What the 2025 Test-Preparation Meta-Analysis Says About IQ Practice

What did the 2025 review measure—and what did it find?

The 2025 review found that preparation can improve performance on the large-scale educational tests it studied. Across 28 experimental or quasi-experimental studies, the pooled difference between prepared and comparison groups was positive: Hedges’ g = 0.26, with a 95% confidence interval from 0.10 to 0.42. Hedges’ g expresses a difference between group averages in standardized units, rather than in the original score units. The confidence interval describes uncertainty around the pooled group estimate; it is not a likely score range for any one person. These results support a measured conclusion about educational-test performance. They do not show that IQ practice raises IQ, and 0.26 is not an IQ-point gain or an individual forecast. The scope matters because “test preparation” in The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies covered varied interventions intended to improve performance on a specific test. These could aim to build relevant knowledge or skills, teach test-taking skills, or do both. The included studies used experimental or quasi-experimental comparisons of preparation with a control condition. The authors identified 92 effect sizes across the 28 studies, so the evidence base contained multiple outcome estimates in some studies; 92 is not the number of independent studies or a count of participants. The tests were educational assessments in real-world educational contexts. Twelve included studies concerned college-admission tests, 11 concerned English-proficiency tests, and five covered other contexts. The review’s question was whether preparation affected performance on these large-scale tests, compared with no preparation or usual practice. It did not test a course in IQ puzzles, nor did it assess whether a gain transferred to a validated measure of general intelligence. The average also hides meaningful variation. The review reports high heterogeneity across study estimates, indicating substantial variation in effects. A pooled mean summarizes that varied body of evidence; it does not promise that each program produces the same gain, or that every student benefits. Read together, the estimate and heterogeneity support the claim that preparation improved average performance across the included educational-test studies, while leaving the size and relevance of any particular person’s gain uncertain. For a reader considering reasoning practice, the useful takeaway is not to dismiss preparation gains, but to name them accurately. If the outcome measured is the prepared educational test, the result supports a claim about performance there. It cannot be translated into IQ units or promised for an online reasoning score. The 2025 review provides evidence about educational preparation; applying it to IQ practice would require direct studies of IQ-practice interventions and suitable outcomes. That distinction keeps a study-level average tied to the test outcome actually observed.

Sources: The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies; Oxford Research Archive record for The Impact of Test Preparation on Performance of Large-Scale Educational Tests

Which preparation features explain the result—and which do not?

The 2025 review found that some preparation features traveled with larger score effects across studies, but it did not show that any one feature caused a gain. In its synthesis of 28 experimental and quasi-experimental studies of large-scale educational tests, workbook use and instruction in test-taking skills were significant moderators. By contrast, the review found little evidence that using sample items or practice tests explained differences in preparation effects. These are findings about patterns across the included studies, not a controlled recipe for an individual learner. A moderator analysis asks whether the estimated effect differs between groups of studies that have a coded feature. It is not the same as assigning otherwise similar students at random to receive a workbook, strategy instruction, or no workbook while holding everything else constant. If studies using one feature also differ in other ways, those differences can contribute to the observed association. The meta-analysis therefore identifies useful candidates for further testing, not isolated causal ingredients. The study categories describe what researchers coded in the available reports; they do not tell a reader which single activity to choose, how much to use, or whether the same association would appear on a different kind of assessment. That distinction matters because preparation is not one uniform treatment. A program may teach relevant subject knowledge, show how to interpret directions, give practice applying a strategy, or reduce uncertainty about the format. The review’s authors proposed domain-specific knowledge and test-specific cognitive skills as possible explanations for score improvement, while noting that further evidence is needed. Such explanations are hypotheses consistent with the pattern, not established accounts of why each participant’s score changed. The fact that a gain is tied to a test does not make the learning unreal; it does limit what the gain establishes about other tasks. Preparation can improve a relevant skill and still leave the broader transfer question unanswered. The results also do not support the simple rule that broader instruction always produces larger gains, or that narrow drilling is always more effective. The review compared studies by instructional breadth, including broad knowledge or skill work, test-specific content, and focus on the same tested items. Those comparisons did not yield a universal prescription. Study mix and design make fine-grained rankings uncertain. For a reader practicing reasoning, the responsible takeaway is modest: features associated with educational-test preparation effects may suggest questions to investigate, but they do not establish a personal plan for raising an IQ score. Learning a relevant operation can be worthwhile even when the evidence only supports better performance on related tasks. To test transfer, researchers need fresh measures and a way to isolate the skill from the wider preparation package. Until then, treat workbook or instruction findings as study-level clues, not promises or mechanisms proven for IQ practice. “The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies” reports educational test outcomes; it does not test IQ practice.

Sources: The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies

Does the review show transfer beyond the prepared test?

