Can Convergent Evidence Support an Intelligence-Test Interpretation?

What claim is convergence being asked to support?

Yes. Convergent evidence can support a bounded interpretation of reasoning performance when related measures behave as a theory predicts in the relevant population. Convergent evidence means an expected association between a measure and other measures of a related ability. This supports the interpretation that performance may reflect that ability. It does not establish that two tests use the same score scale or that their results can be substituted for one another for a particular person. The distinction matters because a test score does not carry one interpretation into every setting. Validity is the degree to which evidence and theory support a specified interpretation of scores for a proposed use. The Standards for Educational and Psychological Testing explain that validation evaluates interpretations for uses, rather than assigning a permanent validity label to a test itself. Each intended inference needs its own support. This gives readers a two-step way to assess claims. First ask whether the pattern of relationships supports interpreting the tasks as measures of a particular reasoning ability. Then ask whether there is evidence that the scores are close and precise enough to stand in for one another for the same person and decision. A study answering the first may leave individual score differences unresolved. The tutorial “Designing and evaluating tasks to measure individual differences in experimental psychology” describes convergence and warns that convergent data alone do not provide sufficient construct evidence. The 2014 Standards require accumulated evidence and theory to support a specific interpretation and use. An ETS report, “Validity of Test Interpretation and Use,” also emphasizes the meaning, relevance, and utility of scores. Together, these sources support a qualified conclusion: convergence can strengthen an interpretation, but it cannot by itself prove equivalent IQ numbers. Before weighing any correlation, identify which of these two claims it was designed to support.

Sources: Standards for Educational and Psychological Testing; Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial; Validity of Test Interpretation and Use

When does convergence support a reasoning interpretation?

Convergent evidence is most informative when a test behaves as theory predicts across related and distinct measures. Here, convergent evidence means an expected association between a proposed ability measure and measures of related abilities. It does not settle what the test measures, how consistently it measures it, or what decisions its score supports. One positive correlation is a clue, not a complete case. Consider an illustration, not a test item or study result. A researcher proposes that two different tasks draw on reasoning: one asks a person to infer a rule from changing visual relations; another asks the person to identify a relationship among quantities described in words. If people who do relatively well on one also tend to do well on the other, the pattern is consistent with a shared reasoning demand. The interpretation gains specificity if a comparison task designed around an unrelated demand does not show the same pattern. This would not prove reasoning is the sole source, but tests more than whether scores correlate. Scores can move together for other reasons. Suppose both tasks use dense written instructions, impose similar time pressure, or require the same response style. Language familiarity or speed under those conditions might contribute to both results. Reusing near-identical questions could also produce covariation through content familiarity. These shared features do not make an association useless; they leave open what it supports. A different method can help distinguish those explanations, provided it still gives a fair opportunity to show the ability under study. The open-access tutorial “Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial” describes convergent evidence as an expected relationship with an established measure of the same proposed trait. Discriminant evidence is the expectation that a measure will not relate in the same way to tasks of independent traits. The tutorial cautions that convergence alone is insufficient. It places these relationships among other considerations: whether task content fits the interpretation, scores are reliable, administration is standardized, norms fit the intended comparison, and theory explains the pattern. Convergence is stronger when predictions are stated in advance and evidence includes varied methods and measures that should not share the proposed ability. That is why comparison tasks should be chosen deliberately. If every task shares language load, timing, and format, several correlations may repeat one design feature. If tasks vary those features while retaining a relevant reasoning demand, agreement is harder to attribute to one shared nuisance. A relationship found only among tasks sharing a format may describe performance under that format, but gives less distinct support for a broader interpretation. This is a question of competing explanations, not a formula for accepting or rejecting a test. A further boundary concerns what the interpretation is meant to do. Evidence that a score reflects reasoning is not automatically evidence that it predicts school performance, work outcomes, or success elsewhere. Those are separate claims about external outcomes and need evidence tied to the relevant population and use. The tutorial distinguishes describing task performance, interpreting it as an ability, connecting that ability to real-life outcomes, and making consequential decisions. The evidence required depends on the intended use, with more stringent support needed as stakes rise. An association between reasoning measures therefore cannot alone justify a prediction or decision rule. The “Standards for Educational and Psychological Testing” treats validity as support for a specified interpretation and proposed use, integrating evidence and theory. A coherent pattern across related measures can make an interpretation more credible, especially when less-related measures and shared task features provide useful checks. It cannot turn correlation into proof of a single cause or establish that every practical use is justified. Credit convergence without overreading it: ask whether several independent predictions survive plausible alternatives, and whether the evidence fits the people and use at hand.

