An economic analysis of Chinese fertility behavior

Pergamon 0277-9536(95)00214-6 Soc. Sci. Med. Vol. 42, No. 7, pp. 1027-1037, 1996

Copyright © 1996 Pubfished by Elsevier Science Ltd Printed in Great Britain. All rights reserved

0277-9536/96 $15.00 + 0.00

AN ECONOMIC ANALYSIS OF CHINESE FERTILITY BEHAVIOR

GORDON G. LIU, 1 TETSUJI Y A M A D A 2 and TADASHI YAMADA 3 tDepartment of Pharmaceutical Economics, University of Southern California, 1540 Alcazar Street, Rm 140-G, Los Angeles, CA90033, U.S.A., 2Rutgers University, Department of Economics, Camden, NJ 08/02 and 3University of Tsukuba, Institute of Socio-Economic Planning, Tsukuba City, Ibaraki-ken,

305 Japan

Abstract--This paper is the first to present a Chinese general fertility model that simultaneously controls for the endogeneity of infant mortality and per capita income determination at county level. Using the 1982 Chinese population census data, comprising 2305 observational units, this analysis improves on existing studies in several ways. First, since all the underlying variables are measured at the Chinese county level, we treat both the per capita income and infant mortality rates as endogenous, as opposed to exogenous as assumed in most previous studies on Chinese fertility. Our testing results strongly reject the null hypothesis of the exogeneity of both infant mortality and income determination within our model. Secondly, concerning the hypothesis of nonlinear income effect on fertility behavior, we examine both the variable income-elasticity and constant income-elasticity models. Strong evidence is obtained in support of the variable income elasticity model, predicting a U-shaped income effect on Chinese general fertility. This suggests that a more equitable income distribution leads to a reduction in the Chinese fertility rates. Thirdly, employing the two stage least squares procedure, we find a much stronger positive replacement effect of infant mortality when the endogeneity of infant mortality and income are both controlled for simultaneously. Our results indicate that Chinese general fertility may well be shaped by optimizing behavior.

I. INTRODUCTION

The issues of Chinese fertility behavior and population growth are of considerable significance not only in shaping China's own development strategies, but also in formulating world-wide population policies, as well as in developing demographic theories. According to the World Bank [1] China's population in 1991 was 1.15 billion, over one fifth of the world's total population of 5.4 billion. Since 1949, when the People's Republic of China came into being, its total fertility rate has fallen precipitously from 6.14 in 1949 to 2.6 in 1981 [2]. General fertility rates, however, still vary greatly across areas ranging from 36 to 232 per 1000 women aged between 15 and 49.

In order to control its population growth, China started its family planning campaign in the 1950s. This was followed by an urban-oriented birth control policy pursued in the late 1950s, which was disrupted by the Cultural Revolution beginning in the middle of the 1960s. In the early 1970s, China launched its second national large scale campaign of "later, spacing and fewer" births (Wan-Xi-Shao). China's latest effort to control its rapid population growth was the one-child policy beginning in 1980. While these family planning programs are widely recognized as important instruments in slowing down Chinese population growth, the profound impact of dynamic changes in public health and socioeconomic conditions are poorly explored. In particular, there is very little empirical research on how, and to what

extent, Chinese fertility behavior has been shaped by the changes in population health status, income and social norms.

In the West it is widely argued that fertility is influenced by infant mortality, a prominent measure of population health status, for both developed and developing countries (e.g. Refs [1, 3-10]). There are two primary forces through which fertility could be influenced by infant mortality: (1) a biological effect--infant deaths and fertility may be biologically linked to a certain extent, e.g. frequent infant deaths truncate breastfeeding, shorten the sterile period, and thus may increase the likelihood of next conception; and (2) a replacement effect--households' fertility behavior may respond to the anticipated risk of child deaths by directly varying their planned births [5]. Further, to replace actual deaths, so as to keep the optimal number of children, households that are exposed to higher child mortality rates, would end up with more births [3, 9].

In the case of China, a few studies have been carried out concerning the relationship between fertility and child mortality (e.g. Refs [2, 9, I 1-14]). Using the 1982 national census data at the provincial level (28 provinces, municipalities and autonomous regions), Gu [1 l] demonstrates how the total fertility rates can be determined by five groups of independent variables: structural development; quality of life; women's status; rural consumption level; and family planning. Gu argues that the prolonged decline in Chinese fertility rate should not be attributed solely

1027

1028 Gordon G. Liu et al.

to birth control efforts. He finds that some socioeconomic factors such as economic development, progress in the provision of health care, culture and education, and women's rights could play a critical role in reducing the Chinese fertility rate. Poston and Jia [2] present a multiple regression analysis of how Chinese general fertility rates are influenced by four socioeconomic development variables, including the infant mortality rate. Their study uses a sub-sample of 1982 census data truncated for rural counties in 6 geographic regions. One of their major findings suggests that the Chinese general fertility rates are determined primarily by infant mortality, coupled with the illiteracy rates and proportion of the industrial population.

Based upon a micro survey data set of 13,307 ever-married women aged 15-49 years, Zhang [9] estimates a fertility model using both methods of ordinary least squares (OLS) and Tobit maximum likelihood estimation (MLE). Zhang's study includes child mortality as one of his exogenous variables and finds a high replacement rate of 0.83. In another study following the approach suggested by Olsen [3], Zhang [14] finds a smaller direct replacement effect of 0.6 when he corrects for the potential upward bias due to the endogeneity of infant mortality.

