Induced Abortion in the Developing World: Indirect Estimates by Heidi Bart Johnston and Kenneth H

Induced Abortion in the Developing World: Indirect Estimates By Heidi Bart Johnston and Kenneth H. Hill

An analysis of Demographic and Health Survey data provides indirect estimates of the preva- partum insusceptibility. They subtracted total marital fertility and births averted by lence of abortion in 21 developing countries by rearranging Bongaarts’s proximate determi- contraceptive use from this estimate to ar- nants model to allow calculation of the index of abortion from the other principal proximate de- rive at births averted by induced abortion terminants of fertility (marriage, contraceptive use and postpartum insusceptibility to pregnancy), per married woman. They then translat- average total fecundity and total fertility. On average, abortion appears to have an influence on ed this result into a total marital abortion rate, using a method proposed by Bon- fertility similar to that of contraceptive use. This influence appears to be particularly strong in gaarts and Potter.10 the four Latin American countries in the analysis, where abortion reduces fertility by 38–55%. The problem with this approach is that In contrast, abortion’s fertility-reducing effect is only 6–19% in the Near East and 0–32% in Africa. it imposes an inappropriate sequencing In five countries for which two sets of DHS data are available, this reductive effect appears to of effects for the proximate determinants have increased over time. (International Family Planning Perspectives, 22:108–114 & 137, 1996) of the Bongaarts model. Specifically, For- eit and Nortman multiplied the index of postpartum insusceptibility by the aver- nduced abortion has long been recog- lence of induced abortion even in coun- age total fecundity rate (TF) to account for nized as one of the principal determi- tries where abortion is legal.5 births averted by postpartum insuscepti- Inants of fertility levels and is thought The shortcomings of direct measure- bility. Multiplying the index of a single to be practiced throughout the world, re- ment of abortion justify exploring the use proximate determinant by TF to account gardless of its legal status.1 Nevertheless, of indirect estimation methods to measure for births averted is likely to overestimate the demographic impact of abortion in de- abortion. Singh and Wulf estimated lev- the fertility-reducing impact of that de- veloping countries remains uncertain. els of induced abortion in six Latin Amer- terminant, so the technique presented by Without estimates of abortion, demogra- ican countries by adjusting data on Foreit and Nortman is likely to overesti- phers cannot accurately model fertility abortion-related hospitalizations—which mate the fertility-reducing effects of post- trends nor can they thoroughly under- represented an estimate of the total num- partum insusceptibility and underesti- stand the relationships between the de- ber of abortion-related hospitalizations in mate the total marital abortion rate. terminants of fertility and the fertility level. the country—for underreporting of in- In this article, we present an alternative Family planning program managers need duced abortion.6 They multiplied the ad- indirect method of calculating the influ- estimates of abortion levels to identify and justed hospitalization data by a factor, ence of abortion on fertility that can be cal- meet the need for modern contraceptives, ranging from three to seven, that repre- culated with data available from standard and policymakers need to be aware of the sented the availability of safe abortion in fertility surveys. Using this method, we levels of abortion and the conditions under a country. The value of the multiplier was provide abortion estimates for 21 devel- which it is provided, particularly in areas larger for countries where relatively safe oping countries. Like Foreit and Nortman, where maternal morbidity and mortality abortion is more accessible. The resulting we use a residual technique based on Bon- are linked to unsafe abortion. value is an estimate of the total number of gaarts’s proximate determinants model.11 Data quality is an important consider- abortions performed in the country. To im- After describing the method and pre- ation in studying the effects of abortion on plement Singh and Wulf’s approach, how- senting the estimates, we discuss the re- fertility. Direct measures can be used only ever, researchers must develop and sults of tests for confounding between the where termination of pregnancy is accu- continually update a wide range of as- abortion estimates and the “minor” prox- rately reported. Abortion, however, is gen- sumptions and adjustment factors.7 Main- imate determinants of fertility not in- erally stigmatized and is thus subject to taining the necessary inputs for calculat- cluded in the model. considerable underreporting.2 Abortion ing abortion rates using this scheme is data typically come from one of two daunting even at the regional level and is Methodology sources: clinic or hospital records or indi- much more so at the global level. Proximate Determinants of Fertility vidual surveys. Abortion data from clin- Foreit and Nortman8 developed an in- The indirect method with which our esti- ic or hospital sources are thought to be ac- direct technique to calculate abortion in- mates are calculated is based on the con- curate in some countries where abortion cidence for married women in three South cept that all socioeconomic, cultural and is legal and accessible,3 but are likely to be American countries. To obtain an estimate biological variables that influence fertili- inaccurate where abortion is illegal, se- of births averted by abortion per married ty work through a limited number of prox- verely restricted or difficult to obtain.4 In- woman, they used an adaptation of Bon- imate factors. The concept was developed dividual surveys underestimate the preva- gaarts’s proximate determinants of fertil- by Davis and Blake12 and later refined by ity model.9 They first estimated the total others, including Bongaarts.13 The seven Heidi Bart Johnston is a doctoral candidate and Kenneth natural fertility rate by multiplying total factors currently accepted as the proxi- H. Hill is a professor in the Department of Population Dynamics, Johns Hopkins University School of Hygiene potential fertility by an index represent- mate determinants of fertility are the pro- and Public Health, Baltimore, Md., USA. ing the fertility-reducing effects of post- portion of females married or in sexual

