3. OUR APPROACH In this study, we rely on consistent units of analysis (at the household level) for defining both ‘work’ and ‘poverty’. The household unit is based on the dwelling unit identified in the 2013 Census and encompasses the following household types: couples without children; couples with child(ren); one parent with child(ren); two or more family households; one-person households; and other multiperson households. 6 Next, we want to approximate the economic circumstances for each household and condense this information in a binary fashion to classify whether a household is poor or not. The prevailing identification strategy is to generate a poverty threshold based on the income distribution of the whole population. This includes households with self-employed individuals and households that include only individuals aged 66 and over (referred to hereafter as pensioner households). We define a household as being in poverty when their monthly net equivalised income (before housing costs) is below 60 percent of the median income poverty line as at March 2013. To estimate the economic circumstances of a household, the common approach in the economic literature is to account for all income sources. For example, when studying British in-work poverty, Bourquin et al. (2019) use data from the Family Resources Survey. The survey provides information on income “from employment and self-employment, savings and investments, occupational and private pensions, (…), as well as any state benefits and tax credits (…), including the state pension for pensioners” (Bourquin et al., 2019, p. 6). As the aim of our study involves portraying in-work poverty prevalence at a particular point in time (i.e., March 2013) and consequently linking characteristics of individuals at that time point (via the 2013 Census), this requires restriction of our income data sources to those that are available monthly. We have identified suitable income sources in the Inland Revenue data (described in more detail in the following section), as well as the Working for Families (WfF) dataset with information on the WfF tax credit and Accommodation Supplement (AS). It is, however, worth noting that we lack monthly-level information on income from self-employment, private superannuation, and other investments, which can constitute an important income source, especially for households with self-employed individuals and pensioner households. These data are available at an annual level, and using annual income information to link with household characteristics as at March 2013 is deemed not suitable, given the assumptions required. A detailed discussion of the potential impact of not accounting for certain income sources on the approximated household income can be found in the Appendix. The income variable on which the poverty classification is founded is therefore income (aggregated on the household level) from Inland Revenue, WfF tax credit and AS. We account for deductions and we divide the household-specific monthly net income by the OECD scale on family size to produce equivalised income. The OECD scale gives a weight to each household member which sum to produce the equivalised household income (the weight assigned the first adult is 1.0; each subsequent individual aged 14 or over is 0.5; and each child under age 14 is 0.3). In subsequent tables and figures, we denote households as ‘poor’ or conversely ‘non-poor’. It is also noteworthy that defining the poverty threshold based on the population-wide income distribution that includes pensioner households can have potential spillover impacts on the estimated in-work poverty prevalence among the population of interest (working-age households). For example, 6 A group of related (e.g., siblings) or unrelated (e.g., flatmates) people living together who do not form a family. Page 14