Another important argument in the field of fertility economics is concerned with the interrelationship between fertility and income determination. On the other hand, if children are not considered to be inferior goods, fertility would increase with rising income. However, this hypothesis lacks solid empirical support [15]. It has been frequently observed that fertility is strongly associated with income in a negative manner, particularly in developing countries with relatively low income levels. Yet, this negative income effect appears to decrease as income rises, and it could turn to be positive beyond a certain level of income [15-17].

Using the Chinese census data base at a provincial level, Lee [17] addresses how income distribution determines the Chinese fertility pattern. According to his OLS estimates, Lee finds that the linear effect of per capita income appears to be negative and the nonlinear income effect to be positive. While Lee's empirical evidence supports the U-shaped income hypothesis proposed by Repetto [15], his study is restricted to a sample of only 28 observational units measured at the Chinese provincial level. Moreover, Lee's model does not control for infant mortality effect, while concentrating on the measurement of income effect on fertility.

It is certain that the existing studies on Chinese fertility, reviewed above, has made a great contri- bution to our understanding of the Chinese fertility behavior, yet substantial problems remain. First, the Western demographic literature argues that fertility and infant mortality are inter-determined. That is, while a major determinant of fertility, infant mortality could also be a function of fertility and

other factors. For instance, higher rates of fertility and population growth may lead to reductions in the availability of and accessibility to given area-wide health resources including maternal and child care on the per capita basis. These reductions, in turn, could have a profound negative impact on infant health. Second, a similar argument could be made for the inter-dependency of income determination with fertility. A faster growth of per capita income may be largely due to a lower rate of population growth for a given level of area-wide gross income. A decrease in fertility could also permit an increase in the ratio of capital to labor and in labor's share of gross output which, in turn, may lead to a further increase in income growth [15].

In the context of econometric analysis, the dependency of both income and infant mortality upon fertility raises the endogeneity issue of the two explanatory variables in a fertility model. Following the fundamental principles of econometrics, failure to control for the endogeneity of explanatory variables in a model could result in very biased estimates. To the best of our knowledge, however, none of the existing studies have addressed the endogeneity issues associated with the determination of infant mortality and income in Chinese fertility models using aggregate data. Motivated by such a gap in the literature, we present the first Chinese aggregate fertility model with variable income effect that simultaneously controls for the endogeneity of both child mortality and per capita income. Our analytic framework stems from the notion of the neoclassical utility theory [18, 19]. We estimate this model using two stage least squares estimation on data from the 1982 Chinese population census. The rest of this paper is organized as follows. In the next section, we present the analytic framework from which the empirical econometric model is derived. Section III discusses the data sources and estimation approaches. Our major empirical findings are discussed in Section IV. We conclude this study with a summary of our findings and policy recommendations in Section V.

II. ANALYTICAL FRAMEWORK

TO gain underlying insights on the determination of fertility behavior in formulating our analytical framework, we conceptually follow the household utility-maximization principle [18]. Consider a weakly separable utility function:

T

O,, -~ X {VI(N'r; Ei') + V2(Q,,; Ei,)} e-a" a

a <<. t ~ T; Ni, = f~,Oi,; O <<. Ni, <<. f~, (1)

where for household i at the time period t, N~, denotes the number of surviving children as the product of total births f~,, and an underlying child survival rate, 0~,, Q~, denotes a composite consumption good; Et, represents a vector of other conditional variables; at,

An economic analysis of Chinese fertility behavior 1029

is a discount factor reflecting household preference; and T indicates the upper limit of parental reproductive age starting from age a. Households are assumed to adjust their consumption behavior in accordance with the number of surviving children. That is, households choose an optimal combination of births and consumption so as to maximize their total utility, subject to a wealth constraint:

y,, = Q,, + Pc,~,0,, + Pg,t(f,.,O,, - 1) (2)

where Pci, is the standardized price of raising a child, and Pgi, is the cost of governmental mandatory penalty for violating the one-child policy. As usual, we assume diminishing marginal utilities of both surviving children, frO#, and composite goods, Q#, namely Ujl is concave with respect to f~,0# and Q#. Maximization of U# subject to equation (2) with respect to f , and Qi, leads to a first order condition:

V'2{y,t-f~,O#(Pci, + P,i,) + P~#; E~,} = 1 . (3) v', (f,,0,,; E,,) Pc,, + P,,t

Equation (3) serves as a necessary condition for households to determine the optimal number of birthsft as a function of the underlying variables such as income, the survival rate of children, the preference for and the perceived cost of children. It is noteworthy that the total perceived cost of children includes both the shadow price of children Pc,',, largely determined by parental socioeconomic and demographic characteristics; and the more-than-one- child penalty Pg#, which varies greatly across areas and populations in practice. There exists no data, however, allowing one to measure the heterogeneity effect of Pgi, directly. In order to capture such a hard-to-measure effect, in our empirical study we substitute it with the region-specific proportion of ethnic minorities which, in fact, serves as a primary determinant of how birth control programs, particularly the one-child policy, are implemented in China [20, 21].*

While equation (3) offers us useful insights about the associations of household fertility behavior with its major determinants, an aggregate fertility model is required in order to employ the Chinese census data which were compiled at county level. Strictly, a number of restrictions must be imposed in order to integrate the individual behavioral model (3) up to an aggregate model [22]. In essence, however, an aggre-

*The Chinese government allows some flexibility in pursuing the one-child policy for its 55 officially recognized minorities, depending upon specifically the historical, economic, natural, traditional and geographic situations. For example, in the provinces of Inner Mongolia, Jiangxi, Guangdong, Sichuan, Guizhou, Yunnan, Gansu, Qinghai and Ningxia, minorities are encouraged to have one child, but are allowed to have two. Whereas, in some areas where the majority of the population is comprised of minorities and minority population growth is slow, the government helps and encourages them to have two children and even permits three children.

gate general fertility model could be derived from solving equation (3) for f~ and then integrating it across individual is within each county. Assuming that the derived demand function for births is subject to a generalized Cobb--Douglas specification, an aggregate fertility function may be written as:

Fj = []oM~" Y~" X~Pke ~j (4)

where Fj is the general fertility rate for thejth county, Mj- -m( l - 0 j ) is the county-specific infant mortality rate, ~ denotes the county-specific per capita income, Xj represents a vector of other area-wide socioeconomic determinants of fertility, caj is a stochastic term capturing the unobservables in the equation, and fl~s are the parameters to be estimated.

