MAKING GENDER VISIBLE: SOME METHODOLOGICAL ''TRAPS''

At the conceptual level, poverty is increasingly seen as a multi-dimensional phenomenon, which includes market-based consumption (or income), as well as the public provision of goods and services, access to common property resources and the intangible dimensions of a good life such as low levels of disease and crime, clean air, dignity and empowerment.⁵ The proponents of the conventional approach argue that the income/consumption measure is still the best single proxy for poverty since it can incorporate non-market goods and services (state-provided education and health) and a wide range of other utilities (clean air, democracy) and disutilities (noise, pollution), through "shadow prices", into a monetary equivalent that is easy to compare over time and across contexts. But their critics argue that common property resources and state-provided commodities have usually been ignored in practice, and the consumption of non-traded goods has also been under-estimated (Baulch, 1996). It is also questionable whether "shadow prices" can meaningfully translate the different kinds of values that are embodied in non-market goods and services into monetary equivalents that are comparable.

⁵ One should note that adding new dimensions to the definition of poverty will automatically increase its observed extent and depth.

The tension between income/consumption methods and alternative approaches also revolves around their different claims to knowledge and the different methodologies employed for assessing levels of poverty and well-being. The income/consumption method determines levels of well-being and ill-being in an a priori fashion, as physical needs deprivation through private consumption shortfalls measured on the household. It is not necessary to repeat here the wide-ranging debates about the poverty line and the head-count measure of poverty.⁶ It is useful, however, to note in passing that despite the claims of some of its proponents, the substantive content of the poverty budget⁷ tends to include an important relative and "cultural" element besides an absolute subsistence element; thus the poverty line in the United States is between 10 and 17 times higher than that in India (depending on the exchange rates used), although both pretend to be absolute in some sense (Scott, 1981). Even in cases where most of the household budget is spent on food, the minimum subsistence element cannot be adequately ascertained nor can it be priced in a consistent manner, since it cannot be assumed that the poor face the same prices as the non-poor (or that they consume the cheapest available kinds of food). There is thus an inevitable degree of arbitrariness about where the poverty line is drawn, and the proportion of population that is classified as poor. In practice, it is not uncommon for political considerations to dictate where the line is drawn (Scott, 1981).

⁶ The head-count measure of poverty is the proportion of the population/households that falls below a certain pre-determined poverty line.

⁷ "Poverty budget" refers to the minimum amount of income required by an individual or household in order to remain above the poverty line.

But a more fundamental drawback of the income/consumption method is the unreliability of the data on which it is based, especially in developing countries. In most African and Latin American countries, household budget surveys tend to be one-off (non-repetitive) exercises, which makes them unsuited as a device for monitoring poverty. This is ironic in view of the claims made by the proponents of the income/consumption approach that the superiority of their methods lays precisely in their ability to make comparisons of poverty over time. As Lustig (1993) has observed, even if a researcher is willing to work with surveys that have less than a complete definition of income, there are only five countries, out of close to 30 in Latin America, for which an analysis of poverty in the 1980s can be made. This gives some indication of the scale of the problem.

In contrast to the conventional approach, what have become known as participatory approaches to poverty determine both the constituents and the sources of well-being through an iterative process involving the PRA (participatory rural appraisal) facilitator and participants. These participatory approaches have a long pedigree, going back to the traditions of participatory action-research and the work of Paolo Freire and his followers, and have gained greater elaboration and international recognition through the work of Robert Chambers. In the late 1980s and early 1990s this "confluence of older streams of research together with new inventions evolved as a family of approaches and methods known as participatory rural appraisal (PRA)" (Chambers, 1998:xv). Participatory methods for information generation are often open-ended, visual (using mapping, diagramming and matrices) as well as verbal (discussing and debating). Their proponents claim that participatory methods of research and policy analysis enable the local people - including the poor, illiterate, women and the marginalized - themselves to appraise, analyse, plan and act.

