Verbal Analogies: Structure Beats Vocabulary
Verbal analogies are deceptively simple: given two related words, choose a third pair that preserves the same relationship. The format dates back to the early Wechsler scales and remains a core verbal-reasoning subtest. What makes analogies a good measure of reasoning, rather than just vocabulary, is that the relationship has to be inferred and then mapped onto candidate pairs.
Verbal analogies are deceptively simple: given two related words, choose a third pair that preserves the same relationship. The format dates back to the early Wechsler scales and remains a core verbal-reasoning subtest in modern batteries including the Miller Analogies Test (MAT), the Graduate Record Examination (GRE) until 2011, and the ICAR Verbal Reasoning subtest.
What makes analogies a good measure of reasoning, rather than just vocabulary, is that the relationship has to be inferred and then mapped onto candidate pairs. The test-taker must (1) identify what relationship links the example pair, (2) hold that relationship in working memory, and (3) compare it to each candidate pair in turn. Vocabulary knowledge is necessary but not sufficient: a high-vocabulary test-taker who fails to detect the relationship type will still answer incorrectly.
Common relationship types include synonym, antonym, part-to-whole, whole-to-part, type-of, instance-of, cause-and-effect, function (X is used to do Y), location (X is found in Y), young-to-adult of the same species, sequence (X precedes Y), and intensity (X is a milder version of Y). Strong test-takers learn to scan for these families when reading the example pair and reject candidate pairs that match a different family.
Analogies items are more culturally loaded than matrix or rotation items because they depend on the specific vocabulary and conceptual structure of a language. Translations often fail to preserve the underlying relationship. ICAR releases its Verbal Reasoning items in English only and recommends against direct translation for non-English populations.
Performance on analogies correlates strongly with crystallized intelligence (Gc) and with reading volume, which is why heavy readers tend to outperform their matrix-reasoning percentile on verbal subtests. The correlation between matrix and verbal scores within an individual is positive but moderate (r ≈ 0.5 to 0.7), reflecting the partial separability of fluid and crystallized abilities described in the CHC model.