Is the Quarter of Birth Endogenous? New Evidence from Taiwan, the US, and Indonesia

Published date01 December 2017
DOIhttp://doi.org/10.1111/obes.12175
Date01 December 2017
1087
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd.
doi: 10.1111/obes.12175
Is the Quarter of Birth Endogenous? New Evidence
from Taiwan, the US, and Indonesia
Elliott Fan,Jin-Tan Liu† and Yen-Chien Chen
Department of Economics, National Taiwan University, Taipei, Taiwan (e-mail: elliottfan@
ntu.edu.tw, liujt@ntu.edu.tw)
Department of Economics, National Chi Nan University, Nantou, Taiwan (e-mail:
yenchien@ncnu.edu.tw)
Abstract
Recent evidence based on US data suggests that the quarter or month of birth (QOB or
MOB) may be endogenous,since f amilycharacteristics can explain up to 50% of the effects
of QOB on the education outcomes and earnings of adult males. In this study, based on a
sample of one million Taiwanese siblings, we examine university admission at age 18 as our
outcome variable and find that at school entry,the oldest (September born) children are 31–
38% more likely to be admitted into university at age 18 than the youngest (August born)
children, indicating strong seasonality in university admission. The inclusion of controls
for family background is found to explain only a small portion of these effects, particularly
for males. Given that such results are at odds with the recent US evidence, werevisit the US
Census data and find that when racial differences are properly controlled for in the estima-
tion, evena rich set of family characteristics is capable of explaining only a minor proportion
of the QOB effects. Furthermore, using data from the US and Indonesia, we find that sea-
sonal temperature variation is unlikely to be an important contributor to the US-Taiwan
disparity. Our findings imply that the validity of using QOB or MOB as an instr umental
variable may be dependent on the population being studied and the sample selected.
I. Introduction
The two decades that havepassed since the pioneering work of Angrist and Krueger (1991,
1992) have witnessed considerable developments in the understanding of the ways in
which seasonal patterns of birth can predict the outcomes of individuals.1Although the
JEL Classification numbers: C10, J11, J13.
1Following Angrist and Kruger, many have sought to eliminate potential biases when estimating the returns to
education – through either earnings or other outcome variables – one strand of the related literature uses ‘quarter of
birth’(QOB) as an ‘instrumental variable’(IV). Recent examples include Lefgren and McIntyre (2006), Cher nozhukov
and Hansen (2006), Leigh and Ryan (2008), Maurin and Moschion (2009), Arkes (2010), Lee and Orazem (2010)
and Robertson (2011). Another line taken within the recent studies is to investigate the ways in which QOB directly
affects the outcomes of individuals in terms of either educational achievement (Bedard and Dhuey, 2006; Crawford,
Dearden and Meghir, 2010) or health (Weber, Prossinger and Seidler, 1998; van Hanswijck de Jonge, Waller and
OXFORD BULLETIN OF ECONOMICSAND STATISTICS, 79, 6 (2017) 0305–9049
1088 Bulletin
key assumption underpinning the validity of the empirical strategies used in the related
studies is that the ‘quarter of birth’ (QOB) is determined exogenously, the legitimacy of
this assumption has been challenged by growing evidence, suggesting that parents may
manipulate the timing of births according to their preferences for certain birth months or
seasons.2
A very recent example of the exploration of such manipulation of birth timing was pro-
vided by Buckles and Hungerman (2013, hereafter BH), who found that the characteristics
of women giving birth in the winter months in the US differed significantly from those of
women giving birth in other seasons, ultimately leading to a strong correlation between the
characteristics of the mothers and the QOB of their children. Currie and Schwandt (2013)
confirmed this correlation using US sibling data, whilst Ram´ırez and C´aceres-Delpiano
(2014) also observed similar correlations in Spain and Chile. Although the strong sea-
sonality of maternal characteristics may imply that mothers select different birth months,
evidence on the extent to which family background can explain the seasonal differences
in the adult outcomes of children is rather mixed. BH found that maternal characteristics
could explain up to 50% of the relationship between QOB and adult education outcomes
and earnings, raising serious concerns about the validity of exploiting QOB to construct
an IV estimation.3However, also using US data, Currie and Schwandt (2013) found that
controlling for the full set of family ‘fixed effects’ (FE) only slightly reduced the season-
ality of birth weight conditional on gestational age, as well as other health outcomes at
birth, thereby essentially eliminating family background as a major explanatory factor of
seasonal differences in health at birth. Solli (2012) found that month of birth (MOB) in
Norway had significant effectson g rade point averages at age 16, with these effects remain-
ing robust to controls for family background characteristics or the complete set of family
FE. Lokshin and Radyakin (2012) examined the waysin which the MOB of Indian children
were associated with their health, with their household FE estimates of the MOB effects on
the health of young children turning out to be substantially larger than the corresponding
ordinary least squares (OLS) estimates.4
It may, therefore, be of interest to further explore the potential endogeneity of birth
seasonality, and more importantly, to examine the extent to which various populations
Stettler, 2003; Costa and Lahey, 2005; Lokshin and Radyakin, 2012), both of which are primary determinants of
earnings.
