POVERTY IN TIMOR-LESTE 2014 O V E R T Y CONTENT PREFACE 01 POVERTY REDUCTION: TIMOR-LESTE IN THE 23 EXECUTIVE SUMMARY 03 INTERNATIONAL CONTEXT INTRODUCTION 09 SENSITIVITY OF POVERTY INCIDENCE 27 POVERTY MEASUREMENT METHODOLOGY I: 11 Calorie requirements 27 CONSUMPTION-BASED WELFARE INDICATOR Festivities and ceremonies 29 Consumption as the welfare indicator 11 Fieldwork team effects 30 Constructing comparable nominal consumption 11 Changes in nominal consumption 12 HOUSEHOLD CHARACTERISTICS AND POVERTY 32 expenditure 2007-2014 Demography and poverty 32 Consumption pattern and poverty 35 POVERTY MEASUREMENT METHODOLOGY II: 13 O R Nutritional status of children 35 POVERTY LINES Ownership of livestock and other durable goods 35 District-level poverty lines 13 Food poverty line 14 MULTIDIMENSIONAL DEPRIVATION AND POVERTY 39 Rent poverty line 15 Annex A: Timor-Leste Survey of Living 42 Non-food (excluding rent) poverty line 15 Standards-3 Overall poverty line 15 Annex B: Consumption-based welfare indicator 45 The estimated poverty lines 17 Annex C: Rental model 49 Annex D: Standard errors and 50 POVERTY ESTIMATES 19 confidence intervals Poverty indices 19 Annex E: Sensitivity analysis: fieldwork teams 52 E S T E Headcount index 15 Annex F: Issue in developing a multidimensional 57 Poverty gap index 19 poverty index for Timor-Leste Squared poverty gap index 19 Results 20 REFERENCES 63 Average consumption and inequality 20 Poverty estimates: national, sectoral 20 and regional District-level poverty estimates 21 E R T PREFACE M O R Poverty in Timor-Leste 2014 This report provides a detailed assessment of the As opposed to the first two TLSLS which both followed methodological approaches and headline poverty periods of instability and upheaval, the intervening results from the Timor-Leste Survey of Living period between TLSLS-2 and TLSLS-3 has been one Standards 3. The survey is the third in a series of of peace, development and stability in Timor-Leste. mutually comparable, detailed surveys to assess a It is therefore important to reflect upon the impact wide range of aspects of living standards in Timor- that a stable country with an ambitious development Leste. Over time, the Timor-Leste Surveys of Living agenda can have on improving living standard when Standards (TLSLS) have become larger to allow for not set back by periods of conflict or disasters. greater precision and depth of analysis. TLSLS-1 was conducted in 2001 soon after Timor-Leste became This report focuses on providing key results from 01 L E S T E an independent nation. TLSLS-1 surveyed 1,800 the TLSLS-3 and a detailed account of the survey households over a period of three months. Six years methods. These is not the end but marks the later, TLSLS-2 began, in 2007, and included 4,477 beginning of an exercise to exploit the rich detail of households surveyed over 12 months. the data-source, and the Government of Timor-Leste in coordination with its development partners and This survey, TLSLS-3, is the latest in the series. It research community will be conducting further work was conducted over a 12 month period from April to assess the drivers of poverty, and help to design 2014 to April 2015 and involved surveys of 5,916 policy and interventions that have the biggest positive households, 30 percent more than the previous impact for the most people. survey. A focus of the series has been to conduct high-quality surveys that provide a sound basis for the monitoring of household living standards, and the critical task of designing public policy to help Dili, September 2016 improve living standards for all. TLSLS-3 marks the highest level of survey design and implementation by Helder Lopes the General Directorate of Statistics, with technical Vice-Minister of Finance support from the World Bank. The result is a very comprehensive, high-quality survey. V E R T Y EXECUTIVE SUMMARY Poverty in Timor-Leste 2014 INTRODUCTION including consumption expenditures, health, Data from the recently completed 2014-15 Timor- education, employment, housing and access Leste Survey of Living Standards (TLSLS-3) show to services. A total of 5,916 households were a significant reduction in poverty in the country interviewed by trained and closely supervised since 2007. At the national poverty line, which enumerators over 12 consecutive months from represents the cost of meeting basic needs in April 2014 to March 2015, and the sample was relation to food, shelter and non-food items in distributed across the country so as to obtain Timor-Leste, the proportion of Timorese living reliable district-level poverty estimates. The in poverty declined from 50.4% in 2007 to an survey was deliberatively designed such that 02 estimated 41.8% in 2014. At the internationally poverty estimates could be directly compared with 03 comparable extreme poverty line of $1.90 (in those estimated from TSLSS-2, the