Changes in Family Structure in China - A Simulation Study Zeng Yi

Changes in Family Structure in China: A Simulation Study
Author(s): Zeng Yi
Source: Population and Development Review, Vol. 12, No. 4 (Dec., 1986), pp. 675-703
Published by: Population Council
Stable URL: http://www.jstor.org/stable/1973431
Accessed: 14-03-2016 14:53 UTC
Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at http://www.jstor.org/page/
info/about/policies/terms.jsp
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of content
in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms of scholarship.
For more information about JSTOR, please contact support@jstor.org.
Population Council and Wiley are collaborating with JSTOR to digitize, preserve and extend access to Population and
Development Review.
http://www.jstor.org
This content downloaded from 212.219.138.14 on Mon, 14 Mar 2016 14:53:42 UTC
All use subject to JSTOR Terms and Conditions
Changes in Family
Structure in China:
A Simulation Study
Zeng Yi
It is widely recognized that remarkable changes in Chinese
family structure have accompanied changes in the rates of marriage, fertility,
and mortality in recent decades. The average age at marriage of females in-
creased gradually from around 18.5 years before 1949 to about 20 in 1970
and increased rapidly thereafter to about 23 circa 1980. The total fertility rate
declined from about 6 births per woman before 1970 to 2.5 in 1982. Life
expectancy of China's population increased from 47 years in 1950 to 67.9
years in 1981-66.4 years for males, 69.4 years for females (Jiang Zhen-hua
et al., 1984, p. 5).'
How do these tremendous changes in demographic factors affect Chinese
family structure? Vital rates and current family structure data alone cannot
answer this question; we need some way to relate these two types of infor-
mation. This article reports the findings of a simulation study of how changing
demographic factors affect family size and family types. As preface to the
model and interpretations of the output, I first present a brief review of the
cross-sectional evidence on the changes in Chinese family structure. A non-
technical account of the simulation model is given in the second section.
Presentation of the simulation results and discussion of the lessons we have
learned from this exercise are the topics of the remaining sections.
A brief review of the cross-sectional
evidence
The average size of the Chinese family is much smaller than it was before the
formation of the People's Republic in 1949. The nuclear family has become
the dominant family form. The extended family with married brothers living
together has become rare; however, the three-generation family remains an
important family type in Chinese society.
POPULATION AND DEVELOPMENT REVIEW 12, NO. 4 (DECEMBER 1986) 675
676 Family Structure in China
Table 1 shows the substantial variation in average family size2 in China
over the last half century. In prerevolutionary China, family size was around
5.5; by 1953 average family size had fallen to 4.3. This sharp decline was
due mainly to an increase in the number of families resulting from the breaking
up of some extended families, especially those of married brothers living
together. However, family size increased significantly to 4.78 in 1973. This
may have been attributable to housing shortages. Housing construction was
seriously disrupted during the Cultural Revolution of 1966-76; as a result,
some married brothers and their spouses and children were forced to live
together. The average family size found in the 1982 census was 4.43-higher
than in the 1953 and the 1964 censuses but below the size reported in 1973.
Table 2 shows that in 1982 the proportion of nuclear families had increased
by about 30 percentage points and the proportion of multigeneration families
had sharply declined as compared with 1930.
In light of the deeply rooted traditional Chinese ideology favoring large
families, such remarkable changes in family structure could only have been
precipitated by significant socioeconomic change. In prerevolutionary times,
well-to-do Chinese made every effort to sustain large multigeneration families
so as to safeguard the land and property they had accumulated over generations.
Parents did not want their children to live separately, because they did not
want the family's land and property to be divided. Poor families owned very
little land and scanty means of production, which were not partible, constrain-
ing them to maintain extended families as well. Under the individual ownership
system and extremely low productivity levels, it appeared profitable in the
traditional Chinese context to maintain large families (Ma Xia, 1984).
Of course, many Chinese could not reach or maintain the traditional
ideal because of the lack of three or more surviving generations in the same
family, or because poverty forced some brothers to migrate to seek employ-
ment. Nevertheless, as observed by Parish and Whyte (1978), given the high
death rates prevailing in China before 1949, the average family size of over
five (see Table 1) indicates that most Chinese who had the opportunity to live
in joint families did so.
TABLE 1 Average size of Chinese families,
selected years, 1930-82
Average
Period Source size
1931 Sample survey of 22 provinces (Buck, 1937) 5.21
1930-40 Sample survey of birth histories covering 7 regionsa 5.58
(quoted from Ma Xia, 1984)
1953 Census 4.30
1964 Census 4.29
1973 Household statistics 4.78
1982 Census 4.43
a Survey carried out by the Population Research Center of the Chinese Acad-
emy of Social Sciences.
Zeng Yi 677
TABLE 2 Percent distribution of Chinese family types,
1930 and 1982
Three-generation
Nuclear families families and
(including one- other extended
Year Source person families) families Total
1930 Survey by Li 51.5 48.5 100.0
Jing Han in
Hebei Provincea
1982 Censusb 81.2 18.8 100.0
1982 One-per-thousand 80.5 19.5 100.0
fertility survey
data from Liaoning
Provincec
1982 One-per-thousand 76.5 23.5 100.0
fertility survey
data from Hebei
Provincec
1982 One-per-thousand 70.0 30.0 100.0
fertility survey
data from Fujian
Provincec
SOURCES: a quoted from Ma Xia (1984). b State Statistical Bureau (1983a).
c Lavely and Li (1985).
Following the Chinese Revolution in 1949, feudal landownership was
abolished and the land was reallocated. Formerly large, well-to-do families
were split up, because they no longer owned vast amounts of land. Once-poor
families gained assets, enabling married brothers to live apart and unmarried
brothers to marry and leave the extended family. Furthermore, as a result of
the cooperative movement, members of families no longer worked their own
land; instead, they worked in production teams under unified direction and
with collective income distribution. The advantages of maintaining a large
family in which married brothers live together, working for private production,
no longer obtained.
Clearly, the Chinese family has become a smaller unit and there are
fewer extended families than in the 1930s. However, the three-generation unit
is still an important family type in China. China's 1982 census indicates that
18.8 percent of Chinese families are extended. Other, more recent surveys
provide confirmation. For example, according to a sample survey of 2,035
peasant families in Sichuan Province, families of more than three generations
account for only 1.0 percent of all families, whereas three-generation families
account for 22.3 percent (Zhao Xishun, 1985). The most recent in-depth fer-
tility surveys in China show that among all interviewed ever-married women,
53.6 percent in Hebei, 54.8 percent in Shaanxi, and 40.3 percent in Shanghai
live with their parents-in-law or parents (State Statistical Bureau of China,
1986). The 1982 one-per-thousand fertility survey found that 25.9 percent,
31.6 percent, and 40.9 percent of all persons in Liaoning, Hebei, and Fujian
provinces, respectively, live in families of three or more generations. Several
other local surveys conducted by Western scholars show similar patterns (for
example, Parish, 1981; Pasternak, 1986).
Evidently the proportion of multigeneration families was dramatically
reduced in the 1950s. Did the pattern of coresidence (parents and one of their
married children living together) differ very much in the early 1980s from the
mid-1950s? Although we do not have national-level data on family types in
the 1950s to answer this question directly,4 the census observations on average
family sizes (4.30 in 1953, 4.29 in 1964, and 4.43 in 1982) can tell us
something. The total fertility rate in 1982 (2.5 children per woman) was less
than half the average over 1950-70 (5.8). A reduction in fertility certainly
reduces family size if other factors remain unchanged. Reduced mortality, by
contrast, increases family size. It seems unlikely that the effects of the mortality
decline exceeded those of such a remarkable fertility decline. (This suspicion
will be verified later through simulation.) Therefore, the average family size
in 1982 would have been reduced if the proportion of parents who live with
one married child (thus forming three-generation families) was significantly
lower in 1982 than in the 1950s and 1960s. But this was not the case. Thus
we may reasonably conclude that the prevalence of coresidence of parents and
one married child remained more or less stable from the mid-1950s until the
early 1980s. (The proportion of families with married brothers living together
declined notably, resulting in an increase in the proportion of nuclear families.)
In sum, compared with the 1930s the proportion of extended families
has decreased sharply, largely due to declines in large families of married
brothers living together, but the three-generation family is still an important
family type in China. Therefore, the model that was constructed for the study
of family dynamics in China, which is also expected to be applicable to other
Asian countries, had to take into account both nuclear families and three-
generation families.
The simulation model
One limitation of the kind of cross-sectional evidence we have just reviewed
is that it cannot depict the relationship between ongoing changes in demo-
graphic variables and changes in family size and structure. How do chang-
ing fertility and mortality affect family type and family composition?
This interesting research question has received considerable attention from
demographers.5
One promising modeling approach entails simulation of family formation
using life-table methods. A standard life table describes how a population
cohort as it ages is diminished by death. In a "multistate" life table, transitions
to other states are also identified-to marriage, and thence to widowhood or
divorce, for example (see Willekens et al., 1982; Rogers, 1975; Schoen and
Land, 1979). Bongaarts (1983) takes this multistate "marital status life table"
and adds to it a variety of maternal (or paternal) states, to make a nuclear
Zeng Yi 679
family status life-table model.6 He defines a marital-parity-maternal-fecundity
status life-table population. (Parity status refers to number of children born;
maternal status, to the number of children living at home; and fecundity status,
to the capacity to bear another child.) Each surviving member of the life-table
population is distinguished with respect to four states. As the member ages,
she (or he) changes her (his) marital, parity, maternal (paternal), and fecundity
statuses. If the member dies, she (he) quits the system. The model inputs are