108 International Family Planning Perspectives unions; contraceptive use and effective- and 1, although with real data there is no (15–49), regardless of marital status. In 10 ness; the prevalence of induced abortion; guarantee that a residual estimate of the surveys—those from countries in which the duration of postpartum insuscepti- index will not exceed 1. few, if any, sexual relationships are bility; fecundability; spontaneous in- There are three major potential sources thought to occur outside of marriage— trauterine mortality; and the prevalence of error in the rearranged equation: the DHS limited the sample to ever-married of permanent sterility. possible measurement error in the indices women aged 15–49. Bongaarts demonstrated, in an analysis of the fertility-reducing effects of the major The TFR is calculated by summing age- of 41 developing, developed and histori- determinants; omission of the effects of the specific fertility rates for the three years cal populations, that four proximate de- three proximate determinants not in- preceding the survey. Data on fertility terminants (marriage, contraceptive use, cluded in the model (fecundability, spon- come from a full pregnancy history asked abortion and postpartum insusceptibili- taneous intrauterine mortality and per- of women of reproductive age. In coun- ty) explained 96% of the variation in fer- manent sterility); and the use of a single tries where the DHS sampled all women tility.14 The small amount of unexplained value for TF, the average total fecundity aged 15–49, regardless of marital status, fertility is, in part, a result of variations in rate of a population in the absence of any the TFR is based on the data for all women the remaining three proximate determi- fertility-reducing effects of proximate de- in that age-group. In countries where the nants—fecundability, sterility and in- terminants. As with any residual method DHS sampled only ever-married women trauterine mortality. These minor proxi- of estimation, any overestimation or un- aged 15–49, calculations of the TFR include mate determinants are thought not to vary derestimation of the contributing values only births to ever-married women. Al- substantially by country or across time. will affect the size of the residual and thus though not free of error, TFR estimates In Bongaarts’s proximate determinants the accuracy of the estimate. from the DHS are thought to be “reason- model, the total fertility rate (TFR) is ap- Bongaarts estimated that TF averages ably complete and accurate.”18 Where the proximated by the average total fecundi- about 15.3 lifetime births per woman for TFR is underestimated, the fertility-re- ty rate (TF) adjusted for indices repre- developed and developing countries and ducing effect of abortion will be overesti- senting the fertility-reducing effect of each ranges from 13 to 17, depending on the fer- mated, and vice-versa. of the four principal proximate determi- tility-reducing effects of the proximate de- The indices for marriage, contraceptive nants: TFR = TF x Cm x Cc x Ca x Ci, where terminants not included in Bongaarts’s use and postpartum insusceptibility are TFR is the average total number of lifetime model.15 calculated according to formulas pre- births per woman if current age-specific In calculating TF, Bongaarts made some sented by Bongaarts,19 except as described fertility rates prevail throughout her child- simplifying assumptions that could influ- below. For each index, we discuss the data bearing years; TF is the mean potential ence our estimates of abortion. First, he as- used to calculate the indices, the estimat- number of lifetime births per woman dur- sumed that the minor proximate determi- ed values of the indices, and the potential ing her reproductive years, estimated at an nants—natural fecundability, intrauterine influence of data error. For the five coun- average of 15.3; Cm is the index for the pro- mortality and primary sterility—have a fair- tries for which we have data from two portionate reduction in fertility caused by ly small fertility-reducing effect, and that DHS surveys—Egypt, Kenya, Indonesia, nonexposure to sexual intercourse; Cc is this effect varies little between populations Morocco and Senegal—we report the the index for the proportionate reduction or over time, thus contributing minimally changing influence of each principal prox- in fertility caused by use of contraceptives; to differentials and trends.16 Second, for all imate determinant over time. Ca is the index for the proportionate re- of the developing countries but South For two of these five countries, there is duction in fertility caused by induced abor- Korea, and all of the historical populations, an important difference between the data tion; and Ci is the index for the propor- he assumed that the effect of abortion on fer- collected in the first and second surveys. tionate reduction in fertility caused by tility was negligible.17 Bongaarts’s estimate The data collected for the later Kenya and postpartum insusceptibility. of TF thus may include some fertility-re- Senegal surveys include information on ex- The TFR and the index for each of the ducing effect of abortion. Residual estimates posure to intercourse; the data collected in principal proximate determinants, with the of the fertility-reducing effects of abortion the earlier surveys do not. For consistency exception of abortion, can be calculated di- from our rearranged equation are therefore of measurement when analyzing tempo- rectly from the data collected in standard relative to the average unmeasured influ- ral changes in the indices, we use data for fertility surveys. Each index has a value ence of abortion in the populations con- ever-married women (including women ranging between 0 and 1. The lower the tributing to the estimated TF of 15.3, rather ever in stable sexual unions) to calculate the value for an index, the greater the reduc- than absolute estimates. index of marriage, and data for contra- tive effect of that factor on fertility. For ex- ceptive use among currently married ample, an index of 1 indicates that the spec- Equation Variables women to calculate the index of contra- ified factor has no fertility-reducing effect, To estimate values for the equation vari- ception. As a result, for Kenya and Sene- and an index of 0 implies that all fertility ables, we use data from 26 Demographic gal, the values for the indices of marriage, is prevented by that particular factor. and Health Surveys (DHS) conducted in contraception and abortion in the tempo- 21 countries in Africa, Asia and Latin ral trends analysis differ from those in Table Estimating Abortion as a Residual America. Most of the data were collected 1 (page 110) and in the main analysis. The index of abortion can be estimated as in the 1990s; however, for five countries •Total fertility rate. The TFRs for the coun- a residual by rearranging Bongaarts’s in which two surveys have been con- tries included in the study range from 2.7 equation as Ca=TFR/(TF x Cm x Cc x Ci). ducted, we use data sets from both the lifetime births per woman in Turkey to 7.7 Like the indices representing the other 1980s and the 1990s to examine temporal in Yemen. The TFR decreased over time proximate determinants, the index repre- trends in abortion. Sixteen of the surveys in all five countries surveyed more than senting the fertility-reducing effects of included in this analysis drew their sam- once. The absolute decrease in the TFR abortion should have a value between 0 ple from all women of reproductive age was most dramatic in Kenya—from 6.7 in