In this study, we focus on the role of infant mortality (M~) and income (Yj) in the determination of Chinese general fertility (Fj). The economics of fertility argues for two primary linkages between infant mortality and fertility. From the perspective of the demand for children, in order to replace the children who have died, parents will produce more births than if their children had lived--direct replacement effect; and in response to an anticipated increase in the likelihood of child death, parents may decide to plan for more children than otherwise---hoarding effect. From the perspective of the supply side, a high infant mortality may be associated with women's poor health and endowment and low fecundity, leading to a reduction in fertility.

An increase in household income directly improves women's health and reproductive capacity (fecundity) which, in turn, could contribute to the potential supply of fertility [23]. Rising income also increases one's purchasing power, which could further induce a demand for more children, provided children are considered to be normal goods. Increased income, however, also generate forces to reduce fertility. As income rises, the benefits to parents from having children tend to decrease, while the opportunity cost of having children tends to increase [6, 19]. Rising income, accompanied by other social changes such as urbanization and industrialization, could also change the social norms towards having fewer children. In general, as income rises the value that households place on children decreases, relative to obtaining other durable goods and investments. Particularly in agricultural societies, parents may depend crucially on their children both as productive assets for their labor when they are young, and as a principle method of saving, security, or insurance for their future in life [24]. Thus, rising income and social progress could substantially reduce the dependency of one gener- ation on the next, leading to a downward trend in demand for children [6].

Vector Xj consists of other explanatory variables which are essentially pre-determined prior to current fertility, or determined largely by the forces such as government actions other than people's own behavior. Subject to our data sets discussed in the next

,~)SM 42/7~F


section, we control for several variables as exogenous included in Xj. First, we include the county-specific population's educational level. While education could influence people's fertility behavior through a variety of ways, generally it leads people's demand for children towards a lower quantity but higher quality. This is because the opportunity cost of raising a child increases with parental educational level due to higher forgone earnings. Well-educated households are likely to put a greater emphasis on child quality as opposed to quantity, leading to a higher cost of raising each child and consequently a demand for fewer children [19]. Education may present an image of new life styles that competes with traditional images of large families by altering cultural norms or shifting tastes in a manner unfavorable to children [23]. And finally education augments people's stock of knowledge and improves the efficacy of fertility planning and control, resulting in fewer unwanted and total births.

We also control for women's contraceptive prevalence rate. Following Easterlin and Crimmins [23], fertility is generally determined by three forces-- demand, supply and fertility regulation costs. The fertility regulation refers to people's deliberate use of contraceptive instruments. While this variable could be determined largely by people's choice behavior in well-developed regions, it could also vitally depend on the availability of and access to the community- based family planning methods and supplies in most developing countries. More importantly, in the case of China, the extent of massive adoption of contraceptives may well reflect how firmly the government birth control campaigns have been pursued in each area [25]. Furthermore, the variable is measurable only at the province level with the data sources available. Taking into account these considerations, we treat the contraceptive prevalence rate as exogenous, as opposed to be endogenous [24].*

Another controlling variable in our fertility model is the proportion of the population under age 14. Since this variable is not measured at the household level, it carries little individually cumulative fertility

effect. Rather, for a given cohort of a population residing in a particular area, this variable may primarily reflect four aggregate effects on their fertility: the population's reproductive capacity or natural fertility, the social norms about children, the community-based public health facilities influencing children's rates, and the historical efforts of government in implementing the birth control polices in that area. Since these effects are very likely to shape current fertility but are not measurable directly, we include the proportion of the population under age 14 as a pre-determined variable to capture these effects [15].~" It is also noteworthy that this variable may share some common determinants with the contraceptive prevalence rate, yet they are apparently not perfectly or highly correlated with each other.

Our model also controls for other exogenous variables including the proportion of industrial labor force, population density, the proportion of the population above age 65, and an ethnic dummy variable. We use the proportion of industrial labor force as a proxy for industrialization. In the economics of fertility, industrialization has been commonly identified as an important driving force for a lower rate of fertility, primarily by reducing the economic benefits of children, and raising the costs of child-rearing to working parents [6, 23]. We include population density because it is associated with the area-specific availability of public health programs, urbanization and effectiveness and intensity of government birth control policies. We also assume that people's fertility decisions would take into account the area-wide availability of child care services. In China, such services are largely provided at home by grand-parents in extended families when available. Therefore, we include the proportion of the population over age 65 to capture in part the supply effect of child care services. Finally, a minority dummy variable is created to control for the well- known differentiated birth control policies in practice areas with mostly the Han nationality, the ethnic majority of Chinese, and those with mainly other ethnic minorities.

