PHN-8709 DETERMINANTS OF FERTILITY CONTROL IN EGYPT, 1979-80 by Richard Easterlin Eileen H. Crimins Ibrahim Khodair May 1987 Population, Health and Nutrition Departmeni World Bank The World Bank does not accept responsibility for the views expressed herein which are those of the author(s) and should not be attributed to the World Bank or to its affiliated organizations. The findings, interpretations, and conclusions are the results of research supported by the Bank; they do not necessarily represent official policy of the Bank. The designations employed, the presentation of material, and any maps used in this document are solely for the convenience of the reader and do not imply the expression of any opinion whatsoever on the part of the World Bank or its affiliates concerning the legal status of any country, territory, city area, or of its authorities, or concerning the deliminations of its boundaries, or national affiliation. PHN Technical Note 87-9 DETERMINANTS OF FERTILITY CONTROL IN EGYPT, 1979-80 ABSTRACT Based on a "supply-demand" theory of contraceptive use this paper empirically investigates variations in deliberate fertility control among Egyptian women aged 35-44 in 1979-80. The results point to the importance, in determining contraceptive use, of motivation, measured as the excess of the potential supply of children over desired family size. This measure typically performs better than alternative measures of motivation suggested by the literature, such as whether the respondent wants more children, the number of living children a couple has, desired family size, and the couple's actual number of unwanted children. Other studies (for Egypt in 1974-75 and 1979 and Sri Lanka in 1975) have yielded very similar results. Along with motivation, access to family planning services also significantly influences contraceptive use among the rural population, for whom the "access" measures are better than for the urban. But the importance of access is considerably less than that of motivation, suggesting that the most important stimulus to contraceptive use is a situation where households envisage unregulated fertility as leading to a family size considerably greater than that desired. A comparison of husband and wife responses on several key variables suggests that so far as the measurement and explanation of fertility control is concerned, wives' data taken alone are better than husbands. Combining husbands' and wives' responses, however, sometimes improves moderately the statistical explanation obtained. Prepared by: Richard A. Easterlin, Eileen M. Crimmins, and Ibrahim Khodair Consultants to the World Bank May 1987 + Determinants of Fertility Control in Egypt, 1979-80 by Richard A. Easterlin, Eileen M. Crimmins, and Ibrahim Khodair* At the start of this decade, among Egyptian women nearing the end of their reproductive careers who had had at least two children, almost six in ten had at some time used some method of avoiding conception, usually the pill or IUD. On the average, users of deliberate fertility control had started about 11 years prior to the survey date, although about 3 percent had begun at least two decades earlier, and another 3 percent had started only in the last year. How can one explain these variations in the use of deliberate fertility control? Clearly, to the extent one can explain past use-, one will be better able to predict future use and devise policies to promote greater use. The explanation of observed variations in use of fertility control is the central con- cern of this paper. Generally speaking, use of fertility control is believed to depend on the strength of a household's motivation to limit family size, the favorableness of its attitudes toward family planning methods, and its ease of access to such methods. In seeking to explain household varia- tions in fertility control, this paper employs a theory that suggests certain measures of motivation, attitudes, and access, and fits an econometric model based on this theory to World Fertility Survey (WFS) data for Egypt for 1979-80, following the lines of a previous analysis * The -authors wish to acknowledge the excellent assistance of Ramesh Amatya, Donna Hokoda, and Pierre Landry, and financial support of the University of Southern California. 2 by two of the author's for two other WFS countries (Easterlin and Crimmins, 1982). As shall be seen, the fit of the model is good by the usual satistical tests. Several procedures are followed to evaluate the results. First, the most distinctive feature of the model is the special measure of motivation for fertility control that is used. One way of judging the results is to see how well models using alternative motivation measures perform. Second, to determine whether the results are specific to this particular data set, comparisons are presented with alternative data sets for Egypt to which a similar model has been fitted. Both analyses strengthen the present results. The responses of women near the end of their reproductive careers do not necessarily provide reliable indicators of the attitudes toward family planning methods or access to such methods that shaped the fertility control decision. Fortunately, several family planning program measures for the rural population included in the community module of the Egyptian fertility survey (EFS) provide better measures of access. The model is re-estimated for the rural population, therefore, using these new measures of access. This analysis sheds tentative light on the importance in promoting use of control of the family planning program vis-a-vis motivation. The conclusion is that motivation is most important, but the family planning program does make a contribution. Finally, using data from the husbands survey in the second round of the EFS, an assessment is made of the reliability of wives' responses for the measurement and explanation of fertility control. The results uphold the value of the wives' data, although the availability of husbands' data as well yields a modest improvement in the results. 3 Theory As mentioned, decisions regarding deliberate fertility control are commonly seen as involving three types of considerations: motivation, attitudes, and access (Freedman, 1961-62; Petersen, 1969). The motiva- tion for fertility regulation is viewed as stemming from concerns about having too many children or having them too soon. Attitudes toward fer- tility regulation embrace both very broad notions of the acceptability of family planning in general as well as feelings about the appropriate- ness of specific practices. Access pertains to the availability (including both time and money costs) of fertility control services and supplies. In general, fertility regulation is viewed as varying direct- ly with the degree of motivation, favourableness of attitudes and extent of access. The present theory formalizes these notions in terms of three con- cepts (Easterlin 1978, Easterlin, Pollak, and Wachter, 1980): 1. Costs of fertility regulation ,(RC): this encompasses a couple's attitudes toward and access to fertility control services and sup- plies. It includes both subjective disadvantages of regulation and the economic costs of control. Family planning programs promote fertility control chiefly via lower regulation costs, by increasing access to fertility control through offering below cost family planning services, and promoting favorable attitudes toward the practice of deliberate family size limitation. 2. Desired family size (Cd): this is the number of surviving children a couple would want in a 'perfect contraceptive society', one where costs of regulation were negligible (see Bumpass and Westoff 1970). It reflects the taste, income and price considerations of the usual .economic theory of household decision making, including both the economic and non-economic returns from children as well as their costs. 3. Potential family size (Cn): this is the number of surviving chil- dren a household would-have if it did nothing deliberately to regu- late its fertility. Potential family size is the product of a couple's natural fertility (N), and its child survival rate (s). Both natural fertility and potential family size may be well below the biological maximum because of cultural conditions that inadver- tently reduce fertility and family size, such as prolonged breast- feeding. The excess of potential family size over desired family size, Cn-Cd, is the number of unwanted children a couple would have in the absence of deliberate fertility control. The larger this excess, the greater is the potential burden of unwanted children, and consequently the greater is the household's motivation to limit its fertility. It is worth stressing that motivation is determined by potential as well as desired family size. Sometimes motivation is simply identified with desired family size and it is assumed that only if this decreases will motivation for fertility control grow. In fact, an increase in poten- tial family size can increase motivation, even if desired family size remains constant, because it increases the potential number of unwanted children. An increase in potential family size might arise from an increase in a couple's natural fertility, improved chances of child sur- vival, or both. The value of Cn-Cd may be negative, indicating that a household is in a 'deficit fertility' situation, that is, that it is unable to pro- 5 duce as many children as it would like to have. In-this case, there is no motivation to limit fertility and a couple would. have as many chil- dren as possible; in other words 'natural fertility' would be a logical outcome of the couple's underlying reproductive conditions. Even if the value of Cn-Cd is positive, however, it does not neces- sarily follow that a couple will deliberately control its fertility. Against the pressure to do so must be weighed the costs of fertility control, RC, that is, the subjective disutility and economic costs attached to the actual use of control. If RC is high and the motivation (Cn-Cd) low, then a couple may feel that the disadvantages of unwanted children are less than those associated with deliberately restricting fertility, and hence may forego fertility control. Again, natural fer- tility may be a rational response to the couple's- basic situation. In general, the probability of adopting control is higher. the greater the degree of motivation (the excess of potential over desired family size) and the lower the costs of regulation. The theory thus leads to comparing households in terms of motivation for control, Cn-Cd, and costs of control, RC, to see if