Volume 22, Number 3, September 1996 109 Induced Abortion in the Developing World

Table 1. Total fertility rates (TFRs) and estimated indices of the principal proximate determi- estimation of contraceptive effectiveness nants and indicators of the minor proximate determinants, by country, Demographic and Health or any underreporting or overreporting Surveys, 1986–1993 of current contraceptive use will con- tribute to inaccurate estimates of C . Ac- Country (and year Observed Index of Index of Index of Index of Indicator of Indicator of Indicator of a of survey) TFR marriage contra- post- abortion fecun- intrauterine sterility cording to an analysis of DHS data, the in- ception partum dability mortality formation on current contraceptive insuscep- 23 tibility prevalence is generally acceptable. We calculate two indices of contracep- N.E. Brazil (1991) 3.66 0.694 0.669 0.935 0.551 0.789 0.974 0.910 Burkina Faso (1993) 6.91 0.942 0.947 0.533 0.949 0.410 0.949 0.969 tion—one for traditional methods and one Cameroon (1991) 5.83 0.951 0.850 0.604 0.780 0.554 0.967 0.897 for modern methods. The product of the Colombia (1990) 2.86 0.696 0.658 0.901 0.452 0.688 0.989 0.939 two indices represents the total influence Egypt (1988) 4.69 0.683 0.672 0.749 0.891 u u 0.956 Egypt (1992–1993) 3.93 0.653 0.577 0.830 0.821 u 0.967 0.957 of contraceptive use on fertility. Tradi- Indonesia (1987) 3.43 0.716 0.574 0.722 0.755 u u 0.953 tional methods appear to have a minimal Indonesia (1991) 3.03 0.717 0.539 0.775 0.661 0.849 0.968 0.951 effect that varies little from country to Jordan (1990) 5.57 0.604 0.724 0.885 0.941 u 0.979 0.960 Kenya (1988–1989) 6.71 0.771 0.824 0.683 1.011 u u 0.972 country. The effect of modern methods, on Kenya (1993) 5.40 0.886 0.780 0.683 0.748 0.583 0.974 0.989 the other hand, varies substantially. The Namibia (1992) 5.37 0.875 0.786 0.746 0.684 0.582 0.969 0.966 Niger (1992) 7.38 0.967 0.976 0.593 0.861 0.691 0.959 0.981 combined index of contraception varies Madagascar (1992) 6.13 0.914 0.919 0.631 0.755 0.686 0.961 0.899 from 0.98 in Niger (minimal importance) Malawi (1992) 6.73 0.918 0.928 0.658 0.785 u 0.959 0.989 to 0.54 in Indonesia (great importance). Morocco (1987) 4.84 0.619 0.718 0.735 0.967 u 0.959 0.951 Morocco (1992) 4.04 0.612 0.662 0.760 0.856 u 0.969 0.958 The average value of the index in the 26 Paraguay (1990) 4.70 0.830 0.753 0.866 0.568 u 0.981 0.951 data sets is 0.79. For each of the countries Peru (1991–1992) 3.54 0.673 0.748 0.743 0.618 0.731 0.975 0.961 with multiple observations except Sene- Rwanda (1992) 6.23 0.764 0.892 0.567 1.055 0.743 0.961 0.994 Senegal (1986) 6.62 0.857 0.942 0.576 0.930 u u 0.948 gal, the fertility-reducing effects of con- Senegal (1992–1993) 6.03 0.853 0.952 0.610 0.795 0.401 0.965 0.976 traceptive use increased between the two Sudan (1989–1990) 4.96 0.628 0.935 0.617 0.894 u 0.956 0.956 Turkey (1993) 2.65 0.614 0.609 0.901 0.515 u 0.971 0.971 surveys. The increase in the fertility-re- Yemen (1991–1992) 7.67 0.816 0.936 0.813 0.807 u 0.963 0.993 ducing effects of contraceptive use was Zambia (1992) 6.47 0.935 0.943 0.662 0.723 0.625 0.960 0.986 greatest in Egypt, where the value of the

Note: u=unavailable index fell from 0.67 in 1987 to 0.58 in 1991. •Index of postpartum insusceptibility. The index of postpartum insusceptibility, Ci, 1988–1989 to 5.4 in 1993. The average TFR index of marriage in the 26 data sets used measures the fertility-reducing effects of in the 26 data sets used in this study is 5.2. in this study is 0.78. The effect on fertility sexual abstinence or amenorrhea follow- •Index of marriage. The index of marriage, of delaying marriage appears to have in- ing a birth. The index is calculated using Cm, is intended to measure exposure to creased over time for four of the five coun- data on postpartum abstinence and post- sexual intercourse. In most cultures, age tries for which we have data from two sur- partum amenorrhea among women who at first sexual intercourse is a more accu- veys. It increased most in Senegal—where had given birth up to 35 months before the rate measure of exposure to risk of preg- the value of Cm fell from 0.86 in 1986 to DHS survey. In the 22 data sets that in- nancy than age at first marriage or at first 0.79* in 1992–1993––and negligibly in the clude the median duration of abstinence stable sexual union.20 For this reason, remaining four countries. or amenorrhea, the median duration is where the DHS provides such data, we •Index of contraception. The index of con- used; elsewhere, the mean is used. As cur- use proportions of women who have ever traception, Cc, represents the extent to rent status data are considered reliable,24 had intercourse, rather than proportions which populations limit fertility through we assume that misreporting contributes of women who have ever been married, use of modern and traditional contracep- minimally to the variation in Ci. as the basis for calculating Cm. tives. To measure the fertility-reducing ef- Estimates of the index of postpartum in- Because not all women who have ever fects of contraceptive use, the prevalence susceptibility range from 0.94 in North- had intercourse are exposed to intercourse and the use-effectiveness of each method east Brazil (showing little effect) to a high regularly, this measure could overestimate must be known. While data on the preva- exposure. For the 10 countries in this analy- lence of contraceptive methods within Table 2. Contraceptive use-effectiveness rates, sis for which data on nonmarital inter- countries are available, data for contra- by method course are not available, we use the pro- ceptive use-effectiveness are less country- Method Rate portion of women who have ever been specific. Although contraceptive use-ef- Oral contraceptives 0.82 married as the basis for calculating Cm. We fectiveness has been shown to fluctuate Injectable 0.96 will continue to refer to Cm as the index of across countries,21 this analysis uses a stan- Implant 0.99 marriage, but it should be interpreted as an dard set of contraceptive use-effectiveness Tubal sterilization 0.99 22 Vasectomy 1.00 index of exposure to sexual intercourse. rates (see Table 2). IUD 0.90 Estimates of the index of marriage (Table In general, the contraceptive use-effec- Condom 0.62 1) range from 0.60 in Jordan (showing a tiveness rates applied represent the lower Vaginal methods 0.80 Periodic abstinence 0.50 moderate effect) to 0.97 in Niger (showing limits of available estimates. By using the Withdrawal 0.38 a minimal effect). The average value of the lower limits, we assume we are more ac- Other 0.10