household 2011 purchasing power parity dollars), poverty survey conducted in 2007. Poverty in Timor-Leste 2014 in Timor-Leste fell from 47.2% to 30.3% over the same period. S T E METHODOLOGICAL STEPS IN POVERTY This Executive Summary provides an overview of: ESTIMATION 1) key design features of the household survey; Even though the understanding of poverty 2) the main steps used in arriving at the above generally differs across people, places, and estimates; and 3) key findings on poverty in social contexts, it is based on some underlying Timor-Leste. notion of deprivation. That is, poverty is defined as having fewer resources than would be needed to meet basic human needs, even though what TIMOR-LESTE SURVEY OF LIVING STANDARDS-3 are considered “basic needs” might differ across TLSLS-3 is the third in the series of nationally countries and across people. Deprivations representative surveys conducted by the General also exist in different dimensions (e.g. food, Directorate of Statistics (DGE). These surveys shelter, health, education etc.), and for practical are designed to help measure and monitor purposes, there is need for a summary measure living standards in Timor-Leste. They do this by that captures these multiple dimensions. collecting information on a broad range of topics This report provides key results using (i) a were directly computed from TLSLS-3 responses. FIGURE 01: NOMINAL CONSUMPTION FIGURE 2: CHANGES IN DWELLING AMENITIES 2007 TO 2014 consumption-based indicator that aggregates For rent, as most dwellings are owner-occupied deprivations in multiple dimensions in monetary and few people actually pay rent, the value of NOMINAL CONSUPTION, 2007, 2014 90 83 2007 % population living in dwellings with terms and (ii) a set of non-monetary indicators rent is imputed with the commonly-used hedonic 80 72 2014 72 Nominal consuption per person per month that directly capture specific deprivations in regression approach. The hedonic model uses the 60 70 15 58 50 60 key dimensions. A consumption-based rather relationship between respondents’ estimates of 49 44 40 50 42 than income-based measure is used because actual rent paid (when available) or how much their 15 36 37 40 30 6 30 information on consumption is more easily and dwelling could be rented for and the characteristics 4 30 22 20 20 17 accurately collected than information on income of the dwelling to estimate market values of 10 10 given the large subsistence and informal sectors in dwellings with specific characteristics. 20 32 10 2 0 2007 2014 0 the economy. In addition, non-monetary indicators Food Rent Other Concrete Metal Flush Solid Improved Has Has Non Food or external roofs toilet Clay pit access access to are used to assess deprivations in specific The consumption-based poverty line is the sum of walls tap electricity Poverty in Timor-Leste 2014 2014 Consumption Poverty Estimates dimensions that are not completely captured by three components: 1) the food poverty line; 2) the monetary measures, such as health, education, rental poverty line; and 3) the non-food non-rent and ownership of essential assets. poverty line. The food poverty line is derived as access to water and electricity (see Figure 2). All the cost of the typical local food basket that yields Given the objective of generating district-level of these factors indicate increased rental value, Consumption-based poverty measures. The a nutrient value of 2,100 calories per person. The poverty estimates, poverty lines were estimated and partly explain the increase in estimated consumption-based indicator is per capita total rental poverty line is the average estimated rental separately for each district, accounting for rental expenditure. Food expenditures, on the household expenditure which consists of three cost of a reference dwelling that has 2 rooms, differences in consumption patterns as well as other hand, grew the most slowly for practically key components: 1) the value food expenditures good external walls, proper sanitation and access commodity prices. Given the smaller sample sizes all households. (purchased as well as own-produced); 2) the rental to electricity. Finally, the non-food poverty line is 04 at district level, the margins of error are higher 05 value of dwellings (actual or imputed); and 3) the specified as the average non-food expenditure of for district-level estimates. The key steps used in value of all other non-food, non-rent expenditures. those households whose food expenditures are estimating poverty prevalence are summarized in POVERTY LINES FOR TIMOR-LESTE Poverty in Timor-Leste 2014 The values of food and non-food consumption close to the food poverty line. the Box 1. Details on these steps are discussed in While nominal expenditures grew significantly Sections 2 and 3. over 2007-2014, so did the cost of living for the poor. The national poverty line, which represents the average cost of meeting basic needs, grew BOX 1 NOMINAL CONSUMPTION GROWTH by 84.5%, from $25.14 per person per month in KEY STEPS IN MEASURING POVERTY Between 2007-2014, per capita consumption 2007 to $46.37 per person per month in 2014. 