PHN-871 2 HOUSEHOLD INCOME AND CHILD SURVIVAL IN EGYPT by John B. Casterline Elizabeth C. Cooksey Abdel Fattah Ismail June 1987 Population, Health and Nutrition Departme 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. FILE COPY t A PHN Technical Note 87-12 HOUSEHOLD INCOME AND CHILD SURVIVAL IN EGYPT ABSTRACT The relationship between household income and child survival is examined utilizing household economic and fertility data from Egypt. Neither total household income nor per capita household income show effects on mortality rates during infancy, but both are inversely related to mortality in early childhood (ages one through four years). The relationships persist with controls for other associated socioeconomic variables. The relationships also seem to exist in roughly the same form for major regional and socioeconomic subgroups. The mechanisms through which the income effects operate are not evident in this analysis; income differentials in sources of household drinking water, type of toilet facilities used by household members, and maternal demographic characteristics (age, parity, birth spacing) do not explain the net impact of income on child mortality. Infant mortality is strongly associated with only two variables: region of residence (the rate is much higher in Upper Egypt) and maternal demographic characteristics. Child mortality is significantly associated with paternal status (occupation and educational attainment) and the type of drinking supply for the household, as well as household income. The absence of effects on both infant and child mortality of size of place of residence and of maternal education is notable. In order to reduce infant mortality in Egypt, policies which encourage childbearing at more optimal ages, parities, and spacings between births would be very desirable. There is also a need for research which explores the sources of the massive regional differentials. Efforts to reduce child mortality must take account of substantial differentials by paternal socioeconomic status and household income. Prepared by: John B. Casterline, Elizabeth C. Cooksey and Abdel Fattah Ismail Consultants to the World Bank June 1987 9, HOU.SEHIOLD INCOME AND CHILD SURVIVAL IN EGYPT John B. Casterline* Elizabeth C. Cooksey* Abdel Fattah Ismail** * Population Studies Training Centre, Brown University, Providence, Rhcde Island 02912, U.S.A. -** Central Agency for Public Mobilization ar.- Statistics, Cairo, Egypt. FV1AL 2,-==12-5,=su----a zth z2 9 C Q 4 ACKNOWLEDGEMENTS Financial support for this research was provided by he World Bank. The research was carried out at the Department of Sociology and the Population Studies and Training Center, Brcun university.' The authors are grateful to Dr. A. Hallouda, President of the Central Agency for Public Mobilization and Statistics, for permission to analyze the Egyptian Fertility Survey data, and to Dr.'Susan Cochrane, the World Bank, for her efforts in arranging for the research to be undertaken. The EFS data files were supplied by the Dynamic Data Base, International Statistical Research Centre, ISI, the Hague, Netherlands. Progra=ming assistance was provided by Kalpana Mehra, the World Bank, and Irene Gravel, Population Studies and Training Center, Brown University. ABSTMACT The relationshi between household income and child survival is examined utilizing household economic and fertility data f: Egypt. Neither total household income 1or -er ca=ita househCl4 income show atfects on infant mortality, but both are inversely related to mortality in early childccld (ages one throu!h four years) . The relationship persists with controls for other associated socioeconomic variables. The relationshim also seems to exist in roughly the same for for major regicnal and socioeconomic subgroups. The mechanisms which underly the inccne effects are not evident from this analysis: income diffarenzials in sources of household drin3king water, type of toilet facilities used by household members, and maternal demgraphic character- istics (age, parity, birth spacing) do M- explain the net iPact of income on child mortality. The type of toilet facility is not associated with either infant or child survival, but the presence of piped water in the dwelling shows a net negative effect on child mortality. Infant mortality is strongly associated with only two variables: region of residen-e the rate Is much higher in Upper Egypt) and maternal demographic characteristics. Zhe absence of effects on both infant and child mortality of tY7e of place of residence and of maternal educai-4- 4s ---able. The results suggest that efforts to reduce infant mortality in gymt require a better understanding of the mechan.sms underlyimg z!.e regional differentials, as well as policies to enciraSe child- bearing at more otimal ages, parities, and spacings betUean children. Efforts to reduce child mortality must take acoun' a substantial differentials by paternal socicecon=ic status an.d household income. TABLE OF CONTENTS I INTRODUCTION...........................................1 1.1 Objectives of the Study...........................1 1.2 Structure of the Report...................... 6 Ir BACKGROUND................................. ...........7 2.1 Mortality in Egypt ................................7 2.2 Economic Differentials in Mortality............... 2.3 Modelling Mortality..............................13 I:II: DATA.... 0.............................................. 15 3.1 The Egyptian Fertility Survey (EFS)..............S 3.2 Variables........... ..... .. ... ................. 19 3.3 Selection of Cases for Analysis..................40 IV ANALYTIC APPROACH .....................................44 4.1 The Model ........................................ 44 4.2 Equations ........................................4 4.3 Statistical Model ........... .. ................. 49 V FINDINGS ..... ...................... 4..................S4 5.1 Bivariate Results................................54 5.2 Multivariate ReSU tS.............................57 VI SUMMARY AND CONCLUDING RF-R:S........................76 z. INoTROc:CIO 1. Obliectives of the Studv Desite substantial declines in mortality thrcugnout =ost zi the developing world in the period since the Second World War, mortality levels renaln high in many devaleping courtries. . 40 percent of the 72 ccuntries classified as low or lower-oidd.e income, by the World Bank, the esti=ated infan =ortality rats exceeded 100 deaths per 1000 births in 19a2. :n 10 of these countries the rate exceeded 150 deaths per 1000 births (World Bank, 1984). That such mortality rates are not a necessary concomitant of low levels of economic developnent is deonszrate by the success of a few countries in achieving narkedly !ower infant mortality ratas; notable exanplas are Costa .ica, Cuta, Sri Lanka, Thailand, Philippines and China. The apparent possibility of achieving signlficant in =ortality has spurred extensive research regard.c !aczcrs responsible for high levels of mortality. Thare is general agreement about the factors which deternine levels cf =crzalv' (See Secretaria-t of the World Ferility Survey, '3 *fr a general. raview) , tut ar lass c==.sansus c=ncern`ng re.a-:- inportance of each fac=cr a:.d at=utwhch faczrs =os= efac=-.4ie :ccal =cints cf c' : ns. =c :- tal condititcs (i- a th preseanca -. and cuallt-,* ':f h=u_~~'----.'of education, and the accessibility and utilization of medical services have each been singled out as critical determinanzs which can be modified by public policies. It is very likely that the significance of each of these determinants varies according to ecological and cultural setting, and thus optimal policy interventions .must be formulated on a case-by-case basis.. Research seeking to identify the critical determinants of mortality can take several forms. In-depth analyses of counzries having experienced the most rapid and substantial declines may yield conclusions about the main causes of the decline (See Secretariat of the UN ESCAP, 1984). However, on the basis of only one or a few cases, it is often difficult to isolate the key factors from among the complex. of factors which changed during the period of mortality decline. Joint and independent roles of multiple factors can be explicitly modelled in cross-naticnal analyses of national-level data from a large number of counzries, but unfortunately measurement of the some of the factors hvpoth- esized to be of central importance is often imprecise or lacking altogether at the national level. A third and rather difJErent research approach is to utilize experimena or as mental designs at the sub-national level. This a=proac. 4s particularly a==ro,riate for inveszigaticn z-e 4==a=':of medical or sanitation interventions, bu:t does :CZ en- 1el : are either no: readi,y manipulaed, ::alr: :e>, .ay: im-acz wi-h substantial time lac (scicec icr ii of parents and hcuseholds). 1oireover, the exmeri4ental apmrza-* can be costly and difficult to 1=ple=ent. A final a=rroach, an. the one adopted in this research, is analysis of cess-s household survey data which contain =easureaen»s of no=ta:.itv along with xany of its direct -and indirect deter=ilnants. Tr- the standpoint of makng inferencas about the causes of morality decline, this aptroach is handicapped by the cross-secticnal nature of the survey data. On the other hand, it is feasib=e to -collect information in some detal'- at the household leval about many of the presumed principal deterinants of norbidity and mortality, which is an advantage of this approach ralative to others. This report presents analysis of the deterinants of infant and early childhood mortality in Egypt, utilizing data collected in the E.gyptian Fertility Surrwy (ZFS). This national survey was conducted in 1930 as part of the World Ferity Survey progra=. The EFS, in com=on with all WS surveys, included a retrospective maternity history, obtained fr== ever-married wo=en of remrcduct_ve age, which prvi-4des =aasura- ment of child survival during the =eriod d tne survay. The ES, in a departure from other WIS surreys, a.so in·: detailed hosehold eccnci i .se 'aztr aza a un.-vua opzznt- oivetg the veaia i · a 7.ee:- househld in:c=e and Child s ual, anz z*-s i-3 c- tne resaarh rezr-d r income and its effects on child survival. The first derives from concerns about the correlates of economic status, and the second from interest in understanding the determinants of mor-ality. Economic status is a major dimension in human 3ccis". In part because of the recognition that economic resources are related to many valued aspects of welfare and wellbeing, the allocation of economic resources has undoubtedly been a central issue in most societies. Nothing is more valued than health and survival. However, the extent to which economic position affects health and survival - for example, probabilities of child survival - is a matter which must be determined through empirical investiqation. The strength of the effect is an indicator of the failure of the society to separate allocation of .economic resources from .allocation of those resources which are essential for cood health and longevity of life. A second motivation for focusing on the association between income and child survival is the desire to understand as fully as possible the determinants of mortality, with the ultimate objective of providing empirical evidence which assists in the design of effective policies to reduce mortality rates. zere are numerous reasons to expect inccme level to be rlae t survival probabilities. These we will s-ellct secticn of the report in the context of a general::~ de':er=_4=an=s of =cr=a_!Jwz. oes~no o.Liist coba hihs =crtali, .: -,s* -_en deter=ine the soe;hof the effc:c n= n=-1 and to identify the mechanisms th=ouch which income exe-s it effect. The relationship between household income and child suvival will be considered in the context of a general mcdel of the determinants of child survival, and thus the research has =ltiple objectives: + Estimation of the effect of household income on child survival pobabilities, both without and with controls for associated socioeconomic characteristics, such as maternal schooling. + Examination of the extent to which income effects on i survival orerate indirectly through cther household charact- eristics, such as sanitation conditions, and tarough parental characterics, such as maternal age and party. + Estimation of the net direct and indirct affects on martality of other socioeconomic characteristics, such as maternal schcoling, and comparison of the size of these effects to these of household income. + Estimation of net effects of interver..c fact=rs, such as household sanitation and maternal deicharacz- istics. j.2 Structure of the Revort The report contains five main sections following this introductory section. In the next section, we provide .a back- ground to the study: a summary of levels and trends in mortality in .gpt in the decades preceding the 1980 EFS, a brief review-of evidence on the relationship between income and child survival, and presentation of the conceptual framework guiding the -analy- sis. Section III contains a descripticn of the design of the ZFS and of the selection of variables and cases for this analysis. Design of the analysis is discussed in Section IV, including specification of the equations to be estimated and of the statistical approach adopted. The findings are presented in detail in Section V with accompanying tables, and their signifi- cance is discussed in the concluding Section V-. 2,1 -Mortalitv in, E=vt Egypt is a=ong those cou.ntries referred to in Section where the Lfant mortality rate exceeds 100 deaths per 10CO live births. The World Bank estinate for 193*2 is 104 (WrId Zank, 1984), which indeed aPpears somewhat lcw against other esti.ates for recent pericds, including estimates frem the 7S (see Table 1). The E7S estimate for the five years preceding the su-rsey 132; adjust=ents for heaping of reported ages at death cn twelve mcnths and for not stated ages at death raises the estinate by roughly five pointz (Ismail A.E. et al, 1983). Althcugh the 7 estimates are scmewhat higher than other recent estinates, the avallable figures all inicate that the infant rtortai e during the past decade has been in excess of 100. Zt is alsc clear that Child survival chances have 1=mrcved draaiall since 1945, as suggested by the tine-series of estiatas in Tatle 1. We are aware of no systenatic analysis of the causes cf this mortality decline. The decline can safelv be ass-ed to be in part the consecuence of iproveents in =utlic san.iza-iz services (water and sewage), in public haalz- sertcss (azcass to vaccinatic.s and cz=atie care), and neral -enens the standard of livin (utritic, :sin. The levels and trends in nor i dur-; rezent dac%des are thius t--iao he exnaer.