Improving the Fit Between Concepts and Methods in Family Research: Or, Why We Have Painstakingly Pursued Multilevel Models

blessinghomoeopathAI and Robotics

Nov 30, 2013 (3 years and 6 months ago)

49 views

Improving the Fit Between Concepts and Methods in Family Research:

Or, Why We Have Painstakingly Pursued Multilevel Models


Acknowledgements: This work was made possible with support from an ESRC
research methods programme grant


Tom O’Connor

Jon Rasbash

Jenny Jenkins


Problems of measuring systems and systemic influences in a
meaningful way

{A}s thinkers we seem to take a connoisseur's delight in grand abstractions.
Among these, first and foremost, has been general systems theory. Imported
early in the development of the field, it has lingered as an intellectual
backdrop for almost all discourse. But there have been other, equally mighty
abstractions that have fueled our thinking: cybernetics, group theory, the
theory of logical types, and now a plethora of "new" epistomologies.
Although I believe that highly abstract schemas can be useful for orientation
and integration of more concrete theories, they are inadequate for generating
specific, testable hypotheses, and when attempts are made to apply them
directly to data gathering, the results can be disastrous. The central, sad fact
is that none of these grand abstractions are adequate for generating specific
hypotheses that could guide enlightened inquiry about family and other social
processes and social change.







D. Reiss (1988)

In other words:


Theories of
family
-
level processes are often
vaguely stated


Measures of
family
-
level processes are
weak and subject to the usual array of limits


Much of the ‘family’ research is actually
concerned with dyad
-
level processes

A Cynical View of Current Family Research


The notion of family system is an introductory or
discussion point that is not seen in the methods


Family research is almost entirely dyad
-
level


BG findings have spurred the study of siblings,
but not of family
-
level processes


There is a focus on context for human
development, but this has shifted to neighborhood
or cultural levels rather than to a family level

Aims



To describe the conceptual
-
methodological
context of the programme grant


Identify several challenges for improving
“family effects” research


Illustrate several examples of how analytic
approaches offer novel insights into family
process

1. Context


Family
qua

family effects were rarely investigated
and theories weren’t that useful


Family effects were illusory:


Behavioral genetic findings implied that genetic effects
were substantial, ‘shared’ environmental effects were
minimal, and most family studies were unable to
deduce cause


Family research has lost its bearings and null
hypotheses


Intervention research happily and defiantly
plodded along

2. Challenges


What was meant by ‘shared’ environment?


did it have anything to do with the family


even it if was only about siblings, was it useful


How can genetically
-
sensitive family research
incorporate families that vary in size and genetic
patterns?


How can genetically
-
informative family research
include sensitive process measures?


Can we identify a really useful analytic framework
that is accessible and offers new conceptual
insights?

3. Illustrations & Strategies


Avon Brothers & Sisters Study (ABSS)


Dunn et al. study of 217 UK families with 2
-
3 children residing in single
-
parent, stepfather, complex step, and non
-
stepfamilies recruited from the
ALSPAC cohort


Nonshared Environment and Adolescent
Development (NEAD)


Reiss et al. study of 720 US families with 2 same
-
sex adolescent siblings
ranging from MZ to unrelated stepchildren


National Longitudinal Study of Child and Youth
(NLSCY)


National probability sample of 12,872 children in 8,719 Canadian families
that included up to 4 children (wave1)

-0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Dyadic mutuality
MZ
DZ
Full
Adoptive
Parent
-
child mutuality in early childhood:

Replication across sample and design

Source: Deater
-
Deckard & O’Connor, 2000

correlation

h
2

from UK twin study=.59

h
2
from CO adoption study=.50

UK

CO

General point #1


Studying larger family units may reveal
something interesting within and between
families

Parent
-
child conflict

Child aggression

Hypothetical Example:

Association between predictor and outcome varies across families


Parent
-
child conflict

Child aggression

Hypothetical Example:

Association between predictor and outcome varies across families


Family B

Family A

Average
association

across sample

Family
C

In other words, each family has its own regression line:

Question: Is there
significant

variation among families in these slopes?

