572 Panel Data Methods
12.2.3.2 Twin studies
Within-sibling variation can control for common factors relating to family back-
ground, upbringing and environment. But siblings are born at different times
and they have different genes. Twin studies take the notion of natural controls
a step further by removing the genetic variation (at least for monozygotic twins).
In the context of research on birth outcomes, using twins means that the sib-
lings share the same pregnancy and are born at the same time. This controls for
unobservable characteristics of their mother and her behavior and environment
during the pregnancy. Almondet al.(2005) show that using variation in birth
weight between twins leads to lower estimates of the impact of low birth weight
(LBW), defined as less than 2,500 g, on short-run outcomes than is typically found
in cross-section studies. They find heterogeneity in the effects of LBW, suggest-
ing a highly nonlinear relationship. Two identification strategies are adopted. The
first exploits variation “within mothers” by comparing outcomes for heavier and
lighter infants for all twins born in the US between 1983 and 2000. Using this
within-variation should control for all observed and unobserved characteristics of
the mother. The second exploits variation “between mothers” in a complemen-
tary analysis of maternal smoking and singleton births. The strategy here is to
attribute the whole effect of smoking to LBW and compare it with the twins esti-
mates. Data are drawn from two sources: linked birth and infant deaths data from
the US National Center for Health Statistics (NCHS), covering the population of
US twins, and data from hospital discharge abstracts from the Healthcare Cost and
Utilization Project (HCUP) state inpatient database. There are some caveats to bear
in mind with this study. Some useful descriptive analysis presented in the paper
highlights the inherent differences between twins and singletons (the latter are
more healthy). This raises questions about external validity of analysis based on
samples of twins rather than the general population. The study only uses short-run
outcomes and may miss long-term consequences (see the studies by Behrman and
Rosenzweig, 2004, and Blacket al.,2007, below). For fraternal twins, genetic dif-
ferences may mean that changes in birth weight may be associated with changes
in unobservables (there is evidence of a negative correlation of birth weight with
congenital defects) and the fixed effects approach may overestimate the impact of
birth weight. Also the data do not distinguish between monozygotic (identical)
and dizygotic (fraternal) twins.
In Behrman and Rosenzweig (2004), variation between monozygotic twins pro-
vides a way of identifying the impact of birth weight on long-term outcomes,
such as measures of adult health, anthropometric measures and adult school-
ing and earnings. Increased birth weight, as measured on the birth certificate,
increases schooling among adults and this effect is underestimated by 50% when
cross-section variation is used to identify the effect. Data were collected through a
survey mailed to monozygotic twins on the Minnesota Twins Registry, the largest
birth certificate based registry in the US. The identification strategy assumes that
difference in birth weight reflects random differences in nutrition in the womb
that are uncorrelated with individual endowments and therefore avoids selection