The decline in the total fertility rate (TFR) of Bangladesh from more than six lifetime births per woman in the mid-1970s to slightly more than three births per woman in the early 1990s is remarkable. Some observers, pointing to a sharp increase in contraceptive prevalence—from less than 10% in the mid-1970s to about 45% in 1993-1994—attribute this decline to a successful national family planning program.1 Others, however, express doubt that the program could have reduced fertility by half without societal change.2

Fertility remained relatively static in Bangladesh between 1993 and 2000, despite a seven-point increase in contraceptive prevalence.3 In Matlab, where fertility was also stable during that period, the effect of an eight-point increase in contraceptive use was offset by the effect of a decrease in abortion.4 On the other hand, recourse to abortion is increasing in the country overall.5

Son preference, which has its roots in the patriarchal form of society, dependence on sons for financial support during old age and continuation of the family name, and the necessity of a dowry for female children, may be an obstacle to further decline in fertility.6 Although several studies have examined the effects of son preference on fertility and mortality, none has looked at its effect on abortion in Bangladesh.

Abortion is one of the four most important proximate determinants of fertility;7 in some countries, it has at times been the principal means of fertility control.8 Abortion is not always used exclusively for fertility control, however. In countries where the preference for sons is strong, such as China and Korea, sex-selective abortion is very common.9

In Bangladesh, induced abortion is illegal except to save the life of a pregnant woman. The government of Bangladesh, however, stated in a 1979 memorandum that menstrual regulation is an "interim method of establishing nonpregnancy" for a woman at risk of being pregnant, whether or not she is actually pregnant. In reality, menstrual regulation is used to avoid unwanted births, and the procedure is usually considered as abortion in Bangladesh.10 Because abortion is a sensitive issue on which data collection is very difficult, no accurate estimates of abortion prevalence and trends are available, and the purposes for which it is used in Bangladesh are not clearly understood. The study on which this article is based investigated whether son preference is a common reason for abortion in Bangladesh and, if so, how much abortion contributes (relative to contraceptive use) to achieving a couple's desired number of sons and daughters, as well as to limiting fertility.

In most of the countries in South and East Asia, people prefer sons to daughters.11 This preference often influences people's behavior and affects both fertility and mortality.12 The effects of son preference on mortality and its proximate determinants have been examined in many studies in Bangladesh. One study revealed preferential treatment of sons in food distribution and use of health care.13 Other research demonstrated that preferential treatment escalated during periods of famine,14 and that excess female child mortality was much higher among girls who had sisters than among those who did not.15 In another study, however, mortality was higher among both girls and boys who had one or more siblings of the same sex than among those who did not.16

Although the effects of sex preference on mortality in Bangladesh and other countries are consistent, its effects on fertility are not. Some observers have argued that son preference would be a strong barrier to the success of family planning programs, and thus would be an obstacle to fertility decline.17 According to one study, given perfect contraceptive use, if all couples desired at least two sons, families would have an average of 3.9 children, whereas if all couples desired at least one son and one daughter, the average would be three children.18

Son preference was not found to have an influence on fertility in the 1960s in Bangladesh and Pakistan.19 In Taiwan and South Korea in the 1970s and 1980s, however, couples with more daughters than sons had higher subsequent fertility.20 Investigators examining the reasons for this inconsistency concluded from the Matlab data that the effect of son preference on fertility would be more pronounced in a population with high contraceptive prevalence than in a population with low contraceptive prevalence.21

Their work explains the relationship between son preference and fertility in countries with low or moderate levels of contraceptive use, but does not explain the situation in developed countries, where contraceptive prevalence is usually high, fertility is low and son preference has little or no effect on fertility. Moreover, their study did not take the other proximate determinants into consideration. In Korea and China, abortion was found to be the principal means used to have children of the desired sex.22

An examination of cross-sectional data on fertility intentions and contraceptive use from 27 countries concluded that sex preference was not likely to have a major impact on contraceptive use and fertility.23 That analysis, however, had several limitations because of the lack of appropriate data. It covered a wide range of fertility intentions and contraceptive use, and the difference between observed contraceptive use and expected contraceptive use in the absence of sex preference was interpreted as the effect of sex preference on contraceptive use. Naturally, this effect will be small if prevalence is low. Furthermore, if a population does not prefer children of a particular sex, the effect will be small even if contraceptive use is high. Thus, at low values, this measure does not provide a clear indication of the effect of sex preference on contraceptive use (regardless of level of use) in a country in which sex preference is strong.