these theoretical determinants are, in fact, systematically associated in the expected way with differences. in the use of fertility control, and what is the relative importance of each. Model For empirical application to household data, this theory is expressed in a two equation model. The first equation links fertility (F) to a set of "proximate determinants," including use of deliberate fertility control, and the second links the use of deliberate control 6 (U) to potential family size (Cn), desired family size (Cd), and regu- lation costs (RC). Thp theory itself leads explicitly only to the second equation, where use of control is seen as a function of motivation (the excess of potential over desired family size, Cn-Cd), and regulation costs, RC. But empirical. implementation of the theory poses a difficult problem, for which the first equation is needed: how can one estimate the number of children a household would have if it did nothing deliberately to regulate its fertility, when, in fact, a number of households did con- trol their fertility? The answer is to estimate from data for both users and nonusers of fertility control an equation that accounts for actual fertility in terms of its proximate determinants, including use of fertility control, and then, by setting the value for use equal to zero for each household, to estimate "natural fertility," that is, fer- tility in the absence of deliberate control -- hence, the first equation of the model. As has been indicated, this equation corresponds in con- cept to what demographers call a "proximate determinants" analysis (Davis and Blake, 1956; Bongaarts, 1978, 1980); to economists, proximate determinants analysis might be thought of as a type of "production func- tion" for children (Easterlin, Pollak, and Wachter, 1980). Specifically, the proximate determinants equation used here is (1) F = ao iX i +8 U + Z i = 1,...7 where the variables are F children ever born, . I duration of marriage in years, X2 first birth interval in months, X3 second birth interval in months, .5 7 X 4 not secondarily sterile (NSS), X months of breastfeeding in last closed interval, 5 X6 proportion of pregnancy wastage, X7 proportion of child mortality, U use of contraception measured either in years since first use or ever use, and s a stochastical disturbance. In terms of Bongaart's (1980) proximate determinants classification, the variable, U, is a measure of deliberate fertility control. The El through X7 variables comprise collectively a set of natural fertility determinants with duration of marriage reflecting the extent of exposure to intercourse; first birth interval, fecundability; pregnancy wastage, spontaneous intrauterine mortality; "NSS", secondary sterility; and the combination of second birth interval, breastfeeding in the. last closed interval, and child mortality, the duration of postpartum infecundabili- ty. As regards the parameters of this equation, it is hypothesized that the cumulative fertility of a continuously married woman near the end of her reproductive career would be greater: 1. the less the use of deliberate fertility control by her or her husband ( < 0), 2. the longer her period of exposure, as measured by duration of marriage (alI > 0), 3. the greater the couple's fecundability, taken to vary inverse- ly with the length of the. first birth interval (a2 < 0), 4. the shorter her period of secondary sterility (a4 > 0), 5. the lower her rate of foetal wastage (miscarriages, sponta- neous abortions, and stillbirths), and hence physiological problems of reproduction (a6 < 0), and 8 6. the shorter the nonsusceptible period, reflected here in a shorter second birth interval, shorter duration of breast- feeding in the last closed interval, and a higher rate of child mortality (a3 < 0, a5 < 0, a7 > 0). Although no specific precedents exist for estimation of such an equation at the micro level, macro level techniques provide valuable guidance (Boagaarts, 1980). It is-unlikely, of course, that individual households form specific numerical estimates of their natural fertility and potential family size. Rather, the present approach should be seen as an attempt at gen- eralizing on how women pick up clues about their childbearing prospects. Thus, the pace of early childbearing, prospective exposure (age at mar- riage, duration of marriage), foetal loss experience, evidence of fecundity problems such as irregular. menstruation, infant mortality experience, etc. all probably contribute to a woman's assessment of her natural fertility and potential family size. The second equation is (2a) U = 00 + 6(Ca-Cd) + yRC + p, where the new variables are Cd desired family size, RC regulation costs, Cn potential family size, the number of surviving children in the absence of contraception (= sN), N natural fertility, total births in the absence of contracep- tion (= a0 + aX ), s the child survival rate (= 1 - X ), and pz a stochastical disturbance. By hypothesis, the coefficient on motivation for fertility control, Cn-Cd, should be positive (6 > 0) and that on regulation costs, RC, neg- ative .(y < 0). In theory, Cn and Cd should have the same coefficient, because the behavior of two couples differing on Cn-Cd by some given magnitude and in other respects identical should be unaffected by whether the source of the difference in motivation is Cn, Cd, or both. This expectation was examined empirically by estimating coefficients separately for Cn and Cd and testing the significance of their differ- ence. The results confirmed the theoretical expectation. The endogenous variables in the two equations are children ever born (F) and deliberate use of fertility control (U); the exogenous variables are desired family size (Cd), regulation costs (RC), and the set of natural fertility determinants (X through X7). Because the disturbance terms in the equations may be correlated due to omitted variables or errors of measurement, a two stage estimation procedure is used. First, the reduced form equation is estimated, (2b) U = A0 +XX - 6Cd + yRC + p, in which use of deliberate fertility control is expressed as a function only of the exogenous variables. If U is measured as time since first use, a tobit procedure with maximum likelihood estimation is employed; if U is measured as a zero/one variable (ever use), a logit procedure is used. The resulting parameters of (2b) are used to obtain estimated values of years since first use (or probability of use), which are then used along with the observed values on X through X to estimate by OLS the coefficients of equation (1). Finally, each household's potential supply of children, Cn, is estimated as the product of its natural fer- tility, obtained as a0 + I X i for the household's X through X7, and 10 its child survival rate. The household estimates of Cn are then used along with the household values of the demand for children, Cd, and regulation costs, RC, to estimate the parameters of equation (2a) by tobit or logit depending on the nature of the use variable. The methodology used here is an advance over that in the previous .WFS study by Easterlin and Crimmins (1982) in that it employs estimation procedures (tobit or logit) more appropriate to the nature of the -use variable, and in allowing for possible correlation between the error terms of equations (1) and (2) by a two stage estimation procedure. When use is measured in terms of years since first use, it can assume only zero or positive values, though there may be some households for whom conceptually a negative value would be meaningful. An example is a household that is unable to have as many children as it would like and, if the technology were available and cheap enough, would increase its fertility. Conceptually, such a household would, if it were possible, have negative "use" (family size "supplementation") rather than positive use (family size limitation). Because the use variable is truncated at zero, however, a number of households with varying degrees of low or negative motivation (Cn-Cd) are clustered at this value and an OLS regression line fitted to the data gives a biased estimate of the rela- tionship of use to motivation. The solution to this problem is to employ tobit estimation (see Berk, 1983; Maddala, 1983). A somewhat similar problem arises when ever use is the dependent variable. In this case the dependent variable can assume only values of one or zero (use or nonuse). A regression line fitted by ordinary least squares may yield predicted values lying outside this range, whereas the logit procedure constrains the predicted probabilities of use to lie within the zero to one range (Pindyck and Rubinfeld, 1981). A compari- son of the methodology used here with both ordinary least squares and two stage least squares (not using tobit or logit) shows that while the signs and significance of the coefficients are almost always the same, the present procedure yields somewhat more plausible estimates of magni- tude (Crimmins and Easterlin, forthcoming 1984; Easterlin and Crimmins, forthcoming 1985, ch. 4). Data and Measurement of Variables The study population is currently married females close to the end of their reproductive careers, those aged 35-44, who have been married only once, are still married, and who have had at least two live births. The restriction to continuous marriages minimizes conceptual and measure- ment problems associated with marital disruption. Childless and parity- one women were omitted to avoid biasing the results in favour of the theory, because this group consists almost wholly of women who have never regulated their fertility and who lack the motivation to do so because they have been sterile throughout their reproductive careers or have severe fecundity problems. The actual measures used in the empirical analysis are approxima- tions to the ideal, inevitably reflecting the constraint of data limita- tions. The construction of the variables is detailed in Appendix A. In the present section we discuss some for which special estimation proce- dures were required and/or some problems with the variables as measured. Use of fertility control, U, refers to the use of contraception or contraceptive sterilization, as reported by survey respondents. Contra- ception includes the use both of what WFS calls "efficient" methods 12 (pill, IUD, diaphragm, condom, and injection) and "inefficient" methods (douche, withdrawal, abstinence, and rhythm). Respondents who report no use of.any method are assigned fertility control values of zero. For those who report use, a rough estimate of years since first use is made by differencing