Sources: Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial; Standards for Educational and Psychological Testing

Why can related scores still disagree for one person?

Correlation describes how two sets of scores vary together across a group. It does not tell you that two scores agree closely for each person. Scores can be strongly associated even when one measure tends to sit higher or the size of the difference changes from person to person. A group pattern may support a shared reasoning interpretation while leaving an individual’s result uncertain.

A raw score is the performance recorded under a test’s scoring rules,. A norm-referenced score is a conversion that compares performance with a specified reference group. Raw totals from different tests do not automatically share a scale, and two scores labelled “IQ” do not become comparable just because they share a label. Interpretation depends on what was counted, how it was converted, and which population supplies the comparison. A reported range or interval must also rest on evidence for that instrument; a label alone cannot establish its uncertainty.

Reliability concerns the consistency of scores under stated conditions; precision concerns how much uncertainty remains around a score or difference. Neither alone establishes that two tests are interchangeable. The open tutorial “Designing and evaluating tasks to measure individual differences in experimental psychology” treats standardization, reliability, norming, and validity as distinct requirements. Performance can reflect both the ability of interest and method-specific influences. Familiarity, timing, language, item mix, and administration can affect observed performance without showing a change in underlying ability.

A ruler analogy makes the statistical distinction visible: two rulers may track length in the same direction while one is marked with a consistent offset. Their readings could correlate across objects, yet disagree about a particular object’s length. Intelligence tests are more complex, but the analogy captures correlation’s limit. Association alone does not show the size or spread of score differences, or whether one examinee’s gap is small enough for a decision.

The empirical comparisons show why those questions need direct evidence. “The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests” examined seven tests in people aged 4–20. Its abstract reports mostly substantial correlations, alongside several significant group mean differences; the researchers also could not reliably replicate exact-score 95% confidence intervals across tests for individuals. Bounded to those tests and ages, this shows group convergence does not guarantee matching individual estimates. In a different, selected sample, “The exchangeability of brief intelligence tests for children with intellectual giftedness” studied 36 children who took three brief tests. Despite strong convergent validity and reliability evidence, the reported dependability coefficient was .80, and test-by-examinee interaction contributed about one-third of IQ-score variance. This is no universal threshold; it shows test choice can matter even when measures are related.

So, when comparing two results, ask for more than their correlation: identify whether each is raw or norm-referenced, the conversion and reference group behind it, the score’s precision, and whether person-level differences were studied for the relevant population and use. Without that evidence, related scores may support a cautious interpretation of reasoning performance, but they cannot establish that the numbers are interchangeable for one person.

Sources: Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial; The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests; The exchangeability of brief intelligence tests for children with intellectual giftedness

What does a direct comparison of seven tests show?

