VI. Robustness Checks
In this section, we provide several checks on the robustness of our basic empirical results. In particular, we are concerned about the possibility that the
introduction of ultrasound technology in a county may be correlated with unobserv- able variables, such as the One Child Policy, that affect the sex ratio in the locality.
A. Ultrasound or and One Child Policy
Although access to ultrasound machines provides a possible technology for sex selec- tion, the One Child Policy provides an incentive to select fetal gender. One obvious
concern over the identifying assumption is that the timing of the introduction of ultra- sound technology to Chinese counties may pick up temporal and spatial variations in
the implementation of the One Child Policy. For example, the One Child Policy may be correlated with postnatal sex selection, which is not related to ultrasound technology.
22
To deal with this issue, a measure of the local enforcement of the One Child Policy is included in the regressions. In particular, the birth rate in a county during the year in
which the mother became pregnant is employed as a proxy for the overall intensity of population control at the county level. The birth rate is defi ned as the number of births
divided by the number of women aged 15 to 49. The birth rate is lower under stricter enforcement. For meaningful comparison of estimates across specifi cations, the birth
rate is normalized with a mean of 0. As the One Child Policy only applies to births after 1979, when it was introduced, the analysis is restricted to the sample after 1979.
23
Controlling for the birth rate does not change our estimates of the effects of ul- trasound technology on sex selection. In Column 1 of Table 6, the interaction terms
of birth rate with birth- order indicators are added into the regression. The estimated effects of ultrasound technology remain essentially unchanged, which implies that
the baseline estimates are not much confounded by local birth- control policies. The coeffi cients of the interactions between birth rate and higher- order indicators are nega-
tive and signifi cant, suggesting that higher- order births are more likely to be male in counties governed by more stringent fertility policies.
Interestingly, an interaction effect between sex- selection technology and incentives is observed. In Column 2 of Table 6, we include triple interactions of birth rate, ultra-
sound technology, and birth- order indicators to allow the effects of ultrasound technol- ogy to vary with the intensity of local policy implementation. The coeffi cients for the
interactions between birth rate and higher birth- order indicators are negative but no longer signifi cant. However, the coeffi cients for the triple interactions of birth rate,
ultrasound technology, and birth- order indicators are negative and signifi cant, at least at the 10 percent level. This implies that local access to ultrasound technology has a
considerable positive effect on male probability for higher- order births, particularly in areas where enforcement of the One Child Policy is stricter. In other words, the results
22. Existing literature suggests that the One Child Policy has been an important contributing factor to the high sex ratio in China Ebenstein 2010, and enforcement of the policy has been highly localized and varied
over time Li and Zhang 2007. 23. In unreported results, the estimated coeffi cients using the entire sample are highly similar, possibly
because the effects of ultrasound technology are identifi ed almost entirely from the rapid diffusion of the technology in the 1980s.
Table 6 The effect of ultrasound availability, and its interactions with One Child Policy
enforcement on male probability births 1979–92: linear probability model results
Dependent variable: child is male 1
2 3
4 First birth × Ultrasound
–0.006 –0.005
–0.004 –0.005
0.004 0.004
0.005 0.005
Second birth × ultrasound 0.010
0.008 0.011
0.008 0.005
0.005 0.005
0.005 Third or higher order birth
× ultrasound 0.021
0.019 0.024
0.022 0.007
0.007 0.007
0.007 Birth rate × fi rst birth
0.038 0.027
0.073 0.081
0.045 0.051
0.050 0.060
Birth rate × second birth –0.100
–0.016 –0.060
0.046 0.056
0.068 0.058
0.073 Birth rate × third or higher
order birth –0.146
–0.054 –0.122
–0.022 0.075
0.093 0.075
0.098 Birth rate × ultrasound × fi rst
birth 0.011
–0.041 0.072
0.087 Birth rate × ultrasound ×
second birth –0.192
–0.247 0.094
0.104 Birth rate × ultrasound ×
third or higher order birth –0.198
–0.227 0.124
0.125 Individual controls
Yes Yes
Yes Yes
County fi xed effects Yes
Yes Yes
Yes Year of conception fi xed
effects Yes
Yes Yes
Yes County- specifi c linear time
trends No
No Yes
Yes Observations
257,499 257,499
257,499 257,499
R - squared
0.0037 0.0038
0.0056 0.0056
Note: Birth rate is calculated as the number of births divided by the number of women aged 15–49 in the county during the year when the mother became pregnant. For meaningful comparison of estimates across
columns, birth rate is demeaned using the sample average. Individual controls include mother’s ethnicity, education, maternal age at conception and its squared term, gestation length and indicators for the timing of
initial prenatal care visits. County fi xed effects are separate indicator variables for each county. Year fi xed effects are indicators that allow for unrestricted differences in year- to- year changes. Reported in parentheses
are standard errors clustered by county. denotes statistical signifi cance at the 10 percent level, at the 5 percent level, at the 1 percent level.
suggest that the observed increase in sex ratio at birth in China is largely driven by the interaction of the One Child Policy and access to ultrasound technology. The results
are insensitive to the inclusion of county- specifi c linear trends Columns 3–4. To further examine the differential effects of ultrasound technology with strong, me-
dium, and weak enforcement of the One Child Policy, ultrasound technology effects at the 25th percentile, median, and 75th percentile values of the birth rate are calculated.