*In their report using the household survey data for Zimbabwe, Thomas and Maluccio [24] find the estimated effect of contraceptive use on fertility to be essentially the same whether or not the usage is treated as endogenous.

tFollowing the notion of general "stock-adjustment" model [29, 30], we also could gain similar insights about the role of the number of living children in the determination of current fertility behavior within a household. In this representation, current fertility is primarily determined by three elements; the actual stock of living children, the desired number of children, and the parental fecundity and the efficiency of their fertility control. Basically, two alternative hypotheses could be made concerning the association of current fertility with the stock of living children, which may lead to either a positive or negative relationship, depending on how the three elements are interacted within the household [15].

III. DATA AND ESTIMATION

The data base used in this study is developed from The Population Atlas of China, a published data source from the 1982 Population Census of China, compiled by the Population Census Office of the State Council of the People's Republic of China and the Institute of Geography of the Chinese Academy of Sciences [26]. The 1982 population census is the third national population census, following those conducted in 1953 and 1964. The census data base is augmented with 1981 and 1982 contraception rates and ethnic minority population proportions by province, taken from The 1986 Year book of Chinese Birth Planning [27]

An economic analysis of Chinese fertility behavior

Table 1. Variable definitions

Variable name Definitions

Y; LY Per capita gross value of industrial and agricultural output 1982

POPDEN; LPDEN Population density (persons/kin 2) POPOI4; LPOPI4 Percentage of total population aged 0-14 POP65; LPOP65 Percentage of total population aged 65 and

over W1549; LW1549 Percentage of total female population of

childbearing age (15-49) IMR; LM Infant mortality rate 1981 (infant deaths per

1000 live births) EDU; LEDU Number of people with junior middle school

and above per 10,000 population POPINDU; LINDU Industrial population as a percentage of

total employed population CON; LCON Average contraception rate for 1980 and

1981 DMINO Minority dummy variable, DMINO = 1 if

mino > 39% otherwise 0 GFR; LGFR General fertility = BIRTH/W1549 L y2 Logarithm of y2

initial L denotes logarithm of the original *Second variable with variables.

and The 1990 China' s 4th National Population Census Data Sheet.

At the time of the 3rd Census, there were 235 cities and 2143 county-level administrative units, across the 29 provinces, autonomous regions and municipalities on the mainland of China. Since both the infant mortality and contraception rates are not reported for Tibet in the original data source, we had to drop all the 72 observational units of Tibet. Another dropout is the Yanshan District, Beijing, due to its unusually high per capita income of 28,475 yuans, which is nearly three times the second highest per capita income level of 9961 yuans, in contrast to the national average of 654 yuans. Thus, the final data set has 2305 observational units at the county-equivalent administrative level.

The dependent variable of our model is measured by the general fertility rate (GFR), which is defined as the total number of living births per 1000 reproductive women aged between 15 and 49. Using the census data base, we construct the GFR by dividing the crude birth rate by the percentage of women aged 15--49. The infant mortality rate is measured by the ratio of infant deaths per 1000 live births. We measure the per capita income (Y) with the per capita gross value of industrial and agricultural output, a popular officially used statistic in China.* The definitions of other exogenous variables and the descriptive statistics for the key variables are given in Tables 1 and 2.

Examining the statistical distribution of GFR for all 2305 observation units, we find that the general fertility rates do not appear to follow a normal

*There are three officially used statistics for income measurement in China: per capita gross value of industrial and agricultural output (GVIAO), per capita gross value of industrial output (GVIO), and per capita gross value of agricultural output (GVAO). The latter two are not reported in the Census data.

1031

distribution as shown in Fig. 1. However, after a logarithmic transformation, LGFR exhibits a well shaped normal distribution, as shown in Fig. 2. We, therefore, present the fertility model equation (4) in a log-linear form as follows:

LGFRj = ~o + ~u LMj + ~yL Yj + ~ ~k LXkj + ~ j. (5)

We first obtain the ordinary least squares estimates (OLS) of the fertility model, reported in Appendix A. Contrary to our theoretical expectations, the OLS estimates show that both the infant mortality rate and per capita income play no statistically significant role in the general fertility model. Moreover, the general fertility rate appears to be negatively correlated with infant mortality rate, and positively with per capita income which, again, contradicts the primary hypotheses.

In order to determine whether the inconsistent OLS results are due to the potential endogeneity of LM or LY as hypothesized, we conduct a Wu-test [28], presented in Appendix B. The result of the Wu-test highly rejects the null hypotheses of exogeneity of both LM and LY at the 0.01% level. Based on these observations, we then re-estimate the LGFR model where both infant mortality and income are treated as endogeneous using the two stage least squares approach (2SLS).

The 2SLS estimates reported in Appendix C, in contrast to the OLS estimates, better identify the impact of infant mortality and income on fertility as expected. Unlike the negative effect predicted by the OLS model the 2SLS model shows the relationship between LGFR and LM to be positive, predicting that high infant mortality leads to high fertility. The 2SLS income elasticity is about 0.14, as compared to only 0.006 predicted by the OLS estimates. Both the mortality and income effects are also highly significant at the 0.01% level.

Although the 2SLS estimates of equation (5) appear far more promising than the OLS estimates, the model specification is still constrained by the nature of its constant income-elasticity specification. Numer- ous studies in the West suggest that income effect on fertility is unlikely to be constant as the level of income changes [1, 15-17]. It has been frequently observed that a negative association between fertility and income is stronger in most developing countries

Table 2. Descriptive statistics of some key variables

Variables Minimum Maximum Mean SD

Y 70 9961 654 771.94 POPDEN 0.1 28357 377 796.43 POPOI4 13.8 48.5 34.97 4.89 POP65 0.2 9.2 4.70 1.13 W1549 18.3 30.4 23.25 2.09 IMR 6 319 39.05 29.07 ED U 188 4384 1708 63.74 POPINDU 0.1 76.7 13.23 14.55 CON 56.6 98.25 82.56 8.82 GFR 36.22 232.54 98.54 32.84


e~

tL tL

36 55 74 93 112 131 150 169 188 207 226 236 GFR

Fig. 1. The distribution of GFR.