A direct comparison makes the distinction between related scores and interchangeable scores concrete. The 2021 primary study “The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests” examined seven current, widely used intelligence tests in participants aged 4 to 20. Its abstract reports that results from the tests were mostly substantially correlated. It also reports significant differences between some test means and says exact IQ confidence intervals could not be reliably replicated across tests at the individual level. Together, those findings support a bounded interpretation: the tests often tracked related differences across people, but the study did not establish that each test would give the same person an equivalent score or uncertainty range. A correlation summarizes how two sets of scores vary together across the people studied. If people with relatively higher scores on one test also tend to have relatively higher scores on another, the tests show an association. That pattern can matter: it is consistent with the measures capturing some shared aspect of performance. But correlation does not require the two scales to have the same center, spread, or meaning at every point. One test can produce systematically higher values than another while preserving much of the same ordering. And a broad group pattern cannot tell us, by itself, how close the two results will be for a particular examinee. The study’s report of differences between some group means therefore matters alongside the correlations. A mean is the average score in a group. When test means differ, the results caution against treating their numerical values as if the tests automatically use one common scale. It reports a more limited result: substantial association was common, while some average levels differed. That combination can support an interpretation of related performance and still leave a scale-equivalence question unanswered. The individual-level finding addresses a different question from the group correlations. A confidence interval expresses uncertainty around a score estimate under the methods and assumptions of a particular test. The abstract says that exact-score confidence intervals were not reliably replicated across tests for individuals. This is a warning against reading a similar group trend as a promise that the tests will place one person at the same point, with the same precision. That distinction does not make group correlations useless. A measure that behaves in expected relation to other measures can contribute evidence for a proposed interpretation, especially when the comparison population and intended use fit. The seven-test result narrows what that evidence can establish. It supports the claim that test scores often moved together in this study; it does not, from correlation alone, settle whether a score from one test can replace a score from another in an individual report or decision. Such substitution asks for evidence directly about differences and uncertainty at the person level, not only evidence about shared movement across a sample. The authors also reported larger differences toward above-average scores in the studied sample. That observation belongs to this comparison and should remain attached to it. It is not a general rule about people who score above average, a basis for ranking individuals, or a finding that can be transferred automatically to other tests or populations. The abstract’s age range is 4 to 20, so the result does not establish how adult scores compare across these instruments. For a reader, the useful lesson is to ask what level of claim a comparison supports. Evidence that scores correlate can make a shared reasoning interpretation more credible within the studied context. Evidence that some averages differ and individual confidence intervals do not replicate reliably blocks a leap from “these measures are related” to “these numbers are equivalent.” A careful account should retain both sides: convergence contributes information, and the reported limits matter when someone proposes substituting exact scores. This study cannot validate a separate browser quiz or its conversion simply because that quiz concerns reasoning. The participants, instruments, and comparison methods described in the abstract do not supply an adult reference population or score conversion for an unrelated assessment. The conclusion is narrower and more useful: the seven-test comparison shows why convergent evidence can support a bounded interpretation while leaving individual score equivalence unproven. Before treating two results as substitutes, a reader needs direct evidence that addresses the instruments, population, precision, and decision at issue.

Sources: The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests

Why can strong convergence coexist with weak substitution?

Convergent evidence and score substitution answer different questions. A set of tests may move together across a group because they capture related reasoning abilities, yet differ enough in what they yield for a particular examinee that one result cannot simply stand in for another. Brief-test research shows how both can be true. In “The exchangeability of brief intelligence tests for children with intellectual giftedness,” researchers administered three brief intelligence tests to 36 children identified as intellectually gifted. Primary examiners scored the tests live, and a second examiner later scored them. They calculated six IQ values per child and analyzed all 216 using generalizability theory, which separates sources of score variation. They reported strong convergent validity and reliability evidence for the brief IQs, but the dependability coefficient was .80, which they interpreted as relatively low exchangeability across tests and examiners. The test-by-examinee interaction contributed about one-third of the IQ variance; the other listed components, including examiner effects and the test effect by itself, contributed little (Irby and Floyd, 2017). The interaction gives the result practical meaning: differences among tests were not uniform for every child. The particular test and examinee combined to affect the observed IQ. Strong overall relationships can still yield a less dependable individual match. Broad patterns may align across the sample without guaranteeing the same score for each child or the same difference for everyone. A group association summarizes co-variation; exchangeability asks whether an individual outcome stays sufficiently similar when the test changes. The study measured exchangeability directly and found limits despite favorable convergence and reliability evidence. The scope matters. The study involved only 36 children in a selected gifted group, three brief tests, and two scoring occasions involving different examiners. It is not a universal rule, a study of adults, or evidence that all short tests are poor substitutes. Its .80 coefficient is not a general pass/fail threshold; it answers a bounded question about exchangeability under the tests, population, and design examined. This is why “the tests correlate” and “their scores can be substituted for this person” should not be treated as synonyms. A different, narrower study shows why the conclusion should not become an absolute ban on short forms. “External validation of the equivalence of two computational formulas for a WAIS-R seven-subtest short form” compared weighted and prorated formulas for estimating scores from the same seven-subtest form with the full WAIS-R in a mixed clinical sample. The PubMed abstract reports that the formulas produced nearly identical results, with validity coefficients for WAIS-R IQs from .95 to .97. The prorated formula had a slight advantage on several comparisons, including fewer differences greater than two standard errors of measurement and fewer full-scale IQ misclassifications (Hilsabeck, Schrager, and Gouvier, 1999). This counters the claim that a shortened assessment can never yield a close estimate. It compares two formulas applied to one short form against its corresponding full WAIS-R in the sample studied. It does not show that unrelated instruments share an interchangeable scale, that every examinee's result will match, or that a short estimate suits every decision. The abstract gives no sample size. That close agreement supports this bounded use, not transfer to another measure or population. Read beside “The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests,” the studies clarify why the evidence can look mixed without contradicting itself. That 2021 study examined seven current tests in people aged 4–20. It reported mostly substantial correlations, alongside several significant group mean differences; its abstract says exact-score 95% confidence intervals could not be reliably replicated across tests at the individual level. The broader comparison and selected 36-child study are not replications: they used different tests, samples, and analyses. Both illustrate the same boundary. Related measures may support an interpretation of shared ability patterns while the evidence for substituting an exact score remains separate. The standard is directness of evidence. If the question is whether a task reflects a proposed reasoning ability, expected relationships with other measures can contribute support, when alternative explanations are considered. If the question is whether one person's score can replace another instrument's result for a consequential decision, the evidence must address individual differences and uncertainty for the relevant population, tests, and use. The 1999 result shows close estimates are possible in a defined comparison; the other studies show why convergence alone cannot establish that possibility. A correlation is a reason to examine an interpretation, not a shortcut around evidence of person-level agreement.

