There is evidence for genetic contributions to reading disability, but the
phenotypic heterogeneity associated with the clinical diagnosis may make id
entification of the underlying genetic basis difficult. In order to elucida
te distinct phenotypic features that may be contributing to the genotypic h
eterogeneity, we assessed the familial aggregation patterns of Verbal IQ an
d 24 phenotypic measures associated with dyslexia in 102 nuclear families a
scertained through probands in grades 1 through 6 who met the criteria for
this disorder. Correlations between relatives were computed for all diagnos
tic phenotypes, using a generalized estimating equation (GEE) approach. GEE
is a recently developed semiparametric method for handling correlated data
. The method is robust to model misspecification and flexible in adjusting
for the subjects' characteristics and pedigree sizes as well as for the asc
ertainment process, while estimating the correlations between related subje
cts. The Nonword Memory (NWM) subtest of a prepublication version of the Co
mprehensive Test of Phonological Processing (CTOPP) and Phonemic Decoding E
fficiency (PDE) subtest of a prepublication version of the Test of Word Rea
ding Efficiency (TOWRE) showed correlation patterns in relatives that are s
trongly supportive of a genetic basis. The Wechsler Scale Digit Span, the W
ord Attack subtest of the Woodcock Reading Mastery Test-Revised, and the Sp
elling subtest of the Wide Range Achievement Test-Third Edition had slightl
y weaker evidence of a genetic basis. Five additional phenotypes (the Spell
ing subtest of the Wechsler Individual Achievement Test, the Accuracy, Rate
, and Comprehension subtests of the Gray Oral Reading Test-Third Edition, a
nd Rapid Automatized Naming of Letters and Numbers) gave suggestive evidenc
e of such a pattern. The results cross-validate in that evidence for a patt
ern consistent with a genetic basis was obtained for two measures of phonol
ogical short-term memory (CTOPP Nonword Memory and WISCIII or WAIS-R Digit
Span), for two measures of phonological decoding (WRMT-R Word Attack and TO
WRE Phonemic Decoding Efficiency), and for two measures of spelling from di
ctation (WRAT-3 Spelling and, to a lesser extent, WIAT Spelling). These mea
sures are thus good candidates for more sophisticated segregation analyses
that can formulate models for incorporation into linkage analyses.