-ence Cof. =an-, Su_~~ h.v C-u==Ud,~- raz-s az ---eazj high levels, levels comparable-to those found in most other countries in the same per capita income range (World Bank, 1984). Discussions of the reasons for these moderately high levels typically stress the most obvious explanations: poor water supplies and environmental sanitation, -resulting in high rates of schistosomiasis and gastrointestinal diseases; reliance on untrained traditional midwives as birth attendents; and inaccess- iObility of effective modern medical care (Ikram, 1980). Serious efforts have been made to tackle each of these problems. For example, the supply of drinking water has been greatly improved and-as of 1980 it was estimated that 80 percent of villages were served by government pumping and water treatment stations. Often, however, the water supply is too far from individual households to be used regularly (Oldham, 1984). Since the Revolution, Egypt has developed an unusually extensive network of health services intended to provide curative and preventative physician-based services free of charge to the majority of the people. The ratio of inhabitants per physician is exceptionally low for a country of Egypt's income level (970 persons per physician in 1980; World Bank, 1984), and zhe requirement that new medical graduates serve two years in rural areas assures access to physicians outside of urban areas. The health service system does not function as well as intended, however. It has been criticized for bein= clinic-cen:ered rather than communit-orien, fn assind =raruraii areas, medicine of the physicians assigned cc rural aeas, andor s policy of f=eTuent transfer of health perscnne! f== c.e c:=n- ity to another, which unde=ines the establishcfnt of wcrkin relationships with ru=al residents (1kra=, 98gs). One analysis, utilizing ErS data, reveals no relationship in rural areas between the accessibility of health clinics or the density o. =edical personnel and the level of 1nfa.nt and child =ortality (Eid and Cazsterline, 1983). The sa=s st=dy identifes a =eder- ately st=ong- negative relationship ketween the presence in villages of trained midwives and neor.atal ortality, suggesting that emphasis on greater density of these particular health personnel might have beneficial i4pacts on infant mortality levels. . A competLig interpretaticn of the sources of high =rtali:t. in Egypt e=phasizes social and econ=ic conditicns rather than the effectiveness cf public health and sanitaticn =rcra=s. Cne basis fer evaluating the validity of this arpe~.t is analys-is cf sociceconcic differentials in infant and child craity. Tw detailed analyses of the ES data reveal su=prisinly weak effects of variables such as type cf place cf residenca, =azerna' schooling, =aternal e:plcyent, paternal scheoli-.g, andatera occuation (sail .E. et al, l93 a:d castri, :n the analyvsis presented in this rw:cer, we ri rcesie2 these effets in c==.1 ti i t-cl:stiati. wi:Z. inccee efects, these la-te CctS n=tr:i- . C: factc=S are =ne CZIef -~ mortality in Egypt: first, demographic characteristics of the mother and child, specifically maternal age, elapsed time since the previous birth, and birth order of the child; second, region of residence (Lower or Upper Egypt) , which plausibly reflects the impact of environmental conditions or, perhaps *more fundament- ally, customs and social behavior which affect the care voung children receive. The evidence therefore is not entirely supportive of the view that improved sanitation and access to modern medical services is critical to reducing infant and child mortality in Egypt, nor of the view that modification of social and econcmic conditions is essential. In our view it is premature to reject these hypotheses, however. The potential impact of public sanitation and health measures needs to be assessed In more controlled studies, especially given the alleged shortcomings of the public efforts to date. Similarly, the significance of social and economic conditions must be investigated on the basis of a balanced set of indicators of the range of pertinent social and economic dimensions. The previous studies did not inccr- pcrate indicators of income or wealth. When such fundamen=a7 sccioeconcmic factors are emitted, interpremtation of diffe,nt ials according to demographic characteristics and rei= is hampered. Lacking -ertinent measures, i s noc tcssit a: explore the extent to Whic emograhic effec=s indirect exressions of: secal and sc.c=i sacffZs. more, cur ability to derive useful o:ncLusicns differentials is limited by the inability to dt::e wheth r er not particular social and econoic variablas account for the differentials. 2.2 I;co=e Differentials in Mortality =n sharp contrast to the proliferation of studies that attempt to establish the role of such sociceconomic detei-.ants as parental education or place of residence upon sur-vival chances, there appears to have been a ccmparative dearth research concerning the effect of household income upcn infant and child mortality. A pri=e reason for lack of knwledge in this area is that household data sets incorpor atinç infrc=azin on both child survival and. income are very rare. (Sullivan, Cochrane and Kalsbeck, 1982). Instead, ether va=-ab=es, such as parental education and ccupation, are equal size, and thus the sample distribution acacrding to Inccme is of little concern. The sample distribution according to income quintiles is shown in Table 2. In restonse to concern about the accuracy of ear n -- income, we also examine two ;roxies fcr incmea, name> , useho exn,endltura and nu=.ber of- modern ccnsum:er drb on household e'pen.dtue was obtaired th%oegh -.a questi.n Cnly on total household expenditure in the =onth precedin tZe Interview. Preliminary analysis indicated that expen.iture is highly correlated with mCst of the incoe measu=es. This is also evidant from analysis presentad in the principal report (see VolUe III, Tables 2.5 and 2.6 and accoranyir.g text CCIZMks, 19833). Further, in analysis of child s-rvival these two types of measure, expenditure and incoe, perfcr=ed in a siilaz manner. Consequently, estimated relationships between househod expenditure and child survival are not presented in this rercrt. The sa=e set of considerations do nct atzly to the other proy zeasure, number of modern consumer durables, which .on bc theoretical and e=pi=ica! bases would see= to remresent a distinct economic dimension. r. the enc=:c survey, the household informant was asked about whezhe= =ne househod possessed each of the focwing ten items: radio; television; gas stove; bicycle; water heater; =etorcycle; tele,hone; =rivate ca=; sewing machine; refrigerator (electri or gas). -n additi4c., whether cr not the dwelling had elecricity was ascTnrand. -te sourca of drinking water and ty:e of toilet facilities ai.e to the household were also ascer=ained, ecause Co the s-ectal ra ea-.c of these ar les fer r:r-a..r- we ::nside: the= sepa=ately below. We Co=Mne vhesa ele:en iters - CCnsumer dura*les plus elcri- - in=o an . z2=s. duratles tv s=tunti-.; t-.i ossessad. -at is, each i4a= ts -eza account is taken of households possessing more than one of any item. An analysis of the household variation in the index suggests that possession of a radio, television and refrigerator are the chief determinants. Analysis using an index based on the ten consumer durables, excluding electricity, was also carried out, but the results proved to be the same in all important respects. The distribution of sample births according to number of consumer durables (Table 2) indicates that roughly one-fifth of the births occurred in- households lacking all of the eleven items, whereas roughly one-quarter occurred in households posses- sing four or more of the different items. It is common practice to develop an index of consumer durables,- such as the one we utilize, as a proxy for household income (Farah and Preston, 1982). Such an approach is appealing because it is much less demanding in terms of data collection:- inquiry about a small number of household possessions requires much less questionnaire space and interView time than does a satisfactory inquiry about income, and the responses are assumed to be more valid due to the greater simplicity and lesser sensitivity of those items. Nevertheless, it is im=orCan: o recognize that an index of consumer durables is only artv: adequate as a proxy for household income - variation in awr.ers- of durables may be substantially affected by cmher varaes, including the price of gocds (a funof o oale), :h -*-e -- household income (cash or non-cash), and :seclpre:erence about hcw to use di-sp=sable_ i Ined w;ih 27 latter point, although a high score on such an index ref:ects a high level of household incoe, at any given incc-e level it alsu reflects a high expenditure on non-child ite=s, i.e. direction of disposable income away frcm resources that are relevan to child survival such as facd, clothing, or quality of hcusir.. Such an index may in fact partly funct_ion as an inverse proxv for parental tastes and attitudes toward invest=ent in children (Farah and Preston, 1962). Thus, from a thecratical stanoint, it dces- not necessarily fo1cw that an index of consu=er dura`Ies should be negatively associated with infant and child =ortality. Other ex-clanaterv variables Additional explanatory variables are selected cn the basis of theoretical considerations and on the basis of eirica evidencefres Zgpt (_small A.E. et al, 1903; Eid and Casterline, 1983) and elsewhere. Naturally the selection is consaid the variables cffered by the Ers. We classify births according to the region and tye of pace of residence cf the household at the tie of the survey. Rec4- has two categorias: lower, which incSzudes the co f Alexandria and Caire, and C.Uper. Type c -place of rs n= three categories: rral, urban, and =-trzzc'.itan, Z_h - consisting of CairO and Alexandria. ,rat of s variables su=*. as education and cccupaticn, rac_-.. . VVZ= =7vz- -=n u=easurad SCC-a and fa=rs of place of residence, on the other hand, should be associated with sanitation conditions and access to medical care. Both region and type of place of residence showed significant effects in earlier analysis of the EFS. The distributions in Table 2 indicate that approximately two-thirds of the sample is located in Lower Egypt and a similar proportion in rural areas. Roughly equal proportions of the non-rural sample are located in urban and metropolitan areas. Maternal schooling is now recognized as a fundamental determinant of child survival in many settings (Caldwell, 1979; Cochrane, O'Hara and Leslie, 1980; Cochrane, Leslie and O'Hara, 1982; Farrah and Preston, 1982; Callum, 1983; Martin et al, 1983; Merrick, 1985). The relationship is far from universal, however .(see Hobcraft et al, 1984). Maternal schooling affects the care children receive in a variety of ways, in particular through nutrition, hygiene, and utilization of modern medical therapies. On the basis of previous analyses of the EFS data, ho-:ever, one is forced to conclude that the impact of maternal schoinc o. child survival in Egypt is rather weak (Ismail A.E. em al, 1922; Eid and Casterline, 1983). Modest negative effects of sc*occ!`- beyond primary level are apparent for pcst-neona=al and ear'y childhood mortality. However, only a small ec Ce. p - occur to women with this level of education (six er:e.h sam.le for this analysis; see nable . i i of .z: effects of maternal schocling is neverheless an ia -- thi_s anal%rsis, fcr = ,a reasons. w.;=*- _ the mode5t negative effects on mortality of sc.ooling eyod a pri=ary level persist when household ino e is cn-olled. sa, this has obvious implications for the value of irasi. female educational attainent levels as one component of policies to reduce mortallty in Egypt. Second, we wish to deter=ine whether maternal schooling and househcld ic=e ccndition each othe='s impact on child survival. One =ight hyctnesize, for example, that income effects are sharper for better educated mothers, as such =others =ake mcre effective use of the addition- al rescurcas available. A number of alternative hypotheses coancerning interactive effects are also plausile a We combine education and cccupaticn of the father into a-. index of paternal sociceconomic status. The categories o the index are defined as follows, ranging fro i (low' status) to : (high stätus): i AgrIcultural ocupation, and 0-6 years of schooling. II fanual occupation (skilled or unskilled) , and 0-6 years of schooling. I Sales or service occupation; agric r or =na- occupation, and 7+ years of sc*-oing; cler4ca.:r professina! occupatin, and 0 years cf 27 carical or rofessional c ation, a .- -tars scnoLing. A~roinoa.yf=**y =er=enz of tita sa=pie' fal azzc-:-- and =3 thar te =een-4n =Itvz-* 77 .:ur h117='ses -essa=.=zt.--:az f=tra szzzz Czýz, almost identical as those for household income. (Both are indicators of economic position, defined most generally). Thus a priori we might not expect net effects of these variables in models which contain both, but in reality the measurement of .each is likely to be far from perfect, and thus we should not t3 surprised if the measured indicators tap different dimensions of economic position. It is of theoretical interest, and of relevance to the design of future data collection efforts, to determine which variable shows stronger and more pervasive effects on child survival. Past analyses of E:S data reveal significant effects of paternal status on child mortality only: births to the best educated and white collar fathers are less likely to die. We include in the analysis a set of four variables more directly related to mortality than the socioeconomic variables just specified. Two of the variables are selected to capture presumed differentials in quality of care and level of investment of resources in children. Sex of the child is meaningful if care and investment of resources vary by gender. Egyp is a strongly patriarchal society, and there is a considarable body of ethnc- graphic data suggesting that male children receive pre i treatment (Wilbcur, 1887; Ammar, 1954; Ayrcut, 1963; Critchfief, 1978; Lynch, 1984; Oldham, l984). Indeed i' is Ssi za those children who are execzed to be more acoina- as adults will receive a larger share of fami4y children, especially in a low income countr suhaig: (Rozenweig and Schultz, 1982). In an effort to mursue futher mortality differentials due to differential invest=enz ir child=en,.we classify births according to parental (naternal and _ - paternal) responses to a Vuestcr. about the ancunt of sc*nc=.4-- desired for a daughter. We regard th!s variable as an indlcatr of willingness to invest resources In children. Attitudes towards the schooling of daughters wculd seem a zore sensitive indicator of this factor than attitudes tow-ards the schcling C- sons, since status atta.nent of sons is =ere hichly sanctioned in traditicnal Egyptian society. Clearly the =easure is a =cugl proxy for the concept of interest. For one thing it is gender- specific; in the analysis we examine whether effects on =ortality are restricted to girls. More generally, we suspect it serves as uch as an indicator of attitudes towards :enale rc`as as o attitudes tcwards invest=ent in children. Attitudes towar4s female roles is not irrelevant to our underlying concern, however, as enhanced female power and status nav well be asscc- iated with the availability of more resources for chIldren and tetter child care (Caldvell, 1979; Tecke and Shnrter, =S data reveal rather high aspiratIe.s for schccin cf ers: both =arents of fort-wo percen: cf the airts desire universitv educat4cn fcr a daucnzar, tcz,- parents :f ot agrae=ent tetween v_Jfa and husband cn =ata 4s - ratta= -- c=!?=e=n vlf zte 211-- - =Za* =,wc * categories, however, as this permits us to consider whether maternal or paternal attitudes show stronger mortality effects. The inclusion of measures of household sources of water and toilet facilities requires little explanation, as these are directly pertinent to sanitation and hygiene conditions. For example, access to internal piped water in the household is likely to reduce exposure to waterborne diseases, especially diarrheal diseases. Furthermore, this may also lead to an increase in the amount of water used, thus contributing to better hygiene. The presence of more modern toilet facilities should lead to reduced exposure to gastrointestinal diseases. However, the empirical evidence is mixed on this matter. Poor water and toilet sanitation are found to be associated with high levels of child mortality, as expected, in a variety of settings (Zachariah and Patel, 1982; DaVanza et al, 1983; Martin et al, 2.983; Merrick, 1985), yet numerous studies find little effect on disease prevalence of the provision of unpolluted water and tailet construction (eg., Stephens, 194.From a review of the pertinent literature it would appear that interactive effects ma% exist between these sanitation variables and other sccacnoni variables. For example, Stephens -found th.