In the course of a methodological investigation of regressions of mean
s of selected tests, computed in stratified levels of a general intell
igence composite, on the mean levels of intelligence within the strata
we encountered curious phenomena at the low end of the distribution o
f intelligence. At about -2.00 SD units of that distribution, the regr
essions of the vocabulary means for both boys and girls turned abruptl
y upwards, while those for mechanical reasoning dropped sharply. To in
vestigate these curious regressions, we formed high and low vocabulary
groups for each sex within the low-scoring subsample of general intel
ligence, and the means of the four groups on a series of cognitive and
self-report scales were obtained. In this sample the high school boys
and girls who are high in vocabulary relative to their low scores on
the intelligence composite have lower means than their low-low control
s on a set of cognitive tests epitomized by visualization in three dim
ensions, but the set also includes verbal tests in which the tasks are
ambiguous and require inference and hypothesis formation in order to
obtain a solution. Correct answers are not directly present in long-te
rm memory as they are for the academic (including mathematics) and non
academic information tests, on which the two high vocabulary groups ar
e substantially superior. The high and the low vocabulary groups also
differ from each other in health histories, personality traits, vocati
onal interests, and biographical data scores, the high groups being co
nsistently closer to the norm for their high school class. Sex differe
nces are minimal in the incidence of the deficit and in its correlates
. A genetic explanation for the problem is plausible, but the locus is
not on the 23rd pair of chromosomes.