Skip to content

Individual Poverty and the Impact of Gender-specific Incomes Olivier Bargain | March 6, 2023

From household to individual poverty

Tax-benefit incidence or distributive analyses are usually performed at the household level or using some measure of per-capita consumption or equivalized (i.e., per-adult-equivalent) consumption. Yet, inequality, poverty and welfare should ultimately be assessed on the basis of individual, not household, consumption. All the more so as there is increasing evidence of a large number of disparities in access to resources within families (World Bank, 2018). Using anthropometric measures with data for the Philippines, for instance, Haddad and Kanbur (1990) showed that standard measures of inequality in calorie adequacy would underestimate actual levels by 30 to 40 percent if intrahousehold inequality is ignored.

It is presumed that one of the main factors behind intrahousehold inequality is an unequal distribution of decision-making power associated with lack of access to resources (income, for example) by certain members. For that reason, many cash transfers programs in low- and middle- income countries are given to specific family members, often women, to try to operate some intrahousehold redistribution in favor of women and children (see Handa et al., 2009). There are also examples of gender-targeted policies on the tax side (for instance a higher tax threshold for women in India, cf. Grown and Valodia, 2010).

Critically, there is limited evidence on whether these policies achieve their objective. For instance, a women-targeted transfer may be shared among household members according to the “usual” resource allocation rule of the household; alternatively, it may disproportionally benefit women and possibly children, but no one knows by how much. To answer this question, one needs to measure individual consumption before and after an intervention that redistributes from men to women without changing total household income. We rarely face this type of policy reform, i.e., one that operates a pure redistribution effect without changing household total resources. A famous exception is Lundberg et al. (1997), which shows how a change of recipient identity of child benefits in the UK, from fathers to mothers, shifted household consumption in a way that benefited women and children. In the absence of this type of natural experiment to assess the role of income control, an approximation would consist in measuring intra-household consumption sharing and estimating how it correlates with the level of income commanded by the wife versus the husband. It is then possible to proxy the role of benefits or taxes (via their effect on net earnings) on individual consumption. An illustration of this type of framework is suggested by Bargain (2023) and motivates this blog.

There is a major difficulty in this endeavor: expenditure data are typically collected at the household level. Indeed, measuring individual-level consumption data on a large scale is challenging and costly.1 Expenditure data are not collected frequently enough to measure the intrahousehold effect of the implementation of (or the changes in) redistributive policies such as gendered-targeted benefits or taxes. Even if we are availed of frequent individualized consumption data, it may still not be enough since consumption is not entirely private, i.e., some goods are ‘public’ (the dwelling) or can be seen as ‘consumed jointly’ (my car is a private good, but I may ride it with my wife half of the time so that there are economies of scale). One would ultimately like to know the individual valuation of the goods consumed jointly and hence a measure of scale economies. So far, however, there is no common framework that allows enriching socio-fiscal analyses with such a comprehensive individual perspective. In what follows, I suggest an illustration of what can be done drawing from the recent literature on ‘collective models’, as shown in Bargain (2023).

Conceptual framework

Empirical methods aimed at approximating intra-household allocation start with a conceptual framework known as the ‘collective household’ model, in which a household consists of different types of individuals (e.g., children, men, women) who have possibly different preferences over goods, and may differ in their decision-making power within the household. An implicit bargaining rule leads to an allocation of resources between family members.2 ”Resource shares”, i.e., the share of household consumption expenditure allocated to each individual, provide an explicit measure of the extent of consumption inequality within the household. With resource share estimates, individual poverty can also be assessed: an approximation of individual consumption can be calculated for each member of the household by multiplying her/his resource shares by total household expenditure, then confronting this individual resources with a poverty line that should ideally be adjusted by type of person.3

Along individual resource shares, we would ideally like to recover economies of scale, in order to augment individual resources by the implicit gains from joint consumption. As we will see, unfortunately, it is difficult to recover both elements with existing methods in the context of developing countries. Also note that, ideally, both resource shares and scale economies should be modelled as functions of socio-demographic characteristics (i.e., different needs and preferences) but with resource shares additionally influenced by ”bargaining factors”: that is, variables that affect the balance of power in the household and, consequently, the distribution of resources. Such factors may pertain to the household’s institutional or socio-demographic environment. For instance, divorce laws may favor women’s bargaining position or the sex ratio (the relative number of men compared to women) in the country, may change the possibilities in the marriage market and ultimately the relative empowerment of women vs. men (see Chiappori et al., 2002).