3.54 3 .74 3 .94 4 .14 4 .34 4 .54 4 .74 4 .94 5 .14 5 .34 5 .54

LGFR Fig. 2. The distribution of LGFR.

than in developed countries. Repetto [15] also argues that the way income is distributed among different classes of people may be even more important than the income level itself.

In order to determine whether the constant income-elasticity model is appropriately specified, we perform two investigations. First, we examine how the residuals {ej} from the 2SLS constant income- elasticity estimates are correlated with income levels {LYj}. Plotting the residual {ej} against the series {LYj} as illustrated in Fig. 3, we find the error terms from the constant income-elasticity model clearly to be associated with the income level in a non-random manner, suggesting the possibility of non-constant income-elasticity for the model. Following this line, we then examine whether and how the variable income elasticity of fertility, /~y, would respond to changes in income level. In doing so, we specify an unrestricted model allowing income LY: to change across three levels: low income group (25% quartile); middle income group (75-25% quartile); and high income group (100-75% quartile). The 2SLS estimates of the unrestricted model are given in Appendix D. Based on these estimates, we conduct an F-test of the unrestricted model against the constant in income-elasticity model (6) as the restricted form of the specification:

Unrestricted model (H 1):

/~0I ~ flOll ~ ~0111 ; ffyI ~1~ /~yll ~;~ ~ylll ; ~kl = ~klI = ~klll ;

Restricted model (H0):

F = (e~er - e'e)/q = 24.99/4 e 'e / (N - k) 76.96/2291

= 185.98 ~ F(q, N - k) = F(4,2291). (6)

The F-test significantly rejects the restricted model. In fact, the estimates of the unrestricted model show explicitly that income elasticity is negative at low income levels, suggesting that for low income areas, rising income reduces fertility. As income rises, however, the income effect becomes positive. Further- more, the positive income elasticity increases as income rises but it is statistically significant only at high income levels. As a consequence, we re- formulate our fertility model by incorporating a quadratic term of the per capita income into equation (5):

LGFR: =/~o + ~LMj + ~yLYj + ~,2Lr~

+~/~kLXkj + E:. (7) k

In this model, the income elaticity with respect to fertility is a function of LY, given by the derivative of LGFR with respect to LY. Controlling for the endogeneity of L Y and LM, we obtain the 2SLS estimates of the model presented in Table 3. Finally, we examine whether the residuals from the variable income-elasticity model are still noticeably correlated with L Y in a non-random manner. In contrast to Fig. 3, Fig. 4 demonstrates that the distribution pattern of the residuals {ej} is highly symmetric around 0 regardless of the income level L Y r With this and other strong evidence shown above, it is

Table 3. 2SLS Est imates o f var iable income-elastici ty L G F R model

Variables Pa ramete r s ~k SF~#k T-test Significance

I n ~ r c e p t 4. I 17 0.976 4.2183 0.0001 LM 0.117 0.014 8.252 0.0001 L f -- 1.269 0.236 -- 5.377 0.0001 L ~2 0.096 0.016 6.021 0.0001 LEDU - 0 . 0 3 4 0.019 - 1.733 0.0832 LCON - 0 . 0 9 1 0.048 - 1.878 0.0605 LPOP14 1.226 0.077 15.955 0.0001 LPOP65 0.346 0.021 16.472 0.0001 LINDU 0.017 0.016 1.094 0.2741 LPDEN -- 0.023 0.005 - 4.401 0.0001 DMINO 0.130 0.019 6.744 0.0001

2 2 R = 0.616; Adjus ted R = 0.614; F-value = 367.31; ,, = 0.0001; N = 2,305; replacement effect of M = 0.30.

X I |

1 0.8 0.6 0.4 0.2

0 -0.2 -0.4 -0.6 -0:8

-1

An economic analysis of Chinese fertility behavior

n l

I I I ~ I i

. - " ' , , : I I R ~ - - ' ~ "

. ~ . J r '

I IJB I m l i m a I l e l e m m m m

Im~ ~ m e

m e

i i m

1033

3 4 5 6 7 Per capita income {LYi}

Fig. 3. Constant income-elasticity model.

8 9 10

clear that the variable income-elasticity model is statistically superior to the constant income-elasticity model. As a consequence, we present a discussion of our major findings drawn from the variable income- elasticity model in the next section.

IV. MAJOR EMPIRICAL FINDINGS

In discussing the results in Table 3, we first focus on the income effect and the role of infant mortality on Chinese general fertility rates. There is strong evidence of the variable income effect. In particular, we find that the per capita income elasticity is an increasing function of L Y at a very high level (~ = 0.01%):

~lGrR, r = ~LGFR/dL Y = - 1.269 + 0.192L Y.

This suggests that rising income reduces fertility when the income level is low. As the income level increases, however, the negative income effect diminishes. And more importantly, the income effect could eventually turn to be positive beyond a certain level of income.

*Using a survey sample of 4758 ever-married women in the Hebei province, China, Zhang [14] reports a total replacement effect of 0.6 on total fertility rate. Yet, some evidence from other countries suggests a smaller replacement effect such as 0.35 for Malaysia and 0.20 for Colombia by Olsen [3, 31].