curately reflecting reality, although we Sources: Modern methods—J.G. Hutchings, G. Perkin and L. *To allow comparison with the estimate of Cm for 1986, may be underestimating contraceptive ef- Saunders, 1987 (see reference 22); Periodic abstinence and withdrawal—S. Thapa, D. Hamill and P. Lampe, 1993 (see ref- we recalculated the estimate for 1992–1993 using data fectiveness for some countries in the erence 22); Other—A. K. K. Sayila, 1992 (see reference 22). for ever-married women. analysis. Any underestimation or over-

110 International Family Planning Perspectives of 0.53 in Burkina Faso (showing a strong Figure 1. Trends in the index of abortion as a predictor of fertility in five countries effect). The average value of the index of postpartum insusceptibility in the 26 data Index of abortion sets is 0.72, making postpartum insus- 1.10 ceptibility, on average, the most influen- tial principal proximate determinant in this analysis. The fertility-reducing effects 1.00 of postpartum insusceptibility decreased over time in four of the five countries with 0.90 two observations.

Abortion’s Effect on Fertility 0.80 Kenya Morocco The estimates generated using the indirect Senegal method presented here suggest that abor- Egypt 0.70 tion has a large effect on fertility in some Indonesia settings and a small effect in others. Abor- tion appears to have a strong effect in 0.60 South America, with index values rang- 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 ing from 0.45 in Colombia to 0.62 in Peru, Year yet seems to have only a marginal impact Note: For consistency in measurement across time, the index of abortion for all 10 observations was calculated using data for ever-mar- ried women (for the index of marriage) or currently married women (for the index of contraception), and not for women who had ever in high-fertility areas of Africa, as sug- had intercourse. For this reason, the values for the index of abortion for Kenya and Senegal differ from those presented in Table 1. gested by values of 0.95 in Burkina Faso and 1.06 in Rwanda. The effect of abortion on fertility ap- ducing effects of abortion appear to be in- when abortions were illegal in Turkey, pears to be geographically clustered, sug- creasing with time—a trend shown in Fig- Tezcan found that when asked directly, gesting that common socioeconomic sit- ure 1 for each of the five countries with 14% of a nationally representative sample uations and cultural heritages influence two observations. The increase was great- of Turkish women reported having had an abortion levels. Abortion appears to have est in Indonesia, where the value of the abortion; when the randomized response the greatest impact on fertility in Latin index fell from 0.76 to 0.66 between 1987 technique was used, 33% of the same sam- America, a strong influence in Southeast and 1991. Figure 1 corrects for the exag- ple reported having had at least one abor- Asia (as measured using data from In- gerated change over time shown in Table tion.28 Since 1975, abortions are thought to donesia), a varying impact in Sub-Saha- 1 for Kenya and Senegal by using consis- have increased, as a result of the legaliza- ran Africa, and the weakest influence in tent definitions across time. tion of abortion in 1983 and of a decrease the Near East (barring Turkey). That We can compare direct and indirect es- in desired family size.29 Turkey is an exception is no surprise—of timates of the index of abortion using data Alternatively, the difference in estimates the 21 countries* included in the study, from the 1993 Turkey DHS. The directly es- could be a result of the proposed residual Turkey has the least restrictive laws re- timated total abortion rate, 0.52, represents technique overestimating the influence of garding access to abortion.25 the average number of abortions a Turkish abortion on fertility. The equations or the Table 1 exaggerates the change over woman will have in her lifetime if she has data used to measure the indices of the time in the influence of abortion on fertil- abortions at the age-specific rates prevail- principal proximate determinants of fer- ity for Kenya and Senegal. For both coun- ing in the five years before the survey. The tility may underestimate the fertility-re- tries, the index of abortion for the earlier estimate is calculated as the number of ducing effect of the determinants and thus period refers to ever-married women (in- abortions recorded in the five-year calen- overestimate the fertility-reducing effect cluding women ever in a stable sexual dar of respondents’ fertility histories, di- of abortion. In addition, missing deter- union), and the index for the second pe- vided by the total number of woman-years minants may contribute to the residual riod refers to women who had ever had represented in the calendar. To obtain abor- and lead to an overestimate of the fertili- intercourse. The index of marriage from tion data, the DHS interviewers in Turkey ty-reducing effect of abortion. the earlier survey probably overestimates asked respondents about their pregnancy The difference between direct and indi- the fertility-reducing effects of delaying histories over the preceding five-year pe- rect estimates probably results from the entrance into sexual activity and under- riod and probed to determine how in- combined effect of underreporting and estimates the fertility-reducing effects of complete pregnancies ended.26 Using the measurement error. At any rate, the two abortion. If this is the case, the use of data direct method of estimating the index of estimates of Ca that exceed 1 suggest that for ever-married women rather than those abortion presented by Bongaarts and Pot- some unidentified error is influencing the for women who had ever had intercourse ter,27 the total abortion rate of 0.52 translates estimates. The factors most likely to in- could be one reason the value of Ca for to an abortion index of 0.89, very different fluence residual fertility are the three prox- Kenya in 1988–1989 exceeds 1. from the indirectly estimated index of 0.52. imate determinants not included in Bon- The average value for the index of abor- The vast difference in the direct and in- gaarts’s fertility model—fecundability, tion is 0.78, indicating that the fertility-re- direct estimates of the effect of abortion natural sterility and intrauterine mortali- ducing effect of abortion is similar to the on fertility could be a result of abortions ty. Misestimating the average total fecun- influence of contraceptive use and mar- being underreported in Turkey; the gen- riage, but less important than that of post- eral consensus is that women, in Turkey *In this article, average values for the 21 countries in- cluded in the analysis are based on 26 data sets because partum insusceptibility for the countries and elsewhere, tend to underreport abor- Egypt, Indonesia, Kenya, Morocco and Senegal are each included in this analysis. The fertility-re- tion, whether it is legal or illegal. In 1975, represented twice.