In our context, the bargaining factors of interest are the incomes received by the household (e.g., net-of-tax salaries, benefits, pensions, etc.) and especially the share of these incomes received by women vs. men. The focus on incomes stems from the fact that it is an important question per se and because income is a “bargaining factor” that is affected by fiscal policy (e.g., transfers targeted to women, differences in taxes, etc.). Thus, the analyst may be able to evaluate how much of gender inequality in consumption, or the poverty depth of men versus women versus children, depend on women’s control over income and, ultimately, on the way tax-benefit instruments alter how much of the income is controlled by women.

The general approach

A set of recent contributions allow recovering both economies of scale and resource sharing. We explain here how they are related to the approach we ultimately use but also why they cannot be employed directly. The most comprehensive method is the one suggested by Browning et al. (2013), for couples, and Bargain et al. (2022), for couples with children. The approach relies on many years of expenditure surveys pooled to estimate resource shares and scale economies upon the different goods consumed by a household.4 Yet, this is not very tractable for operational policy evaluations, which often have only one cross-sectional set of data. Simplifications have been suggested by Lewbel and Pendakur (2008), for couples, and Bargain and Donni (2012), for couples with children: they lead to the estimation of resource shares and of a summary measure of scale economies using cross-sectional data only.

In these four studies, the logic is the same and goes as follows. The econometric identification of resource sharing and scale economies is achieved thanks to expenditure on exclusive goods in the data, namely goods consumed by specific individuals in the household (for instance tobacco for adults and toys for children), or assignable goods, namely goods for which expenditures on each type of person is usually recorded in the data (for instance women’s clothing, men’s clothing and children’s clothing). The trick is that for these goods, there is a simple relationship between what we observe in the household, namely the share of total budget spent on these goods, and a person’s control over total household resources. For instance, the fraction of total budget spent on the wife’s clothes (say, 10% of the household budget) is simply the share of total resources accruing to the wife (say, 50% of household resources) times how much she spends on clothes out of her own resources (say, 20% of her budget). Household budget shares (10% on female clothes in our example) are observed in expenditure surveys while the wife’s own budget share (20% on clothes in our example) can be recovered by the observation of single women if we assume that women’s behavior on clothing consumption does not change when getting married and having children. This ”preference stability” assumption may be strong, but it allows comparing individuals in different demographic configurations and estimating resource sharing within multi-person households. Indeed, in our example, once we know the household budget share (10%) and the woman’s budget share (20%) on clothing, we can infer that her resource share, i.e., her share of total household resources, is 10%/20% =50%. All it takes is expenditure data on clothing for both single individuals and individuals living in families. We now proceed to discuss the validity of this assumption hereafter and the choice of clothing as the most common assignable good found in standard expenditure surveys.

Current framework

A difficulty with this approach is that in developing countries it is not common to find people living alone.5 Hence, Dunbar et al. (2013) suggest a slightly different method of identification, which does not require data on single individuals but still relies on expenditure data for exclusive/assignable goods. Their approach allows the identification of resource shares but unfortunately not that of economies of scale in consumption. As in previous contributions, the empirical approach hinges on the estimation of household budget shares (observed in household expenditure surveys) on assignable goods (typically clothing). Again, household budget shares for a person’s type (say adult women) are modelled as functions of individual resource shares (the share of resources accruing to women) and the person’s own budget share on that good.

The trick is the use of specific functional forms that simplify identification and allow to recover resource shares for each person type (women, men, children) with household expenditure data only, i.e., without data on single individuals. However, this identification strategy requires one to make assumptions that put restrictions on the way individual budgets for assignable goods may vary with individual resources. For a one percent increase in a person’s individual resources, the increase in her or his share of personal budget allocated to clothing is, say, X percent. Identification assumptions put restrictions on X. Under one assumption, known as ‘similarity across types’ (SAT), this X is different for women, men, and children but for each of these persons, it does not change when the demographic composition of the household varies (for instance when the number of children varies). An alternative assumption, known as ‘similarity across people’ (SAP), allows X to vary with the household type but for each type, i.e., for each household composition, X is restricted to be the same for women, men, and children.

