Data for Multidimensional Poverty Measurement

As stated in Chapter 1, the initial step in poverty measurement, even prior to identification, is to select the space of analysis (resources, capabilities, utility) and the purpose of the measure to be constructed.

Multidimensional poverty is measured using micro data. By micro data we mean the unit-level data containing responses that each unit of analysis (such as person or house­hold) provided. This contrasts with macro data, or aggregate indicators or marginal meas­ures such as the mortality rate, literacy rate, mean household income, or the enrolment ratio, which summarize the achievements of a society. The three most common micro data sources are censuses, household surveys, and administrative records—also called register data. New relevant data sources such as mobile telephony and satellite imagery are rising sharply and will deeply enrich future multidimensional poverty analyses.

A census is an enumeration of all households in a well-defined territory at a given point in time (Mather 2007). National censuses are typically conducted every five to ten years and contain information on a strictly limited number of variables: demographic variables such as nationality, age, gender, marital status, place of birth, location, ethnicity or religion, and language; social variables such as literacy, educational attainment, and housing conditions; and economic variables such as activity condition and employment (UN 2008: 112-13, table 1). Special censuses may be implemented for targeting and monitoring certain programmes, again using a few simple variables.

Censuses provide information with negligible sampling error (the whole population is considered) at highly disaggregated levels (municipalities-neighbourhoods).

Administrative data refers to information typically collected by a government department or agency primarily for administrative purposes (birth registration, customs, administration of a social benefit, etc.). One prominent example is population registers constructed through a civil registration system (UN 2001). Population registers consist of an inventory of each member of the resident population of a country augmented con­tinuously by information on vital events (births, deaths, adoptions, marriages, divorces, among others) (UN 2001). Other examples are tax, education, police, and health records.

Some advantages of using administrative datasets are that (a) they typically cover virtually 100% of the population of interest in a continuous form, (b) there are no added data collection costs, (c) there is data for individuals who might not normally respond to surveys, and (d) when linked to other unit-level data sources, administrative data can produce a powerful resource for multidimensional poverty. However, there are also some disadvantages: (a) the information collected in administrative records is limited and may not match the research purpose, (b) any changes in data collection procedures or definitions may prevent comparability over time, (c) serious data quality issues may compromise accuracy, (d) metadata is usually not available,^[205] (e) access to

BOX 7.1 THE MILE AHEAD IN DATA COLLECTION

As Chapter 1 mentioned, enormous progress has been made in data collection worldwide since the 1940s.

In terms of frequency, poverty data continues to lag behind most other economic information. The lack of frequent data makes it impossible to inform policies responding to the impact of certain events such as financial crises and natural disasters on the poor.

In terms of population coverage, household surveys typically exclude certain groups such as nomadic people, recent or illegal migrants or refugees, and the homeless—as well as institutionalized groups such as prisoners, those hospitalized or in nursing homes, the military, and members of religious orders. They may also overlook the elderly within households. Some excluded groups may be particularly marginalized, thus should be considered in poverty measures.

A different though related problem consists of the falling response rates in household surveys over subsequent rounds, even when they are not panel surveys. Such a problem is being observed in some—usually developed—countries, such as the UK, particularly with respect to indicators such as wealth or assets. While such a problem can be partially overcome by reweighting the sample, this is not ideal, and creative ways to deal with falling response rates will need to be devised.

Dimensional coverage is limited, often in ways that would be relatively easy to address. Common missing dimensions that may be relevant for poverty studies include health functionings, safety from violence, the quality of work, empowerment, social connectivity, and potentially time use.

Many surveys seek to define household-level achievements, but do not elucidate intra-household inequalities, gender inequalities, and age inequalities, nor do they cover overlooked topics such as the care economy and household duties.

The nature of the indicators that are collected is another area of potential improvement. Paraphrasing the Report by the Commission on the Measurement of Economic Performance and Social Progress (Stiglitz, Sen, and Fitoussi 2009), the time is ripe to move from the space of resources to the space of functionings. Functionings, as argued in Sen's capability approach, seem to be central to poverty reduction and are of intrinsic importance. Yet even such a central dimension as health functioning is absent from most good surveys that collect income or consumption data.

Last, but not least, given that the aim remains to reduce poverty, not to measure it, improved channels of complementarity are required between censuses, administrative records, household surveys, and other information such as from satellites and cell phones, in order to advance towards an integrated programme of data collection and compilation. Merging GIS data on environmental conditions with household surveys, for example, greatly strengthens poverty measurement and the monitoring and impact evaluation of sustainable poverty reduction programmes. In sum, despite incredible progress in data collection, there is still a way to go so that poverty reduction can be informed by a sufficient depth and frequency of data such as are available for societies' other priorities.

administrative (micro) data varies by country, and (f) linking data sources is rarely straightforward.

Household surveys are the most commonly used data source to study poverty.

The respondents for the survey are selected from what is called a ‘frame' or list, which is usually obtained from the most recent census and is typically a list of households. Different sampling methods such as simple random sampling and complex multistage sampling are used in order for the sample to be representative of the population. Deaton (1997) offers a valuable introduction to each method.^[206]

Household surveys were collected as early as the eighteenth century in England (Stigler 1954). After World War II, household surveys expanded internationally with India being a pioneer. Since the 1970s, several international programmes have promoted and supported the collection of household survey data in developing countries. These include the World Fertility Surveys, which were introduced in the 1970s and became the Demographic and Health Surveys (DHS) in 1984, and the World Bank Living Standards Measurement Study (LSMS) survey programme, ongoing since 1980 (Grosh and Glewwe 2000: 6). In 1995, the United Nations Children's Fund (UNICEF) began its Multiple Indicators Cluster Surveys (MICS), and in 2000 the World Bank launched the Core Welfare Indicator Questionnaire (CWIQ). The World Bank has also intensively supported the development and widespread fielding of household budget surveys and income and expenditure surveys that are used for income and consumption poverty measures and may contain other topics. Multidimensional poverty measures typically rely on multi-topic household surveys, which collect information with one survey instrument using a sample frame that has been defined to capture a diverse set of topics.^[207]

7.2