For example, to determine the ultrasound effects at the 25th percentile of the birth rate, the regression is rerun, in which birth rate is replaced with [birth rate—P
25
birth rate], where P
25
birth rate is the 25th percentile of the birth rate. This yields, from the new coeffi cients for the ultrasound indicators, the estimated effects of ultrasound at
the 25th percentile of the birth rate and the standard errors.
24
The same approach is ap- plied to obtain the effects of ultrasound technology at the median and 75th percentile
values of the birth rate. The results presented in Table 7 show that the effect of ultrasound technology is
larger in areas with stricter enforcement of the One Child Policy. The effects of ul- trasound technology access under stringent enforcement of the One Child Policy are
estimated in Column 1, obtained by evaluating the effect at the 25th percentile of the birth rate. At this level, access to ultrasound technology raises the probability of male
birth by 1.8 percentage points for second births and 3.1 percentage points for third and higher births; both are signifi cant at the 1 percent level. For the median birth
rate Column 2, having access to ultrasound technology increases the probability of male birth by 0.9 percentage points for second births, and 2.3 percentage points for
third and higher births. However, the estimated effects around the 75th percentile of the birth rate are quantitatively smaller and not statistically signifi cant at the 5 percent
level. Overall, these results suggest that the observed effect of ultrasound technology on child gender is predominantly a result of prenatal sex selection in areas that more
stringently enforce birth- control policies.
We also explore an extreme case—those who were exempted from the One Child Policy. The One Child Policy was initially applied only to the majority Han of the
Chinese population, and later extended to ethnic groups with a population larger than 10 million, namely Zhuang and Manchu Li and Zhang 2007. However, the other 53
ethnic minorities were largely exempted from the one- child rule. Columns 4 and 5 of Table 7 separately show the effects of ultrasound technology by ethnicity. For the three
largest ethnic groups with a population of 10 million and above—Han, Zhuang, and Manchu—who faced a more stringent fertility control policy, the estimated coef-
fi cients for the interaction of ultrasound technology with second- birth indicators and the interaction of ultrasound technology with third- birth and higher indicators are
positive and signifi cant, and the magnitudes are qualitatively similar to those in the full sample. For smaller ethnic groups, the point estimates are quantitatively small and
statistically indistinguishable from zero, revealing no evidence of prenatal sex selec- tion. This fi nding echoes earlier research that emphasizes the role of the One Child
Policy in explaining sex imbalance in China Ebenstein 2010.
25
Overall, these analyses show that the effect of ultrasound technology on sex selec-
24. See Wooldridge 2003, Example 6.3. 25. However, this interpretation should be treated with caution as the study is unable to rule out the alterna-
tive explanation that people from smaller ethnic groups exhibit a relatively weaker preference towards sons.
Che n, L
i, a nd M
eng 61
Table 7 The estimated effect of ultrasound availability and implementation of the One Child Policy births 1979–92
Dependent variable: child is male Enforcement of One Child Policy
Ethnicity Strong
1 Medium
2 Weak
3 Han, Zhuang
Manchu 4
Smaller ethnic groups
5 First birth × ultrasound
–0.003 –0.005
–0.006 –0.006
–0.016 0.006
0.005 0.005
0.005 0.013
Second birth × ultrasound 0.018
0.009 –0.000
0.014 –0.001
0.006 0.005
0.007 0.005
0.015 Third or higher order birth × ultrasound
0.031 0.023
0.015 0.028
0.007 0.008
0.007 0.009
0.007 0.016
Evaluation at 25th percentile
median 75th percentile
mean mean
of birth rate of birth rate
of birth rate of birth rate
of birth rate Individual controls
Yes Yes
Yes Yes
Yes County fi xed effects
Yes Yes
Yes Yes
Yes Year of conception fi xed effects
Yes Yes
Yes Yes
Yes County- specifi c linear time trends
Yes Yes
Yes Yes
Yes Observations
257,499 257,499
257,499 264,014
34,623 R
- squared 0.0056
0.0056 0.0056
0.0059 0.0201
Note: Birth rate is calculated as the number of births divided by the number of women aged 15–49 in the county during the year when the mother became pregnant. Individual controls include mother’s ethnicity, education, maternal age at conception and its squared term, gestation length and indicators for the timing of initial prenatal care visits.
County fi xed effects are separate indicator variables for each county. Year fi xed effects are indicators that allow for unrestricted differences in year- to- year changes. Reported in parentheses are standard errors clustered by county. denotes statistical signifi cance at the 10 percent level, at the 5 percent level, at the 1 percent level.
tion is not likely to have been caused by unobserved birth- control policies. Moreover, there appears to be an interaction effect, whereby ultrasound technology has a greater
effect in areas where the One Child Policy is more strictly enforced.
B. Tests for Spurious Trends