-at the-. advanzages cf better sanitation facilities in Ghana were severely liized wn mothers were not educated. Conversely, the coinain f adequate facili:ies and a mcderaze a-cunt of =aena c cn a-t-ea=ad tc redu-e the ris.'- of deatn isan" (1985) also f=und that the effec': cf~zrae water in Brazil in 1976 was somewhat greatar a=cr.c Lw incr.e groups. One explanatlon for the mixed and rather coplex sat ci - - findings is that although sanitary lacilities car. aid nz. t--a maintenance of sanitary con.ditions, they are noct inthezles true indicators of the level of household sanitaticn. T!at is, variables such as source of drinking water and type cf t'.et facility' represent deteinants of the ur.easured varia*les which directly affect child health, na=ely quality of water and the extenz of environmental contamination å:n hunan Zaces. To be more szecif4c, the existence of pied water in a househCld d:s not guarantee its purity, although it may be exected to increase the level of purity by an unknown extent. ,urther, enonarta. containation from human feces is a function of sawera5e dis=csal in the ccmunity, nct merely the toilet facilities in a lar dwelling. In this respect these two dwelling Character-istcs are :Lamerfect indicators of the variables of interest. On t-e other hard, public or private ef forts to improve sanita:izn ccnditions consist, in the first place, of establishen- water systems and sewage disposal systes. tsh standroint of =ublic ;c.ioy it is essen.tiaL t: i.zn assas4r. the impac-. of these systms----. -eilt .eves, an s:se.e. to ex:lore why anticipated impacts ara -.: viert - : -e-- -. becau..se .:dwazaer -'s n--=~ y . Rouch2. one-:hrd oz.te h ~ a c~ or the outside of the dwelling, another one-third frCm a public faucet, and one-quarter from a pump (Table 2). The sources comprising the final category are a well, the Nile or a canal, and other not specified. A Oriri it is importan to distinguish public faucets from those attached to the d,alling, as in the former case large quantities of water may be drawn only occasion- ally and stored for relatively long periods in containers in the dwelling, during which time they can easily become contamInated (Oldham, 1984). There is less distribution of types of toilet facilities. One-half of the households of the sample births rely on a non-flush toilet used exclusively by household members, while about thirty percent of the households have no toilet. The remaining, observations fall into categories of non-f.ush ccmmcn toilets, and flush toilets for exclusive use by household members. The latter is expected to provide the best sanitary conditions. We distinguish non-flush household and nron-flush common in preliminary analysis, despite the small number of cases in the latter, because of a suspicion that a non-flush taie: attached to the dwelling may be associated with worse san.iation conditions than non-flush common toilets, or the lack of a toilet altogether, since disposal of seweraae in these insancs should be on average more distant from the dwellin.. The final two variables affect surial -i i t:ircugh the he ct f the mo.her and th-*e cant-zf chid.Tncw - an cnes~e- impact on child survival of pregnancy at unusually yUng er older ages, higher parity, and shorter elapsed ti=e since the previcus pregnancy (Gray, 1981; Winikoff, 1983, Hcbcraft et al, 1983, De Sweemer, 1984). It is believed that these effects of mazerna. age, birth order, and birth spacing are largely a censeraence o variation in birth weight of the child, itself 1eavily influenced by maternal health. The previcus analy.ses of the 3yS dcCu=ented the significant rOle of these demcgraphic cacteristic i determining variability in mortality expeience in Egypt (s:ai! A.£. et al, 1923; Eid and Casterline, 1933). To maake the present analysis co:putationally more tractable, we combine =aternal age, parity, and length of interval since previous live birth int:, ene index of maternal risk status. The index is obtained by egress- ing whether or not the child survived on the thrae variables (entered as categorical predictors) and then eC=uti rdc probabilities of dying from the estiated e=aicon. "he index thus created Is a weightad sun of the three variaates, with .net effects on the probability of dying serving as the weights. (:he estinated lgistiC regression equations frc= which the index is computed are presented in Apendix 3). The i.dex is cbvicus computed in such a wav as to aximze the estz a=- effe =aternal risk status cn *,a exanied sir incor-oratad in e,uaticns with other si rs, a z. Vt "ndex gives an exavrased s. cf th.e !ac=crs J-t -ja-resentz. But t.ha fac===5 has n=W 'Zeen sr:a o including several on Egypt. In this analysis cur =ain purpose if incorporating these variables is to determine to what extent they serve as mechanisms for income and other sociceconomic effects. The index constitutes an efficient means. of accomplishing this goal. For the analysis, the index is broken into five categories (low to high maternal risk), each containing between fifteen and thirty percent of the sample observations (Table 2). Effects of maternal age are probably largely due to'associa- tions with health and vitality of the mother, but increasing mortality with maternal age also reflects increases in genetic abnormalities. Another determinant of genetic abnormalities of special relevance in Egyptian society is the blood relationship between spouses. About one-half of the sample births were born to parents related as first or second cousins (Table 2), reflect- ing the traditional custom of arranging marriages within the patrilineage (hamula). In their study of child mortality differentials in Sudan, Farah and Preston (1982) found that child mortality among women married to their first cousins was twenty percent higher than among families in which the husband and wife were not blood relations, with more distanz ralaicns havi4n- intermediate child mortality. As these autho-rs ncac, there may well be a genetic component to these differentials. Alcn wih the possible manifestation of inbreeding deressicn, (Schulz, 1972), the social ramifications of marriaqe withi- ed kin should also be noted. -he szatus an-4 pc:e: me. en..:c within the household ay well var acc:r-in :: they have been =matched to a me-ber of the hanula, with crresmt.- ding impacts on the role o young children and the care thev receive (see Caldwell, 1979). Thus esti=ated effects of the parental kin relationship may reflect either bicjcgical or soc,al causation. Measurement of all of the explanatory variables, with the exception of the maternal risk factors (age, birth c=cer, spacing) , k1i relationship of the parents, and sex of the child, refers to the date of interview. As we will srecify i.n the next section, the ortality measures which serve as demender.t variah- les refer to births in the decade preceding the interview. Hence the analysis iplicitly assues stability in the explanat- ory variables over the referenca per.cd. The values cf the variables need not be assumed to remain ccrstant. Rather, the magnitude of change in any partIcular variable =ust be rcugh' the sa=e across sub-groups of other explaratery variables, s= that the pattern of parental and household dierrtials is apprcxinately preserved. (Lack c thange in all sub-grcu=s neets this condition). We are =ost t nhat zhs assu==t . _ valld for regon and type of =lace cf resid-.ca, =a-.a-. schooling, and paternal stav-s, and leas applias to dwelling characerisaics 'aters'., paciities, oresence cf cc~u,er dura'.as a-.d t.se:I. M% Iate varlabIes are %Ize centra f=u . 2 '.ý --ar zn=s fnasrne s& s s.~ are random with respect to mortality - which seems on balance to boe a reasonable hypothesis - will tend to weaken the estimated strength of these variables relative to the others in the analysis (Menta, 1971). Ifortality Measures We consider survival experience during three age intervals:- the first month of life (neonatal mortality); the second through twelfth month (post-neonatal mortality); and the first through fourth year of life (early childhood mortality). In addition we carry out analysis of survival during the neonatal and pcst-necr.- atal periods combined: the first twelve months of life (infant mortality). The mortality measures ultilized in the analysis are binary variables, coded 0 and 1, denoting survival or death respectively during each of the four age intervals. Analysis of each is confined to children alive at the beginning of the interval in question: live births for neonatal and infant mortality; children surviving the first month of life for post-neonatal mortality; and births surviving the first twel.e months of life for early childhood mortality. :n the 175, respondents were asked to report ages at dea.h in c==Ipeed months and years, and thus the mortality measures vield aqzrs- gate-level probabilities which correspond exaczly with i - probabilities of dying ( , , and so forth). The auality of the Z*s morali! data was cazafully ized in the earlier analyses of m ralt (rail ... a 1983; Eld and Casterline, 19823). The fi.rst point to ct is that the EyS estimates of infant =ortality exceed other esti--aes fcr equivalent historical periods (for exa=ple, see Table 1; or compare with the' esti=ate for Egypt, in World Bank, . Evaluation of the data on the basis of a set of cnter I. rteria - sex ratios of births and deaths, age distribution of deaths, and incidence of age at death "not stated" - prov-ides no grounds to doubt the overal! valldity of mortality data for bi-rtns durin the decade preceding the survey, the observations selected for this analysis. There is little evidence of substantial oissicns of births which later died. The distributie. cf age at death suggests imprecise reporting, however, as there is noderate heaping at half-year intervals, twelve =onths of age in rtica- lar. This cautJons against excessive e=phasis on dif according to age interval in the esti=ated effects cf exp'anazcry va-riables, although we are confident that the classificatio! of deaths by age interval is accurate on balanca. 3.:! Selectient of Cases f-er A.navs The analysis is restrictad to irths durin n years preceding the survey. Vo av4oid censcr-nv -as, we a13C --4t births lacking exosure to te enIre ave inoerva'. irsd i. twelve =cnths vrecedi the sur=ev f=r the a =asis - inf=:n =orta'ity (neonata! and pcst-.e.ata-*-,, an-i` duriz sixtv :cnths receding the sur-e* fr the ana'.;s. noroa it'. .Als ex=Iuded are a s -all r: ei -ir h lacking information an year of death: (-- percent are excluded from the analysis of infant mortality and -- percent from the analysis of child mortality), plus additional births lacking information on month of death (4.1 percent excluded from the analysis of neonatal and post-neonatal mortality). Finally, any births lacking valid information on the explanatory variabes under consideration are also excluded. The data analyzed in this report are derived from a complex mixture of components of the EFS and analytical levels. The First Phase Individual Survey and the Second Phase Household Economic Survey provide the information from which the measures are constructed: Survey Variables Individual survey: mortality measures; maternal risk factors; parental kin relationshi-; sax of child; schooling aspirations for daughters; region and type of place of residence; maternal schooling; paternal status. Economic survey: source of water; toilet facilities; nat household income; per camiza househld income; index of consumer durables. Some of the variables assu=e dfenovalu;es fcr chlre f h same parents, some are regarded as fixed characzeristias cf parents, scme as fixed ctaracteristics of househods, ad=c as fixed characzerisc:s of camunities: ?knavtical lev.el. variables Bortality measures; =aternal risk factors; sex of chid. Parenåts parental kin relaticr.shi; schcoo.n aspirations for daug'ters; =ater.al schoolLng; paternal status. Household source of water; tollet facilities; ret household inco=e; per carita husehold Inco=e; index of consu=er durables. Comun.ity region and type of place of reside.nca. It should be rotad that the adjustment of per capita hsehol incoxe in the case of children who die .(see Sectior. 3.1) intrc- duces some variabillty in this easuare a=or.g births ir. the sa-e household. A more general theoretical point is that variables at certain levels .ay wall be powerfully dete factrs oparating at a higher lavel. For exasple, household tssessin of piped water is dependent on the provisicr, cf pi=ed wazer the comunitv. The number cf births, mothers (or aciaentl, carents', and households selected for the aral.sis are shb*-.irn I ae place of residence. The nu="er cf csh s ah.usanr corresecnd ri:ta CIosCly: fau hsehrs-an ===»e z- ave=-=arrIed =CS=Cr.d_nt wi-th a irhdu=rin= =hna rr.apei not s"--vs- - av cf tha nuIez .e - uacue'd s--- e -e'':E72c.n ' :n---- averace =cre thar. cr.a O tt vhe n:,:_=.s nine-year period selected for the analysis of infant mortality, each woman contributes about two and one-half births. This fact has some bearing on the drawing of statistical inferences, as will be noted in the next section of the re3ort. iv. ANTALYTTC APPROACE 4.1 The Model The choice of the explanatory varia.bles described in Secticr. III is guided by the sley-Chen conceptual 2ra*ework (see Figure I) and other literature on the deterinants of child surviival, previcus research on infant and child =ortality in , and the availability of pertinent variables in the .75. .n additicn tc hypothesized effects on =ortality, we expect the explanatr variables to be related to each other in a comlex fash-n. Previcus research provides less assistance in speciying the natura of these relationships among explanatory variables, as =ost analysts have estimated only single ecuaticn =odels for .ortality (but see - )errick, 19S5). We rely again on the Mosley-- Chen fraework, with adaptation to the specifIc variables we have in hand. The zcde! we estinate is-presented in 3ig're -. Tour sets of variables are distingulshed: (1) Backgrcund factors: fncome and other sccicecc=.=ic varia- bles. It is te=pting to specify :-rthar ca'.sa! =ala-- ships among these variables (for exap'e, inc=e an =aternal education, or inco=e an-d paterna'. szazus), given =easure=en. c- eacb fc the data c- Zhe sur-- only, it sees =:st se .i'..e to assUr.e a set =2 2:-3 . associatirs. (2 fotrve-t.i-: ari ses sex of eti', scan.:g s -ar.3 FIGURE 2 RURAL/ MATERNAL PATERNAL URBAN EDUCATION STATUS TOILET SOURCE OF SCHOOLING SEX OF FACILITIES WATER ASPIRATIONS CHILO MATERNAL KINSHIP ENVIRONMENTAL NUTRIENT I R FACTORS RELATION CONTAMINATION DEFICIENCY HEALTH STATUS Y ----- ----- These factors are assued to be dete=ian.arn zr ccrrela=es of uneasurad factors which affact the proiate daer=4n- ants; Sex of the cbild and schooing aspi.rations are assumed to be asscciatad with g,a1ity of c!1d care (.utri- ion, clothing, hygiene) , whl.e t!:e othier two variables are assumed to be detrminants of househcld saritatier a.d hygiene. Sex of the c..1d should be rar.dz w t a res=ect zo the background factors, but t.e other three faces are modelled as functions of the background fact=rs a td thus can serve as indirect mechanis=s through which inc=e a.d oter socioeconcoic variables express their effects. To select a- example of some importance to this analysis: inome effecs zight be due in part to varlaicn acrss i.r.c=e strata in the, qualaity of housebold sanitaticn, a.d hygier. (3) Proximate deeiants: materal risk factors, and pare.tal kinship rationship. We lack di.ect indicatcrs e the other four sats of roxiate detaminats, and ths we =usz assume that direct effects of the bazkground factors and intervening variables persist wit. czntrcls -: ae al risk status.L at . isk status, a ucti ci ra:er-a. age, and le :!th of r assume t: b daza=-.