current age and estimated age at first use. A second measure of use employed in the analysis is ever use of fertility control, a zero/one variable with ever users assigned a value of one. Years since first use may be thought of as an approximation to household differences in duration of use, although it substantially overstates the latter in failing to allow for lapses from use. In the proximate determinants analysis for Egypt, as well as several other countries, the statistical explanation of household differences in fer- tility is consistently improved when fertility control is measured by years since first use rather than ever use. Hence years since first use is the preferred measure of fertility control here, though results are also presented for ever use. Less than 10 percent of regulators used deliberate fertility con- trol prior to their second birth, but for those who did the second and, possibly, first birth interval variables are flawed as indicators of the postpartum nonsusceptible period and fecndability. For these users, the observed birth interval values are replaced by the mean values for these intervals of those who did not regulate before the second birth. In the case of secondary sterility, the obvious choice for a mea- sure is the response to a question on whether or not the respondent thought she could bear another child. However, this measure yields a proportion secondarily sterile, 11 percent., that is unusually low by comparison with the estimates of Henry and Vincent for women aged 40 13 (reported by Pittenger, 1973:115). The measure adopted here, therefore, classifies women as secondarily sterile if (a) they reported a fecundity impairment, or (b) they were not currently regulating their fertility, had had no child in the last five years, and were not pregnant. This measure gives a proportion secondarily sterile, 24 percent, closer to the estimates of Henry and Vincent, 32 to 33 percent. Those classified as secondarily sterile by this definition were assigned a value of zero on the variable; all others, a value of one. Ideally, the measure of secondary sterility should be independent of knowledge about a woman's use or non-use of fertility control, and in that respect this surrogate measure is flawed. The problem of potential bias arises from the fact that current use of fertility control is one factor affecting the estimation of secondary sterility, which in turn is one (of seven) variables entering into the estimate of Cn, a key varia- ble in explaining time since first use, the primary dependent variable in this analysis. However, it seems likely that this potential bias, if it does exist, is of negligible proportions. First of all, among the total population, the likelihood that the measure seriously mis-esti- mates secondary sterility - in particular, that there is likely to be a substantial number of current users who are, in fact, secondarily sterile - seems small. Second, as noted, the percentages yielded by the measure are fairly consistent with more robust estimates for similar populations. Most important, two tests run in a previous study demon- strated that the explanation of fertility control did not depend on the NSS measure (Easterlin and Crimmins, forthcoming 1985, chapter 4, appendix). Therefore, while the measure of secondary sterility is technically less than ideal, it appears that any distortions it may introduce are not substantive in a quantitative sense. 14 The measure of desired family size, Cd, is the response to the fol- lowing question: "If you could choose exactly-the number of children to have ii% your whole life, how many would that be?" The value of the response to a question of this type is sometimes questioned. To the extent that scepticism arises from lack of correlation between observed fertility and desired family size, it is not relevant here. The present framework views desired family size as only one of a number of fertility determinants, and there is no expectation that desired size alone should be highly correlated with fertility. A more serious objection is that the response reflects the respon- dent's state ex post facto, that is after, not before, decisions regard- ing fertility and fertility control. Thus actual family size may bias upward responses to desired family size, because children unwanted before the fact are reported as desired after the fact.. There is, how- ever, some evidence that the magnitude of the bias is not great enough to invalidate the usefulness of responses on desired size (Knodel and Prachuabmoh, 1973). Especially to the point is a recent study of the change in family size preferences of two cohorts of Taiwanese women between 1965 and 1973 (Jejeebhoy 1981). No evidence was found of an increase in desired size, despite the fact that these cohorts were at a stage in their reproductive career when most women were shifting from having fewer children than were desired to having more than desired. McClelland (1983, p. 319), after a critical analysis of the literature, concludes that "it is not unreasonable to treat measures of family size desires as measures of demand." Conceptually, in measuring the costs of fertility regulation one would like data that reflect a household's subjective attitudes towards 15 the use of fertility control, their information about methods of control, and the economic costs of obtaining additional knowledge about techniques of control and of purchasing supplies or services needed for control. Ideally such data would antedate the actual decision on fer- tility regulation, because one consequence of a decision to use control is likely to be a positive shift in users relative. to nonusers with regard to both knowledge of methods and favourableness of attitudes. The measure(s) used must, of course, be available for all households in the study population;.knowledge, say, of nonusers' attitudes toward fer- tility control is of little value unless one knows how they differ from the attitudes of users. The available measures in the core module of WFS fall short of the ideal. In most of this paper, we use as the measure of regulation costs the number of methods known to a respondent and reported without special prompting. This measure proved to be the best in previous studies, and its use here facilitates comparison with those studies. However, this measure is subject to the bias just mentioned, because use may determine knowledge, as well as vice-versa. Fortunately, the community module included in the EFS provides mea- sures of regulation costs that are free from this bias, though for the rural population only, and the latter part of this paper uses a variety of these regulation cost measures. The community survey reports for the rural population a number of conditions reflecting access to family planning services, such as the availability and date of establishment of a family planning clinic in a community, number of personnel of various types, and, if a clinic is not available, the distance to the nearest one. In the analysis of these data, for any particular variable each 16 rural respondent is assigned the value of the variable for his community irrespective of whether the respondent is a user or nonuser, in other words,-access to family planning services is assumed to be the same for all members of a given rural community. This approach, among those recommended for studying family planning program effects -in a recent critical survey by Hermalin (1983), is sometimes referred to as "multi- level analysis," because it combines data on individuals with those on communities. Empirical Results Egyptian Fertility Survey, 1979-80 As one would expect, use of fertility control in Egypt tends to reduce fertility -- each year since starting use counting for a .17 reduction in children ever born (Table 1, column 1). The signs of the other variables in the proximate determinants equation are also in the hypothesized direction and, as in the case of the use variable, statis- tically significant. Thus, children ever born varies directly with extent of exposure, as measured by duration of marriage, and directly with fecundity, as measured by the secondary sterility and first birth interval variables (the negative sign on the latter shows that low fer- tility goes with a long first birth interval, i.e., low fertility). The three indicators of the length of the postpartum nonsusceptible period -- second birth interval, duration of breastfeeding in the last closed interval, and proportion of child mortality -- also have the expected signs. A longer second birth interval and longer duration of breastfeeding contribute to lower fertility, while high child mortality shortens the nonsusceptible period- and raises fertility. Finally, fertility is lower among women with high rates of foetal wastage. 17 The results when the fertility control variable 'is ever use, rather than time since first use, are highly consistent (column 2). The coef- ficient on ever use implies that, on the average, the effect of use is to reduce fertility by 1.8 births. This corresponds closely to the effect of fertility control -implied by the equation in column 1. Thus, if the coefficient in column I of -.17 on time since first use is multi- plied by the average time since users started controlling, 11 years, one obtains an average reduction in fertility of 1.9 children for users, compared with the 1.8 reduction indicated by the ever use variable. The other variables in column 2 have coefficients similar in sign and magni- tude to those in column 1, and are also statistically significant, except for the breastfeeding variable, which is not significant. How- ever, for all of these variables except foetal wastage, the absolute value of the coefficient in. column 1 is greater than column 2, and the fit, as shown by the coefficient of determination, is slightly better for the equation in column 1 than that in column 2. The implication is that, despite its imperfections, years since first use is slightly better than ever use as a measure of household differences in fertility control. According to the theory, use of control should vary directly'with the motivation for fertility control, as measured by the excess of a household's potential over desired family size. To measure motivation, the proximate determinants equations of Table 1 are used to estimate for each household potential family size in the absence of fertility con- trol, and desired family size is then subtracted from this. For the population as a whole, unregulated fertility is estimated to yield an average of 7.7 births per woman, which, taking account of mortality, 