2The validity of using QOB as an IV has, however,also been challenged from other perspectives. First, significant
inconsistency may occur in the IV estimates whenthere is a weak correlation between QOB and education outcomes,
an argument first put forward by Bound, Jaeger and Baker (1995) following their re-examination of the results of
Angrist and Krueger (1992). Second, Barua and Lang (2016) recently pointed out that the use of either QOB or
legal entry age as an IV also violates the monotonicity assumption, leading to inconsistent estimates of the ‘local
average treatment effect’. In terms of the latter challenge, we present evidencelater in this study to suggest that the
monotonicity assumption is essentially satisfied in the case ofTaiwan. Third, the requirement of exclusion restriction
may be violated whenMOB affects adult outcomes through other pathways, such as health or seasonal environmental
factors (Currie and Schwandt, 2013). In this study, wefocus on examining whether the seasonal patterns in educational
attainment may be explained by seasonal patterns in family background characteristics; we do not undertake any
analysis to address the three other concerns about using MOB/QOB as an IV.
3The findings of BH clearly point to the potential endogeneity of QOB.As such, they echo the findings of Bound
and Jaeger (2000) who noted that when used to instrument education, QOB failed to satisfy the requirement of
exclusion restrictions.
4However, as argued by Lokshin and Radyakin (2012), the sibling sample used in their study may have been
subject to both selection bias and measurement errors, which could well explain the differences between their OLS
and FE estimates.
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd
Is the quarter of birth endogenous? 1089
diverge in this area. Evidence from such a cross-country analysis should advance our
understanding of whether selection is a common threat to the related studies in which
MOB or QOB is employed as an IV. In the present study, based on the use of sibling data
from Taiwan supplemented by data from US and Indonesian Censuses, we provide multi-
faceted evidence to evaluate the extent of selection when MOB or QOB is used as an IV
for adult education outcomes or earnings.
In the case of Taiwan, we make use of merged administrative data sets linking the adult
education outcomes of individuals with their family background at the time of their birth
for around one million siblings (co-residing or otherwise). Our empirical work focuses on
estimating the effects of the MOB of individuals on their universityadmission at age 18. We
then evaluate the role of family background characteristics in determining the magnitude
of the estimated MOB effects and compare the results with the evidence obtained from the
US and Indonesian data.
This present study makes two main contributions to the extant related literature. First,
as pointed out by Lokshin and Radyakin (2012), studies investigating the effects of birth
seasonality have thus far been disproportionally clustered in high-income countries. By
considering data from Taiwan and Indonesia, we estimate the MOB or QOB effects on adult
education outcomes and examine the potential endogeneity of birth seasonality within the
context of developing countries. In particular, to the best of our knowledge, our study
is the first to investigate the endogeneity of birth seasonality using data outside of the
developed world. Second, we revisit the US Census data used by BH and carry out an in-
depth examination of whether the BH findings on the endogeneity of birth seasonality hold
within each of the black/white racial groups, as well as for states with different seasonal
variation in temperature. Evidence from these analyses may help to shed some light on
cross-country differences relating to the endogeneity of birth seasonality.
Using data on Taiwan, we begin by graphically presenting the month-by-month changes
in a set of maternal characteristics to examine whether these characteristics exhibit any
seasonal patterns, and if so, whether such patterns are correlated with the probability of
university admission at age 18. The presence of such a correlation would imply that bir ths
may be selected into months predicting better adult education outcomes. In contrast to
the US scenario, our graphical results appear to indicate weak seasonality of maternal
characteristics.
In addition to our graphical presentation, we go on to compare the effects of MOB
obtained from estimating three different regressions. The first is a baseline OLS regression
which controls for only the year-of-birth and county-of-birth fixed effects, and the time
variables for de-trending the outcome variables.The second is the baseline OLS regression
incorporating a rich set of additional controls, including a range of family characteristics.
The third is a family FE regression which is reliant on within-family comparisons, taking
advantage of our sibling data.5
Our baseline OLS estimates of the MOB effects are quite substantial, since the proba-
bility of university admission at age 18 for males born in September is about 37.5% higher
than that for males born in August, whilst the margin for females is around 30.5%. For
5This ‘three-layer’ structure is analogous to that of Currie and Schwandt (2013) in which the role of observable
characteristics was examined by comparing the two OLS results and the role of unobservable characteristics by
comparing the FE results with those from the OLS regression with controls.
©2017 The Department of Economics, University of Oxford and JohnWiley & Sons Ltd

To continue reading

Request your trial

VLEX uses login cookies to provide you with a better browsing experience. If you click on 'Accept' or continue browsing this site we consider that you accept our cookie policy. ACCEPT