1Arguably the most fundamental achievement of historical and contemporary demographic research in the later 20^th and early 21^st century has been to document the enduring heterogeneity of modern population trends, particularly the diversity of secular declines known as “demographic transitions”. Contra the assumptions of the immediate post-war era, not only have the timing, tempo, and quantum of declines been shown to vary radically, but the range of historical variation in levels of fertility and mortality before transition, and the continuing diversity of fertility post-transition, have all been admirably demonstrated. This variation is visible in the most widely discussed macro-units of analysis, usually whole countries. National pictures, moreover, have been shown to be composed of heterogeneous declines amongst provinces, socio-economic strata, and local groups of different kinds [1]. Variations are not confined to trends in “dependent” variables of fertility and mortality. Reductive models of economic development and modernisation were expected by early post-war demographers to provide the basis for a general theory of contemporary demographic change applicable in all places. Regular statistical correlations “explaining” fertility and mortality declines in terms of a limited set of standardised “independent” socio-economic variables were supposed to be readily forthcoming from analysis of national statistical sources and survey samples. The diversity of declines at so many levels, however, has inevitably undermined such confidence. There is no longer a single pattern to be explained. Not surprisingly, grand syntheses in which consistent correlations between demographic, economic, and social trends establish demographic transition theory as a universal at national and provincial levels have given way to thoughtful summaries of heterogeneity and the several approaches to it (e.g. Lee & Reher, 2011). That “development” might bring an increase in inequalities, with consequently differing reproductive adjustments amongst sub-populations, has yet to become a principal object of theory [2]. Experience has nonetheless shown that macro-level trends are composites made up of many local and regional patterns. Observation and analysis will need to be directed to the sub-populations from which these patterns have emerged if this complex compositional demography is to be explained.

2Demography’s remarkable achievement in demonstrating heterogeneity, however, has led to a seeming theoretical impasse. One horn of the dilemma is empirical: there appear to be a great many subpopulations that may be important, and they are likely to be related to each other in different ways. How best to identify and characterise them? The other horn of the dilemma, in consequence, is methodological. The sub-populations in which changes are occurring often do not coincide with the standardised units of population (households, cohorts, provinces, nations, etc.) that administrative data, censuses, and surveys employ, and on which most measurement and modelling rely. Standard data units, after all, reflect the purposes of prior eras of national administrative development more than those of scientific inquiry. The sub-populations in which people participate include, for example: family, kin and community networks, generations, language groups, class structures, migration networks, religious affiliations, social hierarchies, labour sectors, ethnic minorities, regional cultures, and local and national government patronage networks. The issue is not only that reliance on household units compiled according to administrative boundaries does not capture many groups. Memberships interact, people belong simultaneously to several sub-populations, and their participation in them changes over time. This dynamism needs attention, as influences on fertility are well known to reflect the way that class, religious and other memberships, together influence family life. As historians and anthropologists have long argued, the influence of these many groups cannot be treated as if all of them are on par: they exist at differing levels of social organisation and have different power relations. The usual methodological convention that reduces all populations to two levels, “macro” and “micro” is too crude. Nor is imagining a single, intermediate “meso-level” adequate, as different societies and economies give rise to varying configurations of the several meso-level sub-populations, just listed. Specifying which subpopulations and which relationships underpin population changes is thus essential if the differential influence of social structures and inequalities on fertility is to be documented and compared.

3The nub of the methodological dilemma is that constituent groups in society, as populations, differ in critical respects from the units that conventional quantitative data sources on their own provide. The latter are formally bounded because this is essential to precise quantitative description and comparative analysis. But human populations, as social aggregates in their normal functioning, are not formally closed or discrete. This was recognised, at least implicitly, in some early revisions of the transition: amongst the processes that shape fertility are those involving flows of information and practices, sometimes called “diffusion”, within and between groups (e.g. Watkins, 1987). As social demographers gradually developed this perspective, they have focused increasingly on documenting the nature of such flows as network phenomena (e.g. Kohler et al., 2015). However, networks, in marked contrast to the closed aggregates employed in census and survey classifications, are open population memberships. An individual commonly belongs to, and is able to draw on, different networks over time as they open up differing opportunities, including possible engagement in yet further networks. Such memberships overlap to a greater or lesser degree, and the ability to participate in multiple networks, and to vary participation, is often crucial to how individuals and groups adapt to changing circumstances, giving them a flexibility that can be essential to wider social change [3]. The compositions, relations between memberships, and the size and structure of such groups, are all demographic phenomena likely to change repeatedly across the life course as members adapt their needs and opportunities. Surveys, even in panel format, do not track multiple network memberships and the changing relationships between them, for the simple reason that surveys rely on random samples, but memberships are not random.