It turns out that the SAT assumption leads to weak empirical identification and serious estimation difficulties (see Tommasi and Wolf, 2018, and Bargain et al., 2021). SAP is more tractable and is used in most of the subsequent literature and, in particular, in the application by Bargain (2023) that we focus on here. Yet, SAP can be seen as a strong restriction on individual preferences, namely that a marginal change in resources allocated to either men, women or children would lead to the same incremental change in their respective consumption (out of their personal budget) of the exclusive goods, i.e., of male, female, and children clothes. Note, however, that some empirical results tend to support SAP. Bargain et al. (2021) use direct observations of resource shares and tend to reject SAT but not SAP. Other tests of SAP hinge on indirect methods, i.e., start from alternative identification approaches that do not require SAP and test it as a restriction (see Dunbar et al. 2021 and Brown et al., 2021).

In practical terms, almost all the approaches cited so far require standard expenditure surveys that contain assignable goods (e.g., female clothing, male clothing, children clothing). Moreover, the econometric application consists of the simultaneous estimation of household budget shares for these three types of goods and for the different household types. As made explicit above, household budget shares are specified as functions of individual budget shares and individual resource shares. The latter, the shares of resources accruing to women, men, and children, can be modelled using a logistic function. The complete model is therefore a nonlinear system of household budget shares that can be estimated, for instance, using Seemingly Unrelated Regressions (SUR) methods. Alternative estimation approaches have been suggested but raise several questions and are not very useful in the present context.6

An application to Argentina and South Africa

The approach adopted in Bargain (2023) to study the role of women’s control over different income sources – and how socio-fiscal policies may influence intra-household resource sharing by this channel – follows Dunbar et al. (2013). Specifically, it makes use of a version of the model extended to all household types (Dunbar et al. focused only on nuclear households), in a similar vein as Calvi (2021), Bargain et al. (2021), and Bose-Duker et al. (2020). It applies it to Argentina and South Africa, two countries characterized by the presence of gender-targeted benefits. In particular, this study illustrates how women-controlled incomes affect women’s and children’s resource shares and, subsequently, their individual poverty status.

Empirical results are the following. First, the nature and recipient of income sources seem to alter household decisions on resource allocation. Specifically, the amounts of net earnings commanded by women, and sometimes the share of benefits or pensions they control, are positively correlated with their resource shares in both countries. We provide counterfactual simulations to quantify how women’s financial power and its sources modify women’s and children’s actual resources at different points of the distribution. While the role of women’s net earnings is substantial in the upper part, control over benefits is more limited but non-negligible, especially at low levels of household expenditure.

We then quantify how women’s actual control over resources limits the extent of total country- wide interpersonal consumption inequality by reducing intrahousehold inequality. For instance, in Argentina and South Africa), 14% and 23%, respectively, of total interpersonal inequality is due to within-household inequality but it would increase up to 16% and 31% if the incomes received by women were in fact received by men. We also derive implications in terms of individual poverty, i.e., when assessing who is poor in the household according to a person’s own access to resources. In both countries, women in general are worse off than men, and they would be significantly poorer without effective control over net labor income and state benefits. This framework can be used to approximate the impact of tax-benefit policies on the distribution of resources both between and within households.


Further work is necessary to consolidate the general approach described above – i.e., retrieving resource shares on the basis of assignable goods and preference stability assumptions – and to make it fully operational for individualized tax-benefit incidence and individual poverty analyses. We list here the main limitations of the approach and paths for future improvements.

First, relying on Dunbar et al. (2013) implies that the underlying welfare indicator used is not truly complete. Further work should incorporate economies of scale, which seem particularly relevant when one aims at adequately measuring individual poverty and inequality. As mentioned, identification of scale economies for adults is suggested in Browning et al. (2013) and modelled in a more general way as individual-specific functions in Bargain et al. (2022). Yet both contributions require data for single-person households. Economies of scale for adults and children are estimated in Calvi et al. (2023) but their approach requires imposing both SAT and SAP assumptions.

Second, the general approach described above is based on exclusive goods (i.e., goods consumed exclusively by some type of persons in the household, for instance toys for children) or, similarly, assignable goods (i.e., expenditures assigned to male, female, and children for instance clothing). Exclusive goods are rare and the literature often opts for clothing expenditure, since male, female and children clothing expenditures are frequently reported in standard expenditure surveys.7 Clothing may also be the least subject to the pitfalls attached to the Rothbarth approach (see Deaton, 1997).8 It is not necessarily subject to large consumption externalities (see the extensive checks in Dunbar et al., 2013) and in recent studies, the reported level of self-production of clothing is very limited (see Bargain et al., 2021). However, clothing may have several drawbacks, starting with the fact that it is less frequently purchased than other items such as food (see the discussion in Brown et al., 2021), and hence leads to less precise estimates. Clothing expenditure may also be associated with some other decisions that affect income sources and eventually total resources.9 New data could also inform us on the extent to which the good performance of clothing in Bargain et al. (2021, for Bangladesh) as a proxy for an exclusive good is context- dependent, i.e., how much it depends on the local culture and social environment. At the same time, recall that the collection of other individualized expenditure data, such as food, is rare, costly, and probably more challenging.