18 translates into 5.8 surviving children. Since desired family size is about 4.5 children, the implication is that in the absence of deliberate fertilipy control family size would, on the average, exceed that desired by about 1.3 children. The principal question of interest here is whether household differences in use of fertility control vary directly with this measure of motivation, in other words, are the households that started using fertility control earliest those -in which the potential number of unwanted children, Cn-Cd, is above average? Simple bivariate correlations of use and motivation, as measured by Cn-Cd, indicate that this is indeed the case -- earlier use is significantly associated with higher motivation (Table 2, line 1). This is true not only for the total population, but *also for the population of regulators alone. It is also true when the use measure is ever use rather than years since first use. (Ever use can, of course, only be used in an analysis for the total population, because all regulators have a value of one on this measure.) Lines 2-6 of Table 2 present similar correlations with use of a number of alternative measures of motivation, suggested by the litera- ture (for a good overview, see United Nations, 1981; see also CAPMAS, 1983, vol. II and Khalifa et al., 1982). One would suppose that women who want no more children would have greater use of fertility control, and for the total population this turns out to be the case. This mea- sure of motivation, however, does not show a signficant association with variations in use within the regulating population, no doubt because most women in this group report wanting no more (the proportion is .89). One might also expect that use would be greater among those whose desired family size is smaller, Cd, and the correlations in every column 19 of Table 2 confirm this. In-addition, it is sometimes hypothesized that use would be greater among those who have more living children. For this measure of motivation (C), however, the results are not very impressive -- only two of the three correlations are significant, and then only weakly. The actuali as distinct from potential, number of unwanted children, C-Cd, fares better than C for the total population, but is not significantly associated with use among the regulators. In general, then, a number of measures of motivation commonly men- tioned in the literature show significant correlations with use of fer- tility control, though there are some exceptions. However, comparison of the performance of these measures with the theoretically preferred one, potential unwanted children (Cn-Cd), suggests that the latter is somewhat better -- in every column it shows the highest correlation with use. Moreover, this measure performs better than either of its com- ponents, Cn or Cd (compare lines 1, 3, and 4). This is as it should be, because it is the difference between the two that comes closest to cap- turing differences in household motivation. Nor does it seem that the difference measure, Cn-Cd, is dominated primarily by one of the two com- ponents -- both Cn and Cd are significantly associated with use, but neither one is consistently better. Table 2 presents also the simple correlation of use with regulation costs, as measured by number of methods known to a respondent and reported without special prompting. The expectation is that the corre- lation of methods known with use would be positive, because greater knowledge would imply lower costs and hence more use. For the total population, methods known does show positive correlations with use with orders of magnitude approaching those for Cn-Cd. For the regulating 20 population, however, the correlation, though significant, is quite low. This suggests that the high correlation for the total population may reflect the bias mentioned earlier -- that. use leads to greater knowl- edge rather than vice-versa -- because when nonusers are eliminated from the study population there is much less association between use and knowledge of methods. Multiple regressions of use on motivation and number of methods known add little to the simple correlations reported in Table 2 (see Tables 3 and 4). The single best measure of motivation, judged in this- case by the t-statistic and chi-square values, continues to be the theo- retically preferred measure, Cn-Cd. For the total population, number of methods known has t-statistics approaching those of Cn-Cd, but for the regulating population, this measure drops to bare significance. When Cn-Cd is replaced by its two components separately, there is little change in the goodness of fit, and a chi-square test indicates that there is no significant difference between the coefficients on Cn and Cd. Comparisons with Other Studies The 1979-80 results reported here for Egypt are remarkably consis- tent with similar analyses of several other data sets. There is avail- able for comparison two studies based on Egyptian data -- one, an unpub- lished analysis for 1974-75 done by Mahmoud S.A. Issa in collaboration with Crimmins and Easterlin, and one, a 1979 rural sample analyzed by Allen C. Kelley and Robert M. Schmidt (1983). (For a description of the 1974-75 sample, see Issa, 1981.) Also included here is the analysis by Crimmins and Easterlin of Sri Lanka WFS data for 1975, because of the * 21 similarities to Egypt. In the present comparisons, the equations for both the proximate determinants analysis and determinants of use are estimated by ordinary least squares, the technique used in the Crimmins- Easterlin 1982 WFS study. Although less satisfactory methodologically, this is the only option available for comparison. As shall be seen, the quantitative results are usually the same in terms of sign and signifi- cance, although the absolute magnitudes of the coefficients obtained from OLS are generally biased downward. Comparison of the OLS proximate determinants equations derived from the four data sets shows a number of similarities (Table 5). The signs on the coefficients are the same in all four samples, and, with surpris- ingly few exceptions, the coefficients are similar in magnitude, and significant. The coefficient on time since first use, for example, is significant and around -.10 in three of four cases; the coefficient on duration of marriage is significant and around .30 in all four equa- tions; that on first birth interval, significant and around -.026; and so on. The OLS coefficient on breastfeeding is not significant in the WFS data set for Egypt, but it is, although at a relatively low level, in the other two for which estimates are available (rural Egypt, 1979, and Sri Lanka). The coefficient of determination for all four equations lies between .50 and .55. Comparison of the OLS results for 1979-80 in Table 5 with those obtained by more refined methods (Table 1) shows that the signs of the coefficients are the same, but that the improved methodology usually yields absolute magnitudes somewhat greater, especially on the use vari- able. A similar comparison for Sri Lanka confirms this, and, interest- ingly, yields an increased coefficient on use with the improved method- 22 ology like that obtained for Egypt, -.17 (Easterlin and Crimmins, forth- coming 1985). In the Kelley-Schmidt analysis of the 1979 rural Egypt sample-a more refined technique than OLS is also employed, and with that technique a significant negative coefficient on use is obtained. Turning to bivariate correlations between time since first use and various measures of motivation, one is struck by the remarkable consis- tency between the two Egyptian data sets for which such comparison is possible (Table 6). In both 1979-80 and 1974-75, use of fertility con- trol has the highest correlation with the theoretically preferred mea- sure of motivation, Cn-Cd, and this is true of both the total population and regulating population alone. Moreover, the pattern of correlations in the two data sets is strikingly similar, as a comparison of columns I and 2 shows. Essentially the same pattern of correlation results also shows up for Sri Lanka (column 3). Results like those in Table 5 do much to buttress one's confidence in the robustness of the findings from the 1979-80 EFS. Table 7 extends the comparisons to multiple regressions of use on Cn-Cd and number of methods known, for which all four data sets are available. All coefficients are significant, and the signs are as expected, except for that on number of methods known in the 1974-75 Egyptian survey. The magnitudes of the coefficients on motivation tend to range around 1.0, though there is somewhat greater variability here than for the proximate determinants equation. Comparison of the OLS results for 1979-80 in Table 7 with the tobit results in Table 3 suggests, as in the case of the proximate determinants equation, that the absolute values of the coefficients obtained with OLS are biased downward. 23 The Impact on Use of Access to Family Planning Services The results so far point consistently to the importance of motiva- tion, as measured by potential unwanted children (Ca-Cd), in explaining household differences in fertility control. Regulation. costs, as proxied by number of methods known, is usually significantly associated with use of control, but, as noted, the cause-effect relation is ambig- uous. For the rural population, however, there are superior measures of regulation costs available from the community module of the EFS, not subject to this problem of bias, reflecting access to family planning services. In this section, we replace number of methods known by these new measures of regulation costs. The purpose is to see whether access to family planning services significantly affects use, and, if so, how important in determining use such access is vis-a-vis motivation. The means and standard deviations of the new measures are given in Table 8, Panel A, columns 1 and 2. As can be seen, a number of health, as well as family planning, measures are included. The measures fall into four groups: (1) age of family planning place, an indicator of the period a family planning program has been available to the members of a community (whether or not the program is located'in the community), (2) health or family planning personnel (doctors, nurses, qualified mid- wives, and dayas), (3) indicators of availability of specific contra- ceptive services (pill or IUD) at the family planning place, and (4) distance to specified health or family planning facilities. One might expect that wives who started using fertility control earlier would have had earlier access to family planning