one variable related to education or family position carried a larger coefficient. Kanazawa and Perina's own controlled per genre analysis removed classical music's significance while leaving big band's intact, which is difficult to reconcile with any story about musical complexity.

Second, the structure of music preference is dimensional rather than generic, and openness to experience is the personality trait that predicts it most reliably. Rentfrow and Gosling established the four dimension structure across more than 3,500 individuals in 2003 and the same research program replaced it with a five factor model in 2011. A question about genres is asking about a unit the field abandoned.

Third, the most cited artifact in this area is not evidence about people. Griffith's chart matched school level admission test averages to school level popularity across 1,352 schools, and inside it Beethoven sits at rank 1 while Classical sits at rank 119. No conclusion about a listener follows from a dataset in which the same music occupies both positions.

The correct inference from all of this is narrow and worth stating without decoration. Music preference carries a small amount of population level information about ability, most of which is shared with education and social position, and none of which is usable about an individual. Population level effects of a few points are genuine and individually invisible, and both halves of that sentence are load bearing. The myths page catalogues several other claims that survived by repetition rather than by evidence, and this one is a member of that family.

13 Sources Behind This Page

Every figure above comes from one of the following, linked so the arithmetic can be checked rather than trusted.

• Kanazawa, S., and Perina, K. (2012). Why More Intelligent Individuals Like Classical Music. Journal of Behavioral Decision Making, 25, 264 to 275, DOI 10.1002/bdm.730, published online 24 January 2011. Full text. Source of the standardized coefficients of .1245 and .0954 in the American model with n of 786 and 543, the .0990 and .0176 coefficients in the British model with n of 1,160 and R squared of .0619, the mean verbal scores of 107, 103, 101, 95, and 93 by classical music preference with eta squared of .0859, the British comparison of 661 listeners at 107 against 5,000 non listeners at 100, and the Table 4 per genre coefficients including big band at 1.76 and classical at 1.01 with controls on 517 cases. Note the venue: this paper is frequently cited as appearing in Intelligence, and it did not.
• Račevska, E., and Tadinac, M. (2019). Intelligence, Music Preferences, and Uses of Music From the Perspective of Evolutionary Psychology. Evolutionary Behavioral Sciences, 13(2), 101 to 110. Source of the 467 Croatian high school students, the Nonverbal Sequence Test, the regression on 373 cases with R squared of .333 and intelligence at .134, the larger coefficients for cognitive use of music at .290 and music education duration at .175, the parental education correlations of .133 and .121, and the five factor preference structure.
• Rentfrow, P. J., and Gosling, S. D. (2003). The do re mi's of everyday life: The structure and personality correlates of music preferences. Journal of Personality and Social Psychology, 84(6), 1236 to 1256, DOI 10.1037/0022-3514.84.6.1236. PubMed record with the full abstract. Source of the four preference dimensions, the sample of over 3,500 individuals across six studies, and the statement that preferences related to openness, to self views, and to cognitive abilities including verbal IQ. The full text is paywalled and no coefficient from it is quoted here.
• Rentfrow, P. J., Goldberg, L. R., and Levitin, D. J. (2011). The structure of musical preferences: A five-factor model. Journal of Personality and Social Psychology, 100, 1139 to 1157. And Rentfrow, P. J., Goldberg, L. R., Stillwell, D. J., Kosinski, M., Gosling, S. D., and Levitin, D. J. (2012). The song remains the same: A replication and extension of the MUSIC model. Music Perception, 30, 161 to 185.
• Hällsten, M., Edling, C., and Rydgren, J. (2019). School's out forever? Heavy metal preferences and higher education. PLOS ONE, 14(3), e0213716. Source of the 2,232 Swedish adolescents born in 1990, the roughly seven percentage point difference in tertiary enrollment net of grades, background, personality, network, and neighborhood, and the absence of a difference in career aspirations.
• Schäfer, T., and Mehlhorn, C. (2017). Can personality traits predict musical style preferences? A meta-analysis. Personality and Individual Differences, 116, DOI 10.1016/j.paid.2017.04.061. Cited for the standing of openness as the leading personality predictor of musical taste.
• Griffith, V. Music That Makes You Dumb. Archived front page and archived data table, first captured December 2008. Source of the method description, the 133 entries drawn from 1,455 across 1,352 schools, the mean school score of 1071, and the individual entries for Beethoven, Sufjan Stevens, Counting Crows, Classical, Rap, Jazz, Gospel, and Lil Wayne with their standard errors and school counts.
• Robinson, W. S. (1950). Ecological Correlations and the Behavior of Individuals. American Sociological Review, 15(3), 351 to 357. Cited for the ecological fallacy itself; no numeric result from it is reproduced here.
• ACIS technical manual, for the reliability and standard error figures quoted for the Quick tier composites.

The professional framework governing all of this is explicit. The Standards for Educational and Psychological Testing (2014), published jointly by the American Educational Research Association, the American Psychological Association, and the National Council on Measurement in Education, require that a score interpretation rest on evidence for the specific use proposed, that results be reported with their measurement error, and that the limits of the reference sample be disclosed. The APA standards on test use impose the same obligation on anyone reporting a group statistic: say what the number supports, say what it does not, and never let a group average stand in for the measurement of a person.

14 Frequently Asked Questions