Volume 22, Number 3, September 1996 111 Induced Abortion in the Developing World

Figure 2. Regression of the indicator of fecundability on the index of abortion terminants but a greater effect than either sterility or intrauterine mortality. A popu- Indicator of fecundability lation’s TFR is thought to be somewhat sen- 33 0.90 sitive to variations in fecundability. Frequency of intercourse—a proxy for 0.80 fecundability—is available in some DHS data sets. To test the influence of fecund- 0.70 ability on fertility, we developed an indi- cator representing the proportion of wo- 0.60 men who had ever had intercourse who were sexually active in the four weeks pre- 0.50 y= –0.30x + 0.86 ceding the survey. Because the indicator R2= .13 measures frequency of intercourse among 0.40 those who have ever been sexually active 0.30 rather than among women currently in a 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 union, the values of our indicator of fe- Index of abortion cundability may overestimate the influ- ence of fecundability on fertility. Only 13 of the 26 data sets included in dity rate (TF) could also influence the es- indices they represent. Like the indices, our study report data on frequency of in- timate of the effect of abortion, although the indicators have values between 0 and tercourse. In these data sets, the value of the effect would not differ across countries. 1, where 0 represents the maximum fer- the indicator of fecundability ranges from There is also a minor timing inconsis- tility-reducing effect and 1 represents no 0.40 in the more recent of the two Senegal tency in the measures used. The TFR is effect. As the minor proximate determi- surveys to 0.85 in the more recent of the measured for the three-year period prior nants are thought to have a minimal in- two Indonesia surveys. to the survey, whereas the measures of the fluence on fertility, we expect the values Figure 2 shows the scatterplot of resid- other factors in our equation refer to the obtained to be fairly close to 1.* ual fertility against the fecundability in- time of the survey. If fertility is falling, the To test for association between the dicators for the 13 data sets; the solid line TFR will be somewhat overestimated in minor proximate determinants and resid- plots the regression. Fecundability appears relation to the measures of the proximate ual fertility, we regress the indicators to explain little of the variation in the resid- determinants, and the fertility-reducing against the residual estimate of the index ual estimate of abortion (R2=.13; p>.10). effect of abortion will be somewhat un- of abortion. A significant positive rela- The association is negative, the opposite derestimated. The effect is not likely to be tionship between an indicator and the es- of the expected effect, but not significant- large, particularly in view of the tenden- timate of the abortion index would sug- ly different from zero. Fecundability as cy in DHS data for recent births to be gest that the minor proximate determinant proxied in this analysis has no statistical- pushed backward in time.30 has a fertility-reducing effect that should ly significant relationship with the resid- To ensure that the error is not the result be accounted for in the model. ual fertility reduction and does not explain of a systematic influence on fertility of one •Fecundability. Fecundability is the proba- the variability in the abortion estimates. or more minor proximate determinants, bility of conception, assuming exposure, •Intrauterine mortality. Of all the proximate or a miscalculation of TF, we examine the per menstrual cycle. Fecundability is con- determinants, spontaneous intrauterine potential for bias linked to these sources. sidered to have a lesser effect on fertility mortality, or spontaneous abortion, is than any of the four principal proximate de- thought to have the least influence on the Minor Proximate Determinants Bongaarts acknowledges that the three Figure 3. Regression of the indicator of intrauterine mortality on the index of the abortion proximate determinants not included in his fertility model directly influence fer- Indicator of intrauterine mortality tility.31 Only in unusual circumstances is 0.99 the influence of these variables on fertili- ty differentials or trends considered more than minimal.32 However, given the sen- 0.98 sitivity of a residual to measurement error, the minor proximate determinants could bias our estimates of the abortion index. 0.97 To determine the relationships between the minor proximate determinants and the 0.96 estimate of abortion, we develop indica- tors for the minor proximate determinants that can be derived from DHS data. The 0.95 y= –0.04x + 0.997 indicators are expected to covary with the R2= .45

*However, our main interest is not in the absolute val- 0.94 ues of the indicators, but rather in their relation to the 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 Index of abortion residual estimates of abortion.