4Thus, while the information flows that diffusion theorists have emphasised are obviously important to people’s adaptations, communications are only part of relations between such members and between groups because networks reflect relations of power and influence in society. Networks, in other words, while integral to understanding how and why structural changes occur in society, take place in the context of extant structural differences. We may view changing networks and relations between them as building blocks of a social structure, as they have real and potential capacities to alter such structures. Arguably the central issue in understanding how societies continually generate demographic heterogeneity is to find a way to model the fluidity of network relations and relate these models to the closed units and models that demography customarily produces at provincial and national levels. At present, however, there is no consensus in demography comparable to other fields of population research as to how these building blocks, and the way they combine at different levels to make up social structures, should be defined in regular ways for comparative and modelling purposes [4].

5The subject of this special issue – understanding and modelling the social space of fertility differentials – entails several tasks, well described by the editors. One, as just remarked, is the critical exercise of deconstructing the standardised units employed in conventional data sources. To gain traction any such critique must give rise to more realistic population units and compositions, thus making new interpretations possible. The view taken here is that this rethinking involves three methodological steps. The first is combined qualitative and quantitative ethnography of the building blocks, i.e. of local network sub-populations. An advantage of in-depth local research is that a complete mapping of a community’s households becomes possible, providing the basis for rigorous, stratified sampling. This is carried out in a number of communities, chosen to reflect variation in ongoing social and economic changes in wider society. The second step is to develop sampling and survey design informed by the more detailed knowledge of local languages and mores gained ethnographically to devise local surveys which provide data on a community as a whole. This provides a systematic test of ethnographic hypotheses and their revision as necessary. The third step, then, is triangulation. As will be noted in later sections of this paper, ethnography produces several datasets, both qualitative and quantitative. Just as a local survey tests the ethnography, the much greater variation evident in the ethnography – notably life course data – provides insight that informs interpretation of survey findings. Repeated ethnography and survey at later points in time then make this whole structure longitudinal [5].

6The outcome of this methodology is not only to show how socio-economic levels can be differentiated empirically and modelled from the bottom-up [6]. A second outcome is, at least in outline, a potentially different theory of what shapes reproduction. In contrast to the rational choice approaches which the editors have critically remarked – in which reproductive choice is viewed as “decision-making” addressed to whether or when to procreate –, the nature of actors’ agency is geared to the composition and structure of sub-populations in which they participate. Take, for example, a person who participates in a number networks, variously of kin, affines, labour sectors, or religious affiliations. That person’s ability to draw on the resources of one or several of these memberships depends on his or her positions in them, which define the wherewithal available to the person for a purpose like childbearing (in addition, of course, to the person’s own abilities). Decisions regarding childbearing and rearing entail a number of alternative courses of action or support that are opened or closed by resources made (or not made) available by the several local and meso-level sub-populations that compose social structures. Familiar examples include whether kin or labour networks secure a partner’s job; whether migration limits possible network sources of childcare assistance; or whether membership in religious or other organisations provides educational and other services.

7Such decisions are not isolable as “decisions” pro or contra bearing a child, as they are often the fait accompli of other courses of action (preserving social status, career choices, marriage, housing, etc.). There are many feedbacks between having or not having children and the wherewithal, opportunities and constraints of these memberships. Alternative courses of action, in other words, are adaptive properties of networks and the social structures they build – rather than merely idealised individual rationalities or market choices. Their variation can be explored as part of the processes that continue to make groups differ, or may create new sub-populations.

8The following pages illustrate this bottom-up approach and its implications, drawing on a multi-site, longitudinal study of three Indonesian communities, which is now in process of restudy at the twenty year interval [7]. We begin with a brief résumé of the communities and how this methodology has been employed. Principal constituent sub-populations emerge gradually from observation and documentation of network dynamics. The methodology and its findings readily assist identifying limitations of some common demographic models, and in the fourth section of this paper, three are briefly noted: childlessness and low fertility; inter-generational support relationships; and age-structural transitions. The reproductive spaces of the three communities are then illustrated in a quantitative summary of fertility patterns and differences between age groups and socio-economic levels, focusing on two of the Indonesian communities. Four case studies illustrate the longitudinal ethnographic picture of the levels, age groups, and networks. Reproductive variations across the life course of groups are related to resource options that subpopulation memberships in different levels and networks make available. The case studies are supplemented by data recently collected as part of the ongoing research. A more complete quantitative analysis of the communities for the whole twenty year period is, however, not possible at this stage, as the current restudy is in its early phase.