There has been a long-running debate about the claims made by the proponents of PRA that their methods in effect "hand over the stick" to the local people themselves. Critics point out that PRAs involve highly formal and "public" social events, which construct "local knowledge" in ways that are strongly influenced by existing social relationships marked by significant class and gender hierarchies (Mosse, 1994). Questions have also been raised over any expectations that poor village people willingly divulge truthful and accurate information in one-off "participatory" research exercises, as well as the possibility of developing a model of participation in one place and then applying it elsewhere much more quickly (Jackson, 1997a). There are also some fundamental questions as to whether subjective perceptions of ill- and well-being can, on their own, reveal all that there is to know about individual welfare, given the ideological (including gender) parameters within which all perceptions take shape (Sen, 1985). Sen is particularly concerned about what he calls "physical condition neglect" to which women may be even more prone than men, and thus insists on including physiological conditions in accounts of well-being.⁸

⁸ It should be noted that Sen's concern with physical condition neglect is formulated in the context of his general critique of utilitarian approaches to well-being rather than to participatory approaches as such. But given that both utilitarian and participatory approaches rely on subjective perceptions of well-being, his critique is equally applicable.

The "capability framework", pioneered by Amartya Sen (1985), is based on an alternative notion of well-being directly concerned with a person's quality of life, which embraces both basic "functionings", such as longevity and adequate nutrition, as well as more complex ones such as freedom and autonomy. In practice the proponents of this approach have tended to concentrate on the first set of functionings, which are measurable on the individual through a range of social indicators.⁹ They have therefore used "objective" criteria for assessing well-being. But rather than using income or consumption to determine how individuals fare they look at the actual outcome for individuals in terms of mortality, malnutrition, educational levels and other "basic needs" indicators. Far fewer assumptions thus need to be made in the interpretation of results than do household-based measures of income or consumption. We will not dwell on some of the shortcomings of the capability framework here, since they are discussed at some length later, but note in passing that data problems are as debilitating in the area of social indicators, on which the capability framework relies, as they are in the case of economic data.

⁹ See Carter, 1996 for a critique of Sen's work on freedom.

It is frequently argued that from a gender perspective, these alternative concepts of poverty are useful in that they allow a better grasp of the multi-dimensional aspects of gender subordination, than a focus purely on household income levels. It is worth remembering that these conventional household surveys (of income, or consumption) have been an Achilles' heel for gender analysis; even with complex econometric manoeuvrings it has not been possible to make these data sets answer questions about intra-household distribution that they were not originally designed to address.¹⁰ However, despite the methodological difficulties involved in reading individual levels of poverty from household-based measures of income and consumption, it is arguable that some understanding of levels of material deprivation - even if it is household-based - may be useful for the gender analysis of social indicators, or of autonomy and subordination.¹¹ In fact some of the more interesting questions and hypotheses that have been raised about anti-female bias (in terms of nutrition, survivorship and decision making), relate precisely to the ways in which poverty, opulence and social class impact on discriminatory forces and women's personal autonomy and independence in diverse contexts.

¹⁰ See Hart, 1997 for a critique of the claims made by those using collective models of the household to "recover" intra-household sharing rules from survey data.

¹¹ As such, therefore, we would agree with those who have queried UNDP's new (1997) composite index on poverty, the Human Poverty Index (HPI), for not including some measure of income/consumption in the index, which they argue is contrary to common sense and everyday experience even in very poor countries (Lipton, 1997; Krishnaji, 1997).

Both donor agencies and governments, however, continue to rely, almost exclusively, on household surveys to generate easily quantifiable measures of poverty (income, expenditure, consumption), while case study material and other qualitative evidence are sidelined as "anecdotal". A telling illustration of this methodological bias appears from Lockwood and Whitehead's (1998) analysis of the World Bank's Poverty Assessments (PAs).¹² But the relevance of their argument extends to the technocratic circles of other donor agencies, as well as recipient states engaged in the measurement and analysis of poverty (Marcoux, 1998; Robb, 1998).

¹² These Poverty Assessments are country studies carried out by the World Bank in order to examine poverty. See Lockwood and Whitehead, 1998 for further details.