Sources: The exchangeability of brief intelligence tests for children with intellectual giftedness; The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests; External validation of the equivalence of two computational formulas for a WAIS-R seven-subtest short form

What changes when the population or score label changes?

The inference narrows when the people, test, administration, or reference group differ from those in the evidence. A relationship between scores in one population does not establish that it holds for adults taking a browser quiz. And the label “IQ” does not put two numbers on a validated common scale. Before carrying a result to another setting, ask what was measured, in whom, under what conditions, and for which interpretation. The Standards for Educational and Psychological Testing make intended population and use part of the interpretation question. They note that tests aimed at the same construct can differ because their purposes, uses, and examinee populations differ. Their validity framework says evidence must support the propositions underlying a particular interpretation and use. A study may inform a reasoning interpretation without settling whether its result supports an adult comparison, an educational decision, or substitution for another instrument. Each is a further claim with its own evidence needs. Population is not a footnote. Age, sample selection, language, and experience may affect the demands a task places on a person and how scores vary. A result in children, or a narrowly selected group, does not automatically describe adults generally. An association among established tests also does not transfer by analogy to a new measure with different content, delivery, timing, or scoring. The 2024 tutorial “Designing and evaluating tasks to measure individual differences in experimental psychology” treats standardization, reliability, norming, and validity as related but distinct parts of evaluating a performance task. Evidence for one does not supply the others. Administration matters too. A timed paper task, an untimed browser task, and an adaptive activity may ask people to do different things. If instructions, access to tools, item selection, or scoring rules change, evidence for the original procedure may no longer support the same interpretation. The Standards call for the intended population, construct, and uses to be specified, and for new evidence when an interpretation is applied to a new group or changed test conditions. That limits transfer; it does not make every result useless. Score labels also need unpacking. A raw score records points earned under one test’s scoring rules. A norm-referenced score compares performance with a defined reference group through a documented conversion. A domain result may summarize one part of an activity. These answer different questions. Even when two reports say “IQ,” the label alone does not show that their conversions, reference populations, precision, or meanings match. A band generated by placeholder conversion parameters is not a confidence interval: it does not quantify uncertainty using a validated measurement model for an adult population. That distinction is decisive for Test IQ Free. Its 50-question adaptive activity covers pattern reasoning, number patterns, and spatial reasoning, and provides domain feedback. These facts describe the activity and its recorded performance. They do not establish a validated adult norm group or validate the provisional estimated IQ range. Evidence from other instruments cannot fill that gap without direct validation of this conversion for its intended population and use. A practical transfer check is simple: does the evidence concern a comparable population, measure, administration, score meaning, and decision? When one changes, narrow the conclusion and identify what remains untested. Use the browser result as feedback about performance on this activity; do not treat its estimated range as an established adult comparison or borrow support from child studies by analogy. If a consequential decision depends on a score interpretation, the relevant evidence must fit that question and population.