Research setting

9Beginning in April 1999, a joint Indonesian and British research team has studied the populations of three communities: Kidul in East Java, Koto Kayo in West Sumatra, and Witan in West Java [8]. The family systems in the two communities on Java are characterised by normative nuclear / bilateral patterns, whilst the Minangkabau population of Koto Kayo is traditionally extended / matrilineal. Each community is characterised by a mixed economy, drawing on income from migrants, from employment in local government, and from services and small-scale manufacturing, while also retaining the traditional economic base in agriculture and local markets. The communities are thus all, but in different ways, participating actively in Indonesia’s rapidly expanding economy [9]. Of the three, Witan has remained closest to the much older historical pattern in which wealth, status and inter-generational relations are grounded in rich agricultural resources, notably premium rice land (sawah). Koto Kayo likewise draws on historical precedent, in its case sawah plus a migrant diaspora established by the early 20^th century which brings considerable status and a continuous stream of wealth into the community. Family networks in Kidul have neither copious sawah nor long established diaspora links, relying instead on varying combinations of a mixed economy. Proportions of older respondents’ adult children reported in 2000 as no longer resident in the community (46, 75 and 45 percent, respectively) give some idea of the active engagement of family networks in regional, national and international economies. All communities are predominantly Muslim. Languages spoken in the home are respectively Javanese, Sundanese, and Minangkabau, although the national language, Bahasa Indonesia, has become increasingly prevalent. Interviews were conducted in more than one language in each site.

10A combined methodology, beginning with a year’s ethnographic fieldwork, and continued site visits, enabled compilation and analysis of people’s life histories and mapping of kin networks and exchanges over time. Semi-structured interviewing achieved substantial coverage of the population aged 60 or over (the principal focus of the early phase of research), between 80 and 97 percent in the respective communities. Repeated in-depth interviews were conducted with between 20 and 60 older people, complemented by indepth interviews with one or more other adult network members in most cases. Focusing on older cohorts has proven particularly helpful as their life courses entail the history of family vital events and their community networks. Fieldwork makes possible observation of local events and processes, facilitating familiarity with the problems and adjustments normal to changing circumstances that make up much of people’s daily lives. During fieldwork a comprehensive mapping of households in the communities was completed and updated, enabling two rounds of randomised surveys of health, household economy, and inter-household exchanges in each of the three communities. Stratified sub-samples of households with and without elderly were included. Many important sets of relationships are opened up by this combined qualitative and quantitative database, revealing differences in social and economic status within and between networks and generations, which shape economic support, reproduction, and elderly care. Surveys, indepth case study, and observational data have additionally allowed support from absent network members to be included in analysis.

Methodology in situ

11A number of observations can be made about this research strategy as a basis for understanding the social space of reproduction, and in contrast to conventional demographic approaches. First, the constituent groups making up communities, and the units of comparison for analysing them, are not decided in advance. Which sub-populations are fundamental to a given person’s resources and behaviour is an open question. We began, following demographic custom, by focusing on households and talking to their members. The Javanese social norm is for households to be based on nuclear families; the Minangkabau prefer extended families with three generations. However, there is much variation, and household compositions are in constant flux. This reflects the passage of property and support between generations, both upwards and downwards across the life course, and in response to changing needs: education, setting up small businesses, division of labour, marriage, divorce, and migration (Kreager & Schröder-Butterfill, 2008). For practical reasons, it is simply not possible for most families to observe preferred arrangements all, or even some, of the time (Schröder-Butterfill, 2005; Schröder-Butterfill & Syawila Fithry, 2012). Rather than devising a classification of the resulting arrangements as a static set of household types, it became clear that a more realistic picture would take kin networks as basic units, viewing households as nested within them and tracking changing events and compositions over time.

12We found that important networks were usually kindreds, as both kin and affines, and sometimes neighbours, were involved (Kreager, 2006). Three kindred types were readily distinguishable: a total range of recognised kin (the “abstract kindred”, from a family elder’s perspective); a more limited range of kin (“proximate kindred”) defined by material support exchanged across the elder’s life course; and, finally, the small group of kin and neighbours providing significant support or care at the time (“immediate kindred”). Proximate networks evolved as generations aged, since different sets of members of the abstract kindred become relevant as time goes on. The open character of network sub-populations is ever apparent. For instance, support needed in early adulthood (e.g. older kin providing childcare, or helping finance or build a house), did not have to be sustained as young people grew into maturity. In mid-life proximate kin might involve labour and other exchanges between siblings or affines. Of course, adult needs and responsibilities often changed late in life, as needs for care or hospitalisation meant that the membership of proximate and immediate kindreds was revised repeatedly. These changes, for example as made to secure property or care for an older person, commonly had consequences for the social space of reproduction. Network resources available at a given point in time for childcare, educational support, and other concerns are often consequences of past compositional changes in networks that responded to needs other than reproduction per se. They nonetheless remain crucial to available childcare resources – e.g. merely who is on hand to help, or cover costs – and other needs.

