Data for this research come from the Luxembourg Income Study, or LIS (www.lisproject.org
), a collection of national micro-level surveys on household income. All of the data sets that are part of LIS were collected within the respective countries, often by government agencies.2
When they are added to LIS, however, the data are “harmonized” in order to facilitate cross-national comparisons. The LIS is thus uniquely suited to study the household-level determinants of child poverty across nations.
This paper compares child poverty in the United States with that in Australia, Canada, and 12 Western and East European nations (Belgium, Estonia, Finland, France, Germany, the Netherlands, Norway, Poland, Russia, Slovenia, Sweden, and the United Kingdom). Data from most of these nations were collected in or about 2000.3
The definition of child poverty
used here is based on the concepts of “equivalized household income” and “relative poverty.” Equivalized household income
refers to income adjusted for “household characteristics deemed to affect economies of scale and economies of scope as reflected by differences in household size and composition” (Gottschalk and Smeeding 2000:638
). Following a common practice in cross-national poverty research, we use a measure in which
This simple correction to household income reflects the intuition that a given level of income does not go as far when divided among many people, but there are also economies of scale in sharing a home.
In this paper, we opt for a relative poverty measure, defining children as poor
if their equivalized household income is less than 50% of the median in their home countries. This is a very different way to conceptualize and measure poverty than an absolute standard, such as the U.S. government’s poverty line, which remains the same (after adjustment for inflation) as incomes in a society rise and fall. Although debate continues, relative measures are widely considered more appropriate for research on developed economies, where poverty is commonly conceived as a lack of the resources necessary to participate in what might be termed a mainstream lifestyle, rather than as a deficiency in the goods necessary for mere survival (Callan and Nolan 1990
; Sen 1992
). Relative measures are more dependable and more revealing for cross-national research because they avoid the indeterminacy inherent in evaluating whether an income level, or amount of material possessions, that categorize one as poor in a given nation might be adequate in another nation with a different standard of living (Brady 2003
). The most significant drawback of using a relative poverty measure is its essentially arbitrary cutoff point below which children are defined as poor (Callan and Nolan 1990
). However, the specific relative measure adopted here (50% of median equivalized income) holds the twofold advantage of being easily understood and widely used, particularly in literature based on LIS data.
Our typology of children’s living arrangements includes several variables that we hypothesize to affect children’s chances of being poor in at least some nations: whether two or more adults are present in the household, as opposed to one adult; whether the household head is male; and whether the household contains a married couple, a cohabiting couple, or neither. Taking these issues into account results in a five-part typology: (1) households headed by a married couple; (2) households headed by an unmarried cohabiting couple; (3) households headed by a single male4
; (4) households headed by a single female with no other adults present; and (5) households headed by a single, noncohabiting female with other adults present. The data for Australia and Poland are confined to only four categories because cohabiting couples are not distinguished from married couples.
We begin by estimating the distribution of children across these five types of households. We then estimate and compare before- and after-tax (and transfer) poverty rates for children residing in each of the five household types. After-tax poverty is based on net disposable income, which takes into account the income household members earn from the market, the taxes they pay, and the cash and near-cash transfers they receive from the government. Before-tax poverty is based solely upon the income the household receives from employment and from other market sources, such as interest and rents.
We then decompose the difference between the after-tax child poverty rate in the United States (P
) and in a given other country (p
) into the contributions of the tax redistribution scheme, the poverty gradient across household types, and the distribution of children across household types. There are multiple techniques of decomposition, but here we select the decomposition of rates used in Das Gupta (1993
; see also Smith, Morgan, and Koropeckyj-Cox 1996
), which extends the classical two-factor decomposition of a difference between proportions in Kitagawa (1955)
. This approach is particularly attractive for handling the different factors in a symmetrical manner, yielding components that add up to the overall difference in rates, and involving few components (only two in the two-factor decomposition, when all other decompositions yield at least three) that are readily interpretable as the contribution of each factor.