=ed i4n =a=-= =.ý:te cgrudac:, (4) Health status and mortality: neonatal, pcst-neonatal, infant and child mortality. We lack measurement of morbidity, and of behavioural responses by family members to child illness (personal illness control, in the Mosley-Chen framework), and thus can only model effects of explanatory variaz!as on mortality itself. The model serves several important purposes. First, i leads directly to specification of a set of equations to be estimated (see below). Second, it provides a basis for interpreting / differences in estimated effects between equations, for example, effects of the background factors both without and with controls for the intervening variables. This point will become clearer as we discuss the equations and the results. Third, and perhaps of greatest importance, the model makes explicit the gas and shortcomings in the empirical analysis necessitated by data constraints, in particular the lack of measurement of four proximate determinants and health status, and the inability to model relationships among the background factors. 4.2 Eauations We estimate the following sets of ecuations: (0) Mortality = f(Sociceconomic) (1) Mortality = f(Income) (2) Mortality = f(:ncc=e, SociceconC=ic) - Maternal. risk ststus) (5) Mortality = f(1tervening) (6) Mortality a f (laternal risk status) (7) Intervening - f(Incce, Socioeccnc=ic) (8) Maternal risk status f(Icoe, Socieccnc:i) where Mortality consists of binary variables d survival or deatZå durin the neenatal, post-neonatal, in n,and early child:äcod periods. Socoeconomic consIsts of region and type of tlace o residence, maternal scholing, a.d maternal status. Income ciosists of net hous4hold in.cc=e, per ca=-ta net household icene, and the Index c! consumer duaIbles. Intervening ccnsists of sex o2 the child, schoolIr aspDirations for dauçhters, hcuseh.d source of water, a.d tousehod t-le faciliti~s. Maternal conissts of the fex ri4sk status characzaeisz'cs a.d the parental knshJP ( de. s a se= c f vaiates zs s d ad .e() .e. . . - --h entered as categorical, non-linear effects are permitted; and both additive and inter- active effects are tested. The three income variables are tested individually in separate estimations. Equation (0) then, is estimated. four times, for each of the four types of mortality, while Equation (1) is estimated twelve times, for each combination of the four types of mortality and the three types of income. We structure the analysis to focus on income effects. Equation (0) is estimated for comparison with previous analyses which have not incorporated household income data, and to allow assessment of the impact on estimated effects of socioeconomic variables (maternal education, in particular) of controls for household income -(comparisons of equations (0) and (2)). Equation (1) provides gross effects of household income. Comparison of income effects in equations (2)-(4) with efffeczs from preceding equations (1)-(3) indicates the extent to which income. effects are due to associations of income with other socioeconomic variables (comparison of (2) and (1)) and the extent to which income effects ozerate throuch the measur- intervening variables and zroximate deterinants (cr.ariso. - (3) and (2), and (4) and (3)). These c==arsczs-s strongest causal analysis: we are -zeared to arue fcr exa=mla, that disappearance o nccme effects ih czn::s for :he of household sanitacn :acili:ies indicazes that zhese serve a- th mimrvmecnanismns cff incce effacts. C '--a effects which persist net of all other variables are of ccnsider- able policy interest, but we will be unable to identIfy the source of the effects. It is nevertheless of sO-e value to rue out certain factors as the sources of thesa effects. Equations (5)-(s) fil! out the =odel and are essen=ial complements to eqrations'(1)-(4). Equatlons (5) and (7), for example, assist i.n the interpretatio of the im=act en inc=e effects of controllir.g intervening variables, by idicatir.g the nature of the effects of intervening variables o mortality unadjusted for background factors (equation (5)) and of effects of background factors on intervening variables (eVuation (7)). EqUations (6) and (S) serve a similar fnction in the analysis of indirect effects of. ircome through =ater.al risk status. 4.3 Statistica! Medel The demende.t variables i most of the eo-lati.ocs are tiary indicators of survival or death of the child. Sin='e 1inear regression estiation of eq_ations with binary demer.dent vara:- les is inapropriate, for three reasc.s (2:enta, L97). the varance of the dependart 'ariable is a funct---n c-f =he ean (the var±arce s ( where m is the mrcoe=J. s:r in : dy ig), and thus the :ones:adasti:ity assun==z: is : Seco.d, radicted alues ne.ed n== fal- ztzt. a ranv This is oe indicaon tha= a linea= GCua:== i : a eaas= seriv43, 1=: .3s n=z sans~l.oe t2 a3=ta-- -%r distributed. There are a number of statistical models that are more appropriate than the simple linear regression model. We choose the logistic regression model, which may be specified as a general linear model of the following form: E[ln(d/(c-d))] = + X where E denotes the mathematical expectation operation 1n denotes the natural logarithm transformation c denotes the number of children d denotes the number of deaths to these children is an estimated constant is a vector of estimated parameters is a vector of explanatory variables and c and d are specific to sub-groups defined by X. If p, the probability of dying, is defined as d/c, the statistic- al model can be re-written as: E[ln(p/1-p))I = + Ln (p/(l-p)) is the log-adds, or legit, of mortality, and .hus the model can be regarded as a 1cgit-liear model for ncrtaity. The data are assumed to be binomially d i e. The equat- ions are estimated using the P: Prgrim ithin the stais- ical software package (Dixon, 1933). 2 For a fulLer discussion of this s:atisticaL me, reader is direczed to Aldrich and Nelscon, !L24. ALL explanatory variables are entered as categorical variables. Indicator variables are generated for each category, scored i if the child belongs to that category and 0 otherwise. The indicatcr variable for one category (r.c=ally the first) is omitted from the estimated e<7ation. Unider this "dv variable" approach to handling categorical variables l regression ana'v- sis, the coefficients an the indicator variables for the other categories represent centrasts with the gmittad catevcr-r. m. this instance, 'the coefficients represent differences in the loet of dvjrv between particular categories and the omitted category of each explanatory variable. Some flnal points about statistical testing. Standard errors are calculated for each coefficient, and these perit . testing of whether each coefficient is diferent fro zero. 2"7te that this is essen-tially a test cf the difference in net effects (on xortality or other dependent variables) between each category and the omitted category. A global test, of the statistical significance of the variable is provided by a co=pa=-isc- of tne fit to the data of the eviation incudig the var question as cc=pared to the ejatio excludi-.n the variatbe 1 chl-square test cn the differeZce in Le-Like itcds. tratice we carry Cut this test -c =- tn : e:a.: cut m. the asis cf a gital 7-t e rv se c efii representing each *ariab'e, usi-. -.oe estitated asyst:.ti: variae-ce ia-c -rix of the c- -.s '.es,, --- I ion procedure). This procedure is computationally much simpler, not a small consideration given the number of equaticns and number of observations in the analysis. A few comnarisons of the statistical tests from the asymptotic ccvariance matrix procedure against chi-square tests of differences in log-likelihoods between models suggests that the former procedure is, if any- thing, a bit conservative, that is, slightly less likely to shcw a variable significant at any pre-assigned probability level. Two other factors should act in the opp9site direction, to an unknown extent. The two are in formal terms closely related. First, the EFS employed a complex clustered sample design (CAPMAS, 1983, Volume I), but the statistical analysis assumes a simple random sample. Second, as noted in Section III when commenting on Table 3, there is an additional clustering of the observations (children) by mother, which further undermines the regression assumption of independence of observations. The impact of clustered sample designs on statistical inference fror. non-linear models of the type employed here is, as yet, a relatively unexplored topic in the statistical literature. It is well known that in linear regression analysis szandard errors are biased downwards when the sam-le is clustered, resulting in too liberal statistical tests (Yenza, !97LI. Presumably the direction of bias is the same i4. -.-.4Lnear estimation. 1rote that inclusion in the =cdal Cf account fcr hen nc-idenendence of cbservazicns alle'a:=s M'- bias. We utilize several maternal variables 'eda::in : status) which plausibly serve this functicn. V. FINDINGS We begin by considering differentials in mortality rates according to each of the explanatory variables, and then turn to effects obtained from estimating equations (0) through (8). 5.1 Bivariate Differentials Neonatal, post-neonatal, infant, and child mortality rates for each category of each explanatory variable are displayed in Figures 3a to 3d. Differentials according to socioeconomi.c factors other than income are shown in Figure 3a and generally conform with expectations and previous research or, mortality in Egypt. Xortality is higher in rural areas, but the differential between urban and metropolitan areas is slight. We comment on this differential further when examining multivariate results below. The regional differential is very sharp: children in Upper Egypt have markedly lower probabilities of survival at each age interval. Consistent with evidence from many societies, mortality rates decline with maternal schooling: chi41den a women with seven or more years of educazion az;ear to distinctly better survival chances. The -a=tarn of raes according to paternal status is uneven until the chidhood period, when the rates decline mcr-niall by paterna s . At1: all ages the hi*=!hest_ -is~-.. group:- (childran off und=Sed n e==_CI:Ac INFANT AaND C?ILD MORTALY RATES (cer 1000)- (NN: necnatal PN: post-neonatal ZN: infant CH: child) FIGURE 3a ama ___ ____ ttteNfl � :[1 n м (и n W+ -� r л; .ц (v .-t д; а1 [: ;; а �й г •-� ro w и У+ и и �н -- а � д з о и го � .-+ � 4 . . • � � • м д й g � �• и � t i м :: п l �� � � � У • .• • �, � -- , . t . � i � с . . ' и .1 р 1 �Г ` • К•.1 � •л � 0 1 • .. и А . а. _ г г ° i а у У .. ... .. . � l��111�I l.1 �•;� (,1111f 1.! ��;1 1 f 1111 1 1 f� , � �,�I ./ / _ ~•I'" / �� � � � о - � м . �i_ :���_:_�°-_:.:_ � __ _ . �!�1111_1Ш_1.1.1i^i11t1 "����i ���iiiiiь��i� �'� „•;1.`,,.,... ...� ' � � � � . ,,. _-`•.n.�_4I����_�111�111 1,1 ������1������,���� "������ ���`\� . �.�\•.��?\ ') . . . и ( : . . . . °�Tl_.Т����Т11_�I _I.IIr _ ��(1_1/li/1i;� \ '� _ _ _ _ _ _ _ _' - � F� _ ' _ _ _ . . . _ . . r �.� . . . . � . � ��ia.��i_l_I_i.I.il�i•_ii.l�i Ii�i•_i� � _�h��i•ii�� !i\i � ii. � � ;; . .��:i�.���.i��i�����������+��► � w �����i���` � �h � ���11_1� \ \������ �; �,�'� ���`.�. . . • . , . . . •. � . I Н _ I • � • �!' =_77_I_ Т_Т_Т_ _ � -_- _� 1 �II _7.�,� � ; � .. Ч- �- _ --- = а � а �: _ - -- � и `_ °• _ _ __ ' : � �' " ' � Л • I I + . ��Cll1`�f_ 1_ 1^.1. _1 1► �1: f. i!.! .'�1.1. _� .11 ! 1 f 1111 11 �: �1����� �����1 _ ::� . . . . . � _ �[ г г �У 1 / •�I � . � � -- � � ` -=У-`.-'-'-. � • - � � � � � ���.�.�It �YIa r `�. . •�I � � � LI � ( ( I � А� �� l�� ' � . . • � . • . . � (�j��уΡQ q* ��т-а� ��--- 1' 1 1" 1 1 � 1 � � �� ^ � N V g У� V � Ч I 11 � .в• в ,i .. The final two variables ara intended to caPtu=e aspects cf maternal ri.sk status (Figure 3d). Effects of demographic characterlstics of the =at-her (age, parity. elapsed time Since p::evious birth) are combined in the index of mater.nal status. The index is constructed such that rates rise sharply across the five categories. Nevertheless, the magnitude of the differentials is striking: infant mortali-ty 's roughly four times higher in the highest as opposed to the lowest category, and child mortality varies by a factor of almost five. This indicates the large role played by the demcgraph-Lc facto=s summarized in the index-in determining survival chances. Differentials according to the parental kin relationship are slight. We expect this factor to have mos": imnact on neonatal mortality, because those genetic - abnormalities which result in early death often do not permit survival much beyond birth, in, fact at this age mortality is higher among children born to second cousins, not first, cousins. Mortality is lower, however, among children of unrelated parents at all ages, a differen,-4-1 which plausibly reflects a combinat-Jan of hLolcgIcal and ef f ectS. AnaI'VSj_s Estimation of eaua--,*6.cns (0). t.-ArCuq- (a) y,--'ds taszs cf the stat_,st-ZI si,='_ Icance of effectz and sets Cf e ------ indicatIng the -at';=S C-_- thcse ef*facts. We =rasanz r=S**_'-_Z raz!7ie= than =7 F工GUR日3d 唱目口口口口網口細口開目口口自口 ’闐蘿 &&} &“憚黑．__ ’開〕才州 &Za〕忽悵 ’州。，。總陳日、轎 ‘〕匪粹至：:., 臘。鯽調你 S戲開你！j 州 〕、、響、、．!;:: 邢」牌賽響響“&& 矯彥彎工緣斗 9卻斗L常亂為n勰夕（驕 才仁m一 ，三斗．一“ ‘么訌二，- ‘•．•．.｀、州•••■用 ．開畸一｀一·雙 韶二，日認三二 件日日必》日一么日二 ”》斗二弁之二、一＿＝一＝至 :，荊日日、日斤二一三三二： 亡盒〕煙誰一喜’韋＿呈造豐韭 ．一，一一－..．一一一一 ·，’一一一一一－一一一 一一－一一一一一一 →■．.．■啊■■目．••劇．細～．•■■■■．．■口．•■■．.，■•■•一 名． &,&?:’馱，：、’斗’: ..,. more systematic and focused examination. For each varia-Zle,' we n each ecruaticn., consider first its statistical sign4ficance, i * - and then the logistic regression coefficient-s whIch port=-ay the pattern of net affect. To sum,arize 'Ithe stZatistical tests, we present the p-values associated with the F-statistics fo-- each variable. These values represent the probability that the F.-statistic does not differ fro= zero. That is, a small p-val _,e indicates a significant ef f act. Fallowing convention, we s lec- p--values of .10, .05, and .01 as thresholds for incZI-easing levels of statistical significance. 5.2a Variables other than income We begin with the four socioeconomic factors (excluding .income).. . P-values; for the net effects of thase varlabjeS 4n equations (0) and (2)-(4) are presented in Table 4. S t _- ik i n g findings are immediately arparent. First., only region shows- significant net effects on mortality during infancy. once other factors are controlled, there is no "significant variation in infant survival probabilities according to tyz e of --lace of residence, matern-al education, or maternal status. Z x a t C M of the lCgistic rearession coeff ic-'ent's f=cm t.hes-e a Cru a M --',z n S (Table 5) forces only partial quallffaatIcn, Of t1-ese ions. The large ;OS4-4Ve Coeff4-4en S --r ra=4Cn nd4 Ca-:e ,_hat mo aj.4- Ydu=ing infancy is h'=her in even ccnt=cIs fcr inccme lea-zat4--n I:)), the v a r a S S (in.cluding Incuse.hold Sani--azic- status (eçuaticn (4)). The latter cc.trol =cst attenuates zhe regional diefarentlal: apparently a relatively large propr of the births in Upper Egypt are to mothers at higher risk cf child loss dua to dcgraphic status (ace, parit-, bIrth in-ar- val). As expected, neonatal and infant =mortalitv tends to be lcwer in urban areas, and the contrast with rural areas is sinifican: at the .05 level for neonatal mortality in equations (C) and (2) (Table 5). The mcst interesting outco=e with respect to type cf place of residence is the higher =ortality risk of children in the =etrocolitan areas of Cairo and Alexandria, as CC=parad to children in other urban areas. This is especiallv amparent in the post-neonatal period, during which these children suf er higher rates of ortality than urban or rura! children, with the result that rates for the entire in.fant period are rcugnlv the same in metropolitan and rural areas. It should te kept in rind, however, that these are effects Ze, cf other sociceconoc factors; in the sipile bivariate da)erentials (Figure 2a) metroolitan children enjoy a sligh.t advanzaga c-ier childre- in the other two types of areas. Vhere are several peSSiZ-e explanatii.s for the sligh: net disadvantag In ex=er.enced bv rtr=oi an hild'-.r: a r .ar:er tr:r-i : netroplia:•. as cC=rarad to urban res s a ra-rura ~igrancs, : the dIsru==I== cf _'.