Third, given the last point about clothing (and the difficulty to find exclusive or assignable goods more generally), or the difficulties surrounding preference stability assumptions, it is possible to think of alternative ways to identify resource sharing. For instance, Dunbar et al. (2021) take a different path with an approach that does not require assumptions regarding preference similarity across households of different composition but instead exploits variation in so-called “distribution factors”, that is, variables that affect the allocation of resources within households but do not affect the tastes of individuals nor the budget constraint. To recover individual resource shares without making arbitrary restrictions on individual preferences is appealing. Moreover, the literature gives several examples of such distribution factors: for instance, the respective share of each member in total exogenous income (or in the wealth distribution at marriage) or indicators of the state of the marriage market and marriage/divorce legislation. However, while these factors may indeed influence bargaining positions in the household, their exogeneity is also questioned: most of them may also influence individual preferences. Also, even if distribution factors are truly exogenous, the effect they have on the intra-household distribution of resources may be very limited. In other words, these factors may pose the same difficulties as instruments in IV approaches, i.e., being at the same time relevant and exogenous (see Brown et al., 2021). Creative research is needed to design novel methods that may escape these difficulties.

Fourth and last, no one knows if the resource distribution modeled and estimated thanks to the various (but much related) approaches cited in this note – and constituting the state of the art – is correct, i.e., if it truly reflects the reality of resource allocation for the households present in the expenditure surveys used in these studies. Thus, validation exercises of how well resource shares are predicted are urgently needed. Preliminary attempts mobilize the rare datasets containing individualized expenditure and compare the “observed” resource shares, as computed from individualized consumption data, to predicted resource shares (Bargain et al., 2021). Results are encouraging and tend to show good performances of the collective approach on average, for instance in terms of predictions of the number of poor individuals. Yet it does not mean the estimated models allow predicting the “right” poor (i.e. those identified as such according to individualized consumption data).10 As a less direct alternative, some studies use assignable food expenditure and find correlations between the within-household inequality estimated by the model and inequality in terms of nutrition (Brown et al., 2021) or mortality (Calvi, 2020). Overall, however, more efforts are needed to build confidence in the collective approach for it to become truly operational when conducting individualized tax-benefit incidence analyses and individual welfare analyses.

A handful of surveys contain partly or fully individualized consumption information. Two of them, using different datasets for Bangladesh, find that intra-household inequality in consumption is far from negligible (Bargain et al. 2021) and explain the bulk of inequality in caloric intakes (Brown et al 2021).

2 This sharing rule is possibly rationalized by the assumption that allocation of goods across individuals within the household is Pareto efficient, i.e. no individual can be made better off without making another household member worse off (Chiappori, 1992).

For instance, it would recognize that children have lower needs than adults and that these needs may also vary with children’s age. This individual poverty lines are necessarily arbitrary and the approach suggested hereafter cannot identify both the actual needs of children and potential inequality within the households. In other words, if resource shares of children are low – so they are deemed poor while the parents may not be deemed poor – this might not be all due to unfair resource allocation by the parents but also because we overestimate children’s needs.

4 The reason for pooling many years is that the identification of scale economies for each good requires price variation, which is obtained in a reliable way only when there is enough time variation in prices.

5 In some countries, social norms may also attach a strong social stigma to women living alone.

6 This is for instance the case in Bose-Duker et al. (2020), a study that was developed as part of a project sponsored by the World Bank and published as Lechene, Pendakur & Wolf (2022). It hinges on the approach of Dunbar et al. (2013) but the estimation procedure is made slightly simpler by linearizing the nonlinear model described above. The objective is to be able to estimate the model using standard linear estimation techniques. The computing time is faster but the approach raises several issues. First, it is not clear what the approximation underlying this linearization entails (i.e. what the implications are of several terms becoming reduced-form variables rather than nonlinear function of observed characteristics). Also, more critically for our purpose, this simplified approach does not allow retrieving the estimates of specific determinants of the sharing rule and, most importantly, the coefficients on the different income sources controlled by women versus men.

7 For this reason, clothing is typically used for the Rothbarth approach (cf. Deaton, 1997) or for collective model estimations (e.g. in Browning et al., 1994, Bourguignon et al., 2009, and all the recent applications cited before).