the single-age-specific probabilities of marriage, fertility, mortality, and fe-
cundity status changes. The model output is the distribution of the life-table
population by age, marital status, parity, fecundity status, and number of
surviving children. The nuclear family size and nuclear family life course
based on the input demographic rates can be derived from the simulation.
Watkins et al. (1984) and Menken (1985) applied Bongaarts's model to in-
vestigate the implications of the sets of demographic rates on the family status
of American females.
Bongaarts's nuclear family system assumes no more than two generations
in the family. However, the reality in China and in many other developing
countries in Asia and elsewhere is not so simple. Some children leave the
parental home to set up a nuclear family before or after marriage. Some children
do not leave the parents' home at all. Parents and their married children and
grandchildren form three-generation families or other types of extended
families.
How can we model a general family-status system in which the members
of different generations interact in such a complex way? Brass's concept of
the marker-or reference person of the family (see Brass, 1983)-provides a
good starting point. Following Brass, we suggest making the female the marker,
not only because she marries earlier and lives longer than her husband, but
also because parity-specific fertility data are much more easily obtained for
females than for males and because, following divorce, young children usually
live with their mothers.7
In the extended model then, demographic processes operate throughout
a woman's life: marital status, parity, and maternal status change. In addition,
a woman may at some point in her life become a marker of a family (she may
also change her marker status afterwards)-a fourth kind of status change. All
four must be kept track of in the simulation model.
Clearly, each marker represents a family. The marital and maternal status
of markers can be used to help determine their family size. The number of
ever-married nonmarkers with at least one surviving child is equivalent to the
number of three-generation families,8 since nonmarkers by definition are
women living with their mother or mother-in-law. Different types of three-
generation families can be distinguished by various combinations of grand-
mothers' (marker) and mothers' (ever-married-nonmarker) marital statuses.
Family size can be determined by the marital status of the grandmother (marker)
and mother (nonmarker) and the maternal status (number of children living at
home) of the mother.
Three points should be clarified. First, like most family simulation mod-
els, our model assumes stable demographic conditions. Although real demo-
graphic conditions are never perfectly stable, the model output does help us
to obtain a better understanding of the implication for family size and structure
of a given set of period demographic rates. By comparing the implications of
different rates, we can better understand how changes in demographic rates
affect family size and family structure. We can also obtain some summary
measures that are more instructive and interpretable than the rates derived from
cross-sectional data.
Second, the unit of analysis in the model is the individual. We link the
individual unit with the family unit by keeping track of individuals' marital,
maternal, and marker statuses. For instance, if we estimate that a certain
proportion of the members of the life-table population are married nonmarkers
living with three children and a widowed mother-in-law (or mother), we can
infer the proportion of six-person families containing three generations. How-
ever, we do not include unrelated persons and relatives other than spouse,
parents, and children. This simplification should not distort the overall trends
in family structure derived from the model because the proportions of other
relatives and unrelated persons in the family are usually too small to affect the
main results.
Third, ours is a one-sex, female-dominant model,9 which cannot account
for the families that contain no female members. Again, the proportion of
families of this kind, especially in a country like China in which very few
males live alone, is too small to affect the main results.
A description of the data for the family-status life-table simulation study
can be found in Appendix A. We prepared two major sets of inputs, namely,
observed or estimated demographic rates for 1981 and the average rates for
1950-70. The following sections discuss the output of the family simulation
model. 10
Family size
Cross-sectional observations of family size and structure at one point in time
generally differ from life-table output using period rates as input, because the
period cross-sectional observation consists of many cohorts' past experiences
while the period life-table output is "what would be" if a synthetic cohort
experienced the observed period rates. Nevertheless, as a check on the plau-
sibility of the model output, we compare information on family size, type,
and distribution from the 1981 and 1950-70 family-status life tables with
parallel measures from the 1982 census, whenever available.
The profiles of the 1950-70 and the 1981 family-status life tables do
not reflect the real family structure in those two periods, since the Chinese
population in 1950-70 and 1981 was not stable. Again, however, comparing
"what would be" if 1950-70 conditions remained stable with "what would
Zeng Yi 681
be" if 1981 conditions remained stable, we can gain a better understanding
of how demographic change may alter family structure.
1982 census and 1950-70 life-table
output compared
The observed average family size was 4.78 in 1973 and 4.43 in 1982. The
average family size from the 1950-70 life-table model is 4.90 and from the
1981 life-table model it is 4.37 (see Table 3). The fractions of one-, two-, and
five-person families from the 1950-70 family-status life-table models are rather
close to those found in the 1982 census observation. The proportions of three-
and four-person families from the 1950-70 family-status life table are signif-
icantly smaller, whereas the proportion of families of more than five persons
is significantly larger than in the 1982 census, as shown in panels (a) and (b)
of Figure 1. This model output seems plausible simply because fertility declined
greatly from 1970 to 1982. Young couples in 1982 had far fewer children than
in 1950-70. The proportions of three- and four-person families in the 1982
census significantly increased, since more young couples who were not living
with parents had only one or two children. The census observation on the
proportion of families with more than five members is significantly smaller
than that from the model output under 1950-70 conditions because young
couples who are living with parents also have fewer children.
The 1981 and 1950-70 life-table
output compared
The average family size under the 1981 rates is 4.37 as compared with 4.90
under the 1950-70 rates. Why is family size reduced by only 11 percent,
whereas fertility in 1981 was more than 50 percent below the 1950-70 level?
The explanation is as follows. First, decreased mortality has played a role.
The longer life span gives everyone, including children and older parents, a
higher probability of surviving, which compensates for part of the effects on
TABLE 3 Average Chinese family size
from the 1982 census and the model
output
Family type Source Average size
All types 1950-70 simulation 4.90
1981 simulation 4.37
1982 census 4.43
Nuclear 1950-70 simulation 4.41
1982 census NA
Three-generation 1950-70 simulation 6.17
1982 census NA
NA = not available.
FIGURE 1 Percent distributions of family sizes in China,
(a) according to the 1982 census, (b) under 1950-70
rates according to the model output, and (c) under 1981
rates according to the model output
(a) 1982 census
25
20 -
O15
41
04
0 1 2 3 4 S 5ze S 6 7 8+
(b) Model output based on 1950-70 rates
25
20 -
15
10
1 2 3 4 5 6 7 8+
(c) Model output bind on 1981 rates
20-
15 -
10
1 2 3 4 5 6 7 8+
Zeng Yi 683
average family size of the fertility decline. Second, as we will see later, the
proportion of three-generation families will increase with declining fertility if
the propensity for coresidence of parents and one of their married children
continues. The family size of three-generation families is generally larger than
that of nuclear families. The increased fraction of three-generation families
also compensates for part of the effect of fertility decline on average family
size. This argument is supported by comparison of the average size of nuclear
families and three-generation families. The average size of nuclear families is
3.19 under the 1981 rates, compared with 4.41 under the 1950-70 rates: a
decrease of 27.7 percent. The average size of three-generation families under
the 1981 rates is 5.60, compared with 6.17 under the 1950-70 rates: a decrease
of 9.2 percent.
What is the contribution of fertility decline alone and of mortality decline
alone to the changes in family size between the 1950-70 and 1981 simulations?
To answer these questions, we perform two other simulations. One uses 1981
fertility rates but assumes all other inputs to be the same as in the 1950-70
simulation. The average family size from this simulation is 3.97. In other
words, the fertility decline alone would reduce the average family size by 19
percent, in contrast to a reduction of 11 percent when fertility decline, mortality
decline, and changes of marital pattern are considered. Another simulation,
using 1981 mortality rates but assuming all other inputs to be the same as in
the 1950-70 simulation, shows that mortality decline alone would increase the
average family size by 12 percent.
Family types: Nuclear families versus
three-generation families
We first compare the 1982 census observations with the 1950-70 life-table
output. The proportions of nuclear families (including one-person families)
and three-generation families (including families containing more than three
generations) from the 1982 census are 81.1 percent and 18.9 percent, respec-
tively. Comparable figures from the 1950-70 family-status life-table output
are 72.2 percent and 27.8 percent. The fact that the 1982 census observation
on the proportion of nuclear families does not differ greatly from what it would
be if the 1950-70 rates were to prevail is not surprising. Given that no (or
very few) married siblings live together, the proportion of nuclear families
depends mainly on the number of adult children per older couple and on the
propensity for older couples and one of their married children to live together.
Up to 1982, the average number of adult children per older couple reflected
the high fertility before 1970, since the reduced numbers of children born after
1970 had not yet reached adulthood. Therefore, the availability of adult children
per middle-aged or elderly couple in 1982 was not yet reduced by declining
fertility, whereas reduced mortality may have allowed more children to survive
to adulthood and consequently to form more nuclear families. Hence, the 1982
census observation on the proportion of nuclear families was somewhat higher
than that of the 1950-70 family-status life table.
In a comparison between the 1981 and 1950-70 life-table outputs, the
proportion of nuclear families is 51.3 percent under the former versus 72.2
percent under the latter. Demographic changes (mainly the fertility decline)