13Proximate networks have consistent properties, even as memberships change. Usually they involve a division of labour that combines some adult children, nieces or nephews resident in the community together with significant participation of some of those living elsewhere. Although not an explicit strategy that people claimed to follow, this pattern had the obvious advantage that members of different generations were able to call on support from multiple sources. They changed supporting, caring and other roles in the network as the varying needs of different members arose. The high percentages of younger generation migrants, noted above, indicate the common character of the family as a network not simply resident in one household and location. Migration, in addition, was often not a one-off move away from the community; members might come and go, over different durations. (It is perhaps worth keeping in mind that Indonesia, as an archipelago, has a very long history of movement over widely varying distances: see Hugo, 2015; Schröder-Butterfill et al., 2018.)

14The ethnography of multiple, shifting networks reveals the fluidity of alternatives and constraints that can impact on childbearing, rearing, and social status. Members of a network often give different accounts of “the same” events and human relationships. After all, not everyone’s perspective on children and the family is the same, and (pace demographic convention) the household head’s perspective is not necessarily privileged. It is thus often necessary to make some assessment of the relative accuracy of different accounts, and of motivations behind them. Differences are often telling. People, after all, may be more candid on one occasion than another, and this affects many issues, notably: whether they report a child as own or adopted; whether some children have died; whether there have been family disagreements over children, property or residence; whether there was a previous marriage; or whether behaviour considered immoral is mentioned at all. The same member may give a different account on different occasions. More important, comparing different network members’ accounts often provides missing information. Such issues are fundamental to accurate collection of even basic demographic data. Observations that members of one network may make on the behaviour of other families and their networks sometimes provide additional clues. In short, triangulation of findings from different datasets is crucial to data quality.

15Because our evidence was collected in several ways (observation, everyday conversations, life course interviews, surveys), the results formed different datasets that could be compared. In this way the possible influence of research methods on findings could be monitored [10]. For instance, many apparently simple topics in demographic surveys assume that respondents can make accurate estimates. Examples include: Does the household head know the income of a co-resident adult son? Do people know the cash value of their agricultural holdings, or of the labour contributions which different members make in tilling them? Triangulation here arises from the fact that quantitative information does not just arise from surveys: many of the issues discussed in semi-structured interviews turn on incomes, durations, remittances, and so forth. Data files were compiled on quantitative reportage that took place during such qualitative interviews, and this helped to ensure that potentially important aggregate differentials could be compared and assessed between what people reported in surveys and in-depth interviews. The comparison of these data to survey reportage then provided further grounds for assessing accuracy. The ability to make comparative checks such as these obviously is not available to survey analysis and interpretation on its own. Checks are only possible where a multimethod approach is employed in which sustained ethnography precedes surveys. Of course, in-depth data are not authoritative on their own either: where previously unrecognised ambiguities, relationships, and events are discovered in comparing surveys with other datasets, then further interviews and related ethnography become necessary.

16Put another way, research activities take place in the social space of families, their reproduction, and other concerns. Data collected are never completely separable from local realities as understood at the point of their collection. This needs to be kept uppermost in mind when researchers decide how best to specify units and relationships for comparative and modelling purposes. How to answer the question raised earlier – which sub-populations are realistic units of analysis, and what levels of aggregation need to be distinguished – obviously has a profound effect on what research is or is not able to show. Sometimes demographic units may be aligned almost exactly with local social categories. Anak angkat, or informally “adopted children”, identifies a small but significant group of this kind (Schröder-Butterfill & Kreager, 2005). In most cases, however, sub-populations employed for comparison and analysis, like the three kindreds just outlined, are identified through extended observation and in-depth interview backed up by quantitative evidence compared between in-depth and local survey data.