1. Per capita total consumption expenditure is used to measure welfare. expenditure at least doubled in nominal terms Within the country in 2014, the poverty line was 2. A poverty line, also expressed in per capita consumption expenditure, is specified as the for all households except for those poorest 5% the highest in Dili, the most urbanized district, monetary value of a 2,100 calorie per day diet, living in a 2 room home with proper of households. Figure 1 shows the change in reflecting its higher cost of living. sanitation and access to electricity, and a corresponding consumption level of non-food nominal consumption for the average Timorese goods and services. person. The largest component of that increase 3. The following poverty indices are used to summarize the level of poverty: was in rental expenditures, especially for Poverty headcount index: The proportion of the total population below the poverty line. households in the bottom 50%. The increase in Poverty gap Index: A measure of the average amount by which a family’s consumption rental expenditures is consistent with a marked falls short of the poverty line expressed as a proportion of the poverty line, while the improvement in the quality of dwellings, and consumption shortfall of those above the poverty line is taken to be zero. also improvements in public infrastructure that serve these dwellings, as reflected in improved TABLE 1: National poverty lines in 2007 and 2014 (US$ per person per month, current prices) TABLE 4: Real consumption (US$ per person per PERFORMANCE IN NON-MONETARY month, 2014 constant average national prices) INDICATORS 2007 2014 Both TLSLS-2 and TLSLS-3 capture information Food 15.40 25.01 2007 2014 on a range of non-monetary indicators, both at Rent 4.57 11.50 the household and the individual level. These Timor-Leste 53.9 60.1 Non-food (excluding rent) 5.17 9.86 indicators supplement poverty measures based Rural 50.3 55.5 on changes in consumption expenditure over Total 25.14 46.37 Urban 64.2 71.7 time. Analysis of a subset of these indicators, reported in Table 5, shows significant nationwide TABLE 2: Poverty lines in 2014 (US$ per person per month) improvements in access to basic services Food Rent Other non-food Total Inequality is usually measured using the Gini (especially electricity), children’s health and coefficient which varies between 0 and 1 with education. Improvements can also be seen Poverty in Timor-Leste 2014 Timor-Leste 25.01 11.50 9.86 46.37 higher values indicating greater inequality. in access to sanitation facilities and safe Dili 29.07 12.64 14.45 56.16 The Gini coefficient for per capita consumption drinking water. estimated from the TLSLS-3 data is 0.29 reflecting relatively low level of inequality by international There is also a noticeable growth in the ESTIMATES OF POVERTY standards, and only very slightly higher than the ownership of consumer durables between 2007 Estimates based on TLSLS-3 data indicate that in 2014, just under 42% of people in Timor-Leste lived below level of 0.28 estimated for 2007. and 2014. Ownership of mobile phones and the poverty line. While this is still a high proportion of the population, it represents significant progress, with televisions has increased, as also the ownership poverty incidence reducing by almost 9 percentage points from 50.4% in 2007 (Table 3). of motorcycles, and appliances such as fans and 06 TABLE 3: Poverty indicators 07 Percent of population in poverty (%) Poverty gap index Poverty in Timor-Leste 2014 2007 2014 2007 2014 TABLE 5: Overall performance in key poverty dimensions Timor-Leste 50.4 41.8 13.8 10.4 INDICATORS Rural 54.7 47.1 15.5 12.2 (% of Population Living in Households …..) 2007 2014 Improvement With no electricity connection 64 28 56% Urban 38.3 28.3 9.0 5.9 With poor sanitation 58 40 31% With no access to safe drinking water 40 25 38% There was also a significant fall in the poverty gap index, indicating that welfare improvements were not just With poor quality floor in their dwelling 61 48 21% limited to those near the poverty line. The larger decline in the poverty gap index relative to the headcount Using poor cooking fuel 97 87 10% index implies the average shortfall in consumption levels experienced by the poor in 2014 was smaller than With very few assets 83 56 33% that in 2007. Poverty fell in both rural and urban areas, though the decline was larger in urban areas. With at least one child not attending school 42 17 60% Without anyone with at least 5 years school 17 13 24% With at least one underweight child under 5 years of age 35 16 54% REAL CONSUMPTION GROWTH AND CHANGES IN INEQUALITY With at least one