reduced the proportion of nuclear families by about 20 percentage points. This
striking result may be understood from a simple hypothetical example (see
Figure 2). Assume a population of four elderly couples, each with three sons
FIGURE 2 A simple illustration
Mode A High fertility
Old couplesHFo Q 1
(first generation) L 1[
Young couples LQ & LQ & LQ 2 [Q X
(second generation)
(9]& IQj y5]Q&H[Q
Proportion of nuclear families: 10/13 = 77 percent
Mode B Low fertility; propensity for coresidence same as in Mode A
Old couples
(first generation) Q 1 b] F< I LQ
Young couples LQ LQ
(second generation)
Mode C Low fertility; propensity for coresidence reduced by 33 percent
Old couples ~F- i r x
(first generation) (3 01 LY ?1 L 9
Young couples [ 5 [951 [y ] [
(second generation)
Mode D Very low fertility; propensity for coresidence same as in Mode C
Old couples F F F
(first generation) LQ
Young couples -I
(second generation)
Zeng Yi 685
and three daughters. The children of these four couples get married to each
other and form 12 young couples. We assume that three of the four old couples
live with one young couple and the other young couples move out of the
parental home to set up independent families. Thus, there are 13 families in
the population, ten, or 77 percent, of which are nuclear families (Mode A in
Figure 2). However, if each of the four old couples has one son and one
daughter, there will be four young couples. Three young couples live with the
husband's (or wife's) parents, the other couple does not. The nuclear family
will account for 40 percent of families (Mode B in Figure 2). In Mode C, each
of the four couples has two children (one boy and one girl). Instead of three
out of four, we assume only two out of four (50 percent) young couples are
living with parents (in other words, the propensity for coresidence between
parents and married children is reduced by 33 percent). The proportion of
nuclear families would be 67 percent, which is still lower than in Mode A
(high fertility and higher propensity for coresidence). Clearly, the greatly
reduced fertility reduces the proportion of nuclear families. This argument does
not hold, however, when fertility falls below replacement level, as shown in
Mode D, where each old couple has only one child and thus there are only
two young couples in the system; if one out of two young couples (50 percent)
live with their parents, the proportion of nuclear families will be 80 percent;
the proportion of nuclear families is obviously increased as compared with the
situation represented by Mode C.
Actually, the fact that the proportion of nuclear families in the 1981
simulation (51.3 percent) was reduced by 20 percentage points as compared
with the 1950-70 simulation (72.2 percent) resulted from the combined effects
of fertility decline and mortality reduction. What is the contribution of fertility
decline alone? A simulation using 1981 fertility rates, but assuming all other
input to be the same as for 1950-70, gives the proportion of nuclear families
as 55.0 percent. Thus, the fertility decline between 1950-70 and 1981 alone
would reduce the proportion of nuclear families by 17.2 percentage points.
Family types by marital status
Table 4 shows the percent distribution of families by type as well as the marital
status of the mother for the implied stable populations in which the demographic
regimes of 1950-70 and 1981 prevail. Several observations can be drawn from
the table.
1 One-female families account for a small proportion in both implied
stable populations, since few women leave the parental home to set up an
independent family before marriage and few childless widowed or divorced
women live alone. Note that the proportion of one-female families in the 1981
simulation is somewhat higher than under the 1950-70 regime. This is con-
sistent with the greatly increased mean age at marriage in 1981 since a woman
has a longer period during which she is exposed to the risk of moving out of
the parental home before marriage if she marries later.
TABLE 4 Percent distribution of Chinese family type by
marital status according to the model output
One Husband- Widowed Divorced
Family type Year female wife mother mother Total
Nuclear 1950-70 6.9 62.1 2.7 0.5 72.2
1981 8.0 42.5 0.6 0.2 51.3
With lone 1950-70 - 11.8 0.8 0.1 12.7
grandmother 1981 13.5 0.8 0.1 14.4
With both 1950-70 - 14.6 0.3 0.1 15.0
grandparents 1981 33.7 0.5 0.2 34.4
Total 1950-70 6.9 88.5 3.8 0.7 100.0
1981 8.0 89.7 1.9 0.5 100.0
2 Nuclear families with a widowed or divorced mother constitute 0.8
percent of the total under the 1981 rates and 3.2 percent under the 1950-70
rates. Three-generation families with a widowed or divorced mother account
for 1.6 percent under the 1981 rates and 1.3 percent under the 1950-70 rates.
The proportion of families with at least one widowed or divorced woman
(either mother or grandmother) is 15.9 percent under the 1981 rates and 16.3
percent under the 1950-70 rates. Clearly, "incomplete families" have always
accounted for a small proportion of the total.
3 The great majority of families are "complete families" of husband
and wife (plus children and/or one or two grandparents). Families with at least
one couple (i.e., a middle-generation couple, or grandparents, or both) con-
stitute 90.4 percent under the 1981 rates and 89.4 percent under the 1950-70
rates. Families with a middle-generation couple (grandparent[s] either present
or absent) account for 89.7 percent under the 1981 rates and 88.5 percent
under the 1950-70 rates. This surprisingly small difference is consistent with
the very slight difference in the ratios of currently married to ever-married
women aged 20-45 in the 1982 census and in the Chinese farm population in
1930, as found by Coale (1984, p. 55). The finding suggests that, among
women under age 45, the higher incidence of widowhood in 1950-70, which
was the main cause of "incomplete families" for those women, must have
been offset by high rates of remarriage.
Note that the proportion of three-generation families with two couples
(parents and grandparents) under the 1981 demographic regime is 33.7 percent,
which is more than twice as high as that under the 1950-70 rates (14.6 percent).
This is primarily because the proportion of three-generation families is sig-
nificantly higher under the 1981 regime than under the 1950-70 regime, and
secondarily because the life span of grandparents is longer.
In brief, our findings show that the majority of Chinese families are
"complete families" of the husband-wife type, and that this feature has re-
mained constant.
Zeng Yi 687
How will the size and structure
of the Chinese family evolve?
Observers inside and outside China debate whether the Chinese family will
continue to nuclearize or whether the three-generation form will remain an
important family unit. In the popular media, two prospects are usually de-
scribed: Chinese family size will continue to fall and the proportion of nuclear
families will continue to increase. We think the second of these prospects may
not materialize. As shown by our simulations, the dramatic fertility decline
after 1970 reduces the chances of the coming generation of young adults to
move out of the parental home if the traditional preference of most parents to
live with one of their married children does not change dramatically.
This section further explores the evolution of the Chinese family by
analyzing the results of a number of additional simulations. An effort to discern
the likely evolution of family structure is made. For ease of presentation, we
hereafter abbreviate "the propensity for parents and one of their married chil-
dren to live together" as "the propensity for coresidence."
In light of strong and efficient population policies and family planning
programs as well as continuing modernization, fertility is likely to be lowered
further in China. Rapid economic development will reduce the propensity for
coresidence. These considerations underlie the choice of parameter values for
the following simulations. The parameter values are presented in Appendix B.
We summarize the main outcome of the simulations in Tables 5 and 6 in order
to see the effects of varying the total fertility rate (TFR), female life expectancy
at birth (60), and the proportion of parents who have married children but do
not live with any of them (n2).
Table 5 and Figure 3 show that the average family size decreases as
fertility is reduced. If n2 remains at the level estimated for 1981, average family
size decreases only slightly, from 4.37 to 4.36, when the TFR decreases from
2.63 to 2.21 and eo increases to 74 years. The reason for so little change in
family size in this case is that the effect of reduced fertility is compensated
TABLE 5 Average Chinese family size under various
levels of fertility, mortality, and the propensity for
coresidence according to the model output
eo = 69.3 eo = 74.0
TFR = 2.63 TFR = 2.21 TFR = 2.21 TFR = 1.80
n2 = .195 4.37(1981) 4.24 (9) 4.36 (1) 3.65 (2)
4.21 (1') 3.51 (2')
n, = .350 4.01 (3) 3.81 (10) 3.87 (4) 3.28 (5)
3.73 (4') 3.21 (5')
n, = .500 3.69 (6) 3.53 (11) 3.55 (7) 2.97 (8)
3.46 (7') 2.94 (8')
NOTE: Simulation codes are given in parentheses; for other parameters used, see Ap-
pendix B.
FIGURE 3 Average family size under various levels of
fertility, mortality, and the propensity for coresidence
according to the model output
5.0 1950-70 e (n-0
4.5 -
4.0-
< : ;0,:\ N ~~~~~~~~~~~~~~n2 =0. 195
>u 3.5-
ln2 0.350
.......... in and in increased by 2 years,
D increased to S
3.0 - n2 0.500
in, sn_, and D
same as in 1981
2.5 1 -__ __ I -I - -
6.0 5.0 3.0 2.5 2.0
TFR 2.63 TFR = 2.21 TFR 1.8
eo= 69.3 eo= 74.0 eo = 74.0
Total fertility rate
for by the decreased proportion of nuclear families (discussed below) and the
increasing life expectancy. If n2 and eo remain unchanged, a decrease in the
total fertility rate would result in a decrease in average family size. The average
family size decreases as n2 increases when eo and TFR remain unchanged.
As one would expect, Table 6 and Figure 4 show that, other factors
unchanged, the proportion of nuclear families increases with decreased pro-
pensity for coresidence. They also demonstrate some extremely interesting
trends in family structure with changing fertility. Let us look at the three rows
in Table 6. If n2 remains constant and fertility is above replacement level, the
proportion of nuclear families decreases with declining fertility; the explanation
was given earlier. However, a further reduction in the birth rate after fertility
reaches the replacement level will increase the proportion of nuclear families.
For example, when n2 = .195 and eo = 74, the proportion of nuclear families
increases by about 15 percentage points when total fertility rates decrease from
2.21 to 1.8. Why? The reason is fairly simple: when the number of members
Zeng Yi 689
TABLE 6 Percent of Chinese nuclear families (including
one-person families) under various levels of fertility,
mortality, and the propensity for coresidence according
to the model output
eo = 69.3 eo = 74.0
TFR = 2.63 TFR = 2.21 TFR = 2.21 TFR = 1.80
n2 = .195 51.3(1981) 46.4 (9) 40.9 (1) 56.1 (2)
47.2 (1') 62.0 (2')
n, = .350 64.0 (3) 63.2 (10) 60.0 (4) 68.8 (5)
65.6 (4') 72.9 (5')
n, = .500 74.7 (6) 72.5 (11) 70.5 (7) 79.7 (8)
74.6 (7') 82.0 (8')
NOTE: Simulation codes are given in parentheses; for other parameters used, see Ap-
pendix B.
of the children's generation is smaller than that of the parents' generation,