services. How- ever, the bivariate correlation between use and age of family planning place fails to support this expectation (column 3). Greater use might 24 be expected also to result from greater availability of health and fam- ily planning personnel but, with one important exception, this hypothe- sis is,not supported by bivariate correlation either. The exception is the availability of qualified midwives, which does show a significant (though weak) association with use in the expected direction. Surpris- ingly, the availability of dayas shows no significant association with use. With regard to specific contraceptive services, IUD availability shows a significant positive association with use; pill availability does not, but for eighty-five percent of the population the value of this measure is one, reflecting the widespread distribution of the pill. Also, two of the distance measures -- to rural health centers and family planning clinics -- show significant negative associations with use, i.e., the greater the distance to a rural health center or family plan- ning clinic, the less the use of fertility control. One other measure, distance to a rural health unit, shows a significant association with use, but in the wrong direction. The previous measure of regulation costs, number of methods known, has a considerably higher correlation with use than any of the family planning or health measures, a result perhaps indicative of the importance of use in promoting knowledge, rather than vice-versa. Panel B of Table 8 shows the correlations within the rural popula- tion of use with the different motivation measures. The results here look very much like those for the total population in Table 2. Examina- tion of the means, however, shnws that the average level of motivation, as measured by Cn-Cd, is considerably lower for the rural population than for the total -- in the rural population, mean potential and desired family size are about equal, whereas, in the total population, the former is greater than the latter by about 1.3 children (Table 3). 25 This is due both to higher desired size and lower potential size in the rural population compared with the total. Corresponding to the lower level of motivation, use of fertility control by the rural population is considerably lower than- by the total -- only about thirty-eight percent have ever used fertility control. To explain more fully the possible impact of family planning and health services on use of fertility control, regressions of use on these indicators were run, singly and in a number of combinations. As above, the results turned up little not shown by the simple correlations. The indicators that are consistently associated with use are qualified mid- wives per capita, distance to a rural health center, and distance to a family planning unit. In multiple regressions, availability of the IUD and distance to a rural health unit both drop to insignificant levels. There are, then, three measures of access to family planning or health services for which preliminary analysis indicates a significant effect in the expected direction on use of fertility control. How well do these indicators stand up in competition with motivation as an explanation of use? Do they, perhaps, swamp the effect of motivation, or does the latter dominate? The bivariate correlations suggest that motivation is the most important factor: Ca-Cd accounts for about 13 percent of the variance in use, while the three measures of access to family planning or health services account for only one percent each (Table 8). Multiple regression, using tobit estimation as in Table 3, gives similar results. The chi-square value for the regression of use on Cn-Cd alone is 184; adding the three access measures raises the chi- square value to 206 (Table 9, lines 1 and 2). The t-statistic for Cn-Cd is over 11, for the access measures, it is 3 or less. Replacing Cn-Cd 26 by a less satisfactory measure of motivation, wants no more, yields. similar results; that is, this motivation measure too has a more important effect on use than the access measures, though the performance of wants no more is not quite as good as Cn-Cd (Table 9, lines 3 and 4). The conclusion about the greater importance of motivation vis-a-vis access in determining use of control thus does not depend on one parti- cular measure of motivation. Wives' Versus Husbands' Responses Host surveys of fertility and family planning behavior are based on the responses of wives, and, with the exception of the community data in Tables 8 and 9, this is true too of the analyses conducted in the pre- ceding sections. In the second round of the EFS, however, a subsample of husbands was interviewed. This makes it possible to explore a number of new analytical questions relating to consistency of husbands and wives responses on certain key variables, and the reliability of wives' data alone for drawing inferences on the use of fertility control and its determinants. The analysis below is based on a matched sample of 431 husbands and wives still in their first marriage, and a combined rural and urban sample as in the initial two sections. For the reduced sample of 431 wives there are some minor differences in numerical magni- tudes from the full sample of 1728 wives on which Tables 1-7 above are based, but the effect on the conclusions is negligible, and the sub- sample may be taken as representative of the larger sample. Consistency of responses -- The present analysis compares husbands and wives responses on four measures -- use of fertility control, desired family size, wants no more, and number of fertility control methods 27 known.. The remaining measures -- children ever born, duration of marriage, first and second birth interval, secondary sterility, duration of breastfeeding, pregnancy wastage, and child mortality -- relate to facts of the couple's reproductive history on which it is generally felt that the husband is less well informed than the wife. Indeed, for a number of these measures only wives' responses are available. In con- trast, on questions of desired size, wants no more, and methods known, there may be legitimate differences between the spouses -- that is, the facts for husbands and wives may be different. In the case of use of fertility control, although this is in principle a question of fact, spouses may differ in their responses, not only because of a difference in knowledge or recall of the facts, but because questions on use of contraception probe into private sexual relations that respondents may be reluctant to disclose (cf. Koenig et al., 1984). For each of the four magnitudes of interest, Table 10 presents the means and standard deviations of the husbands' and wives' responses, and the correlation between them. For the analysis of use of fertility control, the only measure available for comparison is the less prefera- ble one, ever use, because the required data for years since starting fertility control were not obtained from husbands. The means on ever use are fairly close, .59 for husbands versus .62 for wives, and the correlation is considerably higher than any other in the table, .75. Cross classification of husbands' and wives' responses, however, shows that almost one couple in eight gave inconsistent responses: Number Percent of Total 1. Spouses agree a. On use 235 54.5 b. On nonuse 145 33.6 2. Spouses disagree a. Wife reports use, husband nonuse 33 7.7 b. Husband reports use, wife nonuse 18 4.2 3. Total 431 100.0 28 Our -analysis so far has been based on the reporting of use shown in categories la and 2a, that is, the wife's reporting of use, whether or not the husband does. Is this the best measure of use? To explore this issue we looked into the nature of the methods used and current versus ever use for the problematic categories, (2a) and (2b). For the 33 wives in category (2a) the technique of fertility con- trol that most (26) reported was the pill. Also, considerably less than half (13) report current as distinct from ever use. For this group, therefore, reported use typically relates to a female method used in the past. It seems reasonable to infer that many of the husbands who are reporting no use, have forgotten about the use of control or, conceiva- bly, may never have known. If this inference is correct, then the reporting of use by most, if not all, of these 33 wives is probably accurate. This implies that this group (2a) should be included along with the 235 users in group la, bringing total users to 268, or a proportion using of .62. This is, in fact, the measure used throughout the previous analysis -- wives reporting use, whether or not the husbands reported use. For the 18 husbands in category (2b) who reported use although their wives did not, the predominant method reported (15) is again the pill. In contrast to the findings for group (2a), however, almost all (15) report current use, suggesting that the issue is not one of recall. Two interpretations, with different implications for the facts on use, suggest themselves. One is that the couples really are using, but the wives are reluctant to admit it. Although such reluctance is common in some developing countries, it is noteworthy that Egypt differs from most in that wives, on the average, report higher use than husbands (Koenig 29 et al., 1984). The other possible interpretation is that the husband thinks the wife is using, but, in fact, she is not. There is no obvious basis for preferring either of these interpretations, and clearly each might apply to different couples. One is left, then, with the conclu- sion that for these 18 couples the facts on use are ambiguous. The real issue at stake, however, is whether the results of the previous analy- sis, which treats these 18 couples as nonusers, would be substantially altered if they were treated as users. To test this, a second measure of use is constructed that adds these 18 couples to the previous total of 268. This measure, based on whether husband or wife reports use yields a total of 286 users, and a proportion using of .66. An important implication of the foregoing reasoning is that the measure of use based on husbands' responses alone, 253 ( = 235 + 18) or a proportion using of .59 is the most questionable. This measure excludes category (2a), for which actual use seems highly likely, and includes the ambiguous category (2b). Put positively, the measure of use in which one may have the most confidence is use as reported by wives; use reported by either spouse could be better or worse than this; use as reported by husbands is almost surely worse. Turning next to