Most World Bank Poverty Assessments begin by asserting the multi-dimensionality of poverty, but ultimately, it seems, all give priority to an income and/or consumption definition, a "money-metric" poverty line and a quantitative estimate of the percentage of people in poverty. Lockwood and Whitehead (1998) see this as a fundamental methodological choice, since it locks the Poverty Assessments into reliance on expenditure data from household surveys, which in addition to being narrow, are also very often unreliable and non-comparable. There is little consistency, for example, between different PAs in how the poverty line is established, which means that both cross-national and time-series comparisons are difficult to make. Such methodological inconsistency effectively defeats the purpose of collecting quantitative data, since one of the rationales for using quantitative data is precisely that they are comparable (Lockwood and Whitehead, 1998; Hanmer et al., 1997). At the same time, many of the potential insights about the nature of impoverishment, or poverty processes, which emerge from the participatory research are either marginalized or dropped from the analysis. The reliance on household expenditure data also means that one of the easier ways to make gender visible is by dividing the households into male-headed and female-headed ones, given that the characteristics of household heads (their gender, age, etc.) are invariably collected through these surveys and form a ready basis for sorting the data.

The tendency to equate female headship with poverty has, however, been queried on both empirical and methodological grounds (Chant, 1997; Lloyd and Gage-Brandon, 1993). In contexts where male gender identity is maintained by siphoning household resources away for personal consumption (drinking, going out), female heads of households may be in a better position to protect priority consumption areas (i.e., food and health) than women in male-headed households, as the evidence from urban Mexico seems to indicate (Gonzalez de la Rocha, 1998). In other contexts, households headed, de facto, by women seem to be less poor than male-headed households because of the remittances they receive from migrant males (Kennedy and Peters, 1992).

The trajectories leading to female headship are clearly divergent. But by lumping together these distinct categories of households, generated through different social processes (e.g. migration, widowhood, divorce), and constructing a simple dualism between male-headed and female-headed households, it becomes impossible to interpret the evidence in a meaningful way. At the same time, the identification of certain types of female-headed households - such as lone widows - as poorer begs the question of why some widows end up living alone while others do not. As Lockwood and Whitehead (1998) put it, the characteristics of the poor say very little about the reasons why they have become impoverished, and it is methodologically incorrect to treat these characteristics as independent variables. In this particular case, they argue, the chain of causation may run the other way: it may very well be that it is poor widows whose children leave the household (through labour migration, for example) and that when more economically secure women are widowed they do not end up living alone. In practice, causes and effects always interact and do so differently in diverse contexts of time and place, and it is unhelpful to think that the ambiguities of causality can be resolved by simple correlations and regressions of a few variables.¹³

¹³ This point was made convincingly by Solon Barraclough of UNRISD in a personal communication.

As was noted above, a somewhat different conceptualization of the gender/poverty nexus focuses on well-being outcomes, such as life expectancy or nutritional status, implicitly equating gender analysis with the gender disaggregation of social indicators. In recent years social indicators have gained increasing legitimacy in policy debates. The UNDP's annual Human Development Report, intellectually buttressed by the capability framework, has no doubt had a significant and positive influence at this juncture. This has given greater visibility to some social sectors (health, education), as well as to gender equity and political participation, even though the efforts to translate the renewed sensitivity to social concerns into sectoral and macro policies remain seriously constrained due to political impediments at both the global and national levels.

From a gender perspective, as many have argued, because these "beings" and "doings" (Sen, 1985) are measured directly on the individual, they are conducive to making gender inequalities visible. It has thus been possible to take on board feminist critiques of the unitary household model without having to resort to problematic assumptions about intra-household distribution. Not surprisingly, the framework has inspired a large body of feminist research on well-being outcomes, documenting significant and sometimes alarming incidence of sex bias (Sen, 1992). However, the gender disaggregation method, which is also partly driven by data availability (albeit of social indicators rather than household income), while useful for revealing gender bias in well-being outcomes, is limited in the kind of causal analysis of gender difference that it can generate. Moreover, there is also a problematic tendency, when assessing indicator values, to assume cross-cultural and ahistorical female disadvantage. The existing empirical evidence, however, cautions against such sweeping generalizations.