some parents have to live away from their married children even if they do
not wish to do so because of the shortage of children in the population. This
is, of course, based on the assumption-generally true in Chinese society-
that no married children live simultaneously with their parents and parents-
in-law.
The preceding results imply that if the current strict birth control policy
continues to depress Chinese fertility well below the replacement level, the
proportion of nuclear families will increase due to the rise in the number of
families in which old couples live alone. This tendency has already been
recognized intuitively by many people. But our simulations give a quantitative
account. On the one hand, we should be aware of the extreme importance of
population policy in China; on the other hand, we should also be aware of its
family-level consequences and attempt to monitor the process and to introduce
necessary socioeconomic measures in order to avoid some of the unfavorable
consequences of an older age structure in the future.
Note that simulations 1-11 (see Appendix B) assume nuptiality and
divorce patterns to be constant at 1981 levels. This is obviously not the case
in reality. But in order to identify the effects of foreseeable further declines
in fertility and mortality and in the propensity for coresidence, we need to
control the variables of nuptiality and divorce.
Now let us consider the effects of changing nuptiality and divorce on
family size and structure. We modify the inputs of simulations 1, 2, 4, 5, 7,
and 8 by increasing the mean age at marriage (m) and at childbearing (mi) by
two years and increasing the divorce rate (D) from 1 to 5 per thousand married
couples, while the other parameters remain the same as the ones previously
used. We call these modified simulations 1', 2', 4', 5', 7', and 8', and their
main outcomes are shown, compared with corresponding constant-nuptiality
simulations, in Tables 5 and 6 and plotted as dotted lines in Figures 3 and 4.
FIGURE 4 Percent of nuclear families (including one-
person families) under various levels of fertility,
mortality, and the propensity for coresidence according
to the model output
90
80 _ ... , n2 = 0.500
tl 701950-70 ? 51?
j70 *2 11 n2 =0.350
60 -
O 60 N N,. 1n2 0.-195
1981
50 - / in and ii- increased by 2 years,
D increased to 5
in, im, and D remain
same as in 1981
4CI I - - - -~ ~ - -I
406.0 5.0 3.0 / 2.5 j 2.0
TFR = 2.63 TFR 2.21 TFR 1.8
eo = 69.3 eo = 74.0 eo = 74.0
Total fertility rate
The results show that the average family size declines by 1-4 percent and the
proportion of nuclear families increases by 2.2-6.3 percentage points as m and
mi increase by two years and D increases to 5 per thousand.
Mean age at marriage has fluctuated around 22.8 since 1977 (22.6 in
1977, 23.1 in 1979, 22.7 in 1982). It is unlikely that it will increase rapidly
in the foreseeable future. The divorce level is also expected to increase only
gradually. Therefore, the assumptions concerning possible increases in m and
D in simulations 1', 2', 4', 5', 7', and 8' (m and mi increased by two years,
D increased fivefold) are rather extreme. Even so, their effects on family size
and structure are not major. We can thus assume that the most important
demographic factors affecting Chinese family size and structure will be the
decreasing TFR and decreasing propensity for coresidence, rather than changes
in m, mi, and D.
As we have already emphasized, the simulations are not predictions.
Nevertheless, they do show how Chinese family size and structure will change
Zeng Yi 691
with changes in fertility, nuptiality, divorce, mortality, and the propensity for
coresidence. A number of observations can be made:
1 Taking into consideration the compensating effect of mortality decline,
Chinese family size will steadily decrease with the expected declines in fertility
and in the propensity for coresidence and the gradual increases in age at
marriage and the divorce level.
2 If the propensity for coresidence does not decrease dramatically, the
proportion of nuclear families will decline when the children born in the 1970s
reach the family-formation stage, since young adults with fewer siblings will
have a smaller chance of moving out of the parental home to form an inde-
pendent nuclear family. However, if fertility falls below replacement level, a
further reduction in the total fertility rate will raise the proportion of nuclear
families. On the other hand, the gradually increasing mean age at marriage
and the rise in the level of divorce may partly compensate for the effects of
the dramatic fertility decline on the decreasing proportion of nuclear families.
3 If the propensity for coresidence decreases rapidly, the effects of
declining fertility (as long as it remains above replacement level) on the de-
creasing proportion of nuclear families will be largely or even wholly offset.
4 Changes in fertility and changes in the propensity for coresidence can
occur simultaneously. But the fertility change, unlike the changing propensity
for coresidence, cannot affect the proportion of nuclear families immediately
(although it does affect family size immediately), because the number of chil-
dren per couple mainly influences the proportion of nuclear families by reducing
or increasing the children's chances of leaving the parental home when they
grow up. This lagged effect of changing fertility should be taken into account
in policy considerations.
Socioeconomic factors affecting
coresidence
With an efficient family planning network providing improved contraceptive
services and information, and with people's changing attitudes toward the
desired number of children (more and more couples, especially in urban areas
and some economically advanced rural areas, wish to have fewer but better
qualified children), we may expect fertility to continue to fall. We may
also expect life expectancy, mean age at marriage, and the divorce rate to
gradually increase with the process of modernization. Rapid socioeconomic
development will gradually reduce the propensity for coresidence. Therefore,
observation 1 noted above is likely to materialize: namely, Chinese family size
will steadily decrease with the expected decline in fertility and in the propen-
sity for coresidence.
Is observation 3 also likely to materialize in the near future? In other
words, will the propensity for coresidence decrease dramatically so as to largely
or totally compensate for the effect of the remarkable fertility decline? Our
answer is negative. To illustrate why we suspect this situation is unlikely to
occur in the near future, we will list and compare two categories of socioec-
onomic factors that operate in opposite directions.
The following will reduce the propensity for coresidence:
1 Rapid economic development and improvement in education will in-
crease people's demand for cultural modernization. The growing availability
of modern means of communication (e.g., the telephone) and transportation
(e.g., the automobile) will facilitate contact between family members not living
in the same house or in the same area. With the adoption of a modern lifestyle,
the gap between the young generation and the old in preferences concerning
family structure will gradually widen.
2 Severe housing constraints in urban areas will be gradually relieved
as a result of widespread housing construction. This will allow more young
people to live away from their parents.
3 With the relaxation of restrictive policies on migration, rural-to-urban
and urban-to-urban migration will increase. In fact, the Chinese government
has permitted or even encouraged peasants to move to towns and cities to
invest in shops and other small-business enterprises. These migrants are usually
young people. Their elderly parents either stay behind or join the migrants
after they have settled in their new location. Separation (maybe temporary) of
some family members is expected if migration gathers force.
4 With the process of urbanization, the proportion of old people in urban
areas who rely on pensions or social security will increase. This will reduce
the necessity for these parents to live with one of their married children. Of
course, having a pension does not necessarily lead to living apart from children
in China. The pension system was introduced many years ago for employees
working in state or collective enterprises in urban areas, yet a considerable
majority of elderly urban parents (most of them entitled to a pension) live with
a married child.
Other factors will act to sustain the propensity for coresidence:
1 The collective mode of production and income distribution reduced
the profitability of maintaining a large multigeneration family. However, from
the early 1980s, under the responsibility system, peasant families have been
replacing the production team as the unit of production. According to a recent
survey of 2,036 peasant families in Sichuan Province, the generally better-off,
specialized households have a larger average family size and a higher proportion
of three- (or more-than-three-) generation families (see Table 7).
In urban areas, the number of privately owned small shops, restaurants,
and hotels has grown rapidly in recent years. At the end of 1985 the number
of privately owned industrial and commercial enterprises holding an official
license totaled 10.2 million (People's Daily, 17 June 1986), an increase of
25.5 percent over 1984. The number of employees in privately owned industrial
and commercial enterprises was 17.6 million, an increase of 34.8 percent over
1984. The number of employees in such enterprises exceeds the number in
the state-owned commercial system (not including industry) by more than 3
million (People's Daily, 23 February 1986). The Chinese government has
Zeng Yi 693
TABLE 7 Comparison of Chinese family size and
structure between better-off specialized families and
ordinary families in rural areas
Percent of multigeneration
families
More than
Average 3 gener- 3 gener-
family size ations ations Total
Better-off, specialized
households 6.07 35.1 2.5 37.6
Ordinary households 4.52 22.2 1.0 23.2
SOURCE: Survey of 2036 peasant households in Sichuan Province (Zhao Xishun, 1985,
pp. 26-27).
obviously allocated an increasingly large role to the private sector in the coun-
try's modernization process.
Private enterprises are usually owned and run by members of a single
family or close relatives. Given the very low levels of automation in agricultural
and private industrial and commercial production, it appears profitable for
owners of these enterprises to maintain large families to ensure an inexpensive
and committed labor supply.
2 It is unlikely that the pension system will spread rapidly throughout
rural China in the near future. Most peasants will still depend economically
on their children when they are old.
3 Although no official government policy promotes three-generation
families, clear evidence exists that the Chinese government favors maintaining
the three-generation family for the sake of upholding a Chinese cultural tra-
dition, and because this family type enables the state to spend less on old-age
care. Both the current Marriage Law and the Chinese Constitution state ex-
plicitly that children have full responsibility for caring for their parents in old
age. A sizable majority of the "model families" selected by the local com-
munities each year for special honors are three-generation families (Liu Yin,
1985).
4 As discussed earlier, the expected rural-to-urban and urban-to-urban