Sources: Standards for Educational and Psychological Testing; Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial; The comparability of intelligence test results: Group- and individual-level comparisons of seven intelligence tests

What can this publication's browser result responsibly mean?

It can be treated as feedback about performance on this 50-question activity and its reasoning domains. The estimated IQ range is provisional output from an unvalidated conversion with no validated adult reference population. It is not a normed IQ score, a confidence interval, an established comparison with adults, or evidence that this result is equivalent to one from a professional instrument. The practical distinction is between what a person did on the activity and what the conversion claims that performance means. The first is observable within the activity; the second needs evidence that this conversion does not yet have. The assessment uses original adaptive questions across pattern reasoning, number patterns, and spatial reasoning, and provides feedback by domain. Those details describe the tasks and the product’s feedback. Fifty questions do not, by themselves, show how precisely a score measures an ability, how it compares with a defined population, or whether its interpretation works for a particular decision. Likewise, naming reasoning domains describes intended task coverage; it does not establish that the resulting range shares a scale with a professionally administered intelligence test. The *Standards for Educational and Psychological Testing* frames validity around evidence supporting a specified interpretation and use of scores. Applied here, evidence that other intelligence measures relate to one another cannot fill the missing validation step for this product’s conversion. Such studies concern their own instruments, samples, and scoring procedures. They may help explain why related reasoning measures can show convergent patterns while still differing for individuals, but they do not test this browser assessment or establish an adult reference group for it. The right conclusion is therefore limited: the result can describe performance on this activity, while its IQ label remains a provisional estimate rather than an established population comparison. For a low-stakes purpose, such as curiosity or practice, a reader can review the domain feedback and worked answers to see which operations were involved and where an explanation may help. This can guide what to practise next. It does not show that practice raises general intelligence, and a change on similar questions would not by itself establish a broad ability change. Treat the feedback as a prompt for learning about task performance, not a verdict about identity, worth, diagnosis, giftedness, or fixed potential. The optional $9 report adds explanation, worked examples, applications, and a practice plan, but its point estimate and percentile come from the same unvalidated conversion. Payment therefore adds detail, not accuracy or validity. If a consequential educational, clinical, or other decision depends on an ability assessment, consult a qualified professional who can select an assessment for that question and population and explain its evidence. The browser range should not carry that decision. For open practice, the free test and its worked answers are the proportionate next step; for high-stakes interpretation, seek an assessment designed and supported for the intended use.

Sources: Standards for Educational and Psychological Testing

What should you do with the result now?

Use the score as task feedback for curiosity. On the free 50-question test, review the worked answers and choose an operation to practise, such as tracking a changing quantity or mentally rotating a shape. This responds to performance here; it does not define your identity or potential. The Standards for Educational and Psychological Testing frame validity around evidence for a particular score interpretation and use, so your next step depends on what you intend to decide.

For low-stakes practice, take the free test at /test and use its explanations to guide work with similar operations. Familiarity is a proportionate aim; this evidence does not establish that practice raises general intelligence. The displayed IQ range is provisional, from an unvalidated conversion with no validated adult reference population. It is not a normed score or established comparison. The optional $9 report adds explanation to the same conversion, not validity. If an educational or clinical decision depends on an ability estimate, consult a qualified professional about a suitable assessment. The tutorial “Designing and evaluating tasks to measure individual differences in experimental psychology” emphasizes that evidence must fit intended use. Convergence can support a bounded interpretation; substituting scores requires evidence for that comparison and decision.

Sources: Standards for Educational and Psychological Testing; Designing and evaluating tasks to measure individual differences in experimental psychology: a tutorial

Questions readers ask