Sudha and Rajan's (1998) analysis of Indian sex ratios at birth for the period 1981-1991 provides further evidence of "gender cleansing" (Harriss-White, 1998) on the sub-continent. In some parts of India parents seem to be adding pre-natal sex selection techniques to traditional post-natal ones to create a "double jeopardy" for their daughters, thereby exacerbating the "missing females" predicament.¹⁴ Their findings, based on an analysis of Indian census data, are corroborated by newspaper reports, case studies and literature from NGOs and women's groups campaigning against prenatal sex determination and female foeticide, to produce a fairly reliable account of the increasing spread and "acceptability" of these techniques. But these patterns are far from uniform. The excess masculinity of sex ratios at birth is concentrated especially in the north and north-west of the sub-continent, and in the urban areas of some central states. The southern states appear to have on the whole normal sex ratios at birth - a regional contrast with some historical precedence that has been the subject of intense debate and theorizing (Bardhan, 1974; Dyson and Moore, 1983; Miller, 1981). But while sex ratios at birth show the most masculinity in the north and north-west, and this increasing masculinity is not a nationwide phenomenon, sex ratios of child mortality definitely show increasing female disadvantage over all of India. Only one or two very small areas of the country show less female disadvantage in 1991 than they did in 1981, while areas in the south that had normal mortality ratios in 1981 show female disadvantage in 1991.

¹⁴ There is similar evidence of increasing masculinity of sex ratios at birth for parts of East Asia (China and South Korea, in particular), where pre-natal sex determination techniques and selective abortion of female foetuses are also implicated (see references in Sudha and Rajan, 1998).

Nor does female disadvantage in early age survivorship imply a consistent pattern of anti-female bias in food intake and nutritional status as is often assumed. Even in north and north-western India, where the evidence of discrimination against young girls in terms of survivorship is most compelling, "rogue" findings from nutritional surveys not infrequently show that adult women fare better than their male counterparts (Harriss, 1990; Lipton and Payne, 1994). Jackson (1996) suggests that these assumptions of universal female disadvantage may be "partly a consequence of too ready an acceptance . . . by researchers of articulated nutritional norms as reflecting actual food access without any interrogation of how women's agency subverts norms, e.g. by snack-food consumption, by eating during food preparation and by consumption of 'leftovers'" (p. 496). But given the fact that nutritionists are normally required to sort out and account for food intake through snacking, these "errors of omission" may have more to do with the practical difficulties of capturing "individual consumption" and "snacks", because they tend to be forgotten by respondents when the "recall method" is used. Confirming the problematic nature of these generalized assumptions of female disadvantage, Saith and Harriss-White's (1998) comprehensive review of the micro-level literature for South Asia reaches the conclusion that the evidence on gender differentials in nutritional status is inconclusive, showing no consistent indication of gender bias.

An interesting instance of the inconsistency is captured in Kynch's (1998) village-level study of nutritional status in the north-western Indian state of Uttar Pradesh. Here age and gender interact in complex ways over the life cycles of men and women. Starting with female nutritional disadvantage during childhood, the pattern is reversed among child-bearing couples where adult men seem to be the ones at greater risk of illness or low working ability, because of thinness. Male disadvantage in this particular context is explained by the compulsions on husbands to "provision" their families, which among the agricultural households means undertaking effort-intensive farm work (ploughing and digging) with consequent threats to their physical well-being. A related point is raised by Kasente (1998). Somewhat contrary to the stereotypical view of gender divisions in Ugandan agriculture, which upholds a caricature of women as over-worked victims and men as lazy patriarchs, her village-level evidence shows substantial labour contributions by male household members to smallholder farming, and a gender division of labour that is visible but "much less sharply than has been assumed" (p. 8).

Kynch's (1998) analysis, in particular, echos the point made vehemently by Jackson and Palmer-Jones (1998) that while time use studies have been of great value in making women's work visible, it is important to resist the conflation of all forms of activity other than sleep as "work", and to recognize how male gender roles in divisions of labour can also involve vulnerabilities for specific groups of men. They argue that an explicit concern with the content and character of work - its physical arduousness in particular - may better illuminate the gendered connections between work and well-being for women and men. In fact, in poor households effort-intensive work (high energy expenditure) can, more often than inequitable food intake, be the cause of differences in adult anthropometry; in these cases after adjusting for energy expenditures, energy intakes are often equitable.

