Time and Income Poverty Measurement. An Ongoing Debate on the Inclusion of Time in Poverty Assessment

Abstract

This article aspires to foster the debate around the methods for measuring time and income poverty. In the last fifteen years a few studies (Dorn et al. in RIW, 2023; Harvey and Mukhopadhyay in SIR 82, 57–77, 2007; Bardasi and Wodon in FE 16, 45–78, 2010; Zacharias in LEIBCWP. https://doi.org/10.2139/ssrn.1939383, 2011; Merz and Rathjen in RIW 60, 450–479, 2014) attempted to measure multidimensional deprivation including time poverty in the definition. Some of them (Bardasi & Wodon in FE 16, 45–78, 2010; Harvey & Mukhopadhyay in SIR 82, 57–77, 2007; Zacharias in LEIBCWP. https://doi.org/10.2139/ssrn.1939383, 2011) put unpaid work–and, therefore, gender inequalities in the division of work–at the center. Despite the fact that the Levy Institute Measure of Time and Income Poverty (LIMTIP) was first presented more than a decade ago (Zacharias in LEIBCWP. https://doi.org/10.2139/ssrn.1939383, 2011), the measure was always employed in reports and never empirically discussed in an academic article. Here I want to fill this gap in the debate by comparing the LIMTIP to the other measures and by applying it to a new case–Italy–furthering the exploration around the linkages between gendered time allocation, employment patterns and household wellbeing in a country characterized by an extraordinary low women’s participation in the labor market and an equally extraordinary wide gender gap in unpaid care and domestic work.

Appendices

Appendix 1

The stages of the construction of the synthetic datasets Created for estimating the Levy Institute Measure of Time and Income Poverty (LIMTIP) for Italy in 2008 and 2014 are presented. I use the Italian Indagine Multiscopo sull’Uso del Tempo of 2008–2009 and 2013–2014 (IT-TUS) for time use data and the Italian data of the European Survey of Income and Living Conditions of 2009 and 2015 (IT-SILC) for the information on demographics and income.

1.1 Data and Alignment

Both IT-TUS and IT-SILC are representative at the national level and contain information for individuals for all age classes. IT-TUS 2008–09 has 44,605 observations, representing 59,426,798 individuals when weighted, while IT-SILC 2009 has 51,196 observations, representing 60,108,862 individuals when weighted. IT-TUS 2013–14 has 44,866 observations, representing 60,410,793 individuals when weighted, while IT-SILC 2015 has 42,987 observations, representing 60,843,061 individuals when weighted.

In order to match the most similar observations, I had to select several variables. Following the example of Masterson (2014), I identified a number of strata variables that are relevant for determining the average amount of household production that is required to subsist at the poverty level of income. For this reason, the reference group for household production threshold estimation consists of households with at least one non-employed adult to avoid underestimation of the necessary amount of household production. Strata variables include, at household level, the number of children and of adults, the presence of a non-employed adult, the income category and, at individual level, the sex, and the employment status. Additionally, other variables might be relevant, as, for example, age, citizenship, region of residence, level of education, etc. These additional variables are selected on the basis of their comparability in the two data files.

Therefore, first of all, I extensively worked on the two separate files in order to align the common variables in terms of definition and measurement. For example, in IT-TUS the only income information present is the main source of income at individual level. Therefore, based on the categories provided by the variable in the IT-TUS, I constructed a corresponding variable in the IT-SILC where, instead, I found detailed information about different sources of income, both at household and individual level. I proceeded according to this principle until I harmonized all the definitions of strata and relevant variables.

Then, to maximize the matching quality, I checked that the distributions of the common variables were comparable. I expected comparable distributions because both data sets have a large number of observations and they are both nationally representative. When common variables did not align, then I doublechecked the definitions and harmonized them, where possible. I found an excellent comparability for all the selected variables (see Tables 8 and 9 below). After the harmonization, I adjusted the sum of the attached weights for records, in order to make them comparable.

Table 8 Alignment of strata variables, percent 2008

Table 9 Alignment of strata variables, percent 2014

1.2 Matching

At this point I need to transfer the variables related to time use from the IT-TUS to the IT-SILC. Considering which are the factors that mostly affect the variation of the amount of unpaid care and domestic work, I divided the reference group into 12 subgroups based on the number of children (0, 1, 2 and 3 or more) and the number of adults (1, 2 and 3 or more).

According to the selected strata variables (the number of children in the household, the number of adults in the household, the presence of a non-employed adult in the household, the marital status, the presence of children under 3 years of age in the household, the sex, the main source of income, the activity status and the number of earners in the household), I separated the data within each file in 38,400 discrete cells.