stunted child under 5 years of age 38 24 37% A reduction in poverty reflects an increased real purchasing power for lower-income households – they have With at least one child under 5 years of age with wasting 20 7 65% a greater capacity to consume essentials. However, there has also been growth in real purchasing power overall both in rural and urban areas, as seen in Table 4. FIGURE 3: FIGURE 4: OWNERSHIP OF CONSUMER DURABLES OWNERSHIP OF LIVESTOCK INTRODUCTION 80 90 82 % of population livings in hhlsd that own 79 % of popultion living in hhlds that own 68 70 2007 80 2007 60 2014 70 2014 50 60 39 40 50 30 24 40 19 29 Poverty in Timor-Leste 2014 20 14 16 12 30 22 25 21 While the Timorese economy has shown signs Information was collected on, among other 7 10 5 2 20 14 11 of faster growth in recent years, and public things, consumption expenditures, the health and 0 10 spending has increased substantially, how the education status of households, anthropometric Fans TV Mobile Motor Electric phone cycle rice 0 poor have fared through this period has remained measurements of children, and occupational and cookers Buffalo Cow Pig Goat largely unknown. The successful conclusion employment status of household members. of the third round of the Timor-Leste Survey of Living Standards (TLSLS-3) in April 2015 provides A total of 5,916 households were interviewed by an opportunity to update our understanding trained and closely supervised interviewers over of poverty and of many other economic and 12 consecutive months from April 2014 to March 08 electric rice cookers (Figure 3). However, there Despite this progress, finding pathways out social conditions. 2015, and the sample was distributed across 09 was no apparent increase in the proportion of of poverty for the remaining poor remains a the country so as to obtain reliable district-level the population owning livestock, the principal continuing challenge. Further analysis of TLSLS-3 The TLSLS-3 is a comprehensive multi-topic poverty estimates. Poverty in Timor-Leste 2014 productive asset aside from land in Timor-Leste data can shed new light on the conditions of the survey and the content covered is very broad. (Figure 4), although amongst those who owned poor and the constraints they face in overcoming It encompasses most of the content covered Survey instruments were deliberatively designed any livestock there was some increase in the poverty. This will be important in developing an under more specialized surveys such as the such that the TLSLS-3 poverty estimates could be number of livestock owned. evidence base to identify and implement policies Demographic and Health Survey, the Multiple reliably compared with the TLSLS-2 estimates. and programs for future poverty reduction. Cluster Indicators Survey and a typical labor Consumption estimates for the TLSLS-3 are In conclusion, data from TLSLS-3 indicates that force survey. based on data from three sections of the there has been visible reduction in poverty in questionnaire that remained almost identical to Timor-Leste over 2007-2014: a 9 percentage point decline when Timor-Leste national poverty lines are used, and a 16 percentage point decline when the lower international extreme poverty 1 The sections are: household information (section 1), housing (section 2) and consumption/expenditure on food, line is used. Several non-monetary indicators non-food and durables (section 4). also suggest substantial improvement in living 2 The TLSLS-3 added two new items (mobile phone cards/credit and bottled water) and one item was split into two categories (`Prepared food standards over time. TLSLS data also confirm and drinks’ has become `Foods and drinks prepared and consumed outside the house’ and `Foods and drinks prepared outside and brought that consumption inequality has remained largely to be consumed at home’). The TLSLS-3 also introduced a single reference period for non-food consumption: each non-food item was unchanged over this period. associated with the last month, the last 3 months or the last 12 months rather than having each non-food item associated with two reference periods –the last month and the last 12 months– as it was the case in the TLSLS-2. These changes are relatively minor and should not raise any significant concerns for the comparability of the consumption aggregate. POVERTY MEASUREMENT METHODOLOGY I: CONSUMPTION-BASED WELFARE INDICATOR Poverty in Timor-Leste 2014 the TLSLS 2007. The two surveys followed highly The main methodological consideration in other developing countries with large informal comparable fieldwork protocols, even though constructing new estimates of poverty with the sectors, in Timor Leste, consumption tends to the TLSLS-3 canvassed a substantially larger TLSLS-3 data is to construct estimates that be measured more accurately than income in sample. Further details on TLSLS-3 are provided are comparable