migration may result in the division of some families. On the other hand, it
may result in an increase in the number of extended families, both because
some young migrants may live temporarily with relatives and because rural-
to-urban migration may outpace the ability of urban housing construction to
provide living quarters.
5 The ethical tradition of respect and care for the elderly will continue
to play an important role. The sociologist Fei Xiaotong has described the
Chinese family in terms of a "feedback model," in which each generation
fosters the generation succeeding it and in turn receives financial support in
old age from the children. In contrast, Fei calls the Western model a "continued
linear model," in which the "feedback" of parental old-age support by children
is absent (Fei, 1983, p. 7). Consequently, the typical family pattern in Western
society is the nuclear family consisting of husband, wife, and unmarried chil-
dren. Chinese family structure, on the other hand, is more variable, since
married children do not necessarily leave the parental home. Preferences are
likely to continue to reflect this tradition: most elderly Chinese will continue
to dislike living alone and prefer to share a warm family environment with a
married child and grandchildren (Lin and Bi, 1984).
This argument is supported by the persistence of traditional extended
families in Taiwan Province of China. It is well known that Taiwan is now a
modem industrial society: most families own motorcycles or cars; ubiquitous
mass media (e.g., television) links almost everyone to local and international
communication networks; the education system takes more than 95 percent of
all children to the ninth grade. Such changes might have been expected to
hasten Westernization and nuclearization of the family. According to 1973 and
1980 surveys, however, while there has been a significant decline in the prev-
alence of extended families, about 40 percent of all ever-married women of
childbearing age were still living in such families, and a majority of elderly
parents (80 percent in 1973, 76 percent in 1980) were living with a married
son. A majority of husbands' parents were living with a married son even in
the most modern strata-72 percent among the best-educated and 68-71 percent
in the big cities, as compared with 87-89 percent among those living on farms
(Freedman et al., 1982, p. 405). Rapid economic development has been ac-
companied by changes in coresidential patterns; however, these changes do
not replicate the Western nuclear model. It is always possible that coresidence
patterns are merely lagging, changing more slowly than other aspects of society.
It is also worthwhile to point out that, while the total fertility rate in Taiwan
fell from 5.61 in 1961 to 2.67 in 1979, the effect of this decline on the
proportion of nuclear families was not yet evident in the 1980 survey, since
the smaller cohorts of children had not yet reached the family-formation stage.
The actual changes in Chinese family structure will be determined by
the balance of these two sets of opposing factors. We suspect that the proportion
of nuclear families will decrease in the near future, given the likelihood that
the propensity for coresidence will not fall dramatically. At a later stage, further
reductions in fertility seem likely to intensify the effect of a gradually decreasing
propensity for coresidence in increasing the proportion of nuclear families.
Conclusion
The output of a general family simulation model, using as input data demo-
graphic rates in 1950-70 and in 198 1, as well as various hypothesized future
demographic regimes, demonstrates how changing demographic factors affect
Chinese family size and structure. An interesting finding of this exercise is
that when young people born after the large fertility decline of the 1970s reach
the age of family formation, with a given proportion of parents and one married
child wishing to live together, the proportion of young couples forming in-
Zeng Yi 695
dependent nuclear families will be smaller, since there will be much smaller
numbers of siblings. If, however, fertility continues to fall after reaching the
replacement level, a further reduction in the birth rate will raise the proportion
of nuclear families. In that case, some parents will find it impossible to live
with their married children even if they wish to do so because of the shortage
of children. Thus, an increase in nuclear families among the old population
would occur.
The simulations tell us that the average family size would be reduced
by about 11 percent and the proportion of nuclear families would be brought
down by about 20 percentage points below 1981 fertility, mortality, and mar-
riage rates as compared with what would be found under the 1950-70 rates.
Our findings also show that the majority of Chinese families are "complete
families," with both husband and wife present, and this feature will remain
stable in the near future. The results also show that the most important de-
mographic factors affecting Chinese family size and structure will be fertility
and the propensity for coresidence, while the effects of changing mortality and
nuptiality patterns will be relatively modest.
Taking into consideration the model output as well as the two counter-
balancing sets of socioeconomic factors that determine the propensity for co-
residence, we suspect that the average family size in China will continue to
fall and the proportion of nuclear families will first decrease and then increase
again in the future. To what extent and how quickly the Chinese family will
change are questions that deserve further study by both sociologists and
demographers.
Appendix A: Data and
estimations
The family-status life table can be constructed other ages with the exception of ages zero to
using the following data; (1) age-specific death one and one to five.
rates (or age-marital status-specific death rates,
if possible); (2) age-specific occurrence/expo-
Death rates
sure rates of first marriage, widowhood, di-
vorce and remarriage; (3) age-parity-specific The death rates are derived from China's 1982
occurrence/exposure rates of birth. If there is census data (Jiang et al., 1984). The age in-
no parity birth control in the study population, tervals in published death rates are five-year
the age-specific rates, rather than the age-par- intervals, except for ages zero to one and one
ity-specific occurrence/exposure rates of birth to five. We obtained the single-age-specific
can be used. (4) Proportion leaving parental death rates from ages 15 to 49 by linear
home before marriage, and its age schedule. interpolation.
(5) Proportion of parents who have married Death rates at other ages are taken directly
children but do not live with any of them and from Jiang's life table. For the period 1950-
the schedule of children leaving the parental 70, we adapt two intercensus life tables esti-
home. All the needed age-specific rates and mated by Coale (1984, p. 67). We give the
probabilities are single-age-specific between proper weight to the 1953-64 life table and
the lowest and the highest ages at childbearing the 1964-82 life table, respectively, in order
(15 and 49, say) and five-year-age-specific for to obtain a set of weighted average death rates
for the years 1950-70. The single-age-specific at first marriage for grooms and brides and the
death rates between ages 15 and 49 are derived mortality rates for married males. On the basis
by linear interpolation. The marital status- of China's 1982 census data, it was estimated
specific death rates are not available for this that the difference between age at first marriage
study; we therefore assume that the death rates for males and females was 2.69 years (Li
are the same for all marital states. Since di- Rongshi, 1985, p. 28). For convenience, we
vorce rates are very low in China and widows approximate the difference in age at first mar-
and divorcees remarry quickly if they are not riage in 1981 and 1950-70 as 2.5 years instead
too old, we expect that assuming the same of 2.69 years. If we assume that mortality is
death rate for all marital states does not create equal in all marital states, the male death rate
significant errors. d(x+ 2.5) [which is estimated by the average
of d(x) and d(x + 5) or the average of d(x + 2)
and d(x + 3)] can be said to represent the fe-
Occurrence/exposure rates
male widowhood rate.
for first marriage
China's 1982 one-per-thousand fertility survey
published female single-age-specific reduced
events of first marriage (some authors call it a Divorce rates
frequency distribution or rates of the second
Following a similar methodology for con-
kind) for calendar years 1940-81.
structing model life tables and using the age-
What we need are the occurrence/exposure
specific divorce rates of 48 countries listed in
rates, which are defined as the number of first
the United Nations Demographic Yearbook for
marriages divided by the number of person-
1968, Krishnan and Kayani (1976) constructed
years lived in the never-married state in the
model divorce tables with 13 levels. The di-
age interval.
vorce level is defined as the number of divorces
The number of person-years lived in the
per thousand married couples in a given year.
never-married state in an age interval can be
From Krishnan and Kayani's model di-
approximated as the number of never-married
vorce table, we derived age-specific occur-
women of that age in the middle of the year.
rence/exposure divorce rates in 1950-70 and
Since the denominators of reduced first mar-
in 1981 for Chinese females, using published
riage and the denominators of the proportion
total numbers of divorces and total numbers
never-married are the same when mortality and
of married women in the population as the
external migration are disregarded, as is often
inputs.
done for retrospective fertility survey data, the
occurrence/exposure rate of first marriage is
equal to the age-specific reduced first marriage
rate divided by the proportion never-married
Remarriage rates
in the middle of the year. We took the age-
In China, published age-specific remarriage
specific proportion never-married to be one mi-
rates by previous marital status have not yet
nus Coale's age-specific period estimate of the
become available. We do have data on the
proportion of women ever-married (Coale,
proportion widowed and divorced from the
1984, Table A.4). The estimates of occur-
1982 census; on the divorce levels in the past
rence/exposure rates for 1981 and the average
30 years estimated by the published number
occurrence/exposure rates for 1950-70 are
of divorces between 1950 and 1981 (Li Ning,
plausible except after age 30, because so few
1985); on life expectancy at birth over the past
women are unmarried and so few marriages
50 years using indirect estimation techniques
occur after this age. To reduce the noise, we
based on 1982 census data (Brass, 1984); and
estimate the occurrence/exposure rates over
on first-marriage rates for 1950-81, or for co-
age 30 by extrapolation.
horts whose members were 15-49 years old in
1982. We propose an indirect estimation pro-
Widowhood rates
cedure to derive remarriage rates by previous
Female widowhood rates can be estimated marital status using the available data (Zeng,
from the differences between the average age 1986a, Appendix 2).
Zeng Yi 697
Age-parity-specific occurrence/ birth except for first birth. Since the estimated
proportion of women with parity zero includes