Then I carefully selected the common variables in the logistic regression model for propensity scores in order to maximize the explanatory power. In the end, my selection of relevant variables included, besides the strata variables: age, level of education, being in education, having a second job, citizenship, region, household tenure, head of the household, spouse.³⁰

After running the model, all records for each file were sorted by estimated propensity score and attached weight. For every recipient in the recipient file (IT-SILC), an observation in the donor file (IT-TUS) was matched with the same or nearest neighbour, based on the rank of their propensity scores. In this match, a penalty weight is assigned to the propensity score according to the size and ranking of the coefficients of strata variables not used in a particular matching round (see Tables 10 and 11 below). Under this sorting scheme, I assigned records with larger weights in the donor file to multiple records in the recipient file until all of their weight has been used up.

Table 10 Distribution of matched records by matching rounds 2008

Table 11 Distribution of matched records by matching rounds 2014

1.3 Test of Quality of Matching

In order to check the quality of the matching I compared the marginal and the joint distributions in the matched file and in the donor file (see Tables from 12, 13, 14, 15 below). The constraints of the matching scheme should lead to identical marginal distributions, and the joint distribution of variables not jointly observed should be nearly the same.

Table 12 Average weekly hours of household production in IT-TUS and Matched file 2008

Table 13 Average weekly hours of household production in IT-TUS and Matched file 2014

Table 14 Ratio of imputed values to IT-TUS values by strata variable 2008

Table 15 Ratio of imputed values to IT-TUS values by strata variable 2014

Therefore, I checked that the mean and the median values for the transferred variables by each strata variable were similar in the matched and the donor files. Specifically, I checked if there were discrepancies in time devoted to unpaid care and domestic work by type of household and sex of the individual. The ratio of the average time spent by women and men for different household activities in the matched file, to the average value in the donor file and the distribution of weekly hours of unpaid care and domestic work for each of the 12 cells, differentiated by number of adults in the household and number of children in the household (see figures from 3, 4, 5, 6 below), give confidence that the marginal distributions have been well preserved in the statistical matching process. Divergences are related in particular to the limited number of observations with three or more children.

Fig. 3

Ratio of imputed values to IT-TUS values, average by number of children and sex (2008)

Fig. 4

Ratio of imputed values to IT-TUS values, average by number of children and sex (2014)

Fig. 5

Distribution of weekly hours of household production by cell (2008)

Fig. 6

Distribution of weekly hours of household production by cell (2014)

Appendix 2

2.1 IT-TUS

The Italian Institute of Statistics (ISTAT) regularly collects data on time use from 2002. The IT-TUS (ISTAT, 2017) is carried out every five years and it is composed of three questionnaires: the individual questionnaire contains general information on family members and their household, the daily diary records the daily use of time of all members aged three years or more, and the weekly diary records, and the hours of paid work for all members that hold a job. Individuals are required to fill in the daily diary for week-days, Saturdays, and Sundays randomly. Sample weights are used to obtain statistics representative of the whole Italian population. Activities are classified in 10 groups: physiological needs, professional work, education activity, household activities, voluntary work in organizations and beyond, social life and entertainment, sport and recreation activities, personal hobbies, using mass-media, time spent on moving and transportation. This classification enables a detailed analysis of the time each household member spends on each activity. The Italian time use survey does not include information on income and earnings.

The population of interest in the Italian Time Use survey is composed of households residing in Italy and the individuals who make them up; people residing in cohabitation institutions are excluded. The family is understood as a de facto family, i.e. a set of cohabiting people linked by marriage, kinship, affinity, adoption, guardianship or emotional ties.

The study domains, i.e. the areas with respect to which the population parameters being estimated are referred, are of two different types: territorial-type domains and temporal-type domains.

The territorial domains are as follows:

• the entire national territory;

• the five geographical divisions (North-Western Italy, North-Eastern Italy, Central Italy, Southern Italy, Insular Italy);

• the geographical regions (with the exception of Trentino Alto Adige whose estimates are produced separately for the provinces of Bolzano and Trento);

• the municipal typology obtained by dividing Italian municipalities into six classes based on socio-economic and demographic characteristics:

• A) municipalities belonging to the metropolitan area divided into:

  • A1 central municipalities of the metropolitan area: Turin, Milan, Venice, Genoa, Bologna, Florence, Rome, Naples, Bari, Palermo, Catania, Cagliari;

  • A2 municipalities that gravitate around the central municipalities of the metropolitan are

• B) Municipalities not belonging to the metropolitan area divided into:

  • B1 Municipalities with up to 2,000 inhabitants.

  • B2 Municipalities with 2,001–10,000 inhabitants.

  • B3 Municipalities with 10,001–50,000 inhabitants.

  • B4 Municipalities with over 50,000 inhabitants.

As far as temporal domains are concerned, the estimates produced by the survey are published with reference to four types of day: weekday, day before a holiday (Saturday), public holiday (Sunday) and average weekly day.

The sampling design is complex and makes use of two different sampling schemes, both based on a cluster structure of the population in municipalities and families. Within each of the domains defined by the intersection of the geographical region with the six areas A1, A2, B1, B2, B3 and B4, the Italian municipalities are divided into two subsets on the basis of the resident population:

• 1- the set of self-representative municipalities (which we will indicate from now on as AR municipalities) made up of the municipalities with the largest demographic size;

• 2- the set of non-self-representative municipalities (or NAR) made up of the remaining municipalities.