Moreover, that study used data on fertility intentions, not on fertility. The relationship between contraceptive use, fertility intentions and actual fertility is not clear-cut. Therefore, longitudinal data at different levels of contraceptive prevalence, as well as data on other proximate determinants (particularly on abortion) and on actual fertility, are very important and useful in assessing the role of sex preference on demographic transition. The research reported on in this study used longitudinal data on contraceptive use, abortion and fertility from Matlab, Bangladesh, to examine the issue.

Data and Methods

Data for this study came from the Demographic Surveillance System and the Record-Keeping System of the ICDDR,B: Centre for Health and Population Research in Matlab, Bangladesh. Since 1966, ICDDR,B has been recording data on the vital events of Matlab's population. The data, collected during biweekly household visits by community health workers, are checked at different levels for accuracy before being transferred to the surveillance system database.24 In 1977, a maternal and child health and family planning project began in half of the Matlab surveillance area. The remaining half, known as the comparison area, remained under the government's standard program. Information on the contraceptive use and reproductive status of married women of childbearing age has been collected every two weeks by community health workers in the project area only.

At the beginning of the project, neither fertility nor mortality differed between the two areas.25 Over time, contraceptive prevalence increased* and fertility and mortality declined in each area, but change occurred more rapidly in the project area. By 1995, statistics in the project area were more favorable than those in the comparison area for contraceptive prevalence (67% vs. 45%), the total fertility rate (3.0 vs. 3.7 lifetime births) and the infant mortality rate (51 vs. 79 deaths per 1,000 live births).

The four most important proximate determinants of fertility are contraceptive use, the proportion of women of reproductive age who are married, postpartum amenorrhea and abortion.26 Son preference is unlikely to affect the proportion of people of childbearing age who are married or of women in sexual union in a population, and its effect on postpartum amenorrhea is thought to be small in Matlab because the patterns and duration of breastfeeding of male and female infants do not differ significantly.27 This article, therefore, examines the effects of son preference and the trends in these effects on the other two main proximate determinants of fertility—contraceptive use and abortion—as well as on fertility in Matlab.

The effect of son preference on contraceptive use could be studied only in the project area, as contraceptive use data for the comparison area were available only in two knowledge, attitude and practice survey samples from 1984 and 1990. These sample sizes were not large enough to provide a valid estimate of the fertility effects of numbers of sons and daughters at different parities.

To cover a wide range of levels of contraceptive prevalence, we took contraceptive use data from the project area for three dates at five-year intervals: December 31, 1983; December 31, 1988; and December 31, 1993. We examined the fertility effect of the number of sons and daughters in a family for the periods 1984-1986, 1988-1989 and 1994-1995, and the effect on induced abortion for the periods 1982-1986, 1987-1991 and 1992-1995.

The occurrence of abortion has been found to be underestimated, despite the record-keeping efforts of the community health workers.28 This underestimation is not expected to affect abortion trends over time or comparisons among different groups of people, because the procedures used in collecting abortion data have been the same.

A modified Arnold Index29 is used here to estimate the effects of sex preference on contraceptive use, abortion and fertility. This index provides an estimate of the relative change in a variable caused by an absence of sex preference. It is defined as the ratio of the absolute difference between observed and expected fertility measures (contraceptive use, abortion and fertility) to the observed value, multiplied by 100.§ (When the expected rate was estimated, it was assumed that all couples at a given parity will act in the same manner as the couples at that parity who are currently most satisfied with the number of sons and daughters among their children.) The index is not affected by errors in reporting contraceptive use, abortion or fertility, unless the amount of error differs by the number of sons and daughters within a given parity. Couples were assumed to be most satisfied with their number of sons and daughters at the parity at which contraceptive use and the abortion ratio were highest and at which fertility was lowest.

The assumptions regarding contraceptive use and fertility seem reasonable, but for abortion, the situation is not clear-cut. If the purpose of abortion is to regulate fertility without knowing the sex of the fetus, the assumption regarding abortion seems logical. However, if the purpose is to abort the fetus only if amniocentesis or ultrasound identifies it as being female (as often occurs in China and Korea), then the current number of sons and daughters needs to be considered in combination with the sex of the fetus. In the latter case, the observed sex ratio at birth in Bangladesh should be greater than the expected sex ratio at birth, as is the case in China and Korea.30 In Matlab, the sex ratio at birth (about 104) remains within the normal range. These results imply that induced abortion in Matlab is not related to the sex of the fetus, suggesting that the women who have the highest abortion ratio in a parity do not want more children of either sex.