The 2025 review offers little evidence that preparation gains carried to a different test, even when the second test measured a similar educational domain. Four studies contributing five effects examined whether preparation for one large-scale test helped on another measure. The pooled estimate was small, g = 0.06, and was not statistically significant (p = .502), as reported in “The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies.” This is a much narrower finding than a direct test of whether reasoning practice changes IQ. The studies concerned educational tests and nearby outcomes, not a broad measure of general intelligence. Transfer is a separate outcome because doing better on the practiced test and doing better on another task are different observations. Near transfer means improvement on fresh tasks that share important operations with the practiced material. Far transfer means improvement on more distant tasks or on a broader ability measure. To claim transfer, a study must measure performance beyond the target test, using a distinct task or criterion. The review’s five transfer effects came from just four studies. Its examples were whether an IELTS preparation course helped performance on the Online Oxford Placement Test, and whether preparation for SAT writing skills helped on writing tasks such as personal statements or complaint letters. Those are comparisons across tests or tasks in related language and writing domains. They do not test IQ practice or general intelligence. The result therefore cannot be translated into a claim about whether puzzle practice changes an IQ score. The authors characterize transfer as appearing specific to the target test, but they also caution that this result rests on limited data and call for further research. That caution matters. A small, non-significant pooled estimate is not proof that transfer never occurs. With only five effects, the evidence is too limited to support a confident conclusion about all educational preparation, much less every kind of reasoning practice. The responsible reading is narrower: this review did not establish reliable improvement on other, similar-domain tests, and it did not examine transfer to general intelligence. This boundary changes how a score gain should be described. If someone prepares for a particular test and later performs better on that test, the observation supports a claim about target-test performance under the measured conditions. It does not say, without another outcome measure, whether the person learned relevant content, became familiar with the format, used a more effective strategy, improved on fresh tasks, or changed a broader ability. The analysis does not distinguish these explanations or justify dismissing preparation as mere trickery: learning task-specific operations may be genuine learning while remaining narrower than broader ability change. For a reader deciding what the 2025 result means for IQ practice, the missing comparison is decisive. The meta-analysis provides evidence about preparation for large-scale educational tests and a small, uncertain set of nearby transfer outcomes. It supplies no direct estimate for an IQ test, and no evidence that practice raises general intelligence. So a gain on a prepared test answers one question: did performance on that test change? It cannot answer whether improvement extends to unrelated reasoning tasks unless those tasks were also measured. The review’s transfer result is too sparse to show that educational preparation reliably carries to even similar-domain tests, and its scope stops well short of IQ practice. Readers may use preparation to learn or perform better on a defined task, but should name that outcome precisely. A claim about broader reasoning or intelligence needs separate evidence on fresh, appropriately different measures.

Sources: The Impact of Test Preparation on Performance of Large-Scale Educational Tests: A Meta-analysis of Experimental Studies

Why can cognitive-test scores rise on retest?

Repeated administrations can raise cognitive-test scores, but retesting is a different exposure from taking a preparation course, and the size of the gain varies with the test and retest conditions. A higher second score shows changed performance under repeat conditions; it does not show broad ability change or transfer to another task. “Retest effects in cognitive ability tests: A meta-analysis” examined repeated administrations, not preparation courses. Its evidence search identified 122 studies, 174 samples, 786 test outcomes and 153,185 participants. The analysis found significant retest effects and no further pooled score gains after the third administration. This leveling pattern describes the included evidence overall; it is not a forecast for one person's next attempt, nor does it mean repetition stops affecting every test after three sittings. The synthesis also found that retest effects varied with the cognitive operation and content, the equivalence of test forms, the interval between administrations, and participant age. Equivalent forms are different versions intended to measure the same construct at comparable difficulty; they need not contain identical items. Conditions shape score comparisons: repeated items may become familiar, while a parallel form can preserve an operation but change items. Moderators across studies do not identify one cause for each gain. Recognition of a format, reduced uncertainty about instructions, or a more effective strategy could contribute to a higher score. A gain could include test-specific learning, reduced interference, or change in the ability being measured; the score alone cannot separate them. “Retest effect” names an increase associated with repeated administration, not its cause or proof of lasting change. “Effects of Practice on Aptitude and Achievement Test Scores” synthesized 40 studies and reported score gains after students took practice forms. Gains were larger when practice and criterion forms were identical than when parallel forms were used; effects also varied with the number of practice tests and the population's ability level. These older aptitude and achievement studies do not predict gains on a current IQ test or adaptive online quiz. They do support recording whether items or forms were repeated when interpreting a change. Retesting and preparation can overlap, but they answer different questions. Retesting repeats a cognitive test; preparation may involve instruction, practice materials, test-taking guidance, or relevant content learning. Retest findings show repetition can matter; they cannot establish that a preparation course caused a gain or that either exposure increased general intelligence. When a score changes, note whether the next sitting reused items or used another form, how much time passed, and what operation was tested. Repeated items make familiarity a relevant possible contributor. Fresh items do not automatically make a task distant from prior practice, and the meta-analysis found form equivalence mattered. Without these details, “my score went up” is accurate but incomplete. Keep the conditions beside the result and describe the change as a gain on that test under those conditions. Do not assign all of it to increased ability or dismiss all of it as familiarity. Retest research makes repetition part of interpretation, while the cause and transfer of any individual gain remain open questions requiring direct evidence.