Deciding sub-population units and levels

17Three principal sub-populations emerged from the research: networks (inclusive of households), generations, and socio-economic levels. It is easiest to see the three as referring to different tiers of social organisation. Socioeconomic levels are the largest subpopulations and reflect key economic differences. Network and generational relations, while building the differences between socio-economic levels over time, also cut across them. The focus here will be on the influence of socioeconomic levels in shaping reproductive spaces. As we shall see, such differences arise from the influence of network and generational interactions. However, as these variations are by nature multifarious, they are described more briefly by providing specific examples. Illustrations from case study material nonetheless indicates the basic roles of the three types of kindred networks in shaping the social space of reproduction [11]. A more detailed account of network dynamics and a typology of outcomes is given in Schröder-Butterfill (2015).

18In formulating the model of socioeconomic levels, two respects in which it is an analytical construct deserve emphasis. First, although all three communities are profoundly status conscious, there exists no explicit, commonly shared vocabulary of social castes or classes. People nonetheless remark differences of reputation, property and so forth in many ways. As they consider such differences telling in their lives, devising a coherent empirical framework for describing these differences is critical for understanding and comparing inequalities in social experience. Second, the model of four levels, described below, is basically a closed model, i.e. a summary device bringing together evidence from datasets on networks, households, and surveys so that the four levels can be treated analytically as discrete units. It was developed on the basis of the three-step methodology described above, that is, after the sustained ethnographic and first round survey programme in late 2000 [12]. The key analytical point here is that the model does not stand alone. That is, both building and subsequently interpreting it requires us to consider the open network processes of the three kindred networks, described above, as these are the active processes that define individual and family position in the four levels, and which continue to sustain (or perhaps fail to sustain) such positions [13]. While the model stood up well following a second round of fieldwork and survey, in 2005, one purpose of the current research is to test how well it is holding up after a longer duration, and to revise it as necessary.

19Comparison of the datasets leading to the four-part scheme drew additionally on linguistic data: the recurring but informal descriptive phrases that people use in ordinary language (Schröder-Butterfill, 2015). In the course of fieldwork, some distinctions emerged more clearly than others. All villagers could say which were the richest families, describing them in the two communities on Java variously as orang kaya (rich person), wong sugih (rich person) or benghar pisan (very wealthy), and in Sumatra as urang baharto (person with wealth). In common parlance these terms are synonymous. Patterns of education and religious observance (notably, pilgrimage), together with local survey data on assets, income and expenditure for this group, were consistent with local opinion. Often people of the highest status were descendants of long established local families. Social intercourse in all three communities emphasised deference in everyday social life: poorer villagers commonly avoid visiting their rich kin, as gossip is likely to portray them as looking for handouts.

20Villagers could also readily identify which were the poorest families, using phrases like kurang mampu (less able), wong susah (person who experiences hardship), or urang bangsek (poor person) to describe them. At this level, family income depended wholly or partly on charity, and neither modern productive assets nor consumer goods were common. In the Sumatran site, they were further distinguished by being almost entirely newcomers to the village (pendatang).

21How and whether to make distinctions within the large group who are neither rich nor poor – and which together make up well over half of the community population – was much less straightforward. Differences between them were not immediately recognisable in everyday discourse, as no single phrase recurred consistently to describe this broad group. Contrasting levels of wealth and material assets were, however, indicated in the local surveys, in which some households in the broad band between rich and poor were found to possess twice the average income of the others. Typically, households belonging to the former grouping are nested in networks with reasonably secure and often multiple modest incomes; several household members are likely to work, and incomes may be supplemented by support from elderly parents’ pensions and/or profits from the land. We therefore decided to identify this group as being the “comfortable” stratum, reflecting a number of adjectives like lumayan (fair), manangah (middling) or sadang (moderate) that recurred in people’s observations on them.

22The situation of those households between the “comfortable” and the poor is captured aptly by a phrase common to all three communities, cukup-cukupan, translatable as “getting by”. These families lack the diversified resources and networks of the “comfortable”. Although their labour enables them to be self-sufficient and to participate respectably as full members of the community in most social exchanges, they have no material safety net should a health or other crisis occur. Not surprisingly, as the threat of descent into outright poverty is ever present, quite a number of cukup-cukupan households were the objects of occasional charitable support. The distribution of households amongst the four levels are given in table 1.

Tabl. 1

Distribution of households by socio-economic levels in the three communities (percentages): household survey and fieldwork data, 2000

Distribution of households by socio-economic levels in the three communities (percentages): household survey and fieldwork data, 2000

Exploring the model: Some examples

23The value of a model, of course, lies partly in the extent to which its applications successfully clarify empirical relationships and give rise to further unrecognised, or insufficiently appreciated, ones. In so doing it should enable us to specify limitations in extant models, and provide ways of correcting them. The four-tier model outlined above, although arising via a field methodology atypical to demography, nonetheless is a closed model of a conventional descriptive demographic kind. Exploration and application of the model is straightforward, and to date has provided useful critical leverage over a number of standard demographic approaches. These are: tracking patterns of low fertility and childlessness; understanding and mapping variations in inter-generational transfers; and identifying limitations of dependency ratios for capturing the impacts of age-structural transition. All three are directly relevant to conceptualising and comparing social spaces of reproduction, as they lead on to describing current variations in that space in the three communities [14].