General point #2


There are many types of “family effects”

General point #2a



Mapping variation is a good place to start

Family slopes: Clustering of data in biological, single
-
parent,

stepfather, and complex stepfamilies:

Mother
-
child conflict and Child aggression

Random Effects or Family context effects: Family type

Source: ABSS, Dunn & O’Connor

0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
Variance
Biol (.31)
Stepfather (.12)
Complex(.53)
Single(.56)
Family-Level and Individual Child-Level Variance by Family Type
Family
Child
Source: O’Connor et al., 2001

Figure 1.
-2.0
-1.5
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
household ses
1
2
3
4
5
differential parenting
family size = 2, no marital problems
family size = 2, marital problems
family size >2, marital problems
family size > 2, no marital problems
Source: Jenkins et al., 2003

The Quest for a Family
-
Level Effect



Key features of the NEAD dataset


Observational data provide some guard against method variance


Each dyad observed for 10 minutes in problem
-
solving session


Observation coded by separate observers


Established coding system used, with 2 broad
-
band factors:


Negativity (conflict, hostility, anger)


Positivity (warmth, involvement, communication)


No measure of “family” level process explicitly included in a large battery


Design permits an analysis of alternative explanations of stability, e.g., family,
genetic, or individual characteristics


Family 1…

Actor: c1 c2 m f

Relationship:
c1

c2

c1

m

c1

f

c2

c1

c2

m

c2

f

m

c1

m

c2

m

f

f

c1

f

c2

f

m


Partner: c1 c2 m f

Dyad d1 d2 d3 d4 d5 d6

The structure of process: NEAD observational data

family

dyad

actor

partner

Relationship score

A unit diagram


one node per unit

A classification diagram with one node per
classification

The multilevel social relations model
(after Snijders and Kenny, 1999)

family

dyad

actor

partner

Relationship score

m
f
m
f
is the effect for the
m’th family

jm
a
jm
a
is the effect of
j’
th actor in
the
m
’th family

km
p
km
p
is the effect of
k’
th partner in the
m
’th family

lm
d
lm
d
is the effect of
l’
th dyad in
the
m
’th family

)
,
0
(
~
)
,
0
(
~
)
,
0
(
~
)
,
0
(
~
)
,
0
(
~
2
)
,
,
(
2
2
2
2
)
,
,
(
0
)
,
,
(
e
m
l
k
j
i
d
lm
p
km
a
jm
f
m
m
l
k
j
i
lm
km
jm
m
m
l
k
j
i
N
e
N
d
N
p
N
a
N
f
e
d
p
a
f
y












m
l
k
j
i
e
)
,
,
(
m
l
k
j
i
e
)
,
,
(
is the residual relationship
effect conditional on actor
j,
partner
k,

dyad
l

and family
m

Interpretation of variance components in Multi
-
level model

Family
: the extent to which family level factors effect
all

the relationships in a
family.

Actor:
the extent to which individuals act similarly across relationships with
other family members (e.g., trait
-
like behaviour)

Partner:

the extent of responsiveness or elicitation from other family members.

Dyad:
the extent to reciprocity (i.e., the score M
-
>F is similar to F
-
>M).

Relationship:
residual variation across relationships in relationship quality.

Results of SRM

Pos.

SRM

Neg.

SRM

Family

0.12

0.19

Actor

0.44

0.12

Partner

0.01

0.03

Dyad

0.18

0.41

Relat.

0.25

0.24

-
2loglike

10225.7

17800.9

Variance partition coefficients

For positivity 44% of the
variablity is attributable to actors
indicating that individuals act in a
consistent way across relationships
with other family members. There
is a strong actor trait component to
positivity.

For negativity 0.41 of the
variability is attributable to dyad.
Indicating the dyad is an
important structure in
determining negativity in
relationships. There is a strong
context specific component to
negativity.

There is significant family level variance

independent of individual and dyadic levels.