with the 2007 TLSLS-2 poverty household surveys. This is largely due to the in Annex A. estimates and are consistent across space. This difficulties in defining and measuring income for in turn implies considerations relating to (a) the self-employed who account for a relatively Using these new data, this report presents using consumption as the welfare indicator, (b) large proportion of the work force. comparable estimates of poverty. The primary constructing comparable estimates of nominal focus is on poverty measured in terms of consumption, and (c) constructing a set of poverty As in the TLSLS-2 poverty estimates, per capita 10 household consumption expenditure, an lines for 2014 that reflect, as far as possible, the household consumption is used as the basic 11 important indicator of wellbeing. The construction same standard of living as the poverty lines for measure of individual welfare. While this measure of this consumption-based poverty measures is 2007. Section 2 covers the first two considerations, does not incorporate some important aspects of Poverty in Timor-Leste 2014 discussed in Sections 2 through 5. and the latter will be covered in Section 3. individual welfare, such as consumption of public goods (for example, schools, health services, Of course, consumption poverty provides only public sewage facilities), it is a useful aggregate a partial window on deprivation and well- CONSUMPTION AS THE WELFARE INDICATOR money metric of welfare that reflects individual being of the population. So this assessment is The decision to use total consumption expenditure preferences conditional on prices and incomes, supplemented with a further look at progress in (including some imputed expenditures as and for that reason, is widely used in welfare other “non-income” dimensions of welfare. The discussed below) rather than income as the assessment and poverty monitoring. last section of this report presents estimates for measure of individual welfare is motivated by several such non-income indicators as building two main considerations. First, consumption is blocks for an analysis of multidimensional poverty arguably a more appropriate indicator if we are CONSTRUCTING COMPARABLE NOMINAL in Timor-Leste. Finally, Annex F suggests options concerned with realized, rather than potential CONSUMPTION for future work on constructing multidimensional welfare, since not all income is consumed, nor all Having selected consumption as the measure of poverty measures for the country. consumption financed out of income. Individuals welfare, the first task in constructing comparable use savings and credit to smooth fluctuations poverty measures is to construct comparable in income and therefore consumption provides estimates of nominal consumption for every a more accurate measure of an individual’s household. Household nominal consumption has welfare over time. Second, similar to many three components: (i) food, (ii) rent as the value of housing services consumed by the household, of household consumption in 2007 (World Bank, and (iii) other non-food goods and services. 2008). A similar model is estimated now with TLSLS-3 data to calculate the rental component The food and non-food components are directly of household consumption in 2014. The estimated POVERTY MEASUREMENT estimated from the survey data based on the rental models for 2007 and 2014 are shown in reported value of the food and non-food items Annex C. METHODOLOGY II: POVERTY LINES consumed. This follows the same procedures as in the TLSLS-2 survey (see Annex B for details). CHANGES IN NOMINAL CONSUMPTION However, most houses in Timor-Leste are owner EXPENDITURE 2007-2014 occupied and the rental market in the country Figure 1 shows the growth in nominal is thin. Hence reported rent in the survey is not consumption per capita between 2007 (TLSLS-2) Poverty in Timor-Leste 2014 DISTRICT-LEVEL POVERTY LINES lines for 2014 to the same six domains for which actual rent, but respondents-estimated rent and and 2014 (TLSLS-3). It is notable that at least in In the case of the TLSLS-2, poverty lines were poverty lines were constructed for 2007. The this is subject to measurement errors. For this nominal terms, rent has been the fastest growing constructed for six domains: the rural and TLSLS-3, on the hand, has a 33 percent larger reason, information on estimated rents is not used component of consumption and food has been urban segments of three regions (Table 1) as the sample size so that poverty statistics can be directly. Instead, when constructing the rental the slowest component. Correspondingly, food TLSLS-2 sample size permitted only this degree disaggregated at the district level. Statistics at component, actual rents are used whenever budget shares have declined over the two survey of spatial disaggregation of the poverty lines. the district level have greater policy relevance