112 International Family Planning Perspectives 34 total fertility rate of a population. To test Figure 4. Regression of the indicator of sterility on the index of abortion the influence of intrauterine mortality on fertility, we developed an indicator based Indicator of sterility on the neonatal mortality rate, which is 1.00 available in most DHS data sets. We chose this indicator on the assumption that the 0.98 proportion of conceptions that do not re- sult in a live birth will covary with the 0.96 neonatal mortality rate. This assumption is based on the grounds that many of the 0.94 conditions associated with the component of intrauterine mortality most likely to 0.92 vary across populations or over time, spontaneous abortions occurring in late 0.90 pregnancy, are associated with the neona- y= –0.05x + 0.92 R2= .08 tal mortality rate.* To obtain the indicator 0.88 for intrauterine mortality, the neonatal 0.40 0.50 0.60 0.70 0.80 0.90 1.00 1.10 mortality rate for the period 0–4 years Index of abortion prior to the survey is subtracted from 1. Twenty-two of the 26 data sets include information from which the indicator of in- clude the information necessary to calcu- Colombia, Indonesia, Jordan, Kenya, Peru trauterine mortality can be calculated. For late the indicator. The values of the indi- and Turkey—the six countries included these 22, the values of the indicator range cator of sterility range from 0.90 in both in Bongaarts’s 1982 analysis and in from 0.95 in Burkina Faso to 0.99 in Colom- Cameroon and Madagascar to 0.99 in this study—will show, Bongaarts’s esti- bia. As illustrated in Figure 3, regression of Rwanda and Yemen. The high values sug- mate of TF may be confounded by the ef- the indicator of intrauterine mortality gest that primary sterility has only a small fects of abortion; thus, the abortion esti- against the estimate of the index of abor- impact on fertility in any of the 21 coun- mates presented here may be too low. We tion shows a statistically significant nega- tries considered here. Regression of the explore this possibility by rearranging tive relationship between the two variables sterility indicator against the residual es- Bongaarts’s model to isolate TF, using data (R2=.45, p<.01). Whereas a significant pos- timate of abortion (Figure 4) shows that presented in Bongaarts’s 1982 analysis (in- itive association would suggest that the primary sterility does not explain the vari- cluding observed rather than estimated variation in intrauterine mortality might ation in the abortion estimate (R2=.08, TFRs). For the developing countries in the bias the residual estimates of abortion, the p>.10). Residual fertility is positively as- analysis, Bongaarts used data from the significant negative association indicates sociated with this measure of sterility, as mid-1970s. A residual value for TF of less that some other unexplained relationship would be expected if sterility were a con- than 15.3 would suggest that the fertility- exists between the two variables. founding factor, but the relationship is not reducing effects of a proximate determi- •Sterility. The sterility indicator is intend- statistically significant and thus does not nant, such as abortion, were not included ed to measure the inability of a couple to observably bias our abortion estimates. in the model. reproduce, for reasons other than contra- If we assume that populations using ceptive sterilization, after the woman’s Total Fecundity Rate contraceptives to limit fertility also turn menarche and before menopause. Al- To determine a value for average total fe- to abortion, according to the data pre- though fertility rates in populations are cundity, Bongaarts regressed values for the sented in Bongaarts 1982 analysis, women thought to be moderately sensitive to dif- total marital fertility rate in the absence of in two of the six countries, Colombia and ferent levels of sterility, fluctuations in rates postpartum infecundability against values Turkey, and to a lesser extent, those in In- of primary sterility across countries are for contraceptive prevalence, using data donesia, Jordan and Peru would use abor- considered uncommon.35 from 23 developing countries, eight de- tion to reduce fertility. In Turkey and To examine the influence of sterility on veloped countries and 10 historical pop- Colombia (the two countries in our study fertility, we developed an indicator for ulations. The resulting value for average where abortion appears to have the sterility calculated by subtracting from 1 total fecundity, 15.3, was the total marital strongest effect), the total fecundity rate, the proportion of women aged 45–49 who fertility rate when contraceptive preva- as estimated from Bongaarts’s data, is have ever been married or in a union and lence was 0.0.36 In calculating the values lower than 15.3 (Turkey, 14.5; Colombia, have no children. If, as we assume, all cou- for total marital fertility rates in the absence 14.6), suggesting that abortion or the ples in the 26 data sets in this analysis who of postpartum infecundability, Bongaarts minor proximate determinants reduced are able to have children will have at least employed the simplifying assumption that fertility more in these countries than they one child, the indicator will include only for all but one of the historical populations did, on average, in all countries in the women who are sterile for reasons other and developing countries in the study, the study, and that, in general, 15.3 could be than contraceptive sterilization. The in- effect of abortion on fertility was negligi- an underestimate of TF. dicator does not measure the level of sec- ble.37 If abortion had an effect that was not In Indonesia, Jordan and Peru, the level ondary sterility, that is, the proportion of accounted for in the calculation of 15.3, of contraceptive prevalence suggests that couples who become sterile for reasons then 15.3 underestimates TF, and residual other than contraceptive sterilization after estimates of abortion may underestimate *Factors associated with early pregnancy loss are less like- ly to affect neonatal mortality, but are also less likely to having one or more children. abortion rates. vary between populations or over time, thus having lit- All 26 data sets used in our study in- As our examination of the data for tle effect on our estimates.