Using the census data, such a turning point would be at 742 yuans, about 12% higher than the nation-wide average per capita income level of 654 yuans in 1982. Clearly, our results provide strong evidence of U- shaped income effect as shown in Fig. 5. Such an effect has been predicted elsewhere such as the studies by Repetto [15] using both a cross-national sample of 68 countries and household data from Puerto Rico and Korea; and by Lee [17] using the Chinese 1982 Census data at province level. The U-shaped income effect yields an important policy implication for Chinese government: policies redistributing income from rich areas to poor areas could lead to a decline in Chinese fertility rates.

As hypothesized, infant mortality shows a strong positive effect on fertility. The mortality elasticity of LGFR is about 0.12 and is significant at the 0.01% level. According to this result, in China a 10% decrease in infant mortality rate would result in about 1.2% reduction in general fertility rate. Converting the elasticity to a marginal effect evaluated at the mean values of I M R and GFR, the Chinese infant mortality replacement effect on general fertility rate measured at the county level would be about 0.3.* A policy implication form from this finding is clear: in order to further reduce Chinese fertility rates, promoting public health, particularly the maternal and

0.8 0.6 0.4 0.2

| o -0.2 -0.4 -0.6 °0.8

-1

m - - m l

e m i

, i . ~ ~ . : . . ' ~ . _ _ . .em m

e q b n . I e e L ' m n m m i l l I m

.--Ih .=-- m. e" " m m m

m m e e I i

m m e .m

5 6 7 Per capita income {LYi}

Fig. 4. Variable income-elasticity model.

8 9 10

1034

A n- IA .

(.9 . . J

n

O

O ¢.9

5.2

5.1

5.0

4.9

*,.8

4 .7

4.6

4.5 3

Gordon G. Liu et al.

m

5 6 7 8 9

Per capita income (LY)

Fig. 5. U-shaped income effect.

lo

child health, should be used in conjunction with direct approaches, such as family planning programs and government birth control policies.

Concerning the associations of the Chinese general fertility with its exogeneous determinants, we find most of these variables show consistent roles as expected. First of all, people's educational level, LEDU, is negatively correlated with general fertility. The educational level is measured by area-specific proportion of the people with at least a junior middle school education. The education elasticity with respect to general fertility is about -0 .03 and is statistically significant at the 8% level. Similarly, we find the contraceptive prevalence rate, LCON, to be a negative correlate of general fertility. The contraception elasticity is about -0 .09 and is statistically significant at the 6% level. Population density, LP- DEN, also has a negative impact on fertility. The population density elasticity with respect to fertility is -0.023, and is highly significant at the 0.01% level. In summary, we find the areas with higher rates of educational level, contraceptive usage, and population density generally achieve lower rates of general fertility, other things being equal.

In contrast, we find that general fertility rates tend to be higher in areas with a higher proportion of the population under age 14, a higher proportion of the population over age 65, and within the regions with mainly ethnic minorities. In particular, the elasticity of LPOPI4 is 1.23 and is highly significant at the level of 0.01%. This suggests that in China a higher proportion of children under age 14 in one area is likely to be a consequence of the area-wide population's high fecundity, inefficient fertility regulation, or perhaps the poor implementation of the family planning programs by local government. As a policy implication, areas with such characteristics may be

*Under the strong guidelines and recommends by all gov- ernments, the Chinese birth control policies have been highly differentiated across regions and populations, generally being in favor of minorities. For instance, while the one-child campaign has been strictly im- pleraented for the Han nationality in most regions, having multiple children is permitted for most ethnic minorities [32].

targeted closely as to their fertility regulation. The LPOP65 elasticity is 0.35 and is also very significant at the 0.01% level. As discussed earlier, since the elderly in extended families could substantially reduce the child care costs in China, this result may largely reflect a higher demand curve for children by the populations with lower child care costs due to the higher availability of the elderly. Concerning the hetogeneity effect of the birth control policies, we introduce a dummy variable, DMINO, being 1 for all the counties of Xinjing, Qinghai and Guangxi that have the highest proportion of their population as ethnic minorities (>39%), and 0 otherwise.* As predicted, the DMINO captures a substantial variation in fertility rates due to the policy differen- tiation with an elasticity of 0.13 being significant at the 0.01% level.

Finally, we find no evidence of the general fertility being influenced significantly by industrialization, indirectly measured by proportion of the industrial labor force, LINDU. While somewhat contrary to our expectation, that may be in part due to the inappropriate treatment of LINDU as a poor proxy for industrialization, which is unavailable from the census data base.

V. CONCLUDING REMARKS

This paper provides the first estimates of the Chinese general fertility model using county level data, which simultaneously controls for the endogeneity of infant mortality and income determination. First, using the 1982 Chinese population census data measured at the county level, we find strong evidence that both infant mortality and per capita income are functioning as endogenous, largely influenced by fertility and other variables within the general fertility model. Secondly, we find Chinese income elasticity with respect to general fertility to be U-shaped, suggesting that the negative income effect on fertility diminishes as income rises and could ultimately turn out to be positive as income goes beyond a certain level. This finding yields an important policy implication that a more equitable income distribution could contribute to a reduction in

An economic analysis of Chinese fertility behavior 1035

Chinese fertility rates and populat ion growth. Thirdly, high infant mortality, or poor health in general, is associated with higher fertility. We find the total replacement effect of infant mortality to be about 0.3. This finding implies that improving public health, especially maternal and child health, could create favorable externalities in reducing Chinese fertility rates. That is, well coordinated efforts between birth planning and public health sectors would be superior to independent policy actions.

