A learning rate parameter - PowerPoint PPT Presentation

1 / 64
About This Presentation
Title:

A learning rate parameter

Description:

Dizygotic (DZ) twins share 50% of their segregating genes (genes that make us ... Why study twins (cont) ... Because our twins almost always share schools, ... – PowerPoint PPT presentation

Number of Views:113
Avg rating:3.0/5.0
Slides: 65
Provided by: bby2
Category:

less

Transcript and Presenter's Notes

Title: A learning rate parameter


1
Longitudinal twin study of early literacy and
language Brian Byrne University of New
England with Richard Olson, Stefan Samuelsson,
Sally Wadsworth, Robin Corley, John C. Defries.
Erik Willcutt
2
Why study twins?
  • Monozygotic (MZ) twins share 100 of all their
    genes
  • Dizygotic (DZ) twins share 50 of their
    segregating genes (genes that make us different)
    on average
  • MZ and DZ pairs reared together have similar
    shared environment in the home and school

3
Why study twins (cont)?
  • Therefore, for traits that are substantially
    heritable, MZ twins will be very similar to each
    other and more alike than DZ twins
  • For traits that are substantially influenced by
    early (shared) environment, MZ and DZ twins will
    be highly and equally similar

4
Concordance rates for schizophrenia
  • MZ twins 48
  • DZ twins 17
  • First degree (e.g., sibling) 9
  • Second degree (e.g., aunt) 4

5
Correlations for intelligence
  • MZ twins .85
  • DZ twins .60
  • First degree .45
  • Second degree .30

6
Structure of project
  • Downward extension of existing reading studies to
    preschool-aged children
  • Sites in Colorado, Australia, Norway and Sweden
  • Four test occasions
  • Preschool
  • Kindergarten
  • Grade 1
  • Grade 2
  • To date, approx 1700 children included

7
Aims
  • To model genetic, shared environment, and unique
    environment effects on known foundations for
    literacy growth, prior to complicating influences
    of literacy levels themselves (ie, prior to
    formal literacy instruction)
  • To trace these effects longitudinally as children
    develop in literacy

8
Twins seen in homes or pre-schools prior to
kindergarten
Note Each child assessed by a different tester
9
Twins typically tested at home after kindergarten
year
10
And again after first grade
11
With a final assessment after second grade
12
Some of the preschool test categories
  • Phonological analysis and synthesis (PA)
  • Print familiarity (PR)
  • General verbal ability (GVA)
  • Vocabulary (VOC)
  • Morphology/syntax (GRAM)
  • Working memory (WM)
  • Rapid naming (RN)

13
Sample twin correlations (composites)
  • MZ DZ
  • PA .79 .55
  • RN .68 .40
  • PR .87 .77
  • GVA .85 .68

14
Results of Mx modeling on preschool variables as
latent traits
Phonological awareness
Working memory
Rapid naming
15
Results of Mx modeling on preschool variables as
latent traits (cont.)
Print knowledge
Vocabulary
Grammar/ morphology
16
Summary of univariate preschool
  • Phonological awareness, rapid naming, and working
    memory show substantial genetic effects and
    modest shared-environment effects
  • Print familiarity, vocabulary, and
    morphology/syntax show substantial
    shared-environment effects and modest genetic
    effects

17
Univariate analysis (the story so far)
  • MZ twins
  • Twin 1 Twin 2
  • X X
  • X X
  • . .
  • . .
  • DZ twins
  • Twin 1 Twin 2
  • X X
  • X X
  • . .
  • . .

If correlation for MZ twins is higher than for DZ
twins, evidence for genetic influence on Variable
X
18
Multivariate analysis (new part of story)
  • MZ twins
  • Twin 1 Twin 2
  • X Y
  • X Y
  • . .
  • . .
  • DZ twins
  • Twin 1 Twin 2
  • X Y
  • X Y
  • . .
  • . .
  • If correlation for MZ twins is higher than for DZ
    twins, evidence for genetic influence that is
    common to Variable X and Variable Y
  • Y can be a different variable or the same
    variable at a different time

19
Multivariate analyses
  • Thus, goal is to identify whether genetic and
    environmental sources influence more than one
    variable
  • A commonly-used technique is Cholesky
    decomposition
  • It is analogous to multiple regression where the
    influence of a predictor on a dependent variable
    is assessed after the influence of another
    predictor is accounted for

20
Cholesky decomposition ofpreschool latent
traits additive genetic effects
GVA
.64
A1
.45
PA
A2
.57
.35
RN
A3
.72
27
.20
PR
A4
.33