available, and predicted (imputed) rents are periods, which in view of Engel’s law (income because districts are the key administrative used otherwise. These predictions are obtained elasticity of food being typically less than one) TABLE 1: REGIONS, DOMAINS AND DISTRICTS units. Given that poverty statistics at the district from a hedonic rental model that estimates the is suggestive of improvements in the standards level are best constructed with poverty lines relationship between reported rental values and of living. The substantial increase in nominal Regions Domains Districts 12 determined at the district level, a continuation 13 a number of observable dwelling characteristics rents and the rental share of consumption also EAST East Urban Baucau, Lautem of the legacy of the TLSLS-2 of six domains for (number of rooms, building materials used etc.). point to the need for more attention to the rental and East Rural and Viqueque determining poverty lines appears now both Poverty in Timor-Leste 2014 Such a model had earlier been estimated with the component of the poverty lines. CENTRE Centre Urban Aileu, Ainaro, Dili, undesirable as well as unnecessary. Hence, TLSLS-2 data to evaluate the rental component and Centre Rural Ermera, Liquica, exploiting the larger sample size of the TLSLS-3, Manufahi, Manututo which is representative at the district level, WEST West Urban Bobonaro, Cova Lima poverty lines in 2014 are estimated separately for the 13 districts. % and West Rural and Oecussi 600 The minimum sample size in the TLSLS-3 500 Given the above, two main approaches can be amongst the 13 districts was 254 households in 400 considered for the construction of poverty lines Aileu district and the median sample size was 419 Rent for the TLSLS-3 in 2014: (i) updating the 2007 households. These sample sizes are somewhat FIGURE 1: 300 poverty lines for the six domains using estimates lower than corresponding sizes for the six GROWTH IN NOMINAL Other nonfood of changes in the cost of living for the six domains, domains for 2007 (Table 2), but nonetheless offer CONSUMPTION BY 200 CENTILE, 2007-2014 Total or (ii) constructing a new set of poverty lines using an acceptable level of precision for the estimation 100 (PERCENT INCREASE) TLSLS-3 data and achieving comparability by of poverty line. This new opportunity for further Food 0 using the same methodology as in 2007. spatial disaggregation is the primary motivation Source: 0 10 20 30 40 50 60 70 80 90 100 Updating the poverty lines is restrictive in that it for a move to district-level poverty lines with the TLSLS 2007 limits the spatial disaggregation of the poverty TLSLS-3. Percentile of total nominal comsuption per person and TLSLS 2014. TABLE 2: COMPARING MINIMUM SAMPLE SIZES determined as the average (per capita) quantities the rent poverty lines by domain are calculated of food items consumed by households belonging by dividing the predicted cost of the reference TLSLS-3 TLSLS-2 to the reference group of the poor who live in that dwelling in each domain by the corresponding particular domain. average household size of the poor in Total sample size 5,916 households from 400 PSUs 4,477 households from 300 PSUs that domain . 3 Minimum sample size 254 households in Aileu district 375 households in East Urban domain The domain-specific average food bundles of in district or domain the poor are scaled up (or down) to yield the As the procedure involves making predictions Median sample size 419 households 695 households recommended 2,100 calories per person per day. over samples for two periods, a parsimonious per district or domain The scaled bundles are then valued using median specification with only six dwelling characteristics prices (unit-values) of food items paid by the poor is used4. The reference dwelling for the rent in each domain to obtain the food poverty line for poverty lines is assumed to have 2 rooms, good that domain. external walls, proper sanitation and access to Poverty in Timor-Leste 2014 It is also worth noting that developing district-level The comparability of district-level poverty lines in poverty lines is a more forward-looking approach. 2014 with 2007 is achieved by using exactly the electricity. The estimated models are shown in It is reasonable to presume that the sample size same approach to the construction of poverty Annex C. for future rounds of the TLSLS will grow. Thus, lines in 2014 as in 2007, although at a lower Rent poverty line it will be increasingly inappropriate and less level of aggregation (i.e., for 13 districts for 2014 The rent poverty lines represent the average defensible to continue with the framework of six relative to the six domains in 2007). imputed rental cost per person of a reference Non-food (excluding rent) poverty line spatial domains inherited from the 2007 TLSLS for dwelling in each domain. These lines are The non-food (excluding