Education and adoption of contraceptives are found to reduce the general fertility rates. It is suggested, therefore, that increasing the educational level and promoting mass access to contraceptive instruments could contribute to reductions in Chinese fertility rates. Moreover, in order to pursue its family planning programs more effectively, Chinese government may closely target those areas with a high proport ion of the populat ion under age 14 and above 65. People in these areas may have a higher demand for children, or higher fecundity. It is also noteworthy that Chinese general fertility rates are very respon- sive to variations in the birth control policy regulations. Areas with mainly ethnic minorities or low populat ion density show much higher fertility rates than other areas, other things being equal. This may be partly due to loose policy regulations in these areas.

Several issues in this area, however, require further investigation. First, China 's economic reform and the rapid transition of the traditional, centrally planned economy toward a market economy have certainly had a considerable influence on Chinese fertility behavior. Using the 1982 census data, however, we were unable to investigate how, and to what extent, the economic reform has shaped Chinese fertility behavior. An investigation of this issue may be feasible using China 's 4th national population census conducted in 1990, when available. Secondly, due to data limitations we were unable to adequately measure the effect of variations in the one-child policy at the county level. Nor do our county-level data allow us to control for any cultural factors which may shape the traditional norms about having children. Finally, we are unable to directly quantify the effect of government penalties for having more than one child in most areas at county level. This is a significant issue because the penalties for having more than one child in many places are substantial. They could be multi-dimensional ranging from political criticism, direct fines, to a wide range of economic punishments influencing child schooling, grain rations, subsidized prices, occupation and housing.

Acknowledgements--We are indebted to Professors Michael Grossman at CUNY Graduate Center and NBER, William Hsiao at Harvard, Michael Kendix at HCFA, and Bernard Okun at CUNY Brooklyn College, We also thank the editor and the anonymous reviewers for helpful comments, and Julia Kyle, Jeffrey Joyce, Jin Wei, Tannei Wang and Avery

Guest for their assistance. Any opinions expressed in this paper, as well as any possible errors, are ours, not those of the Universities.

REFERENCES

1. The World Bank. World Development Report 1993: Investment in Health. Oxford University, New York, 1993.

2. Poston D. L. and Jia Z. Socioeconomic structure and fertility in China: a county level investigation. J. Biosoc. Sci. 22, 507, 1990.

3. Olsen R. J. Estimating the effect of child mortality on the number of births. Demography 17, 429, 1980.

4. Olsen R. J. Fertility and the size of the U.S. Labor Force. J. Econ. Literature XXXII, 60, 1994.

5. Rosenzweig M. R. and Schultz P. T. Consumer demand and household production: the relationship between fertility and mortality. Am, Economic Rev. 73, 38, 1983.

6. Fuchs, V. R. How We Live. Harvard University Press, Cambridge, MA, 1983.

7. Wolpin K. 1. An estimable dynamic stochastic model of fertility and child mortality. J. Political Econ. 92, 852, 1984.

8. Yamada T. Causal relationships between infant mortality and fertility in developed and less developed countries. Southern Econ. J. 52, 364, 1985.

9. Zhang J. Socioeconomic determinants of fertility in China. J. Pop. Econ. 3, 105, 1990.

10. Sah K. The effects of child mortality changes on fertility choice and parental welfare. J. Political Econ. 99, 582, 1991.

11. Gu B. On the effects of Socioeconomic development and family planning on the decline of fertility in China. Chin. J. Pop. Sci. 1, 125, 1989.

12. Li X. The utility of reproduction and economic measures for birth control. Chin. J. Pop. Sci. 1, 311, 1989.

13. Lavely W. and Freedman R. The origin of the chinese fertility decline. Demography 27, 357, 1990.

14. Zhang J. Mortality and fertility: how large is the direct child replacement effect in China. J. Pop. Econ. 3, 303, 1990.

15. Repetto R. Economic Equality and Fertility in Developing Countries. Johns Hopkins University Press, Baltimore, 1979.

16. Boulier B. L. Income redistribution and fertility decline: a skeptical view. Pop. Dev. Rev. 8, 59, 1982.

17. Lee B. S. The effects of income level, income distribution, education and urbanization on fertility rates among 28 administrative regions in China. Korea. J. Pop. Dev. 19, 91, 1990.

18. Becker G. S. A theory of the allocation of time. Economic J. 75, 493, 1965.

19. Becker G. S. and Lewis G. H. On the interaction between the quantity and quality of children. J. Political Econ. gl, 279, 1973.

20. Goodstadt L. F. China's one-child family: policy and public response. Pop. Dev. Rev. 8, 37, 1982.

21. Cooney R., Wei J. and Powers M. The one child certificate in Hebei Province, China: acceptance and consequences, 1979-1988. Pop. Res. Policy Rev. 10, 137, 1991.

22. Stocker T. Empirical approaches to the problem of aggregation over individuals. J. Econ. Lit. XXXI, 1827, 1993.

23. Easterlin R. and Crimmins E. Fertility Revolution. University of Chicago Press, Chicago, 1985.

24. Thomas D. and Maluccio J. Contraceptive choice, fertility, and public policy in Zimbabwe. LSMS Working Paper, No. 109, 1994, The World Bank, Washington, DC.

1036 G o r d o n G. Liu et al.

25. Qin F. The impac t o f family p l ann ing on fert i l i ty in China : an evalua t ion . Chin. J. Pop. Sci. 1, 139, 1989.

26. Popu la t i on Census Office of the State Counc i l o f the People ' s Republ ic of C h i n a and the Ins t i tu te of G e o g r a p h y o f the Chinese Academy of Sciences. The Population Atlas o f China. Oxford Univers i ty Press, H o n g Kong , 1987.