`.v .fe Causa=:- =- 'Z vcxrse -4n vhan a=aas- controls for source of water and toilet facilities, in equation (3), have only a minor impact on the urban-metropolitan differen- tial); mothers are more likely to work away from home in metro- politan areas; in part as a consequence of this, and for other reasons, breastfeeding durations may be shorter in metrczolitan areas. We have not pursued these possibilities in this research, but they merit investigation as poorer child health in metrocol- itan areas is certainly a policy concern. One of the important findings of this study is the absence of effects of maternal schooling, without and with controls for household income. The coefficients in Table 5 indicate that although mortality indeed declines with increased schooling, in particular if women receive seven or more years of schooling, the effects are not statistically significant. In this context we should note that a very small proportion of the sample falls into the highest maternal schooling category (six percent of observations in the analysis of infant mortality, five percen- in the analysis of child mortality). Thus despite negative coeffic- ients of some magnitude, particularly for the early childhcc period, the effects are not significant In a sampla of this size. The results are supportive of the potenzial role o maternal schooling, but indicate that in Eqvt mcre primary schooling is required, and that avera=e ls%.els of educational attai.-ment will have to rise subst!2anti '. or maternal schoclinc has a noticea:ne neative lavels. We note,.ina.M, Z the :na=arnalc-- ients are essentiavlly unaffected by controis for houseteld income. It would appear that the effects o:served are -.t proxies for effects of household economic status. The estimates reveal definita effects of pater-al sta (education and occuation) on early ch.ildhood ortaity; ieed among the four socioeconomic factors it e=erges as the =cst powerful dete=minant of childhood mcrtality. :he mattern cf effects is as expected*(Table 5): mortality falls aczoss cater- ies of rising paternal status. (These c:oeffi.cients also indicate significantly lower =ortality in category -V as opposed to category I in the neonatal period). Controlling fcr household income reduces the level of statistical significance of the paternal status effect (Table 4), but the size of the cce-fi- imnts are hardiy diLinlslhed (Table 5). T:is su=gests, a-tire'y against our expectations, that paternal status effects c=erate largely indepedently of household incme (althç add c.. cf income to the eqration does reduca the eficiencv of the estlat- es, hence the lower p-values). 7rom the stand=oint of irpacet- child survIval prcbabilities, it would appear na= mazer-a status and household income reflect distinct dýIenser.s c economic vesitih. te finding of lare:r pater-al status e ir childhead as comared to ican.z=s zz .hesreic' ex:ectaic.s as well as mrevious research -*.7mt a.n .: We t= n-,; e C=sizr ef±aas of:h (Tables 10-13). Risk of death varies according to sex of the child only in the neonatal period, when females are less likely to die. In the post-neonatal and early childhood periods, females are more likely to die, but the differential is .zt statistically significant. These are statistical tests that there is no sex differential (i.e. the coefficient equals zero). A more appropriate test would be against an expected differential based on experience in a range of societies. In the Coale-De.ey. Model West, level 12, life table, the female an4 male infant mortality rates are 132.2 and 156.0 deaths per thousand respecti- vely, implying a two-sex rate roughly equivalent to the E.S rate for the period under consideration (the EFS rate is 142). These rates imply an expected sex differential in the lCgit of mort- ality of -.195 in favor of females. Extracting the standard errors from the logistic regression equations, we can test the estimated coefficients against the expected value of -.195 as follows: (1) (2) (3) (4) Eguation Extected Estimated Estimated t-statistic coefficient coefficient std. errr (:)-ri) (5) -.195 -.067 .085 1.506 (3) -.195 -.059 .086 .531 (4) -.195 -.050 .087 1.,E7 hile only one of the tests yields a diffarana the .10 level, this exercise illustrates that famala Surviv chances are far worse 1-in y a, nCS- Socieies. Parantal educational aspirations ter daughters is a gnifi- cant predictor only in the neonatal period (Table 6) where zortality is lowest for children bor= to parents who toth aspire for thair daughters to be well educated (Ta':Le 7). The patter of coefficiants suggests that maternal aspiraticns are of -ore significance than pa.ernal. If this variable reflects. a-ids towards invest=ents in children and cuality of children, it wcu'd appear that maternal attitudes are rore pertirent. The fact that this variable emerges as significant only in the necnatal period raises doubts about how well it proxies for the uneasured quality of care variable, however, which should be of mcre significance in the post-necnata! and early child'-odcd We susmect that to a larger extent this varialereflects attitudes towards women's roles and status. These attitudes .ay have a large bearing on the health of the mcther (nutritin, medical care) , and thrcugh this on the health of the .ewbcrn an- its probability of surviving through the neornatal period. The two househcld variables assc:iated with santati ccnditions - scurce of driking water and ty.e of tce. 'es - are of scecial =elevance to de*ates aýcut ct-.,=a'. =ents to reduce =orta.ity. We axzect thesa zzi '. : IMact after the n.ara . rri , A year ozf 11i11 whan tnachl tacc=as ricre _ci. a.ýis-: de=enden cn zreasz-Ik as a f=c=:. --`.sa -,-a A2-n:a In...2 Z-=_f - -- - - - is associated with neonatal mortality only, and the nature of the effect is perverse: neonatal mortality is highest in househclds possessing a flush toilet, and lowest in households lacking any toilet. We find these results uninterpretable and thus dism.s them as unmeaningful. In contrast, the association with source of drinking water conforms more nearly to expectations. The estimated effects on neonatal mortality are difficult to explain behaviorally, however, since virtually all Egyptian infants are breastfed for at least one month and are unlikely to receive p other fluids. We hypothesized an impact on childhood mortality, however, and here the significant net effects which emerge are sensible: mortality is lowest in those households with piped water in the dwelling. Interestingly, once income and socicecon- omic factors are controlled, early childhood mortality is hihesz for children in households which rely on a nublic faucet for water, even higher than households relying on "other" sources (wells, pumps, the Nile). We surmised earlier that where the household relies on a public faucet, the water may be obtained infrequently and allowed to stand for long periods, durin= ,nich time it can easily become contaminated. hether this exla the pattern of net effects cannot be deter-ined from these da:a, but it is clear that mrovision of zimed water direct>. :: he household is associated with lower mortality. We ncza a=ain ta results remain, albeit somewhat atenuated, when houseti ine is controlled. :his indicates that, alth holds are more likely to have piped water s a s :1:' the egtiated effects on zortality are not due to source of water supply proxying fo Li.come. Less discussion is reculired of the effacts of the proximate deterzinants (Tables 8 and 9). jaternal risk status Is a powerful deter=inant at each age. As noted earlier, this variable is constructed so as to maxi=ize the st=en=th estimated effects, so the existence ck eacts is less ani.gu than in the case of the other variables considerad ths far. It is interesting to note, however, that the eflects are essential.ly unaffected by controls for income, socoecononic Zactcrs, and the intervening variables (compare equations (6) a-d (4) in Table 9). This indlcates that maternal risk status, although assoc- ilated with these factors (see Table 1,3 and discussion below), exerts its influence on mortality independently' of tha. Parental kln relationship, on the contrary, is not associated with mortality or.ca the backgour.d and intarveni. :actcrs are controlled (Table 8, equation (4), and Table 9). Becausa zte hypothesized effect is biological in nature - incraased a ity of genetic abn=oralities - it shculd =aintai_ its iJpact with contro.s for social and econoc c e co.clu'e za m-ftian te:='o= a::as e`ee das i: - f'ace sigiia ' higher risks of deat in ian=-* - chi.dhood. lie suspect, 'owever, tnat itpctant tex4stin a:taortai±t-. After fata:. c, - and o'scu=es a :n- 7 : =a=antal 44-7 .5.2b Effects of income Having reviewed the effects of all other background factors, intervening variables, and zroximate determiants, we now turn our attention to household income. For the multivariate analysis, the income variables are collapsed into three categories of apprCximately equal size. (We collapse to three categories for reasons of sample size. Three categories permits directions of effects and non-linearities to be detected, without fragmenting the sample to an extent which undermines statistical testing.) In equation (1) effects of income are estimated without adjustment for other variables. All three variables show significant effects on early childhood mortality (Table 10), and in the expected direction: mortality is lower among children in higher income households (Table 11). Significant, although weaker effects also emerge for neanatal mortality and in the expected direction. No effects are apparenz for post-neonatal mortality, nor.for infant mortality excezt fc= the index of consumer durables. The stronger effects cn necnazal than on post-neonatal mortality are unanticipated, but we will see momentarily that the effects on neonatal morali=y disaz;ear once other variables are controlled. The scmeihat la==e= differentials in early childhood mortality are consiszanz zirLar reasoning and previcus research. In ecuations (2)-(4) the incCe effacts are :es:ed. Successively more cC==lex =cde7s, with ccnmrcs f = e socioecono=ic factors (eçuation (2)), interenir.a variales (equation (3)), and the proximate deterinants (e ration (4)) Several important generalizations eterge from this analysis. First, household Incz=a has no impact en =ortality? during the infant period (neonatal or post.-neonatal). Vhe si ic.ar. åffacts are confined to the early childhocd pericd, where we anticipated a st=onger i=pact. Segcnd, once socioecon=.f= factors are cont-olled, the index of ccnsmer durables shows no effects on in:ant or early childhood mortality. We have investi- gated this result further and determined that it is the control for type of place of residence (rural, uZban, netrcpclitan) that eliminates the differentials initially observed. It would see= that possessicn of censu=er durables, while no drutt reflectinç household wealth, is also heavily dependent on residential location. Ths is not su:prising considering that many of the durables comprising the index are =ore eadily. In uzban areas and are likely to assu=e greater imcrtance in urtan lifestyles. Moreover, their ac-.isition cenerally reauires cas earnings, whereas A substan=ial propIrtcn s rs in other fcrns. Zn any case, !t is clear v*-- in tnis ana'.vsi- an index of consu=er duraes does m :er::! as a mr:x. c direc-: measures oe .tcus ehe.. i.e Ä thrd=enera, fi, 4-nr ~s tha- ztes;a:a:r: - ofnet hcuseholl: .==ncoe and,- er caIz net ==e~.. tv CC=rI f2r not~ft the :. ~- factors (equation (2)), reduces the level of statistical signifi- cance (Table 10) and reduces the range of the ccefficients by about one-third (Table 11). This is not unexpected because household income varies by region of residence (CA2MAS, 1932, Volume III) and because paternal status and household income overlap theoretically to a considerable extent. Addition of the intervening variables and the proximate determiants (equations (3) and (4)) results in a loss of statistical signi:i- cance by the income variables, but some loss of efficiency in estimation is inevitable when the number of parameters increases substantially, as is the case here. The unanticipated outcome is the resilience of the parameter estimates (Table 11), which indicates that the negative impact of income on childhood .mortality is not due to effects operating through the measured intervening variables or proximate determinants. The pattern of effects, incidentally, is almost identical for total and =er capita income: the pattern is slightly curvilinear, with children from households in the upper one-third in terms o- income experiencing distinctly better survival Probabilities. A final point is that the estimated effects of a conven.t:n- al per capita income measure (eq. net household incc=e divided by number of household members at the survey) are cons=srens r positive in direction than the effects of the -er ca=4:a -easur= utilized in the analysis presented here. o more pres, conventional measures shw lare and sSas c zositive effects on infant an-d negative effeots on early childhccd mortality (results not shcwn here) . This outcome is consiste t with our earlier rsonir.g concerning the possible pcsItive bias *on the estication of incc=e effects introduced by the ipact of recent =crtality on household membership at the survey (see Section 3.2). (See Append-ix A f: further discussion of this matter.) We are surprised by the resillence of the incc=e efrflects, as we assu=e that variables such as household sanitatin (as determined by source of wate= and type of toilet facility) and maternal risk status (as deteined by de-ograhic cnaracteris=- Ics of the mother) serve as mechanisms for incoe effects. We have pursued this puzzle further by exa=mnr. the indirct paths linking incm=e and =iortality through the i.terveinv and proxl=ate variables. Recall that, a=cnr te .Iere ig varia- bles, only source of water shows net effects cr. =ortalitj (see Table 6), and that these effects are c-.ly of m-oderate . Inco=e is a powerful determinant of source of water, however, (p-values shown in Table 12); examinati c coe.iie - mates (not shown) indicates that higher ino=e hcse'nI-s arn more likely to have pied watar in the d:huing. hus it see= that the 1s2ated idirect effects cf Ir.::=a cn ncra thrnugh hmusakimid sar_itatil=. are w.,eak riaiybecius_= ---h=- rather weak ne: d-act affects of scure cf water and -:-: e tc21e" 2actiities c. 'oi. e stress tiat th d:s nz.c =ula~- v--- -- survival chances.- But apparently these differentials, insoPar as they influence survival chances, are not due to income differen- ials in the source of drinking water or type of toilet. Indirect effects of income on mortality through the prcxi- mate determinants are lacking for a rather different reason. Unlike the household sanitation and hygiene indicators, maternal risk status has a powerful net impact on mortality (see Table 8). In the childhood period, however, household income has no net impact on maternal risk status (Table 13), and thus no signif- icant indirect effect through maternal risk status exists at this age. Parenthetically, we note a slight suppression effect of maternal risk status on the influence of income on mortality during infancy. The p-values in Table 13 indicate that income significantly affects maternal risk status in infancy. Examinat- ion of the coefficients from these regressions (not shown) reveals that the effect is positive: higher income is associated with higher risk status, which in itself will act counter to the expected negative impact of income on mortality. The positive income effects on risk status are small in magnitu-de, hcwever, despite being highly significant statistically, and nss-en : income effects are hardly altered by the contrcl for the mroxi- ate determinants, as observed previously when considerin= :a-as 10 and 11. 