8 These are (i) substitution effects (between own consumption and family size), (ii) the necessity for the relative price of the adult goods not to change across demographic types (the implicit price of food goods may change, for instance, if the returns to scale in food production are not constant), (iii) the requirement for adult goods not to be inelastic with respect to total expenditure (some of the food items are relatively inelastic).

9 For instance, some of the adults may increase or decrease their labor market activity – with consequences on their cloth expenditure if work requires specific clothes – in a way that is correlated with the number of children or other factors that can also change resource sharing.

10 Furthermore, comprehensive validations would need to be done for models including scale economies, which would require more (objective) information on the degree of joint consumption or, possibly, (subjective) information on the individual willingness to pay for goods that are ‘public’ in the household (such as the dwelling, heating, etc.).


Bargain, O. and Donni, O. (2012): “The measurement of child costs: a Rothbarth-type method consistent with scale economies and parents bargaining”. European Economic Review, 56(4), 792- 813

Bargain, O., G. Lacroix, and L. Tiberti (2021). “Intrahousehold resource allocation and individual poverty: Assessing collective model predictions against direct evidence on sharing.” Economic Journal 132:643, 865–905.

Bargain, O., O. Donni, and I. Hentati (2022). Resource Sharing in Households with Children : A Generalized Model and Empirical Evidence from the UK”, forthcoming in the Journal of the European Economic Association

Bargain, O. (2023). Income Sources, Intra-household Allocation and Individual Poverty, in revision for the Review of Income and Wealth

Bose-Duker, T.; I. Gaddis; T. Kilic; V. Lechene; and K. Pendakur (2021). “Diamonds in the rough?: repurposing multi-topic surveys to estimate individual-level consumption poverty.” World Bank Policy Research Working Paper No. 9661.

Brown, C., R. Calvi, and J. Penglase (2021). “Sharing the pie: An analysis of undernutrition and individual consumption in Bangladesh.” Journal of Public Economics 200:C

Browning, M., F. Bourguignon, P.A. Chiappori and V. Lechene (1994). Incomes and Outcomes: A Structural Model of Intra-Household Allocation, Journal of Political Economy, 102, 1067-1096

Calvi, R. (2020).Why are older women missing in india? The age profile of bargaining power and poverty, Journal of Political Economy 128(7), 2453-2501

Calvi, R., J. Penglase, D Tommasi, and A. Wolf (2023). The more the poorer? Resource sharing and scale economies in large families, Journal of Development Economics, 160, 102986

Chiappori, P.-A. (1992): “Collective Labor Supply and Welfare”, Journal of Political Economy, 100, 437-467

Chiappori, P.-A., B. Fortin and G. Lacroix (2002). “Marriage Market, Divorce Legislation, and Household Labor Supply”, Journal of Political Economy, 110/1, 37-72

Deaton, A. (1997), The Analysis of Household Surveys: A Microeconometric Approach to Development Policy, Baltimore: Johns Hopkins University Press

Dunbar, G., A. Lewbel and K. Pendakur (2013). Children’s Resources in Collective Households: Identification, Estimation and an Application to Child Poverty in Malawi, American Economic Review, 103, 438-471

Dunbar, G., A. Lewbel and K. Pendakur (2021). Identification of Random Resource Shares in Collective Households Without Preference Similarity Restrictions, Journal of Business & Economic Statistics, 39:2, 402-421

Grown, C. (2010). Taxation and Gender Equality: A Conceptual Framework, In Taxation and Gender Equity: A Comparative Analysis of Direct and Indirect Taxes in Developing and Developed Countries (Grown and Valodia, eds), New York: Routledge.

Haddad, L., and R. Kanbur (1990), ‘How serious is the neglect of intrahousehold inequality’, The Economic Journal, 100, 866-81.

Lechene, Pendakur and Wolf (2022). Ordinary Least Squares Estimation of the Intrahousehold Distribution of Expenditure, Journal of Political Economy 130, 681-731.

Lewbel, A. and K. Pendakur (2008): “Estimation of collective household models with Engel curves”, Journal of Econometrics, 147, 350-358.

Lundberg, S.,Pollak, R.A., Wales, T.J. (1997). Do husbands and wives pool their resources. Journal of Human Resources 32, 4631l480.

Tommasi, D. and A. Wolf (2018). Estimating Household Resource Shares: A Shrinkage Approach, Economics Letters, 163, 75-78.