Volume 22, Number 3, September 1996 113 Induced Abortion in the Developing World the desire to control fertility existed, but was index of abortion is promising, it is not culating Rates of Induced Abortion,” Demography, weaker than in Turkey or Colombia. We without limitations. As we discussed ear- 29:127–136, 1992. therefore expect abortion to have a slight ef- lier in this article, some amount of mea- 9. J. Bongaarts, “The Fertility-Inhibiting Effects of the fect on fertility in those countries. Because surement error associated with the prin- Intermediate Fertility Variables,” Studies in Family Plan- ning, 13:179–189, 1982. a low level of abortion is included in Bon- cipal proximate determinants contributes gaarts’s estimated average of the total fe- to the residual estimate of abortion. Fur- 10. J. Bongaarts and R. G. Potter, Fertility, Biology, and Be- havior: An Analysis of the Proximate Determinants, Aca- cundity rate, we expect the country-specific thermore, determinants not included in demic Press, New York, 1983. estimates to be close to 15.3. In Indonesia, the model might contribute to the resid- 11. J. Bongaarts, 1982, op. cit. (see reference 9). his estimated rate is 15.2, and in Jordan and ual estimate. However, we found no sta- 12. K. Davis and J. Blake, “Social Structure and Fertili- Peru it is 15.9. The high estimates for Jor- tistically significant relationship in the ex- ty: An Analytic Framework,” Economic Development and dan and Peru imply that fertility-reducing pected direction between any of the minor Cultural Change, 4:211–235, 1956. factors not included in Bongaarts’s model proximate determinants and the residual 13. J. Bongaarts, 1982, op. cit. (see reference 9). (abortion or the minor proximate determi- estimate of abortion, suggesting that the 14. Ibid. nants) have less of an effect than for all the fertility-reducing effects of the minor prox- studied countries as a group. imate determinants do not contribute sub- 15. Ibid. In Kenya in 1976, contraceptive use had stantially to the residual. Nevertheless, the 16. ——, “A Framework for Analyzing the Proximate Determinants of Fertility,” Population and Development a negligible effect on fertility. We suspect proxies we developed may not capture ex- Review, 4:105–131, 1978. that abortion also had next to no influence isting relationships between the minor 17. Ibid. on fertility. (Indeed, according to our esti- proximate determinants and the residual mates, abortion had no effect in Kenya in abortion estimate. 18. F. Arnold, “Assessment of the Quality of Birth His- tory Data in the Demographic and Health Surveys,” in 1988–1989, but had a substantial effect in Despite these limitations, the estimates Institute for Resource Development (IRD)/Macro In- 1993.) As we calculate total fecundity from generated using this technique demon- ternational, An Assessment of DHS-I Data Quality, DHS the data presented in Bongaarts’s analy- strate believable international trends and Methodological Reports, No. 1, Columbia, Md., USA, sis, Kenya’s total fecundity rate equals 15.9, differentials. Although abortion is illegal 1990. implying that abortion or other minor in all but one of the countries included in 19. J. Bongaarts, 1982, op. cit. (see reference 9). proximate determinants had less of an ef- the analysis, the study results imply that 20. A. K. Blanc and N. Rutenberg, “Assessment of the fect in Kenya than in other countries for in many countries women are resorting to Quality of Data on Age at First Sexual Intercourse, Age at First Marriage, and Age at First Birth in the Demo- which data were collected in the 1970s. illegal, possibly unsafe, means of induc- graphic and Health Surveys,” in IRD/Macro Interna- The presence of some fertility-reducing ing abortion. They indicate that abortion tional, 1990, op. cit. (see reference 18). effect of abortion in the estimate of aver- is widely used to control fertility; that the 21. L. Moreno and N. Goldman, “Contraceptive Failure age total fecundity would yield, in our use of abortion is geographically clus- Rates in Developing Countries: Evidence from the De- residual estimation technique, an index of tered; that in settings where fertility is de- mographic and Health Surveys,” International Family 17: abortion that exceeds 1 in countries where clining, abortion has an increasing influ- Planning Perspectives, 44–49, 1991. the true incidence of abortion is negligi- ence over time; and that, on average, the 22. J. G. Hutchings, G. Perkin and L. Saunders, “The Ef- fect of Contraceptive Technology on the Program Envi- ble. Thus, an underestimate of total fe- fertility-reducing influence of abortion is ronment,” in R. J. Lapham and G. B. Simmons, eds., Or- cundity might partially explain the two similar in magnitude to that of contra- ganizing Effective Family Planning Programs, National values that exceed 1 in our analysis ceptive use, the other principal method of Academy Press, Washington, D.C., 1987; S. Thapa, D. (Kenya 1988–1989, 1.01; Rwanda, 1.06). In controlling fertility. Hamill and P. Lampe, “Continuation and Effectiveness of Programme and Non-Programme Methods of Fami- both countries, the minimal use of con- ly Planning in Sri Lanka,” Asia-Pacific Population Journal, traceptives to limit fertility would be con- References 8:73–87, 1992; and A.K. K. Sayila, “Existing Apathy,” In- sistent with negligible use of abortion. 1. L. S. Liskin, “Complications of Abortion in Develop- tegration, June 1992, pp. 38–40. ing Countries,” Population Reports, Series F, No. 7, 1980; 23. I. Shah, “Comparative Analysis of Contraceptive These examples suggest that the estimate S. K. Henshaw, “Induced Abortion: A World Review, Method Choice,” in Proceedings of the Demographic and of 15.3 for total fecundity includes some 1990,” Family Planning Perspectives, 22:76–89, 1990; and Health Surveys World Conference, Vol. 1, IRD/Macro In- T. Frejka, “Induced Abortion and Fertility,” Internation- small effect of induced abortion, and that ternational, Columbia, Md., USA, 1991, pp. 617–639. the use of 15.3 in our rearranged equation al Family Planning Perspectives, 11:125–129, 1985. 24. J. Cleland et al., The Determinants of Reproductive 2. D. Huntington, B. Mensch and N. Toubia, “A New Ap- results in slight underestimates of the ef- Change in Bangladesh: Success in a Challenging Environment, fect of abortion on fertility.* If the extent of proach to Eliciting Information About Induced Abor- Regional and Sectoral Studies, World Bank, Washington, tion,” Studies in Family Planning, 24:120–124, 1993. this underestimation is not balanced by an D. C., 1994. 3. B. A. Anderson et al., “The Validity of Survey Re- 25. S. K. Henshaw, 1990, op. cit. (see reference 1). overestimation caused by factors such as sponses on Abortion: Evidence from Estonia,” Demog- overlapping indices (for example, use of raphy, 31:115–132, 1994. 26. Ministry of Health of Turkey, Hacettape University contraceptives while breastfeeding) or con- Institute of Population Studies and Macro Internation- 4. T. Baretto et al., “Investigating Induced Abortion in al, Turkish Demographic and Health Survey, 1993, Ankara, current use of multiple contraceptive meth- Developing Countries: Methods and Problems,” Studies Turkey, 1994. ods, our estimates will slightly underesti- in Family Planning, 23:159–170, 1992; and L. Remez, “Con- 27. J. Bongaarts and R. G. Potter, 1983, op. cit. (see refer- mate the effect of abortion on fertility. fronting the Reality of Abortion in Latin America,” In- ternational Family Planning Perspectives, 21:32–36, 1995. ence 10). 5. E. F. Jones and J. D. Forrest, “Underreporting of Abor- 28. S. Tezcan, “A Comparative Study of Induced Abor- Conclusion tion in Surveys of U.S. Women: 1976 to 1988,” Demogra- tion in Turkey Using the Randomized Response Tech- While the technique proposed here of re- phy, 29:113–126, 1992; and B.A. Anderson et al., 1994, op. nique Versus Direct Questioning,” dissertation, Dept. of arranging Bongaarts’s proximate deter- cit. (see reference 3). Epidemiology, University of North Carolina, Chapel Hill, N. C., USA, 1977. minants model to estimate indirectly an 6. S. Singh and D. Wulf, “Estimated Levels of Induced Abortion in Six Latin American Countries,” Internation- 29. L. S. Liskin, 1980, op. cit. (see reference 1); and Min- istry of Health of Turkey, Hacettape University Institute *It is reassuring, however, that in the residual estimation al Family Planning Perspectives, 20:4–13, 1994. of Population Studies and Macro International, 1994, op. equation, the value of 15.3 can be replaced with values 7. L. Remez, 1995, op. cit. (see reference 4). of 14 or 16 without having much effect on the resulting cit. (see reference 26). estimate of the index of abortion. 8. K. G. Foreit and D. L. Nortman, “A Method for Cal- (continued on page 137)