The child poor

24Perhaps the first thing to note about table 1 is that it is not a classification based solely on incomes, assets or occupations. There are other issues critical to status that have to be included. Property inheritance, for example, commonly occurs over the course of a person’s later adult life, and in consequence older men and women of recognised wealth and higher status may come to live in a household which, in its material appurtenances and current income, are modest. Their investment, obviously, is in their family network, i.e. much family property is now the possession of other, younger members. Often an individual’s characteristics entail other status resources – e.g. healing expertise, religious and political office – and added resources sometimes flow from these network connections to other family and kin. The table therefore reflects adjustments made in triangulating life course histories and observation with survey data to give an adequate account of social as well as economic status criteria. Even when these adjustments are made, the table reveals an uneven distribution, with over half of households in all three communities falling into the two lower levels. Additionally, there are evident differences between the communities in the number of households in level 1, and also at the bottom of the scale. Households in levels 3 and 4 are more evenly distributed in Koto Kayo, for reasons noted below.

25One of the most fundamental demographic questions for the social space of reproduction is the distribution of children across socio-economic levels. Relative supplies of children have an evident bearing on the size of family networks and their capacities, and hence on the conditions in which childbearing and rearing occur. What is at issue is not just having children and ensuring they are sufficient to rely on, but the greater involvement in the daily life of the community that comes at each life stage as one’s children are growing up (e.g. arranging ritual celebrations to accompany birth, circumcision, and marriage; assisting in costs and celebrating the fruits of education; entry into the job market; election to office; etc.). Involvement at each stage opens up social and economic opportunities for exchanges and forging links to kin and in the community. For this reason it is crucial to look beyond households to the network that family members build across their life courses. Table 2 therefore turns from the survey data to the quantitative picture of family networks resulting from life course interviews. The four levels are here simplified to two, as the evidence shows clearly that marked inequalities affect both the cukupcukupan and kurang mampu, as opposed to the orang kaya and lumayan. A striking association between socio-economic status and the availability of children in this way becomes apparent: in all three communities the better-off are much less likely to be childless or have small family sizes. Indeed, they are much more likely to have many children. Between one quarter and two thirds of elderly people who are economically disadvantaged have either no children at all, or only have one or two surviving children. The latter points to much smaller kin networks, and many fewer opportunities to expand links. Having fewer younger generation members not only increases the need to rely on kin other than children, but to do so with less network resources to offer in exchange. This characterises poorer groups in all three communities.

26From a conventional demographic point of view, the non-Malthusian character of this is striking [15]. More than this, the unequal economic and social opportunities that can confront the child poor raise the issue of their vulnerability. Confining analysis, as is usual in survey research, to households limits what we can know about their access to network alternatives. But how well are networks able to cope? The differentials in table 2 are the product of a number of factors – levels of childbearing, survival, and migration – across the life course of generations and their networks. One implication is that, for lower strata, the picture is one of the inter-generational transmission of poverty: the economic situations of poorer people are bound up with worse health conditions, diets, and means to assist the children they do manage to have; the children, starting and usually continuing in life at a disadvantage, are in turn likely to have much less in the way of support they can offer the older generation [16].

Tab. 2

Availability of children to elderly people by membership to upper or lower two socio-economic levels (percentages): fieldwork data, 1999-2000. For Kidul and Witan the differences are significant (χ² = 22.2, p = 0.008 and χ² = 18.9, p = 0.026); for Koto Kayo the difference is not significant

Availability of children to elderly people by membership to upper or lower two socio-economic levels (percentages): fieldwork data, 1999-2000. For Kidul and Witan the differences are significant (χ² = 22.2, p = 0.008 and χ² = 18.9, p = 0.026); for Koto Kayo the difference is not significant