27. Chinese Bir th P l ann ing Y e a r b o o k Commi t t ee . Year- book o f Chinese Birth Planning 1986, People ' s Hea l t h press, Beijing, 1986.

28. W u D. M. Al te rna t ive tests o f independence between s tochas t ic regressors and dis turbances . Econometrica 41, 733, 1973

29. Lee R. D. Ta rge t fertili ty, con t racep t ion , and aggrega te rates: t oward a fo rmal synthesis . Mimeo , A n n Arbor ,

Popu la t ion Studies Center , Univers i ty o f Michigan , 1977.

30. Schul tz T. P. An economic in te rpre ta t ion of the decl ine in fert i l i ty in a rap id ly deve lop ing country: consequences o f deve lopmen t and fert i l i ty planning. Discuss ion Paper , Economic G r o w t h Center, Ya le Univers i ty , New Haven, CT, 1977.

31. Olsen R. J. Cross-sec t ional me thods for es t imat ing the rep lacement o f in fan t deaths . In Research in Population Economics, Vol. 6 (Edi ted by Schul tz T. P.), pp. 111-136. J A P Press, Greenwich, CT, 1988.

32. X iong Y., Y o u n g Y. X. and Z h a n g T. L. The changes and strategies o f chinese ethnic minor i t ies ' fert i l i ty rates. Chinese Pop. Yearbook 1986. Social Services L i te ra ture Press, Beijing, China , 1986.

APPENDIX A

O L S Estimates o f the Constant-Income Elasticity LGFR Model

Variables Parameters ~k SE¢~ T-test Significance ~¢

Intercept 2.996 0.363 8.262 0.0001 LM --0.015 0.009 - 1.652 0.0986 LY 0.007 0.011 0.620 0.5318 LEDU -0.166 0.017 - 10.01 0.0001 LCON - 0.238 0.045 - 5.274 0.0001 LPOPI4 1.022 0.051 20.00 0.0001 LPOP65 0.240 0.018 13.38 0.0001 LINDU -0.006 0.007 -0.928 0.353 LPDEN - 0.031 0.004 -- 7.852 0.0001 DMINO 0.138 0.018 7.852 0.0001

R 2 = 0.618; Adjusted R 2 = 0.617; F-value = 412.65; ~ = 0.0001; N = 2305; Replacement effect of M = 0.037.

APPENDIX B

Wu-test o f the Constant Income-Elasticity LGFR Model

Variables Parameters ~k SE~#k T-test Significance

Intercept - 0.766 0.493 - 1.553 0.1206 LM 0.143 0.012 11.686 0.0001 L Y 0.142 0.026 5.489 0.0001 LED U - 0.071 0.017 - 4.245 0.0001 LCON - 0.126 0.043 - 2.901 0.0038 LPOPI4 1.322 0.068 19.369 0.0001 LPOP 65 0.298 0.018 16.369 0.0001 LINDU -0.051 0.010 -5.233 0.0001 LPDEN - 0.006 0.004 - 1.417 0.1566 DMINO 0.099 0.017 5.851 0.0001 e* u -0.294 0.017 --17.551 0.0001 e * r - 0.159 0.028 - 5.703 0.0001

R 2= 0.667; Adjusted R2= 0.665; F-value = 417.14; ~ = 0.0001; N = 2305. *OLS residuals of LM and LY in the first stage.

APPENDIX C

2SL S Estimates o f Constant Income-Elasticity LGFR Model

Variables Parameters ~t SE¢#t T-test Significance

Intercept - 0.766 0.579 - 1.323 0.1860 L.~ 0.143 0.014 9.955 0.0001 L] ~ 0.142 0.030 4.676 0.0001 LEDU -0.071 0.020 -3.616 0.0003 LCON - 0.126 0.051 - 2.471 0.0135 LPOPI4 1.322 0.080 16.50 0.0001 LPOP65 0.298 0.021 14.39 0.0001 LINDU -0.051 0.011 -4.458 0.0001 LPDEN -0 .006 0.005 - 1.207 0.2274 DMINO 0.099 0.020 4.985 0.0001

R 2 ~ 0.583; Adjusted R 2 = 0.58l; F-value = 355.81; • = 0.0001; N = 2305; Replacement effect of M m 0.36.

An economic ana lys i s o f Chinese fert i l i ty behav io r 1037

A P P E N D I X D

2SLS Estimates o f the Unrestricted LGFR Models at Different Income Levels

Model I Model II Model III Variables (Low income) (Middle income) (High income)

Intercept 0.841 (I.26) --0.113 ( -2 .12) --1.426(-5.31) LY --0.149 (-1.71) 0.015 (0.22) 0.205 (5.87) LAT/ 0.124 (9.07) 0.124 (9.07) 0.124 (9.07) LED U - 0.045 ( - 2.38) - 0.045 ( - 2.38) - 0.045 ( - 2.28) LCON -0.089 ( - 1.83) -0.089 ( - 1.83) -0.089 (-- 1.83) LPOPI4 1.250 (18.69) 1.250 (18.69) 1.250 (18.69) LPOP65 0.326 (16.03) 0.326 (16.03) 0.326 (16.03) LINDU -0.005 (-0.51) -0.005 (-0.51) --0.005 (-0.51) LPDEN -0.019 (-4.26) - 0.019 (-4.26) - 0.019 (--4.26) DMINO 0.123 (6.42) 0.123 (6.42) 0.123 (6.42)

R 2 = 0.65; Adjusted R 2 = 0.65; F-value = 331.82; ~t = 0.0001; N = 2305; Replacement effect of M = 0.32.

Documents

An economic analysis of Chinese fertility behavior