Specifically, we first write P
is the proportion of children in household type i
in the United States (e.g., D1
is the proportion of children in households headed by a married couple), Pi
is the (after-tax) poverty rate of children in household type i
in the United States, and Σ represents the sum across the five household types.
We then rewrite P
is the before-tax poverty rate of children in household type i
in the United States. If we define Ai
as the ratio of the after-tax poverty rate to the before-tax poverty rate for children in household type i
in the United States, P
then appears as
This first stage allows us to isolate the direct distributional effect of differences in living arrangements (factor D
). It also isolates an overall market earning factor and an overall government redistribution factor. To decompose each of those two factors into a level effect and a gradient effect, we then define
is the before-tax poverty rate of children in household type i
relative to the same rate for households headed by a married couple in the United States, whereas Gi
is the ratio of the after-tax poverty rate to the before-tax poverty rate for children in household type i
relative to the same ratio for children in households headed by a married couple in the United States. P
thus appears as a function of five factors: two scalars (B1
), and three vectors (D
, and G
So written, the difference between P
can now be decomposed into the additive contributions of five factors (Fα
, and Fγ
) among which the one relating to vector D
corresponds to distributional effects across living arrangements (Fδ
), whereas those relating to the vectors E
correspond to the gradient effects of pre-tax market earnings (F
) and of government redistribution (Fγ
Detailed derivations of the five factors are provided in the Appendix
In this decomposition, the contribution made by each scalar or vector to the poverty gap between the United States and each other nation is assessed by calculating a counterfactual poverty gap. Specifically, we calculate a hypothetical poverty gap by substituting, for all but the factor in question, identical variable values and vector distributions in place of the values and distributions that prevail in the United States and that country. The component Fδ
, for instance, represents the change in the poverty gap between the United States and another nation if the two countries had their own prevailing distributions of children by living arrangements but the same values or distributions for each of the other factors. The equations in the Appendix
demonstrate how we chose the counterfactual values and distributions such that the single-factor contributions thus estimated sum to the actual poverty gap between the United States and each country. Because the decomposition is additive, the relative contribution of a factor can be assessed as the ratio of the corresponding component to the overall poverty gap.
The term contribution
does not in this case have the same meaning as in causal analysis, nor as in the common statement that a given proportion of the variance is “explained” by a certain factor. In the classic decomposition of the difference between two crude death rates, one part of the difference is attributed to differences in age-specific mortality rates, whereas the remainder is attributed to differences in age distribution. Mortality changes, however, contribute to age structural changes, and the causal
effect of changing mortality conditions might not be fully captured by counterfactually keeping the prevailing age-specific mortality rates while at the same time changing age structures to a common distribution. Governmental policies, markets, and living arrangements would also be endogenous factors in a causal analysis inasmuch as changes in one of these factors are likely to affect the other factors. For instance, the prevalence of children living with single mothers may create pressure to redirect policy efforts toward their needs, while conversely, specific schemes of means-tested government support may have an impact on demographic behavior (e.g., Moffitt, Reville, and Winkler 1998
The different contributions to child poverty in our decomposition might thus be considered estimates of the first-order effects rather than of the total effects, since it is plausible that extensive changes in the values or distributions of other factors might also affect the distribution of living arrangements that is being kept unchanged in the decomposition. Assessing the full effect rather than the first-order effect of changes in one of the factors requires estimating the elasticity of a factor to changes with respect to other factors, which is beyond the scope of this paper. However, governmental policies, market characteristics, and demographic behaviors depend on a multiplicity of causes interacting in complex ways. We find it unlikely that variation in any one factor alone, such as government tax and transfer policy, would produce anything more than minor change in one of the other factors, such as children’s living arrangements. To the extent this is true, we provide a reasonable approximation of the total impacts of living arrangements, market earnings, and government redistribution through our first-order estimates of those impacts.