These authors also raise some thorny methodological questions about gender-disaggregation, and the difficulties of making meaningful comparisons between male and female well-being when men's and women's bodies are different in form and function. One problematic area, highlighted by Saith and Harriss-White (1998), is that of morbidity, where a significant proportion of the conditions that cause morbidity are sex-specific and defy simple male/female comparisons - reproductive health problems being the most glaring example. Other diseases may be sex-specific due to genetic predispositions, such as rheumatoid arthritis in females, and more controversially, tuberculosis in some male populations such as in India (Dyson, 1984). Another area - explored by Jackson and Palmer-Jones (1998) - is that of physical work capacity. They argue that women and men are differently capable of high physical work intensities, which have both direct and indirect connections with their well-being, and which cannot be captured through the metric of time. However, even in other areas such as malnutrition, meaningful comparisons can only be made once "norms" and "cut-off points" (both deeply contested) have been adjusted for gender difference - a process that is fraught with difficulty (Saith and Harriss-White, 1998; Kynch, 1998).

The point of raising these issues here is to highlight the methodological controversies (and arbitrariness) involved in making well-being comparisons between men and women - issues that tend to be overlooked when global comparisons are made. In the broader scheme of things these technical problems are merely the tip of the iceberg. As those familiar with this field have repeatedly argued, social indicators are probably as unreliable and non-comparable as are income data (Solon Barraclough, personal communication). Very few developing countries, for example, have comprehensive and reliable vital registration systems from which demographic data can be obtained - India being perhaps an exception. And even for those with complete vital registration systems the estimates of mortality and life expectancy produced by international agencies may not be accurate because of the overuse of model life tables (Murray, 1991). UNICEF admits that many of the statistics used for estimating under-five mortality are based on mathematical models rather than recent measurements (1993:8). The same applies to sex differences in mortality, which are very often assumed from a particular family of model life tables rather than estimated directly from real patterns in the data (Murray, 1991). Even for an apparently straightforward indicator like literacy there are few up-to-date estimates; in 1994, for example, for 19 of the 145 (including developed) countries, there were no data on adult literacy since 1970, and for 41 more the data related to a year in the decade 1970-1979 (Srinivasan, 1994). If these data realities are taken seriously, then a number of implications will follow.

Critics suggest that these weaknesses and qualifications need to be reflected in data presentations in order to highlight the true extent of the data gaps (Murray, 1991). It also means that "practical" issues such as ease of measurement and costs of measurement (being affordable) - discussed in Saith and Harriss-White's (1998) paper - are central to the whole project of developing suitable indicators, especially for developing countries. They argue that a disaggregated under-10 female-male ratio (0-4 years and 5-9 years) appears to be a suitable gender-sensitive indicator of well-being, especially for South Asia. But the difficulties with data collection and interpretation reduce the reliability of indicators of nutrition and morbidity.

Data users also need to be realistic about the extent to which findings and policy recommendations that rely on correlations and regressions, between education and income, for example, without enough complementary contextual information can be trusted. A pertinent example is the human capital argument for "investing in women" - a related, and to some extent overlapping, strand of thinking on gender and poverty that has been embraced and developed most enthusiastically by gender advocates in the World Bank (World Bank, 1994). It has also made its imprint on the Bank's Poverty Assessments, as well as on the so-called "New Poverty Agenda" set out in the 1990 World Development Report (World Bank, 1990). The arguments, which are by now fairly well known, suggest that "investing in women is critical for poverty reduction, it speeds economic development by raising productivity and promoting the more efficient use of resources; it produces significant social returns, improving child survival and reducing fertility; and it has considerable intergenerational payoffs" (World Bank, 1994:22). Female education, in particular, is considered to be more efficient than male education in increasing child and household welfare, as well as having broader social returns.