The percentage of married women aged 15-49 years who were practicing contraception on the last day of the year (December 31) in 1983, 1988 and 1993 in the Matlab project area, according to the number of sons at each parity, is shown in Table 1 (page 138). Contraceptive use increased from 40% in 1983 to 54% in 1988 and to 64% in 1993. In each year, the percentage of women using contraceptives increased with parity; within each parity, it increased with the number of sons, except for a slight decrease among women at parities greater than two who had only sons. Contraceptive use increased less with parity (and sometimes decreased) among women with no sons than among other groups. The data also suggest that although the preference for sons was quite strong, couples liked to have a daughter after having two sons.

Women in the comparison area had 17,803 live births during 1982-1986, 18,431 during 1987-1991 and 11,884 during 1992-1995 (Table 2, page 139). In the project area, the numbers of live births in these periods were 15,018, 15,092 and 10,604, respectively. The abortion ratio, defined as the ratio of abortions to live births multiplied by 1,000, was 21 during 1982-1986, 39 during 1987-1991 and 51 during 1992-1995 in the comparison area. In the project area, those ratios were 16, 25 and 24, respectively. In each area and each period, the abortion ratio usually increased with parity, and within a parity it was generally lowest for women with no sons and was often highest for women who had sons and a daughter.

Fertility fell sharply in both areas over time, but the decline differed between areas. In the comparison area, the percentage of women giving birth was 58% in 1984-1986, 50% in 1989-1991 and 27% in 1994-1995; in the project area, those percentages were 47%, 36% and 23%, respectively (Table 3). In both areas, fertility declined with rising parity; within each parity, fertility was highest for women without sons. Among women in the project area with two living children, fertility was lowest among those who had two sons. Among women with more than two children, the lowest fertility in both areas was found among those who had sons and a daughter.

The effects of sex preference on contraceptive use as measured by the modified Arnold Index are shown in Table 4. The proportion of couples practicing contraception in 1983-1993 would have increased by no more than five percentage points (expected minus observed) if there had been no preference for children of a particular sex. Thus, the impact of sex preference on contraceptive use was not great at any time. The contraceptive index declined from 9% in 1983 to 6% in 1993, indicating a decrease in the effect of sex preference on contraceptive use. In other words, the relative importance of sex preference as a determinant of contraceptive use declined as use of contraceptives increased.

The expected abortion ratio was higher than the observed abortion ratio in each year in each area (Table 5), suggesting that the abortion ratio would increase in the absence of son preference. This finding reflects the fact that the abortion ratio among women who had their desired number of sons was higher than the ratio among women who did not. For example, women with two living children preferred two sons to two daughters, and those who had two sons had a higher abortion ratio than those with two daughters (Table 2). In the comparison area, the abortion ratio would have increased by 36-38% in the absence of sex preference, while in the project area it would have increased by 27% during 1982-1986, 36% during 1987-1991 and 55% during 1992-1995 (Table 5). The increase in the abortion ratio as a result of son preference was greater than the increase in contraceptive use; moreover, the effect on contraceptive use decreased over time, but the effect on abortion increased.

The index for fertility and the TFRs during three periods are shown in Table 6. The effect of sex preference in the comparison area increased from 7% in 1984-1986 to 8% in 1989-1991 and 9% in 1994-1995. In the project area, the index increased from 9% in 1984-86 to 10% in 1989-1991, with a further increase to 12% in 1994-1995. Although the contraceptive prevalence rate for the comparison area was not available, there is no question that contraceptive use was increasing in the area over time.31


The great advantage of this study is its use of the largest and most comprehensive longitudinal population data set in the developing world. An estimate of the effect of son preference on fertility and its related variables requires data on these variables according to the number of sons and daughters at different parities. If the overall sample is not large and if there are not enough children in each category, the standard error will be very high. On the other hand, longitudinal data at different levels of contraceptive use and fertility are needed to investigate trends in the effect and to predict future effects. Cross-national data with different levels of contraceptive use and fertility will not serve these purposes, because the nature and intensity of sex preference may vary from country to country, along with fertility and contraceptive use.