5.2c InteractivT affects The analysis thus far has assumed effec=s cf zn 70 nortality which are additive with respecz to other exanator: variables. Many theoretical considaratic-.s lead us to expect inco=e effects interactive with other variables, socIceconc=i~ variables and intervening variables in particu.lar. We Will n.ct develop the full set of theoretical arguents here, as ths wou'.2 be a lengthy exercise. Instead, by way of illustratios, we nota a few of the intaractive effects which 4ght te anticirata on the basis of theory or past resear=h. Consider, to :sin with, =aternal schooling and incoe. We hypothesise that betzar educated women will take fuller advantave, from the sta.dpoint of child care, of the resources provided by higher househol.d income. This reasoning implies that inco=e effects will be sharper for hetter educated women, and that educzti4ra! dIficr- antials will be larger for women in higher incoe hcuseho'ds. A variation in Income effects according to sex of the child also seems plausible, although contrary patterns are equallv olausible a priori: the sex differentials would. be larger for the low income strata .if under circucstances of scarce resources =a:.es were to receive adequate care while fenales were rative. neglected; the differer.ial would be larer fcr the -- -.---- strata I-f urder ci.rcu=stances of ratundanca~~*- :esou:ces were inreszed -. =aes. i.al'., c-.i rz: cf wazr supply and ;ncc=e. "-n - : e :ee. ra: in oter settings su==estin zha= i-r:aete-.s :s:ar Su= ;-ave a ma=t-J--'rI _v nC~I : The estimation of interactive effects is hindered by the rather small number of sample observations (see discussion in Section III). To partially overcome this problem, we colla-se some of the explanatory variables into fewer categories. Even so, the number of observations in some cells of the two-way interactions is small for the analysis of mortality, and there- fore we can anticipate that the patterns of estimated interactive effects will be somewhat erratic. In addition, when carrying out a large number of statistical tests, as we shall undertake in examining interactions, it is important to recall. that statiszi- cal theory indicates that, at any specified critical probability level, an expected proportion of tests will indicate the exist- ence of effects in the population when none exist (i.e. Type I error). When testing at the .05 level, for example, we ex=ect to conclude erroneously that actual relationships exist in one out of twenty tests. For these two reasons - the rather scanty sample base, and the inevitable occurrence of Type I error - we do not place great weight on particular interactions, but instead search for general patterns of significant interactive effects. Statistical tests of interactions between inccme and other socioeconomic factors, and between income and the interveni.- variables, are presented in Tables 14 and 15 resectivel. e do not test interactions with the index of consumer durables, as there are no net significant additive effec(s :able 10)). Parameter estimazes from ecuaicns with szatisti. interactions are shown in Tables 16 and 17. Ivo ineractions with region of residence energe: net household incm=e and regi on for child mortal.ity, and per capita household incC=e and reca for post-neonatal mortality. The pattarns of effects in these two interactlons are not consistant (Table 16) : in toth t*n incoe effects on mortality are more prononced in .Upper Eyp-, but in the first the relationship is negative in mer gyt, and in the second it. is positive. We can offer no resoluti:.cn c this outcome. The significant interactions with type of place of res idance are more interpretable. Both net hCosehold incoe and per capita household income show negative effects on neonatal mortality in urban areas (=etropoltan and urban areas are co=bined here) , but not rral. areas (Vable le). Note that this interaction applies to neonatal mortality cnly, where no -cverall additive effect is evident (Table 10). There are several possible explanations for this result. One whih seeZs s is that in urban areas incea perits murchase cf better se=v-ces for child delivery, whether trained =idives er nospita.l serv- Ices. Such services are less available in rural areas rall regardless of incce leve!. The interactions with =ater-.al exao ar p .e =ost intriging. As we hyp^thtsised, the dac sUg==Ct z'-c acc=e effec-s are sharper for edca:ed othe:s - years of sch=Clin have cat Izasz th -- e= cd a `v.- - e* capita income and maternal education in early childhood). These data do not permit us to explore the mechanisms underlying the interaction. The clear tendency for schooling and income to reinforce each other- would seem worthy of further investigation. On the other hand, the insignificance of the interaction between income and maternal schooling in the early childhood period indicates that the negative impact of income on mortality at this age applies to uneducated and educated mothers alike.* The interactions between paternal status and income are somewhat erratic, like the previous interactions considered, but in general. suggest that negative household income effects are larger where paternal status is higher. Or, expressed differ- ently, effects of paternal status are larger among higher income households. We offer no interpretation of this outcome. Turning to interactions between income and the intervening variables, we observe far fewer statistically significant outcomes (Table 15). At the .10 level we expect one out of ten to test as significant even if no interactions exist in the population, so we should not be surprised by the emergence of to out of thirty-two. Nevertheless, we briefly consider these two. The data suggest that in the childhood perocd females are particularly disadvantaged in the lowest income househcIds (abl 17). But the relationship is reversed in middle i.coe hcuse- holds, and thus we find interoreatn of the ;vera a rn f this interaction el'sive. Of so=ewhat MCrq LnteresS- 4S -teM=aC: household income and water supply. Earlier we ncted evidence from other settings of the existence of such an interaction, but the ErS estimates are not consistent with the previous findings. In Egypt piped water would seem to provide gra-r !enefit to infants from wealthier households (Table 17). Or, exmressed differextly, income effects on infant mortality, absent for the sample as a whole, are apparent in households wizh piped wam2r. Again we observe reinforcing effects of two datar:.ants - i. th.Is case household income and source of water supply - where neither is significant alone (Table 10). However, it should also be noted that these two variables do not interact in affecting childhood mortality where both show significant additive effects. We conclude that, in general, the estimated effects of household income on ear,v childhced ortality apply across subgroups of the population defined by other scciseccne=i4 factors and the intarvening variables, but that effects on t ortalitv exist for only selected subgroups, namely children i urban places, children of educated mothers, children wit. fathers of high scC-oecoamic status, a.nd chil =se=ns with pi-ed water. VI. STARY AND CONCLUDING REMARKS We summarize our findings regarding income as follows: (1) Household income shows a significant negative impact on probabilities of child survival in Egypt. No irpact is evident during the first twelve months of life, but durIng early childhood (ages one through four) the effects are pronounced. (2) The impact of household income on childhood mortality remains when associated socioeconomic factors (region and type of place of residence, maternal schooling, and paternal status) are controlled, and, somewhat surprisingly, when variables intermediary between household income and mortality, are controlled: namely household sanitation (to the extent this is a function of source of drinking wazer and type of toilet facilities), and maternal demographic characteristics (age, parity, elapsed time since previous live birth). (3) The data suggest that the impact of income on infan-: an- child mortality is somewhat greater in ur!an areas, educated mothers, when the father is o: hi2 er scconcric status, and when the household has z:md water dwelling. Although the main c'4ective of =he analysis was effects, provcca-ive findinqs reardin the other variables also e=erged: (4) Infant (but not child) =Crtality is Uzrkedly hiri er Egypt, both without- and with contrls for ir.coe, other socioeconemic factcrs, household sanitatio., and =aternal demographic characteristics. (5) Neither Lnant nor child mortality vary according to tyre of place of residence (ruraL, urban, etropelitan.) c.ce sociceconoic factors are controlled. (6) Maternal schooling shows at best modest etfects en intanz and child mortality, contrary to findings :r-= =anv sezt- Ings, but consistent with pravicus multivariate analyses ct data from Zgypt., The data suggest so=e i==act of scoli bevond the primary leval, but even .this et.fezt is nct substantial in =agnitude, and as of 1saa a vrry s=a- proportion of Zgyptian wo=en of repreductive age had attained this a=cunt of schooling. There is sc=e i.dcatin of a stronger relationship between sch.cling and cid su-rVival amor.g these in the u,er incce st=aza. (7) There is no evidence cf a significant assoiati:n -ee: type c: tcilat tacili-y and -.:ant er chi:. : (3) provision c: =i=ed water to the d-' i- :s ass t i i s,aza is= eaea..- h,C.e ast, car .-s dr--e-in : d"li --aucats- -- e:yaie.c ' -re s-:'r' -r"ai. i du-ir. al child-ed. .e si :t.isra.ct::ai- in containers, often open, in the household. (9) Consistent with previous analyses of the ZES data, demo- graphic determinants of the mother consistently show the most substantial and pervasive impacts on infant and child mortality among the large set of variables examined. Finally, we note two findings of a more methodological nature: (10) An index of consumer durables, constructed to serve as a convenient proxy for household income, shows no net relat- ionship with infant or child mortality, and indeed, does not behave in a manner similar to either total net househcd income or net per capita household income. On the basis of this analysis, it would seem that such an index does not serve effectively as a-surrogate for household income. Direct information on household incc.e, although tedicus to collect and subject to relatively high levels of response error, is necessary if relationships between household income level and demographic responses are to be investi- gated. (11) The estimated relationship between household income Cer capita and child survival varies de=endinq on the consoruc:- ion of the income ter capita measure. The most straiho- ward constructions in all likelihcd yield biased estima:es due to the simultaneity between household size and recan- mortality experience. It can Le de:onszratz'4 z- a- ive measures *ieA ient cu--ms. hra asi S analysis, however, it is nz zossi'a = chczse nez as superior cn enirical grcunds. We conclude by considering two more general questicns. First, what do the findings sugest abut the de rinants o infant and child mortality? Second, wtat do the flndings iplv about the formulaticn of programs to reduce i*.ant and child xortality in Egypt? Beginning with im1Pications of the fin.dings cocernin= the determinants of child survival, it is useful to retain the distinction between the infant and early chil.dhood teriod. In the infant period the maternal risk factors appear to dc=inate. Only region of residence, a=ong the socioeconomic factors considered, shows consistent and intepretable effects. Yat infant mortallty in Eypt is relatively high (roughy. 40 deaths per 1000 births in thls sample) and it can be shcwn that a shift of all births to ore optimal ages and parities, and a enthe.- ing of intervals between previcus births wculd not result J-. infant mortality levels as low as we are accustoed to witressin= in develoned societies (see, for exa=ple, Vrussel and ab.ey, 1984). Eence we feel that we have presented i- this aasis a sericusly !ncoglete picure of the deerinans Jna- noality, althcu=h .4t is cf sC:.e va`ue -n have -ez of variables which are not ss us sseztt .at a c=n=arn... t-h-n.d ------- .=2--a- '~~,at an 2i:4c4 level cf rsc-j==2-, an-- -h regional location is due to cultural variables which bear cn these omitted determinants. We propose that understandin the determinants of high infant mortality demands consideration of factors other than household resource levels an: -r.t' schooling, factors more directly related to, first, health of the mother and, second, the quality of childcare. With respect to the former, it would seem that maternal demographic characterist- ics are of great importance. The findings are more informative concerning- the determin- ants of early childhood survival, for the availability of material resources, as influenced by paternal status and house- hold income, shows a distinct impact. Again the cross-sectional differentials are not substantial enough to.permit us to describe satisfactorily the determinants of high and low childhood mortality regimes. But the differentials do provide strong indications that socioeconomic factors play an important role. In this respect the pictures which emerge of infant and of child mortality are at odds: for the latter, but not the former, material conditions at the household level (parental socio- economic status, income, sanitation facilities) seriously affecm the probability of survival. Several implications emerge from these findi4ns a*cum policies likely to show impact on mcrtaliz- in . 'mrst, substan=ial gains in infant surviva!. c*anCes culi bo e zd Women avoided chldbearing a- e::-s. Z ges, , shortly felowing a previous birth. The a as: mis:::h s0 fa=ily planning p=ogam. :n this specif.ic =anner, t-e gcoals f family Planning and health progra=s are the sa=e. Secc-d, further research is reTuired to explore the scurce of the lar;e regional differential in infant rortality. 7ha dIffa=rntia1 is not explain.ed by conventicnal exlanatory variah.es, at least as offered by the ES. It is difficult to conceive of designing programs targetted at alleviating the regional disparity given the current level of ignc-ance alout the speci!ic causes of tha disparity. Third, there is no evidence that increases in average levels of =aternal education will have an i,mact on =crzalitv of any magnitude, and thus reliance or. this as a focus of =ortality policy would seem mIsdi-ected. On the contrary, and the fourth implication of our findings, househoLd econo=ic status wculd appear to be an iportant deter=iant of cildhocd =ortalitv, and thus policies which foster economic growth and developent (income growth, and shifts of ecCno=ic actjvity 4-to the =edern sector) can be exected to show impact or ortality. S uh policies n.aturally are a nationaj goal, but are no easily ccnceived or i.plemented. 7ifth, the provis.cn of wiaed waer to a greater nu-:ber of hcuseholds appears to be cne spec-4i-c acz-. which should raduca childhod :rality. BIBLIOGRAPHY Aldrich, John H. and Forrest Nelson. 1984. Linear Prcbability, Loit, and Probit Models. Sage Publications, Beverley Hills. Ammar, Hamed. 1954. Growing to in an Ec--tian Villace. Routledge and Kegan Paul, London. Anker, Richard and James C. Knowles. 