Sources: Retest Effects in Cognitive Ability Tests: A Meta-analysis; Effects of Practice on Aptitude and Achievement Test Scores

What does separate cognitive-training evidence say about transfer?

Separate cognitive-training reviews offer a useful comparison, but they do not answer whether preparation for educational tests improves IQ. Across these reviews, improvement is more consistently supported on trained and related tasks than on distant outcomes such as broad reasoning. Transfer is a separate question to measure, not an automatic consequence of a score gain. Their findings clarify inference limits; they are not part of the 2025 pooled estimate.

Near transfer means carrying a skill to a new task that shares important features with practice; far transfer means carrying it to a more distant domain with less overlap. A learner who practices comparing changes across adjacent terms might do better on fresh sequences requiring that comparison. That would show improvement on a related operation, not by itself a change in general intelligence. The outcome must be measured directly before a claim can reach that level.

The 2019 article “Near and Far Transfer in Cognitive Training: A Second-Order Meta-Analysis” combined earlier meta-analyses rather than running one new training trial. It covered working-memory training and, in a wider model, programs including video games, music, chess, and exercise games. The authors reported near transfer to other memory tasks, while their estimates for far transfer were small or null. After adjustments for publication bias and possible placebo effects, they reported practically zero far transfer. This is their synthesis under particular inclusion and correction choices, not a universal finding that no training can ever generalize.

Controls matter because participants may improve for reasons beyond the trained operation. An active control gives comparison participants an engaging, thought-demanding activity, accounting for general engagement and expectations. The 2019 authors analyzed active-control comparisons and applied methods intended to address publication bias. Their conclusion should be read as evidence against assuming broad gains from these programs, not proof that every possible transfer effect is exactly absent.

A second review, “Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of ‘Far Transfer,’” focused on working-memory training. Its accessible record describes 87 publications and 145 experimental comparisons and reports no convincing evidence of reliable far-transfer improvement over treated controls. This strengthens caution about one type of training and broad outcomes. It does not study all reasoning practice, rule out improvement on related tasks, or settle whether another intervention could transfer under different conditions.

Together, these reviews support a burden-of-evidence rule. To claim more than task-specific improvement, assess fresh tasks using different operations and use a suitable comparison that accounts for repeated testing and general engagement. A new sequence using the same comparison operation can show that performance carries to another example; it remains near transfer. A claim about IQ or general intelligence requires a direct measure with evidence supporting that interpretation for the population and use. A higher practice score alone cannot supply that evidence.

This keeps cognitive-training findings in their proper place beside the 2025 educational-test meta-analysis, which examined preparation for large-scale educational tests and had only a small set of transfer comparisons in similar domains. The cognitive-training reviews studied different programs and outcomes. Neither is a trial of this site's reasoning test or proof that practice will raise an online estimate. Practice can still help someone learn an operation, gain familiarity, or prepare for a defined test; describe the benefit at the level observed.

Sources: Near and Far Transfer in Cognitive Training: A Second-Order Meta-Analysis; Working Memory Training Does Not Improve Performance on Measures of Intelligence or Other Measures of “Far Transfer”

How should a reader test whether practice transferred?

