Behavior Genes
597
where the correlation between genotypes
β
=
1
.
0forMZtwinsand0.5forDZtwins.
Therefore, if we calculate the correla-
tions based on observed measures for MZ
and DZ twins,
r
MZ and
r
DZ
,wehavea
set of simultaneous equations that can be
easily solved for
a
2
and
c
2
, giving:
a
2
=
2
(
r
MZ
r
DZ
)(
6
)
c
2
=
2
r
DZ
r
MZ
(
7
)
Although path analysis is most commonly
carried out with twin data, the approach
is perfectly general and can be applied to
other types of relative pairs. It can also
incorporate other types of observations
such as direct environmental measures.
2.6
Structural Equation Models
Structural equation modeling enables for-
mal testing of the extent to which the
genetic and environmental effects de-
scribed above contribute to the variance of
the phenotype, or the variance of liability
of categorical phenotypes. This is typically
performed using computer programs such
as LISREL or Mx, which use an iterative
process to test which estimates for each
of the parameters best explain both the
observed data and the expected variance
and covariance values. These expected val-
ues are calculated by the program using
information regarding the causes of co-
variation, for example, that the covariance
between MZ twins results from genetic
and common environmental effects, and
that between DZ twins is due to half the
genetic effects and all the common envi-
ronment. The best Ftting model is the one
in which there is maximum agreement
between the expected and observed co-
variances. This is achieved by maximizing
a likelihood function or minimizing the
goodness-of-Ft
χ
2
.
A principle of parsimony is adopted
whereby there is an attempt to explain
the observed data with as few parameters
as possible. Thus nested models, in which
parameters are dropped, are tested and
compared with the full model to see if
the remaining parameters can account
for all the similarity between the twins
without signiFcantly worsening the Ft of
the model.
Model Ftting need not be restricted to ex-
plaining the cause of phenotypic variance
and covariance. It can also be used, for
example, to identify the existence of qual-
itative and quantitative sex differences,
identify the causes of phenotypic comor-
bidity, assess sibling interaction effects,
and tease out gene–environment interac-
tion effects.
3
Molecular Genetics
3.1
Animal Studies
Animal models are useful in the identiFca-
tion of genes with a role in behavior since it
is possible to modify and manipulate both
the genome and environment of labora-
tory animals in a way that is not possible
in human samples. Other mammals have
a high level of genetic homology with
humans, and also display certain behav-
iors that are analogs of human traits. ±or
example, the mouse genome shares ap-
proximately 85% homology with humans
and mice show behaviors analogous to
such human behaviors as anxiety, aggres-
sion, and hyperactivity.
Animal studies use inbreeding and se-
lection studies to identify the presence of