NB. We can extend the basic BG model to our structure. The extended model gives
heritabilities (genetic variance)/(total variance) of 0.42 and 0.16 for positivity and
negativity, respectively.

Stability


results of two bivariate SRM

Positivity

Negativity

w1 vpc

w2 vpc


12

w1 vpc

w2 vpc


12

family

0.11

0.12

0.77

0.20

0.17

0.8

actor

0.44

0.46

0.87

0.11

0.11

0.67

partner

0.01

0.01

1.5??

0.03

0.04

0.88

dyad

0.17

0.12

0.15

0.42

0.41

0.34

relat.

0.26

0.29

0.11

0.25

0.27

0.16

The basic

patterns of the
vpc’s found in wave 1 are
repeated in wave 2 for
both positivity and
negativity.

Family effects are
very stable over time
for both positivity (

12

= 0.77) and negativity
(

12
=0.8). Family
effects are a bit
stronger for negativity.

Effects of Family type as fixed and random effect (negativity)

param(se)

param(se)

fixed

intercept

0.661(0.034)

0.590(0.034)

a_mother

-
0.379(0.030)
-

-
0.381(0.030)

a_father

-
0.516(0.030)

-
0.510(0.030)

p_mother

-
0.322(0.028)

-
0.319(0.028)

p_father

-
0.625(0.028)

-
0.616(0.028)

a_moth*p_fath

0.355(0.040)

0.348(0.040)

a_fath*p_moth

0.564(0.040)

0.556(0.040)

step_fam

0.121(0.036)

0.126(0.036)

random

family

0.140(0.012)

0.144(0.012)

actor

0.087(0.006)

0.087(0.006)

partner

0.018(0.005)

0.018(0.004)

dyad(nuclear)

0.239(0.09)

0.239(0.009)

dyad(step)

0.070(0.017)

relationship(nuclear)

0.162(0.005)

0.162(0.005)

relationship(step)

0.030(0.009)

-
2loglike

17187.19

17149.3

Significant fixed effect of stepfamily.

We find more between dyad and
between relationship variances in
stepfamilies, e.g.:

dyad variance nuclear = 0.239,
dyad variance step = 0.239+0.070,
that is between dyad variance is
23% greater in stepfamilies

}

Relationship negativity by actor role, partner role and
biological relatedness

Green line: biologically
related individuals;

Blue line: nonbiologically
related individuals.

Other example findings


Parent
-
child relationship quality:


Within
-
family variation in warmth & conflict is
explained by child characteristics and biological
relatedness


Single
-
parent family membership moderates peer
influences on children’s disruptive behavior


‘Shared’ family stresses such as divorce may
increase within
-
family variation in children’s
adjustment


Family
-
level negativity in stepfamilies is higher,
even as dyadic
-
level negativity is lower

Family and Parenting Research in Pre
-

Peri
-

and Post
-
Genomic Worlds


Pre
-
Genomic model:


Family ‘effects’ were assumed to be entirely
environmentally mediated


Peri
-
Genomic model:


There is substantial differential parenting and this partly
explains sibling differences


Many

measures of the parent
-
child relationship show
genetic influence


Studies that ignored genetic mediation over
-
estimated
environmental ‘effects’


Post
-
Genomic model:


The behavioral genetic model is a useful design, but has
led to confusion


The relevance to clinical practice is uncertain


Implications for Treatment


Are there compelling data showing how
genetic findings are relevant for
psychological treatment?
No


Are the mechanisms by which parenting
interventions have their effect necessarily
wholly environmental?
No


Might the variation in treatment response be
linked with genetic risk?
Maybe


Should current parenting interventions be re
-
shaped?
Not on account of the genetic data


Do family factors moderate genetic risk?
Maybe

General Conclusions


Study designs that accommodate 2 children are
still quite constrained


The test of
family
-
level processes needs to rely on more
than just the 2 targeted siblings


There are instances in which truly novel findings
re: family process have been obtained


Translating the method and findings (but not the
rationale) has proven challenging


The search is for a general analytic framework that
is flexible, accessible, and improves the fit
between concept and analysis