According to Lockwood and Whitehead (1998) the World Bank's analyses of gender gaps in primary education in sub-Saharan Africa are either purely economistic (about expected future returns) or they are anodyne (both "parental reluctance" and "gender bias" cited in one of the PAs on Uganda is so self-evident as to be virtually meaningless). These analyses, they argue, provide an extremely partial and misleading picture of the causal dynamics behind low levels of female education in sub-Saharan Africa. In this connection they also make the critical methodological observation that the relationship between female education and household income may be spuriously being read as causal, when both income and education may be affected by underlying patterns of wealth organized through families, i.e., social class. The alternative hypothesis, which sees educational outcomes as the result of patterns of wealth and poverty rather than the other way around, seems to be more consistent with the qualitative evidence on causes of poverty that is documented in the PAs. It is interesting that a similar conclusion emerges from Saith and Harriss-White's (1998) review paper based on their reading of the limited micro-level analyses of female education for South Asia and sub-Saharan Africa. They suggest that "family poverty in rural and urban areas is probably the most important reason holding girls back from school or withdrawing them earlier" (p. 39). They also make the interesting observation about sub-Saharan Africa that the "same factors that are considered responsible for gender equity in nutrition and health care may be responsible for anti-female bias in enrolment in school, i.e., the higher economic worth of girls and women" (p. 39).

The main message here, again, is that it is impossible to interpret the interlinkages between female education and poverty (income levels) through correlations and regressions alone. But this kind of over-reliance on simple econometric techniques is not limited to international policy-making institutions, government statistical bureaux, or even economists using conventional household surveys (of income and consumption). A similar weakness marks some of the emerging micro-level feminist research, which uses interview techniques to capture different aspects of female autonomy such as intra-household decision making, mobility in the public sphere and even domestic violence. As Kabeer (1998) usefully illustrates, here too the results can be uncontextualized, single-stranded and difficult to interpret, with a heavy reliance on simple correlations and regressions using a few variables. It is arguable that the methodological problems and the difficulties of interpretation are even more daunting in this kind of research. Indicators not only compress a great deal of information into a single statistic, but also embody assumptions about what this information means. And in the case of complex and culturally embedded notions like autonomy, it is impossible to have any faith in whether or not the indicator means what it is intended to mean without contextual evidence to support the assumptions that are being made.

Several contributors to this debate highlight the need for cross-checking and "triangulation" if the estimates and analyses that are arrived at are to inspire any confidence - this is not a new insight (Scott, 1981), but bears reaffirming in view of its significance to poverty analyses. This is presumably the reason why participatory research techniques have been included in the World Bank's Poverty Assessments; their proponents argue that "multi-stranded approaches are more robust than single-stranded approaches such as a free standing household survey" (Holland et al., 1998, cited in Lockwood and Whitehead, 1998:17). But this raises many difficult and unresolved questions about what the relations between different methodologies should be, and how conflicting results should be handled or reconciled (Baulch, 1996).

These questions are taken up by both Kabeer (1998), and Lockwood and Whitehead (1998), whose contributions provide some insights into this difficult methodological area, as well as some useful suggestions for future research. On the basis of a very detailed and critical analysis of existing micro-level research, Kabeer offers a framework for assessing women's empowerment, which entails checking of different kinds of evidence across three different dimensions of power: "resources" (defined broadly to include not only access, but also future claims, to both material and human and social resources); "agency" (including processes of decision making, as well as other - less measurable - manifestations of agency such as negotiation, deception and manipulation); and "achievements" (well-being outcomes such as longevity, education, etc.). These three dimensions, she argues, are indivisible in determining the meaning of an indicator and hence its validity as a measure of empowerment. Lockwood and Whitehead argue that qualitative data and existing case studies need to feed into definitions and analyses of poverty in particular country settings and not be used simply to illustrate points derived from survey data, which basically turns current practice on its head. There is little evidence, however, that the Poverty Assessments are actually tackling these questions in a sustained manner; instead the non-quantitative, non-survey type of information that the "participatory" exercises are able to produce appears to "sit very uneasily within the Poverty Assessments" (Lockwood and Whitehead, 1998:17).