exposure birth rates
never-married women, we divide the rates of
China's one-per-thousand fertility survey pub-
first birth by the proportion ever-married to
lished singe-age-parity-specific reduced events
estimate the occurrence/exposure rates of first
of birth, which are defined as the ratio of the
birth for ever-married women. We do not have
number of births classified by parity and the
the same problem for birth rates of orders
age of the mother to the total number of women
higher than one, since the numbers of never-
of the corresponding age. The period parity
married women with parity one or higher are
progression ratios based on a synthetic cohort
negligible.
life table using the same survey data were pub-
Step 2: Improving the estimated rates The

lished recently (Feeney et al., 1985 and Ma
age-parity-specific birth rates in 1981 derived

Yingtong and Wang Yanzu, 1985). However,
what we need is the age-parity-specific occur- from step 1 need to be further improved since
rence/exposure birth rates, which take the ex- they were not estimated from the same survey
posure into account. The numerators of the data. We need to adjust the estimates in step
1 using the published period parity progression

occurrence/exposure rates are the same as
ratios (Feeney et al., 1985). Applying the es-

those of the reduced birth rates, namely, the
number of births classified by parity (p) and timated age-parity-specific occurrence/expo-
sure rates of birth and first marriage to a

age of the mother. The denominators of the
occurrence/exposure rates differ from those of synthetic cohort, we obtain a set of parity pro-
gression ratios. Comparing these ratios with

the reduced rates; they are not the total number
the published period parity progression ratios,

of women of a given age, but the number of
which are also based on a synthetic cohort life

person-years lived in parity p - 1 by women
table using age-parity-specific occurrence/ex-

of the corresponding age. These required age-
parity-specific occurrence/exposure birth rates posure birth rates from the survey (but the oc-
are not yet available for this study. We there- currence/exposure birth rates used have not
been published), we proportionally reduce (if

fore estimate 1981 age-parity-specific oc-
the estimated ratio is higher than the published

currence/exposure birth rates by a two-step
one) or increase (if the estimated ratio is lower

procedure using available data.
than the published one) the age-parity-specific
Step 1: Estimating the preliminary occurrence/exposure rates of birth estimated in
age-parity-specific occurrencelexposure birth the first step. This procedure is repeated until
rates Multiplying the age-specific propor- there are no further improvements on the es-
timated rates. We finally obtain a set of age-

tional distribution of the number of surviving
parity-specific occurrence/exposure birth rates

children of women from the one-per-thousand
fertility survey by age-specific ratios of the that produce, in combination with nuptiality
number of women classified by number of chil- rates, the parity progression ratios that are ex-
dren ever born to the number of women clas- actly the same as the published period parity
sified by number of children surviving from progression ratios. Although the timing of
the 1982 census, we obtain the single-age-spe- parity-specific fertility may only be an ap-
cific proportional distribution of parity at the proximation due to the indirect estimations
procedure, the estimated total fertility level and

survey time (mid-1982). Assuming that the
age-parity-specific reduced events of birth implied final parity distribution are fully con-
from mid-1981 to mid-1982 are the same as sistent with the published ones.
those throughout 1981, and combining them Using the single-age-parity-specific occur-
with the age-parity-specific proportional dis- rence/exposure rates of birth for 1981 as a basic
tribution in mid-1982, we estimated the age- schedule, we shift all the curves of birth rates
parity-specific proportional distribution in with different birth order to the left by three
years since the mean age at first marriage in

mid-1981. Dividing the observed single-age-
parity-specific reduced events of birth by the 1950-70 is about three years lower than in
estimated single-age-specific parity distribu- 1981. Then we follow the procedure described
tion, we obtained the estimates of single-age- in step 2 and estimate a set of single-age-parity-
specific occurrence/exposure rates of birth.

parity-specific occurrence/exposure rates of
These rates produce precisely the mean age of family size and structure of the remarkable
the first-marriage schedule and period average changes in such demographic variables as nup-
parity progression ratios for the period 1950- tiality, fertility, and mortality from 1950-70
70, which are approximated from the 1982 to 1981.
According to a survey among 709 elderly

census observation of parity distribution of
women aged 50-59 since the fertility experi- persons (males over 60; females over 55) in
Lanchou, a city with about 2.3 million inhab-

ence of this age group represents the fertility
itants, 36.5 percent of the respondents do not

level in 1950-70.
want to live with their married children (Lin
and Bi, 1984). We therefore assume that n2 is
Sex ratio at birth
36.5 percent among the urban population.
In rural areas, about 97.5 percent of the

The one-per-thousand fertility survey of China
elderly are economically dependent on their

held in 1982 reported a total of 818,876 live
children (Yun, 1985, p. 106). This clearly in-

births by sex from 1930 to 1981. There were
dicates that a great majority of old parents live

425,904 male births and 392,972 female
with a married child. We assume that in rural

births. The overall mean sex ratio at birth was
areas the old parents who have married chil-

108.4 males per 100 females (Li and Tuan,
dren but do not live with any of them accounted

1985).
for 15 percent.
Using the proportions of the rural and ur-
Proportion ultimately leaving the
ban population in 1981 and 1950-70 (State
parental home
Statistical Bureau, 1984, p. 81), we estimated
the weighted average of n, for the whole coun-

The estimator of the proportion ultimately
try in 1981 and 1950-70, respectively.

leaving the parental home can be found in Zeng
For both periods, we adapted a timing

(1986b). The required data on the total fertility
schedule of leaving the parental home by mar-

rate, mean age at childbearing, proportion of
riage duration, derived from the survey report

baby girls among all births, and the probability
of China's in-depth fertility survey (State Sta-

of surviving from birth to the mean age at
tistical Bureau, 1986).

childbearing can be directly obtained from the
We assume that 5 percent of all cohort

observed data described in the previous sec-
members who survive at least up to age 18

tions of Appendix A. The proportion of women
leave the parental home before marriage but

who do not give birth to a live child during
after age 18, for both 1981 and 1950-70. A

their lifetime has been derived from census
time schedule is also assumed.

data.
It should be noted that the arbitrarily as-

For reasons discussed in the first section,
sumed proportion and time schedule of leaving

we assume that the proportion of parents who
the parental home before marriage, but after

have married children but do not live with any
age 18, may not be true. But it is unlikely that

of them, n2, remains stable from the mid-1950s
this creates significant errors, because the pro-

to 1981.
portion is so small and because the period

Another important reason for assuming the
between age 18 and marriage is not long.