rent)5 poverty lines are the future. Establishing a new baseline of district- Poverty lines for both years are determined using constructed using a hedonic rental model estimated in terms of what the poor actually level poverty lines now will assist in monitoring the cost of basic needs approach (Ravallion 2008). where the actual or estimated rents reported spend on non-food items. For any given domain, 14 district-level trends in poverty in the future. This method effectively calculates the poverty by households are modelled as a function of the non-food poverty line corresponds to the 15 line as the cost of a consumption bundle that is a number of the dwelling characteristics and average per capita non-food consumption of the Before describing the new approach in detail, it is (i) consistent with the consumption pattern of domain fixed effects. This is the same model as population whose actual combined per capita Poverty in Timor-Leste 2014 also worth noting that in moving to the district as the poor and (ii) deemed adequate for meeting that used for estimating the rental component of food and rent consumption is within plus/minus the level of disaggregation for poverty lines, we, basic needs. The poverty line has three main household consumption as discussed in section 5% of the sum of the food and rent poverty lines in the process, lose the urban-rural split. The components: food, rent and non-food. 2.1 above. The model uses similar specifications for that domain. TLSLS-3 sample size is simply not large enough for 2007 and 2014, the only difference between to disaggregate by both district and urban- the two years is that the fixed effects refer to the rural segments. However, this does not imply Food poverty line six domains for 2007, while they refer to the 13 Overall poverty line that urban-rural cost of living differentials are The food poverty line is anchored to the districts for 2014. The overall poverty line for a domain is the sum of totally ignored under this approach. To the extent recommended nutritional norm of 2,100 calories The estimated parameters are then used to derive the food poverty line, the rent poverty line and the districts differ in their degree of “urbanity”, the per person. For each of the six domains in the cost of a reference dwelling that is kept fixed non-food poverty line for that domain. district-specific poverty lines will build in cost 2007 and for each of the 13 districts in 2014, across domains and over the two surveys and of living differentials due to higher or lower representative food bundles for the poor are representation of urban areas across districts. constructed to correspond to the average food For instance, a higher poverty line for Dili will consumption pattern of the poor in that domain. reflect, in part, the higher urban cost of living for A national reference group representing the poor 3 The average household size of the poor by domain is estimated taking into account only households that belong to the same national its largely urban population. is identified, and the food bundle for a particular reference group of poor households used for the estimation of the food poverty lines. domain (6 in 2007 and 13 in 2014) is then 4 A parsimonious specification helps ward against “out-of-sample” forecasting errors that may result from the inclusion of variables that are only marginally significant or insignificant in one of the two periods. 5 Hereafter, non-food always refers to remaining non-food excluding rent. There is one practical issue in implementing the using rescaled bundles and median prices than 5%. For both 2007 and 2014, it took only the rent poverty line is estimated by valuing the above procedure because it relies on the initial from steps (2) and (3). two iterations for the poverty lines to converge rental cost of a fixed reference dwelling at the identification of a reference group representative 3. Generate the value of the rent poverty lines by to the final estimates. two dates. As noted before, rent has been the of the poor to determine the poverty lines. Yet, domain by dividing the predicted cost of a fixed fastest growing component of nominal household without the poverty lines, we do not know who reference dwelling in each area by the average 7 The estimated poverty lines consumption. Between 2007 and 2014, the the poor are. To get around this circularity, an household size of the households that belong The poverty lines resulting from the application average value of rent per person tripled in nominal iterative method is used. In the first iteration, the to the reference group of the poor. of the methodology set out above are shown in terms. By comparison, the rental valuation of the reference group for determining the food poverty 4. Estimate the remaining non-food component TABLE 3 for 2007 for the 6 domains and TABLE 4 reference dwelling rose two-and-a-half times, line is taken to be the bottom 40% of the national of the poverty line using non-parametric Engel for 2014 for the 13 districts. thus still indicating a substantial increase in the population ranked according to constant-prices functions for each domain: rental component of consumption in real terms. per capita consumption. 