Measuring poverty within the household

Standard approaches to poverty measurement

Income poverty measurement can be thought of through the lens of a consumer choice problem, where a so-called “utility threshold” that separates the poor from the non-poor is determined first, and then the consumption necessary to reach this threshold is set as the poverty line. There are two challenges to this approach: the first involves translating welfare into something that can be reliably measured, and the second requires determining where to set the utility threshold and hence the poverty line.

Income or consumption are typically used as proxies for welfare: the higher the income or consumption, the higher the welfare. For poverty measurement in a developing country setting, consumption expenditure is often preferred, as income fluctuates frequently, particularly in agricultural areas, and can be difficult to measure properly. Consumption, however, has its own measurement issues, particularly when it comes to measuring prices, but is considered a better indicator than income for standards of living and well-being [2]. Nevertheless, the problem remains as to whether the way to translate consumption expenditures into welfare is the same for each person. Specifically, if basic needs vary with demographic characteristics, such as age, gender, occupation, or geographic location, then a one unit increase in consumption expenditure will yield different increases in welfare for different individuals. In other words, some individuals may require fewer resources than others to reach the utility threshold.

Determining the utility threshold (which is often referred to as the “referencing problem” [2]) is similarly challenging. Poverty is a subjective concept, depending highly on context and culture. Individuals within and across countries are unlikely to agree on what constitutes a poor household and poverty lines can vary considerably across countries. One possible solution to this issue could be to ask people about their subjective welfare and use this information to infer a subjective poverty line. For example, the often-used minimum income question (MIQ) asks: “What income level do you personally consider to be absolutely minimal? That is to say that with less you could not make ends meet.” While poverty lines set using subjective welfare-type questions may be appealing, they also have a number of limitations, a key one being that there are almost certainly latent factors influencing how people answer such questions that cannot be accounted for (such as personality or social norms).

An alternative, perhaps more objective, approach is to fix a basket of goods that an individual might need in order to meet some specified level of welfare (below which they are considered poor). Often, this basket will include food needed to support recommended dietary intakes and may be augmented with an allowance for non-food goods (essentially anchoring the utility threshold to biological needs). The value of this basket pins down the poverty line. One problem here lies with determining what should be included in the basket, and how large the non-food allowance should be.

Any poverty measure, either objective or subjective, still faces the same issue: whether or not it is consistent in the space of welfare. Household-level measures exacerbate this challenge: not only do households differ substantially in demographic composition and size, they may also not allocate household resources equally within the household. It may be possible that some household members have a substantially lower welfare relative to others. By using the standard per capita approach, which assigns every household member the same value of consumption, and comparing every individual to the same poverty line, it is unclear whether welfare-consistent comparisons can actually be carried out. Furthermore, household-level measures do not allow for variation in poverty status within the household: so, if one member is poor, then all other members must be poor. As shown later on, this may not hold true in many circumstances.

Equivalence scales

Different household members, along with different household types (e.g. with different demographic compositions or different sizes), are likely to require different levels of consumption. On the one hand, due to economies of scale in consumption, larger households may need fewer per capita resources than smaller households to attain the same level of welfare. For example, more people jointly consume goods that can be shared, such as energy goods. On the other hand, household members may differ in their needs. Children and the elderly, for example, may require less food than working-age adults to maintain a healthy diet, and individuals working in high-activity jobs may need more calories than those who do not. In households where food is a substantial portion of the budget or with higher than average dependency ratios, these differences are likely to have a non-trivial effect on poverty measurement.

Equivalence scales can be used to rescale total household consumption to address these issues. To account for both economies of scale and differences in needs in poverty measurement, it is possible to calculate “equivalent consumption” as household expenditure multiplied by a specific factor, the inverse/reciprocal of the equivalence scale. This equivalence scale depends on the sum of the number of adults in the household (nₐ ) plus the number of children (nₖ ) whereas the number of children is multiplied by a factor beta (as children's needs are set to be a fraction of adults’ needs the scaling factor beta is set between 0 and 1). To get to the equivalence scale this sum is raised to the power of a factor alpha that captures economies of scale in consumption (again, alpha must be greater than 0 but no greater than 1), that is, the equivalence scale equals (nₐ–βnk )ᵃˡᵖʰᵃ.

This whole calculation can be adapted to include different person types, such as the number of elderly members or working-age male and female adults, and yields the per capita household expenditure (with alpha and beta set to 1). For a four-person household with two adults and two children, with beta=0.5 and alpha=1 (i.e. no economies of scale), the equivalent consumption would be just one-third of household expenditure, hence significantly higher than per capita expenditure. The presence of scale economies (e.g. alpha=0.6) would result in even higher equivalent consumption (roughly one-half of household expenditure).

The challenge is in setting accurate values for alpha and beta, which requires knowledge of the needs of different individuals as well as the extent of consumption sharing within the household. An additional complication is that alpha and beta are usually related: for instance, larger households tend to have more children relative to adults, as children cannot live alone but adults can. In practice, alpha and beta are often chosen based on ad hoc assumptions. A common example of equivalence scales is the OECD scale (or Oxford scale), which equals 1 for the first adult household member, plus 0.7 for each additional adult, plus 0.5 for each child. So, a typical family of four, with two parents and two children, would be assigned an equivalence scale of 2.7. The so-called square-root economies of scale parameter, which sets alpha=0.5, is also common. An alternative three-parameter equivalence scale is used in the US and is given by (1+0.8+0.5(nₖ –1))˄0.7 for single-parent households, and by for (nₐ +0.5nₖ )˄0.7 households with multiple adults.