114 International Family Planning Perspectives Induced Abortion… minantes inmediatos del modelo de Bongaarts timation indirecte de la prévalence de l’avorte- (continued from page 114) para permitir el cálculo del índice de abortos ment dans 21 pays en voie de développement, 30. F. Arnold, 1990, op cit. (see reference 18). usando los otros tres principales determinan- par le réarrangement du modèle de détermi- tes inmediatos de la fecundidad (matrimonio, nants immédiats de Bongaarts pour permettre 31. J. Bongaarts, “The Relative Contributions of Biolog- ical and Behavioral Factors in Determining Natural Fer- uso de anticonceptivos y falta de susceptibili- le calcul de l’indice d’avortement en fonction tility: A Demographer’s Perspective,” in R. Gray, H. Leri- dad al embarazo), más el promedio de la ferti- des autres principaux déterminants immédiats don and A. Spira, Biomedical and Demographic lidad total y la fecundidad global. En prome- de la fécondité (le mariage, la pratique contra- Determinants of Reproduction, Clarendon Press, New York, dio, el aborto parece influir en la fecundidad en ceptive et l’insusceptibilité du post-partum à 1993. forma similar al uso de anticonceptivos. Esta la grossesse), de l’indice synthéthique de la fer- 32. J. Bongaarts, 1982, op. cit. (see reference 9). influencia parece ser particularmente sólida en tilité moyenne et de l’indice synthétique de la 33. Ibid. los cuatro países latinoamericanos estudiados, fécondité. En moyenne, l’avortement semble ex- donde el aborto reduce la fecundidad en un ercer sur la fécondité une influence compara- 34. Ibid. 38–55%. Por el contrario, el efecto del aborto ble à celle de la pratique contraceptive. Cette in- 35. Ibid. en la reducción de la fecundidad es solamente fluence semble particulièrement forte dans les 36. J. Bongaarts and R. G. Potter, 1983, op. cit. (see refer- del 6–19% en el Medio Oriente y del 0–32% quatre pays d’Amérique latine soumis à ence 10). en el Africa. En cinco países de los que se dis- l’analyse, où l’avortement réduit la fécondité 37. J. Bongaarts, 1978, op. cit. (see reference 16). pone de dos juegos de datos de la EDS, este efec- de 38 à 55%. Par contre, l’effet amoindrissant to de reducción debido al aborto parece haber de l’avortement sur la fécondité se limite à une Resumen aumentado a través del tiempo. plage de 6 à 19% au Proche-Orient et de 0 à El análisis de los datos de las Encuestas Demo- 32% en Afrique. Dans les cinq pays pour gráficas y de Salud ofrece cálculos indirectos Résumé lesquels deux ensembles de données EDS sont sobre la prevalencia de aborto en 21 países en Une analyse des données d’Enquêtes démo- disponibles, cet effet réducteur semble avoir aug- desarrollo. Esto se logra reordenando los deter- graphiques et de santé (EDS) procure une es- menté avec le temps.

Volume 22, Number 3, September 1996 137