A related point worth bearing in mind is that there is nothing intrinsically dynamic about non-survey techniques (as opposed to the conventional, large-scale household surveys). They can be used to produce quite static accounts of poverty. One example is the tendency to use the results of the participatory Poverty Assessments to construct what Lockwood and Whitehead (1998) refer to as "a composite picture of the poor", which basically suggests that the poor are those without clothes, money or food, but says very little about why they suffer from deprivation. The point made about participatory methods, namely that they can be as gender-blind or gender-aware as their practitioner (Kabeer, 1996), also applies to the static/dynamic distinction. The potential is there for these non-survey techniques to produce a dynamic account of poverty, but whether this potential is realized in practice largely depends on how they are used.

The need for "contextual" analyses, which is reiterated by these authors, hints at some of the inherent and fundamental limitations of indicators, and highlights the need to supplement data from indicators with other kinds of information. This point is succinctly captured in the Report on the World Social Situation, which notes that "social as well as economic indicators can provide no more than partial and incomplete glimpses of the complex and sometimes contradictory realities that determine and shape the quality of life of both individuals and societies" (United Nations, 1993:117). These "glimpses" say very little about how - through what social and institutional mechanisms - different social groups come to enjoy, or alternatively be deprived of, these valuable capabilities. A similar point is reiterated by Saith and Harriss-White when they argue that the comparative use of indicators of well-being, while necessary for revealing bias, "discloses nothing about their social meanings or about the social processes giving rise to gender differentials" (1998:54). For example, while the main reason for the relatively poorer educational attainments of males in Southern African countries (Botswana, Lesotho, Namibia and South Africa) is thought to be early school-leaving and migration to search for work in mines and commercial agriculture, the same cannot be said for the Latin American and Caribbean countries or countries in other regions (such as Sri Lanka, the Philippines, Qatar and the United Arab Emirates), which are also characterized by the same phenomenon of female advantage in education but for very different reasons.

The problems of interpretation, as we have already noted, become even more complex when we move beyond the very basic achievements, such as education or longevity, and onto the thorny area of autonomy and empowerment - astutely explored by Kabeer (1998). Issues of empowerment and autonomy have entered poverty debates through a number of different channels. As was noted above, both the capability framework and other post-opulence understandings of poverty have attempted to broaden the notion of well-being beyond its basic physiological aspects, in order to include issues such as power, self respect and dignity to which intrinsic value is attached. At the same time certain strands of policy discourse on poverty have subverted the notion of female empowerment, seeing it as an effective means of reducing poverty or meeting other policy goals. Its value here tends to be more instrumental and the aim has been to establish the nature of the association between the "degrees of autonomy" permitted to women in different contexts and certain demographic, economic or social outcomes (e.g. fertility reduction, child welfare, etc.), hence the search for easily quantifiable indicators that can be regressed or correlated against these other variables.

Kabeer's (1998) understanding begins from the former position. She sees empowerment as being inescapably bound up with disempowerment and as being about the process by which those who have been denied the ability to make choices acquire such an ability. However, in order to make the notion of choice relevant to the analysis of power, she qualifies it in a number of ways - highlighting in particular the importance of including the context, content and consequences of choice within the analytical framework. She argues that it is only through grounded analyses that issues of power and disempowerment can be meaningfully assessed, because it is only at that level that these contextual dimensions of choice can be understood and interpreted. This is clearly an area where the search for indicators is likely to be futile. Statistical perspectives on intra-household decision making, she suggests, should be seen as what they are: "simple windows on complex realities" (p. 28). They may provide "a brief glimpse of the processes of decision making" (p. 28), but they say very little about the subtle negotiations that go on between men and women in their private lives - which are difficult to capture even through grounded analyses.

To return to our introductory remarks, if by the gender analysis of poverty we mean the intra-household distribution of resources, and whether women suffer disproportionately from poverty and are thus "over-represented" among the poor, then social and economic indicators have been able to provide some, albeit partial and incomplete, glimpses. But if the gender analysis of poverty is to unravel how the processes leading to impoverishment, and the escape routes out of destitution, work themselves out, and maybe differently for women and men, then social and economic indicators clearly need to be used in conjunction with other kinds of analyses.