Use fresh tasks that were not taught or repeated, and say how closely they resemble the practice. Better performance on new examples of the same operation supports a narrow claim: the learner applied that operation to unfamiliar items. It does not by itself show that general intelligence changed. Name the skill and outcome, and include the distance between practice and check. Imagine a learner has practiced noticing how neighboring terms change in a sequence. A check uses unseen sequences and asks whether the learner identifies the relationship without repeating examples or teaching the pattern beforehand. This tests use of a familiar operation on fresh material. A second check could present information in a different reasoning format. Improvement there extends the evidence, but depends on what that format measures and how it was administered. These illustrative checks are not test items or an assessment protocol. Neither establishes broad ability change. Raw performance means the observed number or quality of responses under stated conditions, such as the number of fresh items answered correctly or whether a strategy was applied accurately. A normed interpretation compares a score obtained under specified test rules with results from a defined reference group. Improvement in raw performance is not automatically a change in a normed score. That conclusion needs an appropriate measure and comparison basis; the practice result alone supplies neither. Do not translate an item count, faster completion, or domain-level change into IQ points or a percentile without a relevant, supported scoring basis. Keep conditions similar: use unseen tasks, consistent instructions, and no hints during only one check. Record what changed—accuracy, speed, or strategy—and whether the task was close to practice or meaningfully different. If tasks are close cousins, report improvement on fresh examples of the same operation. If the format differs, describe performance on that format. General-intelligence claims require evidence beyond a small set of practice-related tasks. “Near and Far Transfer in Cognitive Training: A Second-Order Meta-Analysis” reports stronger support for near than far transfer in the cognitive-training evidence it reviewed. It does not test every form of reasoning practice or establish that transfer is impossible. “Retest Effects in Cognitive Ability Tests: A Meta-analysis” addresses repeated test administrations; its reported moderators include test content, form equivalence, interval, and age. Repeating a test is therefore different from checking fresh-task performance after practice. Together, the sources support a reporting rule: state the task, conditions, and distance between practice and check. For a browser reasoning quiz, report observed performance and its limits, not a higher estimate as proof of generalized change. Test IQ Free’s 50-question result is a provisional estimated IQ range from an unvalidated conversion with no validated adult reference population. It is not a normed comparison, confidence interval, or established population position. Fresh examples can support low-stakes exploration of an operation, but cannot validate that conversion. If a consequential decision depends on an ability score, ask whether the assessment’s evidence supports that use and population.

Sources: Near and Far Transfer in Cognitive Training: A Second-Order Meta-Analysis; Retest Effects in Cognitive Ability Tests: A Meta-analysis

What standard should govern interpretation of a practice-related result?

A practice-related score should be interpreted only for a use that evidence supports for that test and the people who took it. If evidence does not support a broader conclusion, keep the interpretation tentative. A gain on a prepared educational test can show better performance on that test; it cannot, by itself, establish that a person’s IQ or general intelligence increased. First ask what was measured, under what conditions, and what decision the result is supposed to inform.

The Standards for Educational and Psychological Testing frame validity around the interpretation and use of scores. Validity is not a permanent badge attached to a test’s name: evidence must support the particular meaning assigned to a result and the decision made from it. The Standards say that when evidence for an intended use is limited or absent, users should make the limits clear and treat hypotheses as tentative. Evidence about one test, population, and outcome does not automatically travel to another.

That distinction keeps the findings discussed here in their proper lanes. The 2025 meta-analysis supports a qualified statement about preparation and performance on the large-scale educational tests it studied. Its small transfer subset concerned other tests in similar educational domains, not general intelligence. The cognitive-test retest meta-analysis supports caution when comparing repeated administrations, because scores may change with test content, form equivalence, interval, and age. Neither source validates a separate online IQ estimate. They answer different questions and do not supply evidence for that product’s score interpretation.

Test IQ Free’s adaptive 50-question reasoning test reports a provisional estimated IQ range from placeholder conversion parameters. It has no validated adult reference population. The range is not a normed score, a confidence interval, or an established population comparison, and it should not be treated as interchangeable with results from a professionally developed instrument. The free test can offer low-stakes reasoning practice and worked answers, but its estimate does not establish a person’s standing, diagnosis, potential, or fitness for a consequential decision.

The optional $9 report explains the same provisional conversion and adds interpretation and practice guidance. Payment does not supply a validated reference group or make the estimate more accurate. For curiosity or practice, a reader can take the free test, review the worked answers, and describe improvement at the level observed. If an education, clinical, or other consequential decision depends on an ability score, consult a qualified assessor and ask what evidence supports that specific interpretation for the relevant population and use. The question is not simply whether a number changed; it is whether the evidence warrants the conclusion drawn from it.

Sources: Standards for Educational and Psychological Testing

What can I responsibly do next?

Practice a specific reasoning operation if learning it is your goal, review its explanations, and check your work on fresh examples. Describe the result at the level measured, such as more accurate responses on new tasks using a familiar operation. If the check repeats items or closely copies the practice format, say so; it cannot establish broader transfer. A changed online estimate is not evidence that IQ or general intelligence increased.

The Standards for Educational and Psychological Testing make intended use central: evidence must support the interpretation a reader wants to make. When it does not, keep the conclusion tentative. Test IQ Free’s free adaptive 50-question test offers low-stakes exploration and practice, with worked answers to review. Its estimated IQ range comes from a provisional conversion with no validated adult reference population. It is not a normed score, confidence interval, or established population comparison. The optional $9 report explains that same conversion; payment adds no validation or accuracy.

For a consequential decision, seek a qualified assessment and ask what evidence supports using that test for that purpose. Name the practiced operation, check it on fresh tasks, and report only what they show. Take the free test for low-stakes practice, or choose professional assessment when a decision needs a defensible score interpretation. A practice gain or browser estimate does not define your worth, diagnose you, or fix your potential.

Sources: Standards for Educational and Psychological Testing

Questions readers ask