27Exploring variation in childbearing has helped to reveal the nature of network coping strategies and their limitations, as well as the extent to which standard demographic and economic measurement obscures fertility-related causes of vulnerability (Schröder-Butterfill & Kreager, 2005). One thing which is clear from in-depth interviewing is that the poor did not wish to have fewer children, nor try to (Indrizal, 2004; Schröder-Butterfill, 2004). For the oldest generation, the childbearing pattern in table 2 reflects the much less favourable situation of the proximate determinants of fertility and mortality in their prime childbearing years, particularly in the 1940s and 1950s. This was true not only in the communities we studied. The era of Japanese wartime occupation and subsequent war of independence saw major disruption of local agricultural production, spousal separations and deaths, and increased levels of STDs owing to enforced wartime prostitution (Van der Sterren et al., 1997). As late as 1971 infant mortality rates remained over 100, by which time total fertility had returned from wartime lows around 4 to an estimated 4.7 (Indonesia: Demographic and health surveys, 2002-3); infant and child mortality were doubtless much higher in the wartime and the early independence era, for which data remain incomplete. As table 2 shows, the considerable impacts of disruption on vital rates are visible not only amongst the poor. But that is not the end of the matter, as it is here that networking options become significant.

28Greater status and wealth provide alternative avenues to gaining children, which is not apparent in standard demographic measures, such as Children Ever Born (CEB). If we take the East Javanese community, Kidul, as our case in point, CEB calculated from fieldwork data on the four levels show that the mean number of children born to villagers aged over 60 in 2000 was 4 in the two higher strata, and 3.25 in the two lower (Schröder-Butterfill & Kreager, 2005); the former figure is comparable to CEB for that cohort in East Java as a whole (Indonesia: Demographic and health surveys, 1991). Overall CEB, however, occludes the marked reproductive shortfall shown in table 2, namely childlessness reaching nearly one in five in upper, and one in three in lower strata. Without attention to key sub-population differentials, CEB is likely to be misleading. There are, however, alternative routes to having children, which include remarriage to someone with children and the longstanding custom of informal adoption which is common throughout Southeast Asia (Koentjaraningrat, 1957; Carsten, 1991; Sriono, 1992). Indeed, even people who have children often adopt further children. Additional issues that affect the availability of younger generations arise from migration, particularly as it reduces network size in the lower two socio-economic groups (Kreager, 2006). Put very simply, where poor people have fewer resources and can contribute very little to their children’s life prospects, those children may be more likely to leave the community and cease all contact with their elders. This has affected older generations, for example, subsequent to the Indonesian government transmigration schemes in which children moved to distant islands. When these several factors are taken into account, it becomes clear that the social space of reproduction – i.e. of having and retaining younger generation members – depends not only on having babies, but on their subsequent distributions.

29Put another way, conventional demographic childlessness (i.e. nil fertility) needs to be compared to actual childlessness. The latter is composed of only some of those who did not succeed in bearing children of their own. To that group must be added those who have ceased to have any contact with children who have left the community, or owing to divorce and family conflict. But to identify those who actually lack access to children and their support, we also have to subtract from those with nil childbearing those who have subsequently acquired children through adoption or remarriage. The outcomes are illustrated in table 3 for the East Java community. Differences of socioeconomic level are again critical: alternative routes to acquiring children improve the availability of children for those at higher levels, and to a lesser extent for those “getting by”, but the situation of the very poorest level is considerably disadvantaged. The limited resources and reputations of the poor leave them with much less chance to adopt or make strategic marriages.

Tabl. 3

Demographic and actual childlessness by socio-economic level, Kidul, East Java (percentages): fieldwork data, 1999-2000

Demographic and actual childlessness by socio-economic level, Kidul, East Java (percentages): fieldwork data, 1999-2000

30Looking at childlessness and low fertility thus raises important questions for the ability of standard fertility measures to reflect central concerns in the social space of reproduction, notably the threat of remaining childless. As a historical and contemporary problem, childlessness and the stigma that attaches to it has been remarkably widespread, over long periods [17]. Reliance exclusively on household figures which do not take account of network adjustments may hide the actual extent of vulnerability, and the coping mechanisms used to deal with it.

Reproductive spaces in younger generations

36Discussion to this point has focused on socio-economic differences and networks built up over individual and family life courses, relying on older persons’ networks, which provide the most comprehensive picture across generations. While striking reproductive differences are evident between levels, it is no less evident that their explanation requires understanding the networks that are the building blocks by which social and economic statuses are formed, sustained, or decline. In other words, a socioeconomic stratification model – whether constructed “top-down” (cf. note 6), or “bottom-up” (as in the methodology developed here) – is not necessarily an adequate guide to fertility differentials on its own. As a model, it has the familiar “snapshot” properties of closed models in demography more generally: it reports outcomes, not processes. To interpret it, and to ensure that the basic demographic data on which it is based are accurate, the processes that give rise to differences of status and wealth need to be tracked over time and understood. Their observation requires multiple methods that enable data quality checks. The model then becomes useful for critical analysis as well as substantive description, as noted in the preceding section.