Within the set of AR municipalities, each municipality is considered as a separate stratum and a design known as cluster sampling is adopted. The primary sampling units are represented by registry families, systematically extracted from the registry office of the municipality itself; for each family included in the sample, the characteristics under investigation of all the de facto members belonging to the same family are recorded.

Within the NAR municipalities, a two-stage design is adopted with stratification of the first-stage units. The first stage units (UPS) are the municipalities, the second stage units are the households (USS); for each family included in the sample, the characteristics under investigation of all the de facto members belonging to the same family are recorded.

Municipalities are selected with probabilities proportional to their demographic size and without repetition, while households are extracted with equal probabilities and without repetition.

For the definition of the overall sample size and its allocation among the different territorial domains, it was decided to adopt a mixed perspective based both on cost and organizational criteria, and on an assessment of the expected sampling errors of the main estimates with reference to each of the territorial domains of interest.

The theoretical sample size at the national level was essentially set at based on cost and operating criteria and is equal to approximately 21,000 families and 500 municipalities. The allocation of the sample of families and municipalities among the various regions was then defined adopting a compromise criterion such as to guarantee the reliability of the estimates both at the national level and at the level of each of the territorial domains.

EU-SILC

EU-SILC (Eurostat, 2016) is a multi-dimensional dataset focused on income but at the same time covering housing, labour, health, demography, education and deprivation, to enable the multidimensional approach of social exclusion to be studied. It consists of primary (annual) and secondary (ad hoc modules) target variables, all of which are forwarded to Eurostat. The primary target variables relate to either household or individual (for persons aged 16 and more) information is grouped into areas: at the household level basic/core data, income, housing, social exclusion and labour information; at the personal level basic/demographic data, income, education, labour information and health. The secondary target variables are introduced every four years or less frequently only in the cross-sectional component. Data are based on a nationally representative probability sample of the population residing in private households within the country, irrespective of language, nationality or legal residence status. All private households and all persons aged 16 and over within the household are eligible for the operation. According to the Commission Regulation on sampling and tracing rules, the selection of the sample is drawn according to the following requirements:

For all components of EU-SILC (whether survey or register based), the cross-sectional and longitudinal (initial sample) data shall be based on a nationally representative probability sample of the population residing in private households within the country, irrespective of language, nationality or legal residence status. All private households and all persons aged 16 and over within the household are eligible for the operation.

Representative probability samples shall be achieved both for households, which form the basic units of sampling, data collection and data analysis, and for individual persons in the target population.

The sampling frame and methods of sample selection shall ensure that every individual and household in the target population is assigned a known and non-zero probability of selection.

Depending on the country, micro-data could come from:

One or more existing national sources whether combined or not with a new survey;

A new harmonized survey to meet all EU-SILC requirements.

The only constraint is that for both the cross-sectional and longitudinal components, all household and personal data will be linkable.

In terms of the units involved, four types of data are involved in EU-SILC:

• (i) Variables measured at the household level;

• (ii) Information on household size and composition and basic characteristics of household members;

• (iii) Income and other more complex variables termed ‘basic variables’ (education, basic labour information and second job) measured at the personal level, but normally aggregated to construct household-level variables; and

• (iv) Variables collected and analysed at the person-level ‘the detailed variables’(health, access to health care, detailed labour information, activity history and calendar of activities’).

For set (i)-(ii) variables, a sample of households including all household members is required. Among these, sets (i) and (ii) will normally be collected from a single, appropriately designated respondent in each sample household–using a household questionnaire for set (i) and a household member roster for set (ii). Alternatively, some or all of these may be compiled from registers or other administrative sources.

Set (iii) concerns mainly, but not exclusively, the detailed collection of household and personal income–this must be collected directly at the person level, covering all persons in each sample household. In most countries, i.e. in the so-called ‘survey countries’, these income variables will be collected through personal interviews with all adults aged 16 + in each sample household. This collection will be normally combined with that for set (iv) detailed variables, since the latter also must also be collected directly at the person level. By contrast, in ‘register countries’, set (iii) variables will be compiled from registers and other administrative sources, thus avoiding the need to interview all members (adults aged 16 +) in each sample household.

Set (iv) variables will normally be collected through direct personal interview in all countries. These are too complex or personal in nature to be collected by proxy; nor are they available from registers or other administrative sources. For the ‘survey countries’, this collection will normally be combined with that for set (iii) variables as noted above–consequently both normally based on a sample of complete households, i.e. covering all persons aged 16 + in each sample household.

However, from the substantive requirements of EU-SILC, it is not essential that–in contrast to set (iii) variables–set (iv) variables be collected for all persons in each sample household. It is possible to do this collection on a representative sample of persons (adult members aged 16 +), such as by selecting one such person per sample household.