The use of Matlab to represent Bangladesh as a whole may be questioned. Matlab is slightly better off socioeconomically than Bangladesh in general. However, there is no evidence that Matlab is an atypical area in Bangladesh; rather, trends and differentials in fertility and mortality in Matlab are similar to those in the country overall. The differences found in some studies are thought to be due mainly to the inferior quality of data in the national surveys.32 The project and comparison areas in Matlab were found in 1993-1994 to be virtually the same socioeconomically, except for the level of children's education, which was higher in the project area.33 The Matlab comparison area in most respects is similar to Bangladesh as a whole, while the project area is demographically a few years ahead of the rest of the country.

The long-term effects of son preference, as calculated here, may be underestimated. An examination of the distribution of women with one or two children according to the number of sons shows that the proportion of women with sons is greater than expected. For example, in Table 1, the number of women in the Matlab project area in 1982 with two sons and no daughters was 633, while the number with two daughters and no sons was 461, a ratio of 1.37. However, assuming a sex ratio at birth of 104, the ratio of women with two sons to the number with two daughters should be 1.08 (or 0.512/0.492). This disparity occurs mainly because women with two sons and no daughters move to the next parity less often or more slowly than women with two daughters and no sons, and partly because of higher mortality among female children than among male children. In this article, however, the calculation of the effect of sex preference was based on the existing distribution of women, so it will yield an estimate of the immediate effect of sex preference on fertility and its related variables. If a situation in which parents preferred neither sons nor daughters persisted over a long period, the distribution of women by number of sons within a given parity would eventually change, and the long-term effect would be somewhat greater than the short-term effect found in this study.

The effect of sex preference on contraceptive use was never high, and it decreased over time as contraceptive prevalence rose (Table 4). The absolute increase in contraceptive prevalence in the absence of sex preference would be no more than five percentage points at any time. Results from previous research for other years in the project area and the comparison area were almost identical.34 This study confirms results of earlier research indicating that sex preference is not a constraint to contraceptive use in Bangladesh.35

The same cannot be said, however, in the case of fertility. The effect of sex preference increased consistently over time as fertility declined. It is reasonable to assume that the change in the proportion of married women giving birth was the same as the percentage change in the TFR. Thus, the effect of sex preference increased almost linearly with the decrease in the TFR. This finding is consistent with the hypothesis that the effect of son preference is stronger in low-fertility situations, because couples have to have their desired number of sons and daughters within a smaller overall number of children. The TFR in the Matlab project area in 1994-1995 (three lifetime births per woman) would have decreased by 12% in the absence of son preference. However, this effect is much weaker than the effect estimated in a population that uses contraceptive methods perfectly.36

A comparison of the effect of sex preference on contraceptive prevalence, abortion and the TFR reveals some important and interesting features. Here, the effect on contraceptive prevalence is smaller than the effect on fertility (see Table 4 and Table 6). If the effect of sex preference on fertility were mediated by contraceptive use alone, the effect on contraceptive prevalence should be greater than the effect on fertility.** This finding suggests that the effect of son preference on fertility is mediated not only by contraceptive use but also by one or more of the other proximate determinants of fertility.

Among the other three important proximate determinants of fertility, abortion is the most likely candidate. As Tables 4 and 5 show, the effect of son preference on abortion increased over time, while the effect on contraceptive use decreased; moreover, the effect on abortion was much greater than the effect on contraceptive use. These results suggest that abortion was used more liberally than contraceptives to maintain the desired number of sons and daughters. It is true that fetal sex identification and sex-selective abortion did not exist in Matlab at the time of this study, and that the Matlab maternal and child health and family planning program was successful in reducing induced abortion.37 But if son preference remains strong in the area, facilities that identify the sex of a fetus may become available, resulting in an increase in the abortion of female fetuses.

The literature includes no studies on sex-selective abortion in Bangladesh. In neighboring India, however, about one million female fetuses were aborted in 1981-1991,38 and about 70% of all abortions in Delhi were performed because the fetus was female.39 Sex-selective abortions were so common that the Indian government announced a ban on the abortion of healthy female fetuses identified during permissible genetic tests.40

Abortion is legal in India, but not in Bangladesh. Yet, about 750,000 abortions occur in Bangladesh each year.41 Islam, which is thought to discourage abortion, is the religion of 85% of the people in Bangladesh, while Hinduism is the religion of 85% of India's population. But the religious and other cultural differences between India and Bangladesh do not seem to lead to much difference in the prevalence of abortion or son preference in the two countries. Policymakers thus need to find how to reduce both son preference and recourse to abortion in Bangladesh. An improvement in the status of women and female children should be helpful in reducing son preference,42 and an improvement in maternal and child health and family planning services should be helpful in reducing the number of abortions in the country.43