the two attitudinal variables, desired family size and wants no more, for both of these the mean values for husbands and wives are quite close, and the correlation between spouses' responses is positive and significant (Table 10). However, for desired family size the correlation is less than half that for wants no more, .22 versus .48. Cross-classification of husbands' and wives' responses on desired family size brings out an important reason for this -- in a number of cases where one spouse reports quite high family size desires, say, 9 30 children or more, the other spouse reports a desired size of 5 children or less, that is, around or below the mean value. When couples in which either-spouse reports a desired family size of 9 or more are eliminated, the correlation coefficient rises to .37, closer to the magnitude for wants no more; the sample size, however, drops to 362. There is no obvious reason for preferring either husbands' or wives' responses on either of these variables -- indeed, the difference between the responses gives us the opportunity of studying whether husbands or wives attitudes explains better the decision on use. However, an alternative attitudinal measure also suggests itself, the average of the spouses' responses, and for both variables we include this as well in the subse- quent analysis. Finally, in regard to knowledge of fertility control techniques, the mean values indicate that wives are better informed, knowing an average of almost two methods compared with 1.5 for the husbands (Table 10). This is a plausible difference in a country where female methods are the dominant ones in use. Basically, what the means reflect is that most wives report knowing both the pill and IUD; in contrast, most husbands report knowledge of the pill, but knowledge of the IUD is considerably less prevalent among husbands. As the correlation coeffi- cient (.44) shows, it is generally the case that if the wife has better knowledge of techniques the husband does. As in the case of the atti- tudinal variables, however, it seems desirable to add to our analysis a measure of knowledge based on the average of the spouses' responses. Determinants of use -- We turn now to the interesting question of whether use of fertility control is better explained by data for wives, husbands, or wives and husbands combined. The preceding discussion 31 suggested two possibilities 'as the best measure of the dependent varia- ble: (1) use reported by wives, whether or not the husbands gave consis- tent responses, and (2) use reported by either spouse. However, for comparison, we include as well the less defensible measure, use based on husbands' responses alone, whether or not the wives reported use. For the independent variables we examine three measures of motivation -- Cn-Cd (potential unwanted children), Cd (desired family size), and wants no. more -- plus the measure of regulation costs used earlier, number of methods known. For each of these we construct measures based on (1) wives' responses alone, (2) husbands' responses, and (3) an average of the two. The analysis presented here is based on simple correlations like those in Table 2. We also estimated regression equations similar to those in Table.4, but found, as in the earlier analysis, that the con- clusions suggested by the simple correlations were the same as those yielded by the regressions. Given the large number of possible combina- tions of dependent and independent variables, simple correlation analy- sis seemed the most efficient way of conveying the results. Table 11 presents the bivariate correlations between the various measures of the dependent and independent variables. In the first column, the first entry in each panel corresponds conceptually to the correlations reported in Table 2, column 3. A comparison indicates some differences between the two tables due to the shift to a subsample of wives (e.g., the correlation between ever use and Cn-Cd drops from .49 to .46), but the size of the differences is quite small. Considering first the correlations between ever use and Cn-Cd (Panel A), one finds the same pattern irrespective of the measure of 32 use. In explaining use the motivation measure based on husbands' responses is substantially inferior to that based on wives or on the average for the spouses. As between the latter, there is little to choose, though the measure based on the spouses' average is arguably marginally better than that based on wives alone. The reason for this pattern is brough: out in panel B, which shows the correlations between-use and desired family size -- the pattern of differences here essentially replicates that of Panel A. The implica- tion is clear ** that decisions on use of fertility control are less dependent on husbands' family size preferences than on wives'. In this case the measure based on the spouses' average is no better than that based on wives alone. If the motivation measure is wants no more, however, the relative performance of the measure based on husbands' responses is improved -- it is as good or better than that.based on wives' responses. However, the relative performance of the husbands' measure is best for the questionable use measure in column (3). If one focuses on columns (1) and (2), perhaps the most important result shown by Panel C is that the average of the spouses' responses on wants no more is better than that based on husbands or wives alone, though not greatly so. What of the correlation between use and knowledge of fertility control? Panel D shows that for the two preferred measures of use, wives' knowledge is somewhat better correlated with use than husbands' (columns 1 and 2), but the positions reverse for the poorer measure of use (column 3). The last line of the panel shows that the average of the spouses' responses is somewhat better correlated with use than the responses for either husbands or wives alone. 33 One of the important conclusions of the earlier analysis was that the Cn-Cd measure of motivation explained use better than wants no more, although the difference in performance was considerably less for the dependent variable available for the present section, ever use, than for years.since first use (compare. Tables 2-4). How does this conclusion stand up here? To answer this, we compare for each measure of use the highest correlation with Cn-Cd (Panel A) with the highest correlation with wants no more (Panel C). Looking at column 1, for example, one finds that the highest correlation between use and Cn-Cd is .48 and between use and wants no more is .46, and that in both cases the measure of the independent variable that performs best is the one based on the average of husbands' and wives' responses. This pattern holds for the other two columns as well -- the highest correlations in the two panels are virtually the same, and the average response of the spouses yields the best measure of the independent variables. Thus, there is a differ- ence here from the earlier results -- as a measure of motivation, wants no more performs as well as Ca-Cd in explaining ever use. This is not to say, however, that the same conclusion would apply if years since first use were the measure of use. For example, the correlation with years since first use (based on wives' responses) is .47 for Ca-Cd and .35 for wants no more (spouses' average used for both of the latter measures). The improvement just reported in the relative performance of wants no more compared with Cn-Cd is not bought at zero cost. To see this, suppose one had, to start with, only data for wives. As shown in column 1 of Table 11, one would obtain the following correlations of the two motivation measures with use (based, of course, on wives' responses alone): 34 Cn-Cd .46 Wants no more .40 If now, one were, additionally, to collect data from husbands, by how much could one improve the prediction of use? If the measure of use were that based on wives' responses alone, as in the preceding compari- son, one obtains (from column 1): Cn-Cd (spouses' average) .48 Wants no more (spouses' average) .46 If the measure of use were that in column (2) of Table 11, which is, according to the reasoning above, a possible alternative to that in column (1), one obtains: Cn-Cd (spouses' average) .49* Wants no more (spouses' average) .50* - In other words, by including data for husbands the explanatory power of Cn-Cd has been raised from .46- to .48 or .49, hardly an impressive improvement. For wants no more the improvement is greater -- from .40 to .46 or .50 -- but the predictive power is no better than for Cn-Cd. Thus with data for wives alone one does almost as well as with data for both husbands and wives. It should be underscored that this statement refers only to the present data relating to fertility control and reproductive behavior; on other matters, such as financial affairs, the conclusion might well be different. Summary and Conclusions In theory, use of fertility control should vary directly with the motivation for control and inversely with the costs of regulating fer- tility, where costs include both subjective drawbacks of control and 35 lack of access to family planning methods and services. The empirical analysis of wives' responses in the 1979-80 EFS in this paper consistent- ly points to the importance in determining fertility regulation of motivation, measured as the excess of potential over desired family size. The implication is that households that envisage unregulated fertility as leading to a family size considerably in excess of that desired are under greater pressure to regulate their fertility. Through- .out most of this paper, this measure (Cn-Cd) performs better than alter- native measures of motivation suggested by the literature, such as respondent reports on whether or not she wants more children, the number of living children a couple has, the wife's desired family size, and the actual -number of unwanted children. This is true in analyses for both the total and regulating population, and also when the measure of fer- tility control is ever use, rather than time since first use. Replica- tion of the 1979-80 EFS analysis using OLS with data from Egyptian sur- veys for 1974-75 and 1979 (rural population only), and from the 1975 Sri Lanka WFS yields virtually identical results. Given motivation, a household is more likely to use fertility con- trol if its subjective feelings about family planning methods are more favorable or if it has better access to such methods. Unfortunately appropriate data on subjective attitudes are not available to test this proposition, but two types of "access" measures