stability of n2 is to ascertain the pure effect on
m m - t - \C fl 00o o m -
aE ? N N o ~\O mtCI T? t Clm
- C- o o o 00
lo . ('l Cl o C-Ns m. -
o} , - C- o v 00 ~ 0k) 0 -
Cl Cl, N. \O O 0 0 N
ON\ -- Cl Cl oo
ON~~~~~O
o o o csE ? v~~~~k) W o ) 00 W-)
00 oo N ?O 0 (t o 00
00 Cl \O-Cfl lf - 4
* * 00) Cl * \Co ~ / 't C) o 0
00 O ?? N 0 0 0 0 o ?00
000 0~~~~~~~0
_~~~~ 0 oC~tI ?? 6 Cl ?N,
00 0 ON C 000 0 00
00 - C- 0 0 0 0 0) NN N o Un N o _ I00
s~~~\ Cl :N0 _ t n0 -nX n
0 O~ ~ ~ ~ C- nCl~C
1r~~ C-C- - I 666 ~6 6 C
E? * 00 C o o _ -, oo
C-C- Cl *~~~~~\0 a- ~ 6 Cl
rn rn rn 00000 0
L0 V00 CVO0 0 0 00 - E
* E 00o El * yC-C/N LN -,, ' t s i, o c,k
00 - C- 0 000~~~rn 0 0 00
C2
CU 00 2l 666 6
E 00 0 ON L/NOL/N 00 at a EN E00 -
Notes
This article is based on a part of the author's and Culter, 1983; Brass, 1983; Bongaarts,
doctoral dissertation presented to the Inter- 1983; Smith, 1983; Watkins et al., 1984; Men-
university Programme in Demography, Brus- ken, 1985; Goldman, 1986.
sels Free University. The research was carried
6 Bongaarts's original model and my ex-
out while he was a visiting research fellow at
tension of it used here are life-table models for
the Netherlands Interuniversity Demographic
family simulation study. Therefore, the terms
Institute. The author thanks Frans Willekens
family simulation model and family-status life
for encouragement and supervision of research
table are used interchangeably in this article.
for his Ph.D. thesis and Ron Lesthaeghe for
Readers interested in the technical details of
invaluable advice. Comments provided by
the construction and estimation of the models
John Bongaarts, Nico Keilman, and Stanley
are referred to Bongaarts (1983) and Zeng
Wijewickrema and computational assistance
(1986a and b).
provided by Yves de Roo are gratefully
7 How can the individual unit be made to

acknowledged.
represent the family? Brass generalizes the idea
1 More detailed discussion of the trends
of the headship of the household into the con-
and age patterns of marriage and fertility in
cept of "marker" of the family. The marker
China may be found, for example, in Coale
is the reference person of a family. Brass ar-
(1984) and Zeng et al. (1985).
gues that on the whole it would be better to
2 The average family size obtained from
take the senior female, whenever possible, as
Chinese censuses and surveys refers to the jar
the marker, because females marry at an earlier
unit, that is, the domestic household, exclud-
age and live longer than men (Brass, 1983).
ing collective households. The concepts of
In my model a girl is not born as a marker.
"domestic household" and "collective house-
She becomes one if and when she leaves her
hold" used in Chinese censuses and surveys
parents to set up an independent family. If she
are analogous to the concepts of "household"
lives with her surviving mother or mother-in-
and "group quarter" used in US censuses. The
law, she is not a marker. If a marker dies, her
word "family" as used throughout this article
position as marker will be given to a daughter-
is equivalent to the jar unit-in other words,
in-law who lives at home or, if there is none,
to a group of coresidential persons who are
to an adult daughter (over 18 years old) living
related through marriage, blood, or adoption.
at home. If she has neither a daughter-in-law
3 Surveys conducted in April 1985 by the

nor an adult daughter living at home, any re-
State Statistical Bureau of China, with tech-

maining family members will be attached to
nical assistance by the Research Center of the

another female marker, through remarriage of
International Statistical Institute.

the father or by joining the related family of
the orphans.
4 A recent family survey in the five largest
cities (Beijing, Shanghai, Tianjin, Nanjin, and
8 In my model three-generation family re-
Chengdu) shows that the proportion of three-
fers to a stem family with grandparent(s), par-
generation families has remained stable-20.4
ent(s), and unmarried children. No married
percent in 1954-57 and 20.2 percent in 1977-
siblings are assumed to live together.
82-while the proportion of joint families with
9 Although a one-sex, male-dominant
married brothers living together fell signifi-
model can readily be developed similar to our
cantly-from 7.9 percent in 1954-57 to 3. 1
model for females, we did not do so due to
percent in 1977-82. The proportion of nuclear
the lack of fertility and nuptiality data for
families increased from 64.2 percent in 1954-
males.
57 to 72.4 percent in 1977-82. Five Cities
Family Research Project Group, 1985, Table
10 A family-status life table can be ap-
102, p. 484, and Table 110, p. 508.
proached from two perspectives. The first
views the family-status life table as a descrip-

5 See for example, Coale, 1965; Burch,
tion of the life course of members of a cohort.

1970; Ryder, 1974; Goodman et al., 1974;
The second views it as a description of a stable