21
Cholesky decomposition ofpreschool latent
traits shared-environment effects
GVA
.76
C1
.61
PA
C2
.23
.22
.28
RN
.58
.58
PR
22
Cholesky decomposition ofpreschool latent
traits unique environment effects
GVA
.13
E1
.23
PA
.
.20
RN
PR
23
Summary of multivariate preschool data
  • A single genetic factor affects verbal ability,
    phonological awareness, rapid naming, and print
    familiarity
  • A second genetic factor affects phonological
    awareness and print familiarity
  • Two additional genetic factors uniquely affect
    rapid naming and print familiarity
  • Thus, each of the four latent traits is
    genetically complex

24
Summary of multivariate preschool data (cont.)
  • In contrast to genetic influences, where four
    factors affect the latent variables, only two
    shared environment factors are in operation, each
    affecting all or most latent traits

25
Variables assessed after one year of formal
schooling (kindergarten)
  • Word and nonword reading efficiency (TOWRE)
  • Spelling (words and nonwords)
  • Phonological awareness
  • Rapid naming
  • Grammar (TROG)

26
Results of Mx modeling on kindergarten variables
as latent traits
27
Results of Mx modeling on kindergarten variables
as latent traits (cont.)
28
Summary of kindergarten univariate data
  • Reading, phonological awareness, and rapid naming
    show substantial genetic influence and modest
    shared-environment effects
  • Spelling is equally affected by genes and shared
    environment
  • Grammar is mainly affected by shared environment,
    with modest genetic influence

29
Cholesky decomposition ofkindergarten latent
traits additive genetic effects
PA
.81
A1
.34
RN
A2
.61
.59
.34
READ
A3
.51
.59
.29
SPELL

30
Cholesky decomposition ofkindergarten latent
traits shared-environment effects
PA
.51
C1
.47
RN
.42
READ
.63
SPELL
31
Cholesky decomposition ofkindergarten latent
traits unique environment effects
PA
.28
E1
.27
RN
E2
.43
.26
READ
.32
SPELL
32
Summary of kindergarten multivariate data
  • Reading and spelling are influenced by genes that
    also affect phonological awareness and rapid
    naming, and in addition by a third set of genes
  • A single shared-environment factor affects all
    four kindergarten latent traits

33
Longitudinal multivariate analyses
34
Cholesky decomposition ofphonological awareness
development
PA1 preschool phonological awareness PA2
kindergarten phonological awareness
.
A1
PA1
.82
.56
C1
.65
A2
PA2
.55
35
Where do the new genes for PA at kindergarten
come from?
  • Possibly reciprocally from orthographic
    processes, as implied by the position that there
    is mutual influence between PA and orthography.

36
To test this in our genetically sensitive design,
we entered kindergarten spelling prior to
kindergarten PA
PA1
A1
.79
.
.44
.63
A2
K spell
.61
.42
A3
PA2
NS
There is no genetic influence on PA2 that is
independent of the genetic influence on K spelling
37
Cholesky decomposition ofrapid naming development
RN1 preschool rapid naming--colours,
objects RN2 kindergarten rapid naming--colours,
letters, digits
.
A1
RN1
E1
.79
.46
.13
.59
A2
E2
RN2
53
.47
38
Summary of developmental data for two single
variables
  • Phonological awareness and rapid naming both show
    both genetic continuity and genetic change across
    the period of development that we have tracked.
    In the case of PA, the second genetic source may
    be genetic influence on orthographic processes.
    In the case of RN, the change may be due to
    introduction of letters and digits in kindergarten

39
Cholesky decomposition ofmultivariate
preschool-kindergarten development additive
genetic effects
PR1
.47
A1
preschool
.61
PA1
A2
.50
.41
RN1
A2
A3
.58
.
.72
A4
READ
.41
kindergarten
significance uncertain with current sample size

40
Cholesky decomposition ofmultivariate
preschool-kindergarten development shared
environment effects
PR1
.85
C1
preschool
.52
C2
PA1
.28
.40
.24
RN1
.
.31
kindergarten
READ

41
Cholesky decomposition ofmultivariate
preschool-kindergarten development unique
environment effects
PR1
.23
E1
preschool
.16
PA1
E3
.40
RN1
.
.15
kindergarten
READ

42
Summary of developmental multivariate data
  • A single set of genes influences preschool PA, RN
    and PRINT and kindergarten READ
  • A second genetic factor affects PA but not RN or
    READ
  • A third genetic factor affects RN but not READ
  • Thus, PA and RN only share genetic variance with
    READ through genes shared with PRINT