1977. "An Empirical Analysis of Mortality Differentials in Kenya at the Macro and Micro Levels". Ponulation and Emoloyment Workinz Pacer S60 International Labour Office, Geneva. Antonovsky, Aaron. 1967. "Social Class, Life Expectancy and Overall Mortality." Millbank Memorial Fund Ouar:erly Volume XLV No.2. Part 1. New York. Ayrout, Henry Habib. 1963. The Eavetian Peasant. Beacon Press, Boston. Butz, W.P., Habicht J.P. and J. DaVanzo. 1984. "Environmental Factors in the Relationship between Breastfeeding and Infant Mortality: the Role of Sanitation and Water in Malaysia." American Journal of Enidemiolov, 119(4):516-525. Caldwell, Pat. 1977. "Egypt and the Arab and Islamic Worlds." In John.C. Caldwell, ed. The Persistence of Hich Fertility, 11. AN, Canberra:593-616. Caldwell, John C. 1979. "Education as a factor in mortality decline: An examination of Nigerian data." Plti Studies 23:395-413. Caldwell, John and Peter McDonald. 1982. "Influences of Maternal Education on Infant and Child Mortality: Levels and- Causes." Health Policv and Education (2). Callum, Christine. 1983. "Results of an Adhcc Suzrey on :nfant and Child Mortality in Sudan". World Healzh S ii Quarterly 36. Carvalho, J.A,.M. and C. Wood. 1978. "Mrai incce bution and rural urban residence in =razil.' c' Develcment ReviewVi 4:504-420. Central Agency for Public Mobilisaticn and Stazismics. 933. 2. Ecvmtian Pa-rtii; SurveY 19c. CA;XAS, Caire. Cochrane, Susan H., O'Hara Dcnald ., and 1canne Les!.`s. "he 71'ects c:- :inon C-e~h Cochrane, Susan H-., Ccnn lasLa and ence." Health Poliev and Educaticn 2:213-230. Co ittee on Population and Demography. 982. T e i c Recent Orands in :ertilitv and Mcitac-zalitr in _Eatioa Academy Press, Washington D.C. Critchfield, Richard, h7h. Stat:t Ar evttian. Syacu.-..r University Press. DaVanzo, J., Butz, W.P. and J-P Eabicht. 19 ":. 'içaEc and Behavioural Influences on Mortality in Xalaysia vary during the first year of life". Peculation Studies 37 (3) :381-402. DaVanzo, J., Eabicht, Jean-Pierre, and Wilia . Eu=. 193. "Assessing Socioecoercnic Correlates of Birth.eigh in Peninsula Malaysia: Ethnic Differences and Changes over Time." Social Science and medlcine 18(2):387-404. De Sweemer, Cecile. 1984. "The Influence of Chi-ld Smacing on Child Survival. " Pomulaticn,. Studies 38 (1) :47-72. Dixon, W.J. 1983. BUDP statistical scftware. University of California, Berkeley. Eld, Ismall and John 3. Casterline. 1983. "Di4ne:entials i. infant and child zortality in Egy=1". Paper resenred at the International Conference on Ter=ilit, in y _gypt, Cair, Decemher. Farrah, Abdul-Aziz and Sam Preston. 1982. "Child lxzrta.i'-. Differentials in Sudan". Population a. nde - 8(2) :365-383. Gray, R.E. 1981. "Birth Intervals, Pstartn Sexual Atstrnan and Child Eealth." In Page, H.. and Ies=haeghe R. (eds), child-SracIn.a in Trtca Af:ica: Meiie. '-' ch-.',: - York, Acadenic Press. Eckcraft, .,, XcDonald, Z., and S. Rutstein. 3. "Ci. a- Eftects on 1nant and Z.ar'.v cld: 2.. . 'c - IZdex 49(4) :585-,13. 4-8a=t and C*ldrrr.L41.4 ~ ---- Czn:!r=--.sez: "eca_'-..-.* .- - - ,* tf Kuznets, S'imon. 1981 (July). "Children and Adults in Distribution". Econom-nic Growth Ceriter Discuss*.:-atr 379. Yale Univeristy. Lynch, Patricia D. and Hoda Fahmy. 1984. Cräftswoen. i- __,esa, Egvt. Women, Work and Development Series No. 7. : 1O. Martin, Linda G., Trussell, J., Salvail F.R. and N. 3.... J%. "Co-variates of Child Mortality in the Philippine=. Indonesia and Pakistan: Analysis Based on Hazard Erd.- - Poulation Studies 37(3):417-432. Merrick, Thomas W. 1985. "The Effect of Piped Wa-tE.: cEZ.ly Childhood Mortality in Urban Brazil, 1970-76". - athv 22(1) :1-23. Mosley, W. Henry and Lincoln C. Chen. 1984. "An Anal-:-. 7rame- work for the Study of Child Survival in Develop*'.:r .ocnt- ries." Poulation and Develorment Review, Sup;::nt to Vol. 10, edited by Mosley and Chen. Mosley, W. Henry. 1985. "Biological and Socioeconomtic . trnin- ants of Child Survival. A Proximate Determinants z.eork Integrating Fertility and Mortality Variablec'. Inter- national PoCulation Conference Volume 2. Florence -.&PF Oldham, Linda. 1984. "Child Nurturance and its e at Manshiet Nasser, Cairo: Based on a Family Pre:'e.' :he Ponulation Council, Regional Paper, West Asia : crt Africa, Cairo. Preston, Samuel H. 1985. "Mortality in Childhood: Lasa..:.i- from WFS" in Cleland J. and J. Hobcraft (eds.), Re-._ ocie Chancre in Develooing Countries. Oxford Universit-.- Rosenweig, M. and T.P. Schultz. 1982. "Market Oo::---:is, Genetic Endowments and intrafamily Res ource Dist::. .: - Child Survival in Rural India. American Ec-.cmi '- .2- (4):803-815. Secretariat of the Econcmic and Social Com7ission f o . :: and the Pacific. 1984. "ortalitv decl`ne and overview of developinc countries of the As- region." Mcrtality and -ealth Pcliv F -i e Expert Group on Morzality and *ea-t Policy, Ivha 3 June 1983., pm 126-143. New York, United -a=i--- Secretariat of the World _`erzility Surv. . 4. the World crii S:.:eyi n renAs, Dffe--- Deter=inants of :ormality in Develoing C: -: ." cr.ro1o and :at c:y OS rm.-~ Schull, William J. 1972. "Genetic 1=plicaticns of Pcp":a&;--. Breeding Str-cture." in G.A. Earison and A.i. =_yce (eds) The Structure of Euran Poftlation-s Claren.dn Press, Cxforf.. Stephens P. Wolanya. 1984 (April). "The Relaticnsni- etween the Level of Hcusehold Sanitation and Child Mortality - an Examination of Ghanlan Data". Aårjcan Denc-a=h- ~n-~~ Pamer 910. PoMulaticn Studies Center, Univrsity oz Pennsylvania. Sullivan, J.X. Cochrane, S.M. and W.D. Yalsbeck. 1982. Procedures :o; Collectiie- and Ar.alv,sinc Moralitv data in . Working Paper #16. The World Bank. Tecke B. and F.C. Shorter. 1984. "Deteminants of Child Mortality: A study of Sguatter Settle=ents in jordan". Poulation and Develoonent Review Supplemenz to Volume 10. Trussel James, and Anne R. Pebley. 1984. "The Poten=ial Ingact of Changes in Fertility and Infant, Child, and r Xortality." Studies in Fatilv Plan-nc Vol,=e 15 6(1) :267-- 280. Wiltaur, C.B. 18a7. Of !Evmtian tomen. Paris. Winikoff, Beverly. 1983. "The Effects of Birth Spacin5 on Child and iMaternal nealth." Studies in 1anfi' ?ar.rn:te .14(10) 231-45. World Bank. 1984. World Dev-lcement Remcrt 1984- CxfOcrd University Press, New York. Zachariah, X.C., and Sulekha Patel. 1982. "Trends and Datern- inants of Infant and Child Mortality in -Xerala." - and funuan Rescuces DTiisicn DisC.:ssl.n =aer Washingtcn, D.C.: The World 3ank. TABLE I Infant Mortality Rates (Deaths per 1000 births): Egypt, 1930 - 1979 NAS EFS Dates Estimates Estimates- 1930-34 248 1935-39 242 1940-44 247 1945-49 218 1950-54. 203 191 195.5-59 185 166 1960-64 163 151 1965-69 160 141 1970-74 141 146 1975-79 132 1. SOURCE: Committee on Population and Denography. 19S2. Taoie 2. Rates shown are unweighted means of single year estimates. 2. SOURCE: CAPMAS. 1983. Volume II Table 5.3. TABLE 2 Distribution of Sample ObservaioTs According to Predictor Variables Analysis of Infant ',or-.alicy Analysis of Child Mortaliey Births Deaths Births Deaths Predictor Variable Number Percent Number Number Percent Number TOTAL 4963 100.0 710 2183 100.0 195 Net Household Income I (owest) 990 19.9 137 433 20.1 .50 II 992 20.0 153 423 19.4 S3 I 996 20.0 1Z6 441 20.2 ;2 IV 995 20.0 140 451 20.6 28 V 994 20.0 124 430 19.7 22 Per Capita Household Income I (lowest) 1002 20.2 129 453 22.1 62 987 19.9 162 441 20.2 45 995 20.0 149 436 20.9 45 IV 997 20.1 144 423 19.4 29 V 987 19.9 126 350 17.4 14 Index of Consumer Durables No items 977 19.7 174 398 15.2 50 1 item 1370 27.6 217 .595 27.2 6$ 2-3 items 1464 29.5 200 650 29.5 .6 4-10 items 115T 22.3 119 540 24.7 21 Region of residence Lower 3152 63.4 347 1480 .. Uacer :1 66- Type of Place of Residence 1U.-an 537 165z. Me=opolitan 9' ' ' '10 .1- .0 Maternal Educarzon None-i Births Deaths Births Deaths Number Percent Number Number Percent Number Paternal Status I (lowest) 2087 42.0 334 916 42.0 110 II 1168 23.5 152 513 22.5 47 .I 1153 23.2 167 523 24.0 01 IV 558 11.2 57 230 10.5 6 Sex of Child Male 2590 52.1 382 1100 .50.4 9.5 Female 2378 47.9 328 1063 49.6 100 Educational Aspirations for Daughter Both parents low - 1722 35.2 284 743 34.2 84 Mother univ/Father low 665 13.6 91 290 13.3 27 Father univiMother low 442 9.0 60 204 9.4 16 Both parents univ 2058 42.1 217 937 43.1 63 Source of Water Other 469 9.5 79 186 8.5 10 Public faucet 1688 34.1 243 741 34.0 83 Pump 1266 25.5 220 541 24.8 63 Residential faucet 1528 30.9 166 709 32.6 33 Toilet Facilities None 1451 29.2 238 614 2S.2 75 Non-flush common 406 8.2 49 170 7.8 1S Non-flush HH 2486 50.0 364 1102 .50.7 1 Flush HH 602 12.1 55 289 13.3 10 Maternal Risk Status I (lowest) 1115 22.4 90 492 22.5 II 771 15.5 56 3. 5 1.2 20 m 977 19.7 109 690 3i.6 - IV 1156 23.9 204 268 :2.3 V 919 is.5 251 355 :6.S Parental Kin Relationship 1sz cousins 1327 26.7 206 .50 2. 2nd cousins 987 19.9 163 Not reiated 2650 .53.4 335- 1:55 5-.. 1. Distrioutionz according co ;redic:ors do :ot always sum :o t. due 5u::: -2. An alysis of~ n fant -%Ior~a' rest-..ctac '.03 or':Z,S OCCU'r-:== In 1 0 7zt':t 5r' -zU,- 3. Anaivsis of C. i or:ami:. res.'r-cze; -0 0,r-s o~z-r~. 2 TABLE 3 Numbers of Observations, Mothers, and Households. by Place of Residence Analysis of Analysis of Infant Morcalitv Child Mor:alityv Number of: Number or Births Mothers HEs Children .1ochers IH Urban 1749 76 761 504 53S 531 Rural 3219 1248 1127 1379 573 535 Lower 3152 1325 1255 1460 942 920 Upper 1816 709 633 723 469 .46 TOTAL 4968 2034 lss3 253 '4. 356 1. Analysis of Infant Mortality restrictad to births occurrinz ;e:: 13 * 120 months preceding the survey. 2. Analysis of Child Mortality restricted to births occurrinz ir. erod 61 - 120 months precedi-g :e survey. TABLE4 Statistical Sigiificance of Effects of Socioeconomic Fqctors: P.Values from Logistic Regressioi Estimations- Dependent Variable: Type of Mortality Predictor Variable and Equation Neonatal Post-Neonatal Infant Child Region of Residence Equation (0) .000O" .0001"" .000""" .072" Equation (2) .0000" .000", .000"" .242 Equation (3) .000** .000w" .000W .453 Equation (4) .00"* .000-" .000"* .736 Type of Place of Residence Equation (0) .118 .156 .135 .919 Equation (2) .129 .142 .141 .924 Equation (3) .689 .157 .313 .826 Equation (4) .618 .172 .292 .783 Maternal Education Equation (0) .426 .268 .143 .204 Equation (2) .465 .222 .140 .282 Equation (3) .670 .366 .259 .226 Equation (4) .534 .401 .218 .316 Paternal Status Equation (0) .261 .822 .495 .005.3 Equation (2) .308 .S19 .476 .006 Equation (3) .287 .884 .503 .0285 Equation (4) .345 .925 .572 .062, denotes siznificant at .10 level denotes sicnificant at .05 level - denotes significant as .01 level 1. Probability :hat -he d-s:a:s::c does not ci.-er From zero. 2. Equations ;2-.4; contain net household .ncome as the income var::bia. TJABLE 5 Logistic Regressionz Coeff.cer-.s g Effects of Socioeconomic Factors Dependent Variable: Type of Mortaliry Predictor Variable and Equation Neonatal Post-Neonatal Infanc Child Region of Residence (contrast with Lower) Equation (0) Upper .722`" .66 .709,- .292^ Equation (2) Upper .700.. .705m- .714` .191 Equation (3) Upper .640- .672" .676-" .140 Equation (4) Upper .529" .59" .591,11 .065 Type of Place of Residence (contrast with Rural) Equation (0) Urban -.375" -.073 -.204 -.054 Metro -.205 .298 .096 11. Equaion (2) Urban ..363-m -.082 -.205 -.005 Menro -.219 .302' .092 -.106 Equation (3) Urban ..134 .023 -.082 .1 Metro -.101 .3s3* .177 .174 Equation (4) Urban -.209 -.003 -.057 .ii Metro -.096 .367 .1$5 Ma:erna Educaion conrast with None) Equation (0) 1-syrs -.69 --Z4 .50 7 -.216 .76 6.. 7.guation 2: L-6-s -.64 -.:.. - 7-yrs ~ .., .5. .. 37 -.. F.quation 3) t-6yrs -.'25-.9-'.3.2 Neonatal Post-Neonatal Infant Child Paternal Status (contrast with I:Lowest Equation (0) II -.010 *.14- .012 -.222 III - .085 -.0.59 .039 -.665 IV -.466V -.122 -.215" .209 Equation (2) II ..022 -.144 ..075 *.252 M .077 -.059 .036 -.677 - IV -.446 -.144 -.227 -1.134"' Equation (3)U -.144 -.133 -.121 -.294 M .134 -.091 .040 ..610" IV -.326 -.127 -.175 -1.0207 Equation (4) II -.054 -.106 -.079 -.240 M .187 -.059 - .065 ..553 IV -.254 -.140 -.164 -.947" denotes significant at .10 level denotes significant at .05 level denotes siznificant at .01 level 1. Equations (2-(4) contain net household income as the income variable. T.ULE 6 Scadsdeal Sigj,ificance of Effects of Intervening Varigbles: P-Values from Logiszic Regression Es:imations' Dependent Variable: Type of Mor:aliry Predictor Variable and Equation Neonatal Post-Neonacal Infant Child Sex of Child Equation (5) .035`** .377 .434 .627 Equaion (3) .041" .376 .45 .47 Equaion (4) .067v .371 .584 .390 Educational Aspiracions Equation (5) .000w* .832 .006"' .575 Equation (3) .022" .360 .69S .545 Equation (4) .040** .327 .794 .749 Source of Water Equation (5) .000"" .340 .041 0 EQuazion (3) .069" .566 .5.53 Equation (4) .0678 .554 .653 .046" Toilet Facilicies Equaton(5) .156 .034" .455 .477 Equarion (3) .0121 .167 .297 .960 Equation (4) .027" .240 .511 .938 denotes signX..cant .10 ;evei denotes sinant at .05 ei denotes signicanz at .0 . abilit7t .at - he -shae tic does :ot d:Ter :ro, :er". 3.Teec,-,a-on3 cartai=n!ý ::..osecd::mea nnco-nev-ti. TABLET Logistic Regression Coefficients f r Effects of Intervening Variables Dependent Variable: Type of Mortality Predictor Variable and Equation Neonatal Post-Neonatal Infant Child Sex of Child (contrast with Male) Equation (5) Female -.254" .101 -.067 .075 Equation (3) Female -.246" .102 -.059 .110 Equation (4) Female -.222* .104 -.048 .135 Educational Aspirations (contrast with Both Parents Low) Equation (5) M. Univ/F. Low -.542" .114 -.179 -.235 M. Low/F. Univ -.417* .075 -.170 -.114 Both Univ -.766"" -.046 ..387"u ..25S Equation (3) M. Univ/F. Low -.411, .305 -.011 -.225 M. Low/F. Univ -.221 .247 .021 -.093 Both Univ .5000* .215 ..119 .144 Equation (4) M. Univ/F. Low -.445" .280 *.004 -.274 M. Low/F. Univ -.150 .286 .066 *.056 Both Univ -.440"' .240 -.075 -.177 Source of Water (contrast with Other) Equation (5) Public Faucet -.146 .192 T.173' .202 Faucet in Dwell -.774"' -.021 -.311. Equation (3) Public Faucet .030 -.051. -.022 44 Faucet in Dwell -.5051. .046 1.63 -.33$ Equation (4; Public Faucet .0.51 -.040 -.024 .260 Faucet in Dwell -.90" .063 -.407 Neonatal Post-Neonaral Infant Child Toilet Facilities (conrast with None/Other* Equation (5) Non*Flush .256' *.234w .014 -.151 Flush .405 -.702 *.25.49 Equation (3) Non-Flush .418... -.140 .13S .044 Flush .677" *.566 -.029 .114 Equation (4) Non-Flush .362mm' -.142 .113 .020 Flush .718* *.11 .009 .168 denotes signiicant at .10 level "" denotes significant at .05 level denotes significant at .01 level L The equations contain net household income as the income variable. TABLE 8 Statistical Significance of the Effects of Proximate Deter.inants: P-Values1 from Logistic Regression Estimations Dependent Variable: Type of Mortality Predictor Variable and Equation Neonatal Post-Neonatal Infant Child Maternal Risk Status Equation (6) .000""" .000,'1 .0001"1 .000 Equation (4) .000*** .000""" .0001" .000- Parental Kin Relationship Equation (6) .013** .598 .034" .447 Equation (4) .268 1.000 .4s4 .758 denotes significant at .10 level denotes significant at .05 level *w denotes significant at .01 level 1. Probability that the F-statistic does not differ from zero. 2. The equations contain net household income as the income variable. TABLE 9 Logistic Regression CocffEciants fo Effects of Proxirnace Determinants Dependent Variable: Type of Mor.aity Predictor Variable and Equation Neonatal Post.Neona:al Infant Child Maternal Risk Status (contrast with LLow) Equation (6) II .547" -.227 -.164 .734"" I .767," .359" .480" 1.227" IV L2950" .8360"1 .S12 1.359"" V 1.7940" 1.432"' L.417-, 1.6901"" Equation (4) 1I .651" -.174 *.111 .737"" III .794"0 .293 6""" 1.135""" IV 1.337. . .709" . V 1.16 -1.34" 1.73"" 1.630"" Maternal Risk Status (contrast with 1st Cousin Equation (6) 2nd Cousin .204 -.040 .097 -.060 Not related ..20S ..12S -.163 -.216 Equation (4) 2nd Cousin .102 -.004 .054 -.029 Not related ..141 -.008 -.053 ..121 " denotes si-tican: at .10 0evei "" denotes signific-: at .05 evel " denotes six.dIc:a:t az .01 levei 1. 