Wachter et al., 1978; LeBras, 1982; Martin
Zeng Yi 701
population in which certain fertility, mortality, the same model and data as used here, such
and nuptiality regimes prevail. We may call it as the proportion of a typical lifetime spent in
a stable population analysis of the family. Due different marital states, parity states, states of
to space limitations, this article presents only being a daughter, being a mother, and being
the findings of the stable population perspec- responsible to both old parents (say, over age
tive, that is, the simulation study of the family. 65) and young children (say, under 18), can
The findings of the life course perspective from be found in Zeng (1986a and c).
References
Bongaarts, J. 1983. "The projection of family composition over the life course with family
status life tables," forthcoming in Bongaarts et al. (eds.), 1987.
, and S. Greenhalgh. 1985. "An alternative to the one-child policy in China," Population
and Development Review 1 1, no. 4: 585-617.
, T. Burch, and K. W. Wachter (eds.). 1987. Family Demography: Methods and Ap-
plications. Oxford: Oxford University Press. Forthcoming.
Brass, W. 1983. "The formal demography of the family: An overview of the proximate de-
terminants," in The Family, Occasional Paper 31. London: Office of Population Censuses
and Surveys.
. 1984. "Mortality in China over the past fifty years: Indirect estimates from the 1982
census," paper prepared for the international seminar on China's 1982 population census,
Beijing, 26-31 March 1984.
Buck, J. Lossing (ed.). 1937. Land Utilization in China, 3 vols. Nanking: University of Nanking.
Distributed in the United States by the University of Chicago.
Burch, T. K. 1970. "Some demographic determinants of average household size: An analytic
approach," Demography 7, no. 1: 61-69.
Coale, A. J. 1965. "Appendix: Estimates of average size of household," in Aspects of the
Analysis of Family Structure, ed. A. J. Coale, Lloyd A. Fallers, Marion J. Levy, Jr.,
David M. Schneider et al. Princeton, N.J.: Princeton University Press, pp. 64-69.
. 1984. Rapid Population Change in China, 1952-1982. Committee on Population and
Demography, National Research Council, Report No. 27. Washington, D.C.: National
Academy Press.
China Population Information Centre (ed.). 1984. Analysis on China's National One-per-
Thousand-Population Fertility Sampling Survey. Beijing.
Encyclopedic Yearbook of China. 1982. Beijing: China Great Encyclopedia Publishing House.
Feeney, G., Yu Jingyuan, and Chi-hsien Tuan. 1985. "Parity progression measures of fertility
in China," paper presented at the international symposium on China's one-per-thousand
national fertility sample survey, Beijing, 14-18 October 1985.
Fei Xiaotong. 1982. "Changes in Chinese family structure," China Reconstruction (July):
23-26.
. 1983. "Problem of providing for the senile in the changing family structure," Journal
of Peking University (philosophy and social sciences), no. 3: 6-15 (in Chinese).
Five Cities Family Research Project Group. 1985. Family Types in Five Cities. Shandong
People's Press (in Chinese).
Freedman, R., Baron Moots, Te-Hsiung Sun, and Mary Beth Weinberger. 1978. "Household
composition and extended kinship in Taiwan," Population Studies 32, no. 1: 65-80.
, Ming-Cheng Chang, and Te-Hsiung Sun. 1982. "Household composition, extended
kinship and reproduction in Taiwan: 1973-1980," Population Studies 36, no. 3: 395-
411.
Goldman, N. 1986. "Effects of mortality levels on kinship," in Consequences of Mortality
Trends and Differentials, Population Studies No. 95. United Nations, Department of
International Economic and Social Affairs, ST/ESA/SER.A/95. New York, pp. 79-87.
Goodman, L. A., N. Keyfitz, and T. W. Pullum. 1974. "Family formation and the frequency
of various kinship relationships," Theoretical Population Biology 5, no. 1: 1-27.
Hu Huanyong. 1982. A Brief Survey of China's Population Geography. Shanghai Foreign
Language Education Press.
Jiang Zheng-huna, Zhang Wei-min, and Zhu Li-wei. 1984. "The preliminary study to the life
expectancy at birth for China's population," paper prepared for the international seminar
on China's 1982 population census, Beijing, 26-31 March 1984.
Keilman, N., A. Kuijsten, and A. Vossen (eds.). 1987. Modelling Household Formation and
Dissolution. Oxford: Oxford University Press, in press.
Keyfitz, N. 1985. Applied Mathematical Demography, 2nd ed. New York: Springer-Verlag.
Krishnan, P., and A. K. Kayani. 1976. "Model divorce tables," Genus 32, no. 1-2: 109-
126.
Lavely, R., and Li Bohua. 1985. "A tentative survey on family structure in Liaoning, Hebei
and Fujian provinces," paper presented at the international symposium on China's one-
per-thousand national fertility sample survey, Beijing, 14-18 October 1985.
LeBras, H. 1982. "Evolution des liens de famille au cours de l'existence," in INED, Les dges
de la vie. Presses Universitaire de France.
Lesthaeghe, R. 1973. "The feasibility of controlling population growth through nuptiality and
nuptiality policies," proceedings of the international population conference of the Inter-
national Union for the Scientific Study of Population. Liege. 9.1 .b, pp. 319-341.
Li Bohua and Chi-hsien Tuan. 1985. "The Chinese sex ratio at birth (1930-1981)," paper
presented at the international symposium on China's one-per-thousand national fertility
sample survey, Beijing, 14-18 October 1985.
Li Ning. 1985. "How does China deal with divorce?" Beijing Review No. 5 (4 February 1985):
18-21.
Li Rongshi. 1985. "A tentative analysis of the age at first marriage in China," Population
Research, No. 1: 28-32 (in Chinese).
Lin Xiao and Bi Ke. 1984. "A sample survey of 709 old urban citizens." Social Science, no.
6: 61-70 (in Chinese).
Liu Yin. 1984. "A tentative analysis on three-generation families in China-Survey report of
model families in Beijing," Sociology Bulletin (March): 45-50 (in Chinese).
. 1985. "Development and changes of urban families in China: A tentative analysis on
the family survey of Beijing, Tianjin, Shanghai, Nankin and Chengdu cities," paper
presented at the second international conference on modernization and Chinese culture,
Hong Kong, November 1985 (in Chinese).
Martin, L. G., and S. Culter. 1983. "Mortality decline and Japanese family structure," Pop-
ulation and Development Review 9, no. 4: 633-649.
Ma Xia. 1984. "An analysis of the size of family household and family structure in China,"
paper prepared for the international seminar on China's 1982 population census, Beijing,
26-31 March 1984.
Ma Yingtong and Wang Yanzu. 1985. "A study on the first marriage and fertility status of
Chinese women," paper presented at the international symposium on China's one-per-
thousand national sample fertility survey, Beijing, 14-18 October 1985.
Menken, J. 1985. "Age and fertility: How late can you wait?" Demography 22, no. 4: 469-
483.
Parish, W. L. 1981. "Marriage and changes in family in the People's Republic of China," in
Research on the Population of China: Proceedings of a Workshop, ed. R. J. Lapham and
Rodolfo A. Bulatao. Washington, D.C.: National Academy Press.
, and M. K. Whyte. 1978. Village and Family in Contemporary China. Chicago: Uni-
versity of Chicago Press.
Pasternak, B. 1986. "Marriage and fertility in Tianjin, China: Fifty years of transition," Papers
of the East-West Population Institute, No. 99.
Rogers, A. 1975. Introduction to Multiregional Mathematical Demography. New York: John
Wiley.
Zeng Yi 703
Ryder, N. B. 1974. "Reproductive behaviour and the family life cycle," in The Population
Debate: Dimensions and Perspectives. Papers of the World Population Conference, Bu-
charest, 1974, Vol. II, pp. 278-288.
Schoen, Robert, and K. C. Land. 1979. "A general algorithm for estimating a Markov-generated
increment-decrement life table with applications to marital status patterns," Journal of
the American Statistical Association 74, no. 368: 761-776.
Smith, J. 1983. "Computer simulation of kin sets and kin counts," in Bongaarts et al. (eds.),
1987.
State Statistical Bureau of China. 1983a. 10 Percent Sampling Tabulation on the 1982 Population
Census of the People's Republic of China (computer tabulation). Beijing: China's Sta-
tistical Press (in Chinese).
. 1983b. Statistical Yearbook of China 1983. Hong Kong: Economic Information and
Agency.
1984. China's Major Statistics- 1984. Beijing: China's Statistics Press (in Chinese).
Department of Population Statistics. 1986. Preliminary Report on In-depth Fertility
Survey in China, Phase 1. Beijing.
Tien, H. Yuan. 1983. "China: Demographic billionaire," Population Bulletin 38, no. 2 (special
issue).
Wachter, K., E. Hammell, and P. Laslett. 1978. Statistical Studies of Historical Social Structure.
New York: Academic Press.
Watkins, S. C., J. Menken, and J. Bongaarts. 1984. "Continuities and changes in the American
family," paper presented at the 1984 annual meeting of the Social Science History
Association, California, May.
Willekens, F. J. 1983. "Multistate life table analysis of marriage and family," paper presented
at the IUSSP workshop on Family Demography: Methods and Their Applications, De-
cember 1983, New York.
,I. Shah, J. M. Shah, and P. Ramachandran. 1982. "Multistate analysis of marital
status life tables: Theory and application," Population Studies 36, no. 1: 129-144.
Yun Kalin. 1985. "A primary research on the way of providing for the aged," Social Science
Front 8, no. 1: 106-111 (in Chinese).
Zeng Yi. 1986a. "Family dynamics in China: The model and its application," unpublished
doctoral dissertation, Interuniversity Programme in Demography, Brussels Free Univer-
sity, May 1986.
. 1986b. "The family status life table: Extension of Bongaarts' nuclear family model,"
Netherlands Interuniversity Demographic Institute Working Paper, forthcoming, The
Hague.
. 1986c. "Fertility and family dynamics in China," Working Paper of the Interuniversity
Programme in Demography, 86-6, Brussels.
, J. Vaupel, and A. Yashin. 1985. "Marriage and fertility in China: A graphical anal-
ysis," Population and Development Review 11, no. 4: 721-736.
Zhao Xishun. 1985. "An analysis of rural family structure and its transition in China," paper
presented at the second international conference on modernization and Chinese culture,
November 1985, Hongkong (in Chinese).
Zhao Xuan. 1983. "State of women's first marriage in forty-two years from 1940 to 1981,"
in China Population Information Centre, 1984, pp. 106-117.
. 1985. "Report of a sampling survey (conducted in 1983) on female first marriage and
fertility status in 1982," Population and Economy, no. 4: 27-34 (in Chinese).

Changes in Family Structure in China - A Simulation Study Zeng Yi

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Changes in Family Structure in China - A Simulation Study Zeng Yi

Uploaded by

Copyright:

Available Formats

Changes in Family Structure in China: A Simulation Study

It is widely recognized that remarkable changes in Chinese

family structure have accompanied changes in the rates of marriage, fertility,

expectancy of China's population increased from 47 years in 1950 to 67.9

et al., 1984, p. 5).'

How do these tremendous changes in demographic factors affect Chinese

cross-sectional evidence on the changes in Chinese family structure. A non-

technical account of the simulation model is given in the second section.

Presentation of the simulation results and discussion of the lessons we have

A brief review of the cross-sectional

together has become rare; however, the three-generation family remains an

important family type in Chinese society.

POPULATION AND DEVELOPMENT REVIEW 12, NO. 4 (DECEMBER 1986) 675

Table 1 shows the substantial variation in average family size2 in China

up of some extended families, especially those of married brothers living

together. However, family size increased significantly to 4.78 in 1973. This

may have been attributable to housing shortages. Housing construction was

seriously disrupted during the Cultural Revolution of 1966-76; as a result,

by about 30 percentage points and the proportion of multigeneration families

had sharply declined as compared with 1930.

In light of the deeply rooted traditional Chinese ideology favoring large

precipitated by significant socioeconomic change. In prerevolutionary times,

well-to-do Chinese made every effort to sustain large multigeneration families

system and extremely low productivity levels, it appeared profitable in the

traditional Chinese context to maintain large families (Ma Xia, 1984).

Of course, many Chinese could not reach or maintain the traditional

family, or because poverty forced some brothers to migrate to seek employ-

in joint families did so.

TABLE 1 Average size of Chinese families,

selected years, 1930-82

Period Source size

1931 Sample survey of 22 provinces (Buck, 1937) 5.21

1930-40 Sample survey of birth histories covering 7 regionsa 5.58

(quoted from Ma Xia, 1984)

1953 Census 4.30

1964 Census 4.29

1973 Household statistics 4.78

1982 Census 4.43

emy of Social Sciences.

TABLE 2 Percent distribution of Chinese family types,

1930 and 1982

Nuclear families families and

(including one- other extended

Year Source person families) families Total

1930 Survey by Li 51.5 48.5 100.0

1982 Censusb 81.2 18.8 100.0

1982 One-per-thousand 80.5 19.5 100.0

data from Liaoning

1982 One-per-thousand 76.5 23.5 100.0

data from Hebei

1982 One-per-thousand 70.0 30.0 100.0

data from Fujian

SOURCES: a quoted from Ma Xia (1984). b State Statistical Bureau (1983a).

c Lavely and Li (1985).

Following the Chinese Revolution in 1949, feudal landownership was

brothers to marry and leave the extended family. Furthermore, as a result of

the cooperative movement, members of families no longer worked their own

with collective income distribution. The advantages of maintaining a large

provide confirmation. For example, according to a sample survey of 2,035

peasant families in Sichuan Province, families of more than three generations

live with their parents-in-law or parents (State Statistical Bureau of China,

provinces, respectively, live in families of three or more generations. Several

example, Parish, 1981; Pasternak, 1986).