are. The one employed throughout much of this analysis is number of methods known to a respon- dent and reported without special prompting, the idea being that lack of knowledge would imply poor access and hence lack of use. For the total population number of methods known performs as hypothesized, although it is not quite as strong as motivation in explaining use. However, there 36 .is a problem with this measure because it is not independent of use -- one would expect, not only that better knowledge would promote use, but also t-bat use would promote better knowledge. To some extent this prob- lem can be circumvented by confining the analysis to the regulating pop- ulation alone; when this is done, however, the effect on use of number of methods known drops to bare significance. The community module of EFS provides several unbiased measures of access to family planning services, although for the rural population only. Analysis of these data indicates that access to such services has a significant effect on use of fertility control. Specifically, con- trolling for motivation, use tends to be greater among women who have greater access to qualified midwives or to family planning or rural health centers. But the importance of such access turns out to be con- siderably less than motivation in determining use of fertility. control. The foregoing conclusions are based on an analysis of wives' responses (except for the community measures of access to family plan- ning services). Using survey data for a subsample of husbands inter- viewed in the second round of the EFS, an analysis was done of the con- sistency of husband and wife responses on several key variables, and the reliability of wives' data for drawing inferences on the determinants of fertility control. This analysis suggested that so far as the measure- ment and explanation of fertility control is concerned, the wives' data taken alone are better than the husbands' alone. Combining husband and wife responses sometimes improves moderately the statistical explana- tion, but for the analysis of fertility control probably not enough to justify the cost of collecting the additional data. 37 References Berk, RIA., 1983. "An Introduction to Sample Selection Bias in Sociolo- gical Data," American Sociological Review, Vol. 48, June, 386-398. Bongaarts, J., 1978. "A Framework for Analyzing the Proximate Deter- minants of Fertility," Population and Development Review, 4:105-32. Bongaarts, J., 1980. The Fertility-Inhibiting Effects of the Interme- diate Fertility Variables, Center for Policy Studies Working Paper 57, New York: The Population Council. Bumpass, L.A. and C.F. Westoff, 1970. "The 'Perfect Contraceptive' Population," Science, 169:1177-82. Central Agency for Public Mobilisation and Statistics, 1983. The Egyptian Fertility Survey 1980, Volumes I-IV , Cairo: Central Agency for Public Mobilisation and Statistics. Crimmins, E.M. and R.A. Easterlin, forthcoming 1984. "The Estimation of Natural Fertility: A Micro Approach," Social Biology, 31:1. Davis, K. and J. Blake, 1956. "Social Structure and -Fertility: An Analytic Framework," Economic Development and Cultural Change, 4:211-35. Easterlin, R.A., 1978. "The Economics and Sociology of Fertility: A Synthesis," in C. Tilly, ed., Historical Studies of Changing Fertility, Princeton, New Jersey: Princeton University Press. Easterlin, R.A. and E.M. Crimmins, 1982. An Exploratory Study of the 'Synthesis Framework' of Fertility Determination with World Fertility Survey Data," No. 40, (November), The Netherlands: International Statistical Institute. Easterlin, R.A. and E.M. Crimmins, forthcoming 1985. The Fertility Revolution: A New Approach. Easterlin, R.A., R. Pollak and M.L. Wachter, 1980.. "Toward a More General Economic Model of Fertility Determination: Endogenous Preferences and Natural Fertility," in R.A. Easterlin, ed., Population and Economic Change in Developing Countries, Chicago: University of Chicago Press. Freedman, R., 1961-2. "The Sociology of Human Fertility," Current Sociology, 10/11:35-68. Hermalin, A.I., 1983. "Fertility Regulation and Its Costs: A Critical Essay," in Rodolfo A. Bulatao and Ronald D. Lee (eds.), Deter- minants of Fertility in Developing Countries, Vol. 2, New York: Academic Press. . -38 Issa, M.S.A., 1981. "Modernisation and the Fertility Transition, Egypt: 1975," Ph.D. dissertation, University of Pennsylvania. Jejeebhoy, S., 1981, "Cohort Consistency in Family Size Preferences," Studies in Family Planning, 12:229-32. Kelley, A. C. and R.M. Schmidt, 1983. "Family Planning, Socioeconomic Change and Population Policy in Egypt: An Exploratory Methodology," paper presented at the Seminar on Egyptian Population Policy, Cairo, Egypt, October 16-17, organized by the International Union for the Scientific Study of Population. Khalifa, A.M., H.A.A.H. Sayed, M.N. El-Khorazaty, and A.A. Way, 1982. Family Planning in Rural Egypt 1980: A Report on the Results of the Egypt Contraceptive Prevalence Surve, Cairo: Population and Family Planning Board (December). Knodel, J. and V. Prachuabmoh, 1973. "Desired Family Size in Thailand: Are the Responses Meaningful?" Demography, 10:619-38. Koenig, M.A., G.B. Simmons and B.D. Misra, 1984. "Husband-Wife Inconsis- tencies in Contraceptive Use Responses," Population Studies, XXXVIII (2):281-298. Maddala, G.S., 1983. Limited-Dependent and Qualitative Variables in Econometrics, New York: Cambridge University Press. McClelland, G.H., 1983. "Family-Size Desires as Measures of Demand," in Rodolfo A. Bulatao and Ronald D. Lee (eds.), Determinants of Fertility in Developing Countries, Vol. 1, New York: Academic Press. Petersen, W., 1969. Population, Toronto: Collier-Macmillan. Pindyck, R.S. and D.L. Rubinfeld, 1983. Econometric Models and Economic Forecasts, 2nd edition, New York: McGraw-Hill Book Co. Pittenger, D.B., 1973. "An Exponential Model of Female Sterility," Demography, 10:113-21. United Nations, 1981. Variations in the Incidence of Knowledge and Use of Contraception: A Comparative Analysis of World Fertility Survey Results of Twenty Developing Countries, New York: United Nations. 39 TABLE 1 .Regression Results for Proximate Determinants Equation with Use of Fertility Control Measured in Two Ways Regression Coefficient (absolute value of t statistic in parentheses) U=Years since first U=Ever Mean Standard Variablea use of fertility control use Deviation F, Children ever born - - 6.70 2.56 U, Use of fertility -.1688* -1.7863* 6.32b 6.63b control (12.60) (8.87) 11, Duration of marriage, .3393* .2987* 21.43 4.73 years (30.29) (28.18) X2, First birth interval, -.0307* -.0261* 24.15 20.42 months (12.79) (10.88) X3, Second birth interval, -.0383* -.0340* 25.49 15.84 months (12.77) (11.33) X4, Not secondarily sterile 2.1940* 2.1140* .76 .43 (=1; others = 0) (18.04) (16.24) X5, Months breastfed, -.0100* -.0054 18.42 12.00 last closed interval (2.50) (1.29) X6, Proportion of pregnancy -1.9905* -2.3218* .09 .14 wastage (5.84) (6.69) X7, Proportion of child 2.6242* 2.5664* .22 .21 mortality (11.11) (10.10) Constant .3049 .9867* (1.15) (3.42) Number of cases 1728 1728 i2.48 .45 * Significant at .05 level or higher. a. Dependent variable is children ever born. b. For ever use, the mean and standard deviation are respectively, .58 and .49. 40 TABLE 2 *Rivariate Correlation Coefficient Between Specified Measure of Fertility Control and Various Independent Variables Dependent Variable and Population Coverage Independent Years since starting Ever Variable fertility control Use Total Regulating Total population population population Cn-Cd (potential unwanted .49* .27* .49* children) Wants no more .31* -.06 .40* Ca (potential family size) .35* .22* .33* Cd (desired family size) -.35*- -.15* -.38* C (living children) .01 -.10* .07* C-Cd (actual unwanted .33* .05 .38* children) Number of methods known .41* .09* .46* Number of cases 1728 996 1728 * Significant at .05 level or higher. * * * TABLE 3 . Tobit Regression Equation for Years Since First Use of Fertility Control and Specified Independent Variables, Total and Regulating Population (absolute value of t-statistic in parentheses) 2 Equation Cn-Cd Wants Ca Cd C C-Cd Number Constant 2 no more of methods known A. Total population (number of cases 1728) A-1 1.6674* 3.0192* -4.9683* 782 (19.39) (14.42) (10.18) A-2 7.1406* 3.8666* -9.5928* 522 (11.69) (17.03) (14.40) A-3 1.9105* 4.0931* -15.8010* 588 (14.30) (19.12) (16.84) A-4 -1.3835* 3.5800* 2.5628* 552 (12.54) (15.98) (3.42) A-5 .4064* 4.6214* -7.6512* 396 (3.24) (20.23) (8.85) A-6 1.2473* 3.8855* -5.0203* 572 (13.32) (17.58) (9.68) A-7 1.9772* -1.4336* 3.0783* -7.9079* 796 (15.68) (13.85) (14.71) (8.07) Hean 1.34 .74 5.87 4.53 5.04 .51 1.88 (std. dev.) (3.26) (.44) (1.80) (2.70) 01.95) (2.99) (1.10) B. Regulating population (number of cases = 996) A-1 .6110* .3796* 8.2782* 76 (9.26) (2.16) (11.85) A-2 -1.4646* .5388* 10.8600* 14 (2.92) (3.18) (18.01) A-3 .7195* .5633* 4.9207* 54 (7.28) (3.14) (6.15) A-4 -.4380* .3558* 11.6020* 26 (4.98) (2.04) (19.93) A-5 -.2467* .4030* 11.1370* 14 (2.66) (2.25) (15.06) A-6 .1198 .5386* 9.3748* 10 (1.57) (3.08) (20.09) A-7 .7663* -.4786* .4229* 6.6994* 80 (7.81) (5.79) (2.37) (7.98) Hean 2.70 .89 6.37 3.66 5.15 1.49 2.32 (std. dev.) (2.42) (.32) (1.68) (1.93) (1.90) (2.25) (1.00) * Significant at .05 level or higher. TABLE 4 Logit Regression Equation for Ever Use of Fertility Contiol and Specified Independent Variables, Total Population (absolute value of t-statistic in parentheses) Cn-Cd Wants Cn Cd C C-Cd Number no more of methods 2 Equation known Constant x A-1 .3748* 1.0002* -1.9829* 724 (14.53) (13.50) (13.50) A-2 1.5801* 1.0661* -2.7625* 584 (11.47) (15.06) (16.27) A-3 .4162* 1.1596* -4.1810* 580 (11.04) (16.42) (15.65) A-4 -.2694* 1.0657* -.3719* 567 (10.36) (15.01) (2.07) A-5 .1102* 1.1959* -2.3710* 456 (3.81) (17.43) (11.70) A-6 .2936* 1.0905* -1.8020* 610 (11.65) (15.25) (12.97) A-7 .4944* -.3247* 1.0169* -2.9232* 739 (12ll (114? (I 7 113.67) (10.0n) Hean 1.36 .74 5.88 4.53 5.04 .51 1.88 (std. dev.) (3.22) (.44) (1.71) (2.70) (1.95) (2.99) (1.10) * Significant at .05 level or higher. a Number of cases = 1728. 