43
Summary of developmental multivariate data (cont)
  • A shared-environment factor affects preschool PA,
    RN and PRINT and kindergarten READ
  • A second preschool factor affect PA and RN (only)
  • There is no new shared environment factor
    emerging in kindergarten

44
Results of Mx modeling on Grade 1 literacy
variables as latent traits
45
We see that word reading and reading
comprehension are both substantially affected by
genesbut, are they the same genes?
46
Cholesky decomposition ofGrade 1 word reading
and reading comprehension
.
A1
word
E1
.91
.40
.27
.90
comp
47
Another longitudinal analysis Cholesky
decomposition of kindergarten and Grade 1 word
reading
.
A1
kind.
C1
.82
.50
.77
E paths omitted modest effects at all three
possible paths
A2
Gd 1
.44
48
Grade 1 summary
  • Substantial genetic effects on word reading,
    reading comprehension, and spelling in Grade 1
  • Genetic effect on word reading accounts for all
    of the genetic effect on reading comprehension
    (Cholesky decomposition, plus genetic correlation
    of .97)
  • New genes kick in for word reading in Grade 1
    on top of those shared with kindergarten

49
Grade 1 summary (cont)
  • No reliable shared environment effects in
    evidence at Grade 1
  • Because our twins almost always share schools,
    little evidence of a differentiating school
    effect, therefore

50
Summary and implications
  • Reading is substantially affected by genes as
    early as kindergarten
  • Most of these genes are in play prior to school,
    influencing processes foundational for literacy
    growth
  • Genes are beyond our control, therefore, we need
    to pull on the environmental levers as
    intelligently as possible

51
Summary and implications 2
  • There were no new shared environment effects on
    reading in Kindergarten, and none were detectable
    at all in Grade 1, in our sample
  • Twins almost always share schools, if not
    teachers
  • Thus, schools may not be contributing to whatever
    differential reading scores exist between schools
  • So public policy measures that penalise schools
    for poor reading levels are misguided, perhaps by
    around 180 degrees

52
The future (funds permitting)
  • Completion of existing cohorts (completion in
    2009)
  • Extension to 4th grade
  • More detailed comprehension measures
  • More detailed print exposure measures (diary
    methods?)
  • Data from other school subjects
  • Integration with imagingfMRI, MEG?

53
Acknowledgements
  • Australian Research Council, Australian Twin
    Registry, National Institute of Child Health and
    Human Development, Stavanger University College
    and Research Council of Norway, Swedish Research
    Council, and our many testers, coordinators,
    database managers and of course the twins and
    their families

54
Addendum
  • So far we have focused on literacy
  • What about spoken language in our data?
  • Recall that there were modest but significant
    genetic effects on grammar, morphology and
    vocabulary
  • Where do they come from?

55
An example
  • of a test item from the TROG

56
point to the boy the dog chases is big
57
A hypothesis
  • Genetic differences among 4-5 year-old children
    on tests of grammar reflect performance rather
    than competence factors, based on the
    observation that the structures tested are
    already present in the repertoires of normal
    children of this age group

58
Another hypothesis
  • Genetic differences among 4-5 year-old children
    in vocabulary derive from genetic differences in
    ability to fix phonetic forms (as assessed by
    nonword repetition)

59
Cholesky decomposition of preschool working
memory, vocabulary and morph/syntax
A1
wm
.76
.51
voc
(No other genetic path close to significance)
.52
mor/syn
60
Cholesky decomposition shared environment
effects (preschool)
C1
wm
.63
.63
C2
voc
.58
.65
mor/syn
61
Cholesky decomposition of preschool working
memory composite and vocabulary and kindergarten
TROG additive genetic effects
A1
wm
.78
.52
voc
(No other genetic path close to significance)
.40
TROG
62
Cholesky decomposition shared environment
effects (Year 1)
C1
wm
.62
.64
C2
voc
.56
.24
.19
TROG
63
Summary and implications
  • A working memory composite accounts for all of
    the genetic influence on our vocabulary and
    morphology/syntax measures
  • Thus there is no evidence in our data for
    independent genetic influence on higher
    language functions
  • The shared environment picture is more complex,
    with a second factor affecting vocabulary (and
    TROG)

64
Summary and implications (cont)
  • We speculate that working memory influences
  • vocabulary via fixation of phonetic form
  • TROG via performance demands of test itself
  • In any case, research into genes, environment and
    language needs to be sensitive to complex nature
    of language variables
Write a Comment
User Comments (0)
About PowerShow.com