The question is: How much do changes in the values of alpha and beta affect poverty measurement? A recent study addresses this question in the context of poverty comparisons between male- and female-headed households (MHHs and FHHs, respectively) in sub-Saharan Africa [3]. The authors find that, based on per capita expenditure, FHHs have lower poverty rates than MHHs. However, even small adjustments for scale economies can change these comparisons: the value of α at which FHHs would have higher poverty rates than MHHs ranges from 0.95 in East Africa to 0.15 in West Africa. Such sensitivity reflects the fact that FHHs are substantially smaller than MHHs, on average. By contrast, variations in adult equivalence scales matter less in this setting, as dependency ratios are relatively similar in FHHs and MHHs.

Intra-household inequality in resource allocation and poverty measurement

Equivalence scales can be useful to capture differences in needs across individuals as long as these differences are stable within person types: for example, if all children need less than working-age adults and by the same amount. However, the usefulness of equivalence scales breaks down when differences in needs are diverse within person types, or if there is discrimination against certain individuals within the household. Importantly, equivalence scales assign the same value of consumption to each household member. As a result, they cannot account for circumstances in which some household members have welfare values below the threshold while others do not. In other words, they cannot account for intra-household inequality and for the fact that poor and non-poor individuals may reside in the same household.

How critical is intra-household inequality? As previously mentioned, assessing the scope of intra-household inequality in consumption is complicated by the fact that consumption data are typically collected at the household level. A handful of surveys have exceptionally collected data on individualized consumption, and intra-household inequality has been found to be far from negligible using these data. For example, examining Bangladesh, an upcoming study finds that adults have substantially higher levels of food and clothing consumption relative to children, and women have slightly lower consumption levels relative to men [4]. Focusing on food consumption in Bangladesh, another recent study finds that intra-household inequality accounts for almost 50% of total inequality in caloric intakes, which is only partially accounted for by differences in caloric requirements by age, gender, and type of occupation [5].

Given the scarcity of individualized consumption data, researchers have relied on differences in nutrition-based anthropometric indicators (i.e. whether an adult is underweight or a child is stunted or wasted) to measure the extent of intra-household inequality. For example, a 2019 study finds that around three-quarters of undernourished women and children in sub-Saharan Africa are not found in the poorest 20% of households [6]. In other words, undernourished individuals are more likely to be found in households that are not poor rather than poor. Furthermore, it is not the case that all household members share similar nutritional outcomes: the majority of undernourished women and children are found in households where the male head is not underweight.

The collective household model

Empirical applications

Given its ability to shed light on within-household consumption, several recent studies have applied the collective household model to measure individual-level poverty.

The authors of one such study, which focuses on measuring child poverty in Malawi, estimate resource shares separately for men, women, and children, and document extensive inequality within the household [8]. They find, for instance, that in a nuclear household with two children, the father, mother, and each child are allocated 42%, 22%, and 18% of the total household expenditure, respectively. The wide scope of intra-household inequality in Malawi results in per capita measures dramatically understating poverty rates for children. A study from 2021 extends this approach to a setting with only partially assignable goods to estimate intra-household allocations between foster and non-foster children in Malawi, finding little evidence of inequality between them [9]. Another recent study examines resource allocation in Indian households and shows that women's resource shares relative to men's decline steadily after the age of 45, when on average women get as low as 65% of men's resources [10]. As a result, at post-reproductive ages, poverty rates are significantly higher for women than men, which can help explain the disproportionately high mortality rates of older women in India.

A caveat to these works is that they do not estimate economies of scale in consumption. In other words, they provide estimates of individual-level “expenditure” rather than individual-level “consumption.” This distinction is not necessarily problematic, as households in both contexts primarily consume private goods such as food. Nonetheless, to accurately measure individual consumption, it would be necessary to account for both intra-household inequality and joint consumption.

This is what motivates recent work on the topic, which successfully identifies both dimensions of consumption [1]. The authors’ method relies solely on observing expenditures on a single private assignable good and, unlike much of the existing literature, does not require data on single-person households. As a result, their approach is ideal for applications in developing country settings, where individuals rarely live alone. Their empirical analysis focuses on Bangladesh and Mexico, with the goal of illustrating differences in consumption sharing across countries at different stages of economic development. In Bangladesh, they find a modest amount of consumption sharing, suggesting that ignoring economies of scale is not likely to lead to a large degree of error in individual consumption estimates. By contrast, they estimate significant economies of scale in consumption in Mexico, which, for example, increase individual consumption in extended families or nuclear families with multiple children by 10% relative to nuclear families with one child.

Comparing standard and model-based poverty rates

The above-mentioned study also compares poverty rates for Bangladesh and Mexico obtained using different consumption measures [1]. Specifically, the authors juxtapose poverty rates based on per capita expenditure, equivalent consumption, individual expenditure, and individual consumption. Figure 1 summarizes their findings. Each sub-figure features the estimates of individual consumption on the y-axis and alternative measures on the x-axis. Each dot in the scatter plot is an individual and the vertical and horizontal lines lines correspond to the US$1.90/day poverty line. For individuals falling in the upper right or lower left quadrants, the two poverty measures coincide. That is, individuals who are classified as poor using individual consumption are also classified as poor when using the alternative measure. On the other hand, individuals falling in the lower right quadrant would be considered poor using individual consumption, but not the alternative measure. As shown in the figure, there is a large degree of poverty misclassification, especially when using per capita expenditure and equivalent consumption. While the degree of poverty misclassification may vary substantially across contexts, these results show that accounting for intra-household inequality and economies of scale in consumption may matter greatly for poverty measurement. The collective household model can be used to compute poverty rates that encompass both dimensions.