43 TABLE 5 Comparison of OLS Regression Results for Proximate Determinants Equation, Egypt (Three Samples) and Sri Lanka Regression Coefficient (absolute value of t-statistic in parentheses) Variable Egypt WFS Rural Sri Lanka 1979-80 1974-75 1979 1975 Years since starting -.09* -.11* -.01 -.11* fertility control (11.83) (17.38) (1.31) (14.04) Duration of marriage .32* .29* .30* .32* (31.08) (36.31) (19.89) (37.28) First birth interval -.026* -.028* -.025* -.026* (11.91) (17.19) (11.26) (11.35) Second birth interval *.034* -.021* -.036* -.036* (11.96) (13.38) (10.50) (16.50) Not secondarily sterile 1.85* 2.00* 2.56* 1.81* (17.05) (21.91) (14.83) (18.26) Months breastfeeding -.001 n.a. -.010* -.016* (.19) (2.15) (4.57) Proportion of pregnancy wastage -2.45* -1.69* -.53 -2.77* (7.59) (3.93) (1.00) (7.45) Proportion of child mortality 3.19* 2.38* 2.60* .98* (14.60) (13.61) (8.16) (3.28) Constant .023 .083 .14 .61 (.09) (n.a.) (.36) (3.01) Summary statistics ft2 .51 .50 .54 .55 Number of cases 1728 2159 703 1613 * Significant at .05 level or higher. n.a. Not available. 44 TABLE 6 Comparison of Bivariate Correlation Coefficients Between Years since First Use and Specified Motivation Measures, Egypt (Two Samples) and Sri Lanka, Total Population and Regulating Population Motivation measure Egypt, 9FS a Sri Lanka 1979-80 1974-75 1975 A. Total Population Cn-Cd (potential unwanted .48* .56* .38* children) Wants no more .31* .38* .21* Ca (potential family size) .34* .42* .23* Cd (desired family size) -.35* -.32* -.15* C (living children) .01 .02 -.07* C-Cd (actual unwanted children) .33* .35* .09* Number of cases 1728 2063 1611 B. Regulating Population Cn-Cd (potential unwanted .29* .42* .38* children) Wants no more -.06 .01 .00 Ca (potential family size) .25* .35* .29* Cd (desired family size) -.15* .18* -.12* C (living children) -.10* ...16* -.07* C-Cd (actual unwanted children) .05 .02 .03 Number of cases 996 903 896 * Significant at .05 level or higher. a. The correlation coefficients for Ca-Cd and Cn here differ slightly from those in Table 2, because Cn has been re-estimated by OLS for consistency with the other studies. 45 TABLE 7 Comparison of OLS Regression Results for Determinants of Use Equation, Egypt (Three Samples) and Sri Lanka Regression Coefficient (absolute value of t-statistic in parentheses) Variablea Egypt WFS Rural Sri Lanka 1979-80 1974-75 1979 1975 Cn-Cd .81* 1.38* .66* 1.01* (17.83) (27.48) (8.21) (15.07) Number of methods known 1.71* -.54* .91* .76* (13.25) (8.39) (8.22) (8.91) Constant 2.35* 7.66 2.12* 1.85* (8.80) (n.a.) (5.84) (7.86) Summary statistics 12 .30 .34 .20 .17 Number of cases 1728 2063 709 1607 * Significant at .05 level or higher. a Dependent variable is years since starting fertility control. 46 TABLE 8 - Community Measures of Availability of Family Planning and Health Services and Measures of Motivation: Mean, Standard Deviation, and Bivariate Correlation with Time since First Use, Rural Population Variable Mean Standard Correlation Number deviation with of time since cases first use A. Measures of family planning and health services Age of family planning place, years a 11.53 9.54 .05 942 Doctors present daily (per 1,000 pop.) 0.16 0.12 .04 942 Nurses present daily (per 1,000 pop.) 0.16 0.24 .05 942 Qualified midwives present daily (per 1,000 pop.) 0.16 0.27 .12* 942 Dayas present daily (per 1,000 pop.) 0.16 0.25 -.02 942 Pill available in 1979 (=1; other = 0) 0.85 0.36 -.02- 935 IUD available in 1979 (=1; other = 0) 0.25 0.66 .10* 935 Distance -to rural health unit,km. 2.22 2.93 .12* 782 Distance to rural health center,km. 2.52 4.15 -.09* 931 Distance to combined unit,km. 4.63 4.72 -.01 871 Distance to family planning clinic, km. 3.28 5.33 -.11* 892 Number of methods known 1.54 0.99 .40* 942 B. Measures of Motivation Cn-Cd 0.00 3.37 .36* 942 Wants no more 0.67 0.47 .25* 942 Cn 5.36 1.62 .27* 942 Cd 5.36 2.98 -.26* 942 C 5.37 1.98 .16* 942 C-Cd 0.01 3.44 .32* 942 * Significant at .05 level or higher. a. Equals 1980 minus date of establishment of family planning place; equals zero, if no family planning place in community. 47 TABLE 9 a Tobit Regression Equation for Years since First Use of Fertility Control and Specified Independent Variables, Rural Population (absolute value of t-statistic in'parentheses) Motivatioi Measure Access Measure Midwives Rural Family Wants per 1,000 health planning 2 Equation Cn-Cd no more pop. center center Constant x 1 1.9385 -4.1587 184 (12.07) (7.28) 2 1.7814 . 3.7161 -.3112 -.2959 -2.9009 206 (11.26) (2.56) (2.60) (3.07) (4.20) 3 10.7390 -10.9590 106 (9.63) (10.03) 4 9.7831 5.4080 -.3648 -.3302 -9.1412 136 (8.95) (3.51) (2.93) (3.26) (7.89) Number of cases = 883. * Significant at .05 level or higher. o48 TABLE 10 -Comparison of Wives and Husbands Responses on Four Measures froi Hatched Sample of Husbands and Wives in their First Harriage Mean(standard deviation in Correlation parentheses) based on: coefficient Husbands' Wives' responses responses Proportion ever using .59 .62 .75* (.49) (.49) Cd (desired family size) 4.64 4.70 .22* (2.49) (3.04) Wants no more .70 .75 .48* (.46) (.44) Number of methods known 1.49 1.90 .44* (1.10) (1.10) Number of cases 431 431 431 * Significant at .05 level or higher. 49 TABLE 11 -, Bivariate Correlation between.Ever Use and Specified Independent Variable Using Measures Constructed from Indicated Groups of Respondents Ever Use Reported by: Independent variable and Wives Either Husbands respondent group only spouse only (Mn=.62) (Mn=.66) (Mn=.59) A. Cn-Cd (potential unwanted children) based on: Wives' responses (Hn=I.15) .46* .47* .47* Husbands' responses (Mn=1.18) .31* .32* .37* Spouses' average (Mn=1.27) .48* .49* .51* B. Cd (desired family size) based on: Wives' responses (Mn=4.70) -.35* -.35* -.36* Husbands' responses (Mn=4.64) -.14* -.14* -.20* Spouses' average (Mn=4.67) -.33* -.32* -.36* C. Wants no more based on: Wives' responses (Mn=.75) .40* .40* .40* Husbands' responses (Mn=.70) .39* .45* .52* Spouses' average (Mn=.72) .46* .50* .53* D. Number of methods known based on: Wives' responses (Mn=1.90) .44* .43* .41* Husbands' responses (Mn=1.49) .34* .41* .46* Spouses' average (Mn=1.70) .46* .50* .52* * Significant at .05 level or higher. r 1 & 50 APPENDIX A Definition and Measurement of Variables Variable EFS Definition and Measurement variable F, Children Ever Born V208 Number of children ever born. U, Ever Use of Fertility V634 Reported ever-use of any method Control of contraception, 1 = yes, 0 = no. U, Years Since First use of S201 The difference between wife's Fertility Control V007 current age and the age at first B012 use is the years since first use thru of fertility control. Age at first B112 use is the wife's age at the birth V010 of the child after which she first V109 used family planning plus two years. If the wife used fertility control before any children were born, her age at first use is her age at marriage. For a small number of cases in which the initial estimate of time since first use was negative, the value was set equal to zero. X1, Duration of Marriage V010 The difference between current age V109 and age at first marriage. X2 First Birth Interval V228 First birth interval in months. The mean first birth interval for regulators who did not regulate until after the first birth is substituted for the observed first birth interval of those who regulated before the first birth. X , Second Birth Interval B022 The difference in months between B012 the date of birth of the second child and the date of birth of the first child. The mean second birth interval for regulators who did not regulate until after the second birth is substituted for the observed second birth interval of those who regulated before the second birth. -' 51 APPENDIX A (cont'd.) Variable EFS Definition and Measurement variable I Not Secondarily Sterile V206 Two-category variable: 1 = fecund; V402 0 = sterile. If currently V637 pregnant, respondent is fecund. .V225 If respondent reports fertility impairment, respondent is sterile. If respondent is not a current user of contraception and reports no birth in the past five years, respondent is sterile. X5, Length of Breastfeeding V302 Number of months breastfed in last closed birth interval. For- 41 "not stated" cases, duration of breastfeeding was set equal to the mean. X6, Proportion of Pregnancy V201 The difference between the number Wastage V208 of wasted pregnancies and the number of induced abortions divided by the sum of the number of wasted pregnancies plus the number of live births minus the number of induced abortions. X7, Proportion of Child V213 The difference between the number Mortality V208 of children ever born and the number currently living, divided by the number of children ever born. RC, Number of Methods Known V601 The number of methods of fertility V602 control known to the respondent V603 and reported without special V604 prompting. Sum of "2" responses V605 on variables listed. V606 V607 V608 V609 V610 V611 V615 Cn, Potential Surviving - See text. Children 52 APPENDIX A (cont'd.) Variable EFS Definition and measurement variable Cd, Numb'er of Children V511 Answer to question, "If you * Desired could choose exactly the number of children to have in your whole life, how many would that be?" C, Number of Living V213 Reported number of living chil- Children dren. Wants No More V502 If respondent is fecund and wants no more children, wants no more = 1; if respondent is not fecund or wants more children = 0. (An alternative measure, using X501, that made it possible also to classify not fecund women according to whether or not they wanted more children, did not yield significantly different results.) Measures from Community Module Age of Family Planning DATEST Equals 1980 minus year family Place planning clinic was established in community. Doctors, nurses, qualified DRS1 Specified personnel group divided midwives, or dayas NURSE1 by total population of the present daily (per MIDWI community. 1,000 population) DAYAS1 NTOTAL Pill available in 1979 APIL79 If available = 1; otherwise = 0. IUD available in 1979 AIUD79 If available = 1; otherwise = 0. Distance to rural health RHU If facility is available within unit, k.. administrative borders of the village, equals zero; if un- available, distance to the nearest community in which service is available. Distance to rural health RHC As above. center, km. Distance to combined COMBU As above. unit, km. Distance to family planning FPC As above. clinic, km. 53 APPENDIX A (cont'd.) Variable EFS Definition and measurement variable Measures from Sample of Husbands Number of times married Q201 Used to select once-married men (Q201 = 1). Ever use of fertility Q315 Yes = 1; no = 0 (recode from 2). control Number of children desired Q427 Same as V511 above. Number of methods known Q304K The number of fertility control Q305K methods known to the respondent Q306K and reported without special Q307K prompting. Sum of zero responses Q308K on variables listed. Q309K Q310K Q311K Q312K Q313K Wants no more Q416 If wants no more ("2" responses Q421 on variables listed), = 1; otherwise, zero.