37We turn therefore to evidence of differential fertility across younger generations, giving some indicative examples from interviews of how the network inequalities noted in the preceding pages are currently evolving. To begin with, an overview is given in table 4, which presents the situation at the time of the 2000 round of the survey in the East Javanese community. Of course, much of the fertility of interest today was incomplete at that time, as some women had not completed their childbearing years, and others had not begun. The data simply provide a representative baseline for discussion of the community as a whole.

Tabl. 4

Fertility of married women by age group, Kidul, East Java: 2000 household survey

Fertility of married women by age group, Kidul, East Java: 2000 household survey

38The right-hand column gives the completed fertility, by level, of women born in the 1941-1950 cohort, and the 40-49 column indicates the substantially completed fertility of women born 1951 to 1960. The figures for the oldest cohort are broadly comparable to the mean number of children ever born to women in these cohorts (4.24) as reported for the whole province of East Java (Indonesia: Demographic and health surveys, 1991), but one child less for the 1951-1960 birth cohort (4.0) (ibid., 1991, national data). Local data, as with provincial and national data, are likely to reflect the wider impact of war-time on childbearing, noted earlier. Sample sizes are smaller in the older two cohorts, partly reflecting the cumulative mortality experience of those ages. The economy and health conditions improved only gradually as the new national government established itself during the 1950s. By the early 1960s it was still the case that more than one in ten babies died (Hugo et al., 1987). The cohort in the left hand column, women born 1971-1980, is smaller for different reasons, i.e. not all had entered marriage and thus childbearing. Their fertility levels, and those for the 1961-1970 birth cohort, were moreover subject to some very different influences than their elders, notably the promotion of family planning under the “two child only” policy of the Soeharto regime which is credited with facilitating a decline in fertility at the national level from around 5.5 to 2.6 over a 25-year period (Niehof & Lubis, 2003). The data have been compared to kin network records from fieldwork, with which they are consistent.

39The figures in table 4 take no account of variations in the distribution of children across families and networks, as presented in tables 2 and 3. In effect, table 4 averages out those variations: the situation of people with nil or low childbearing is effectively swamped by the larger numbers of those having 5 or more children. In short, data on levels taken on their own give only a very partial account of inequality of access to the younger generation. While data for younger age groups are incomplete, an overall trend of fertility over time across the four age groups seems clear. On one hand, groups following those aged 50-59 appear to be on their way to lower reproductive levels. It is very unlikely that the over 40s would all on average have at least one more child, so that their completed fertility is likely to be closer to 3.0. Whether those under 39 will reach the level of 3 children is an open question, to which we return, below. Differences between levels, however, appear lessened over time, possibly reflecting the impact of the national family planning programme (which was very active by the 1970s). The steeper fall in fertility in the upper two strata of these cohorts could plausibly reflect this, given their greater participation in wider economy and society.

40Table 5 gives parallel data from the 2000 survey for women in Koto Kayo, the Minangkabau community on Sumatra. Once again, a clear gradient by age is visible, although level differences are much less, and fertility is higher at all levels and age groups than in Kidul. The variation in child availability presented in table 2 is again not visible in these averages. Such variation is important as table 2 shows that lower strata in Koto Kayo have access to significantly more children than their peers in Kidul. Whether this makes their situation significantly different is, as we shall see, an interesting question. To assess these continuing, if gradually smaller, variations in the fertility of successive age groups in the two communities, we return to the action of local networks, i.e. the dynamics of the open sub-populations that constitute socio-economic levels. Case study data also enable us to consider whether the incipient fertility decline implied by the data from 2000 in tables 4 and 5 has indeed proceeded as transition theory would lead us to expect.

Tabl. 5

Fertility of married women by age group, Koto Kayo, West Sumatra: 2000 household survey

Fertility of married women by age group, Koto Kayo, West Sumatra: 2000 household survey

41Ethnography of these groups is ongoing and interviewing is not yet complete, especially as the randomised survey of the communities in the current research round will not take place until late 2019. Current findings can therefore only be suggestive. Four case studies, comparing members of upper and lower socio-economic levels and their networks in two of the communities will at least help to indicate how variation is evolving. The case studies are particularly helpful in describing how some networks come to be composed of families belonging to different strata and lead to differing childbearing patterns. In Kidul, the cases contrast generational changes in networks that are typical of local family economies, relying in particular on shop keeping. In Koto Kayo, the cases contrast higher status networks able to draw on incomes from premium agricultural land together with the migration economy, with lower status networks having neither resources.