7he eocuations cornairn et iwusehoid inrne~r as -:ie mcorne TABLE 10 Statistical Sip'ficance of Effects of Household Income: P-values from Logistic Regression Estimations Dependent Variable: Type of Mortaliny Predictor Variable and Equation Neonatal Post-Neonatal Infant Child Net HE Income Equation (1) .01023 .871 .138 .0O0 Equation (2) . .321 .584 .677 .004.. Equation (3) .431 .812 .440 .0117 Equation (4) .381 .873 .6.55 .00 Per Capita HE Income Equation (1) .094W .723 .347 .000 Equation (2) .930 .333 .268 .009 " Equation (3) .787 .457 .274 .012 Equation (4) .708 .538 .38,6 .016"; Index of Consumer Durables Equation (1) .000""" .240 .000... .000 Equation (2) .456 .666 .861 .34 Equation (3) .640 .463 .721 .909 Equation (4) .721 .531 .777 .93S denotes significant at .10 level denotes siznificant at .0.5 level denotes significant at .01 levei 1. Probability that the F-statistic does not difFer From zero. T.ABLE li Logiszic Regession CoefMciens for Effecs of Household 1ncorme Dependent Variable: Type of Morcalicy Predicor Variable and Equation Neonatal Post-Neonacal Infant Child Net HH Income (contrast with Lowest) Equation (1) Level E .007 -.046 .011 -.321 Level = ..378""" -.069 ..165' Equaxion (2) Level 11 .099 .017 .0S7 -.230 Level M -.121 -.133 .061 -.682" Equation (3) Level 1 .148 .053 .131 ..253 Level 1 -.029 .097 .089 -.632 Equation (4 Level E .103 .049 .094 -.293 Level 1 -.112 .0s0 .025 -.71 Per Capita H Incone (contrast with Lowesz, Equation (1) Levei ..110 ..059 -.001 Level M -.301,- ..043 -.125. Eaua:ion (2) Level ri .010 .155 .L05 -.253 LevelIl .057 .203 Equaion (3) La-el l .093 Levelim .054 .127 ..4- Euation av~ei ~..- -. Neonatal Post-Neonatal Infant Child Index of Consumer Durables (contrast with Lowest) Equation (1) Level II -.343'* .167 -.247-. Level III -.767"* -.256' -.464... -.908"' Equation (2) Level II -.142 .023 -.051 -.121 Level I -.227 .150 -.008 -.385 Equation (3) Level II *.131 .00S -.071 -.070 Level M -.174 .210 .012 - Equation (4) Level I1 -.100 -.037 ..066 -.0-50 Level II -.170 .149 -.002 -.109 * denotes significant at .10 level denotes significant at .05 level denotes significant at .01 level TABLE 12 Statistical Significance of Effects of' Determinants of Intervening Variables: P.valuesi from Multivariate Estimaions- Dependent Variable: Intervening Variables Equations with Sex of Educational Source of -Type of Net E Income Child Aspirations Water Toilet Region of Residence .32S .00o". .0001"" .000". Type of Place of Residence .082 .0001" .0001"" .000... Maternal Education .)40 .000""" .00011, .000Q" Paternal Status .655 .000""" .00O.w .000""" Net HH Income .699 .000""" .000"" .000""" Equations with Per Capita EHE Income Region of Residence .405 .000""* .000". .000"a" Type of Place of Residence .082 .000""" .000".. .0001"" Maternal Education .890 .000"0" .000-" .000""" Paternal Status .764 .000""" ..0" .0001-" Per capit. E Income .10S .000" .00 .0001"" denotes significan: at . 0 1eve; d denotes simiZa:t a: .0.5 .vei denotes Si.Tic1c.: at .0. :evei ?-Obabilitv . "le does not d~.~i.-r:z.3n TABLE 13 Statistical Significance of Effects of Determinants of Maternal Risk Factors: P-vahies from Linear Regression Estimations Dependent Variable: Risk Status by Type of Mortality Equations with Post- Net HH Income Neonatal neonatal Infant Child Region of Residence .000"" .000-1" .000"" .003 Type of Place of Residence .280 .75S .203 .231 Maternal Education .000"" .009"" .000*" .000u Paternal Status .027' .S04 .207 .31 Net HH Income .003*" .150 .008"" .582 Educational Aspirations .194 .244 .3s8 .303 Equations with Per Capita HE Income Region of Residence .000"". .000""" .000""" .007" Type of Place of Residence .236 .625 .152 .245 Maternal Education .000" .00S". .001z" .001... Paternal Status .039" .739 .202 .3.53 Per Capita HH Income .547 .00-, .151 .560 Educational Aspirations .234 .237 .444 .255 denotes siznificant at . 0 levei denotes signifcant at .0.5 levei denoces sit:i!cant at .0i ieve; . ?*obability that the F-szacisnc does not Fi:er n zero. TABLE 14 Stadstical Significance of Interactive Effects ofSocioeconomic Factors and Income Variables (Equation 2): P-values from Logistic Regression Estimarions Net BE Income Per Capita El Income Region of Residence Neonatal .183 .422 Post-Neonatal .77 4 .046" Infant .204 .212 - Child .074 .640 Type of Place of Residence Neonatal .0171" .059" Post-Neonatal .423 .983 Infant .106 .397 Child .563 .915 Maternal Education Neonatal .040" .235 Post-Neonatal .266 32 Infant .0341" .272 Child .124 .010- Paternal Status Neonatal .026"" .014"' Past-Neonatal .945 .549 Infant .231 .04, Child .343 .056" deates signifcan it '_0 'e7e. " denotes siantiles: a: .05 Leve. """ denotes signxic=: at .0' 'eve; -. ?aoaoiir ~na :n .~i.c ot 4:a: :%rl. TABLE 15 Statistical Significance of Interactive Effects of Intervening Variables and Income Variables (Equation 3): P-values1 from Logistic Regression Estimations Net HH Income Per Capita MH Income Sex of Child Neonatal .291 .753 Post-Neonatal .695 .947 Infant .540 .6S1 Child .064' .103 Educational Aspirations Neonatal .322 .951 Post-Neonatal 1.000 .922 Infant .586 .924 Child .277 .870 Water Supply Neonatal .126 .343 Post-Neonatal .594 .125 Infant .077' .201 Child .231 .8S3 Toilet Facilities Neonatal .459 .501 Post-Neonatal .409 .112 Infant .499 .444 Child .857 .5 25 4denotes significant at 10 1evel " denotes significant at .0. level denotes sianiflcant at .01 level i. Probabilitv that the F-statistic does not difler ^rom zero. TABLE 16 Logisic Regression Coefficients for Interactive Effecs of Income and Socioeconornic Factors Post-neonatal Child Mortality Morrality Region Region Loer . Upper 1.wer UCpe" Net I .000 .210 Per Capit- 1 ' .000 .796 EH I -.363 .179 HH I 9 .719 Income Ll -.499 -1.142 Income Il Neonatal Moraliry Neonatal Mor:aliv Type of Place T-pe Or Pace Rural Urban RU?:: Wre. Net 1 .000 .169 Pr Canil:2 .00 09 Incomne ZZll .L03 .69I.oe tl ., . Neonatal Mortality Infant Mortality Maternal Education Maternal Education None 1+ yrs None r Net I .000 .035 Net 1 .000 ..076 HE II .095 .124 HH II .042 .OSS Income III .084 -.630 Income iII .1s6 -.320 Child Mortality Maternal Education None 1+ yrs Per Capita 1 .000 -.513 HH II -.579 .101. Income III -.852 -.683 Neonatal Morality Neonatal Moraliry Pacernal Stacus Paterna! Stanu 1.11 uri-tv 1-t .II Net 1 .000 .121 Per Capita I .000 .450 .038 .2S2 HII .199 ..055 IncOme arI .05T .4 Income r11 .250 ..1 Infant Mor:ality Child Mor-:aliry Paternal Status . Pasernal S:acus 1 .11 III-IV -. -fl IIH-IV Per Capita 1 .000 .239 Per Capi:a 1 .000 EK .205 .087 -.090 -1.252 icome ..121 nm ..35 - TABLE 17 Logistic Regression Coefficients for Interactive Effects ofIncome and Intervening Variables Child Mortalirv Sex of Child Male Female Net 1 .000 .481 HH I .156 -.194 Income III -.397 -.323 Infant Mormality Source of Water Other Pub Fauce. Dwell Fc:,:: Net I . 000 .115 .003 II .356 .043 9 Income I11 .297 .260 -.350 APPENDZX A ALT=NATIV IASU=3 OF PER CAP:TJA HOCSIOD TICVECI Two issues complicate the const=ucticn of measu=es of mer capita household income. Both concern the dencmirnator of the measure (household me=bers) .1 First, which ousehod teters should be included in the dencminator, and wat weiJgh they be given? Second, household size and structure is in part determined by =ortality, introducing pctential bias in the effeco of income on =ortality. Ecw can this source of bias be raduced or eliminated? Response to the first c-.estion is detprmined in oart by the analytlc cbjectlves of utilizing a per capita =aasure in analysiS of child survival. -n our view, the priary prose of suCh a measure is to serve as an indicator of rescurces aailable cr consumpton by children. Thus the measure should take i.t: account, first, all household .ers and .ct =erely adults, ar.d, second, the expectad relative cns=tie=.rn level cf eacn =e-. zr, as deterei.aed for exmle ty their age and4 se:. ?:- -hi purpose, ar cp=i_al =easure wculd inica the a=zunz ;=: es, by ty-, tha= are availa't~ to c.ce ~.u_~~~.----- .at are a--=red- -- - ar e- :.a - ==- C ca'cu l--. s-- =e-ass Z.h£ e .. .... .. ..... .- - ......- ion of three measures. The first measure is the simPlesz to construct and sorves primarily for comparison with the more complex second and third measures. The denominators of the three are as follows: A. Number of household members. B. Number of household members, weighted by age: Ase Weicht <5 .50 5-14 .75 15-59 1.00 60+ .85 * C. Number of household members, weighted by age and sex: Weiht Age alesFemales <5 .60 .40 5-14 .85 .65 15-59 1.00 .85 60+ .90 .80 The second issue identified earlier - reciprcal causai. between mortality and per capita household income, induced by Zhe impact of mortality on household size - raises ccm,ex an- basically intractable problems. We have ccnsiderad se-eral resmonses. Since mor-ality. c i.fan=s and ciren is 7ne =a:- source of bias, hCusehold incr.a zer adult =an of income per capita. Such a measure 4=as n: reCUire.=enz seci"ied abCve of rslatin -'; resou-r - - for consuption by children. Nevertheless, it might be justified as a means of avoiding bias. Other approaches alleviate the bias by taking into account children who died. One approach is to add to the denominator all children dying ga spcified ref er- ence period (five or ten years befara the survey, for exa=ple) . This seems an over-adjustent, however, since in all likelihood most children who died were not alive during the lifeties of other children in the household who also died. ndeed, this adjustment would see likely to induce a counter bias- towards observing the expected negative relationship between household income per capita and mortality. Addition of the child who died to the denominator, in calculation of the value of per capita household income for that child alone, would see= to yield a household size which reflects =re acurately the situaion far the child in cuestion. In the same spirit, through utilization of the birth histories of E7S respondents, a household membership for each birth can be calculated by adjusting household me=ber- ship at the survey date by subtracting out suvVn children barn subsequent to the child in questicm and adding children assmed to have left the hcusehcld after the =fth of the child but be2cre the survev data. :his appr=ach raqu-=- heroic sirpifyin5 assu;mtions ahcut the tir:i f dc arfr fr:: the household (we assu=e age 25 fzr =ales and ave Is for fe- mae),and is ;eiul ncmlt ea nc daza are ava:__~.. on entrances and exlzs of adults cr enire . =z= z.e househo.ld. further four per capita income measures, with denom=inatcrs as follows: D. Number of household members aged 15 and over. E. Number of household members, weighted by age and sex, including all children dying in the ten years preceding the survey. F. Number of household members, weighted by age and sex, plus the child under analysis if it did not survive to the survey. G. Number of household members, weighted by age and sex, plus the child under analysis if it did not survive to the survey, with adjustment on a child-by-child basis for siblings born subsequent to the child and for siblings assumed to have departed the household after the ch-ild's birth but prior to the survey. Before examining empirical data on the seven per caita household income measures, we should indicate our preference f=r the measure with denominator F. We prefer this measure fz= several reasons. First, it takes into account all househcld members at the .survey date as well as presumed varlazion in consumption behavior by age and sex, both, in our view, =hecram- ically important features of a per capita measure --r uss in analyses of child =crtalitv. Second, Mo=. sensible adjustmant for f he nas inS mortaliv on household size. Because we lac,% reconst.uction of past composition in G is crude and Inc=plete), we are forced to assu=e that changes in househcld c=position over the period have not been associated to any marked extant with total household income. Addition of the dead ch. the household me=bership at su=,ey is meraly an adjustr.ent for the reciprocal causation bias described above, not an effort to reconstruct household membership. For this purpose it seems a zodest and entirely appropriate adjustment. Product-moment correlations among the seven per capita zeasures and net household income are presented in Table A.l. Zach of the per capita measures shows a correlation with net Income of roughly .75, with the excepticn of the measure based on adults only (denominator D) which shows a correlation of .89. Correlations between this particular =east,e and the other per capita measures are also about .90. 0therwise, all corralations among the per capita income measures excaed .95. This would suggest that the choice of a per ca=ita =easura is of little significance, as long as all household =embers are included in the denominator. •That is, weighting by age and sex, and adjustents for children who died, would see- =o =aka :.J==!a difference. .stimations of effects on =or=a.-ty shc z-e conorary, hcwever, sc=ewtat to cur szr=Izse. As feared, -s which fail to adjust for the 1mact on househol zeaship o recent =ortallt of children (=easz=es based o-n den:=-ntors A. and C) tend to show cost.e effects on rort , :t7 Cdnoiatr.n r f:r - (denominators E, F and G). Furthermore, measures based on denominators D and E (adults only (D), and household members plus all non-surviving births in thew past decade (E), show the largest and most resilient negative effects on mortality. We take this as supportive evidence of the view that use of these measures biases the analysis towards estimation of negazive income differentials. As noted above, the problems inherent in devising per capita measures for the study of mortality resist straightforward resolution. We see no firm guidelines for accepting or rejecting measures on an empirical basis, although the empirical results are consistent with our a iriori reasoning. We therefore opt for the measure which seems most satisfying on theoretical grounds, namely the measure based on denominator F. TABLE A.1 Product-;moment Correlations•of Income Measures VARLBLEI (1) (2) (3) (4) (5) (6) 7) (s (1) Net H income 1.00 (2) (1)/A .78 1.00 (3) (1)/B .77 .96 1.00 (4) (1)/C .77 .98 1.00 1.00 (5) (1)/D .39 .39 .90 .90 1.00 (6) (1)/E .77 .98 .99 .99 .90 1.00 (7) (1)/F .77 .98 1.00 1.00 .90 .99 1.00 (8) (1)IG .74 .95 .98 .98 .91 .98 .9S 1.00 1. The denominatnrs. A through G, are defimed in the te.c of the appendi-. APPENDIX B Logistic Regression Equations Containing Demographic Predictors Only Dependent Variable: Type of Mortality Predictor Variable and Category Neonatal Post-Neonatal Infant Child Maternal Age (years) 20-27 ..322' *.717" *.554"" -.562' 28-34 -.682*** -.771 .. -.754... -.635-" 35-39 ..095 -1.191-" -.648" -.273 40+ -.377 -.116 -.188 -.081 p-value .013* .000"" .0000* .133 Birth Order 3-4 .207 .003 .068 .578" 5.* 6 .938"** .247 .549"" .662" 7-9 .724*" .193 .452" .447 10+ 1.264"" .967-A -132" .596 p-value .0000" .043"* .000"" .17.3 Length of Previous Birth Interval (months) 12-17 -.919"" -.293 ..630" -.619 18-23 71.237... -1.107 -1.229... *.537- 24-35 -1.765""" -1.436"' 1.654"-0 36-59 -2.249"" -1.64S"" -1.97"- 6 60- *2.3S6".. .1.565"" -990" -2.024" First birth -1.026'" ..195 -.9 -1.010" p-value .0001"" .000""" .0001" .0001 Constant *1.2S7 1 .251 -7 Number ofcases 4762 4421 496$ 2? Lagz-likeiihood -155 10.5-91E 22 denotes significant at .10 level denotes significant at .05 level denotes significant at .01 level NOT Coefficients represent contr.ss with omitted c.aegor.4s: Maternal ae* <20 years. Birth order* 1 - 2. Leng:h of previous birzh intervaL <12 mon:hs.