6 The 40% cutoff is a. Estimate the non-parametric relationship The estimates in Tables 3 and 4 indicate that while However, despite the large increase in the rent chosen based on the TLSLS 2001 estimate of 40% between non-food consumption per person poverty line, its contribution to the increase in the Poverty in Timor-Leste 2014 there is variation across the six spatial domains poverty incidence (World Bank, 2003). From the and total consumption per person using a and across districts, on average, the new food total poverty line was limited by the initial (2007) second iteration onwards, the reference group locally weighted regression. poverty line for 2014 increased by 62.4% since low share of the rental component of the poverty of the poor is the population consuming below b. Predict non-food consumption per person 2007, which is comparable with the increase in line of about 18%. the level of the poverty line developed in the from the non-parametric regression. food CPI of 69.5%. However, the average total previous iteration. c. Derive predicted food and rent consumption poverty line for 2014 rose by 84.5%, which is The estimates in Tables 3 and 4 indicate also that as the difference between actual total appreciably higher than the total CPI increase the cost of living for the poor has risen faster The iterative method thus involves the following consumption per person and predicted of 66.5%. The estimated poverty lines are thus in the Eastern districts than in the Centre or series of steps: non-food consumption per person. indicative of the cost of basic non-food having the West. 1. Use the temporal price indices to express all d. Derive the non-food poverty line as the average risen significantly more than the increase implied 16 17 consumption values in constant prices. Identify predicted non-food consumption of the by the non-food component of the CPI. the reference group of the poor in the first iteration population whose predicted food and rent An important factor behind this is the large Poverty in Timor-Leste 2014 as the bottom 40% of the national sample ranked consumption lies within plus/minus 5% of increase in the rent poverty line. Recall that by consumption per person at constant-prices. the sum of the food and rent poverty lines. 2. Generate food poverty lines by domain: 5. Derive the first iteration total poverty line for a. Estimate the quantities of food items any domain as the sum of the food, the rent and TABLE 3: POVERTY LINES PER PERSON PER MONTH, 2007 consumed per person per day amongst the the non-food poverty lines for that domain. Food Rent Other Total reference group by domain. 6. Calculate the poverty incidence with the first b. Rescale to ensure that the rescaled bundle iteration total poverty lines, and those identified East rural 11.10 3.25 4.07 18.42 for each domain provides 2,100 calories per as poor will serve as the reference group for East urban 13.02 2.87 4.95 20.84 person per day. the next iteration, which repeats step (2) to 16.55 Centre rural 4.54 4.50 25.59 c. Estimate median prices for food items (5). The algorithm stops when the average 18.89 Centre urban 5.71 6.33 30.92 amongst the reference group by domain. of the absolute percentage change in the total d. Generate food poverty lines by domain by poverty lines over the previous iteration is less West rural 14.38 4.97 6.10 25.44 West urban 15.92 4.98 6.99 27.88 15.40 Timor-Leste 4.57 5.17 25.14 6 Constant prices imply monetary values deflated over time by the temporal price index, but not spatially. Note: At average prices of January 2007/January 2008 of each domain. 7 The cost of a fixed reference dwelling in each area is estimated before the iterative method starts and hence does not change Source: TLSLS 2014 over iterations TABLE 4: POVERTY LINES PER PERSON PER MONTH, 2014 Food Rent Other Total POVERTY ESTIMATES EAST Baucau 24.86 10.64 7.83 43.33 Lautem 21.29 12.94 7.34 41.57 Viqueque 24.06 12.37 8.07 44.51 CENTRE Poverty in Timor-Leste 2014 POVERTY INDICES the headcount index has some well-known Ainaro 24.38 11.15 8.75 44.28 Using per capita consumption as the measure of limitations. It does not take into account how Aileu 23.89 8.80 8.45 41.14 individual welfare, members of a household are close or far the average consumption levels Dili 29.07 12.64 14.45 56.16 considered poor if the per capita consumption of the poor are relative to the poverty line. Nor Ermera 21.74 8.82 7.42 37.97 of the household is below the poverty line. The is it sensitive to the distribution of consumption poverty line is “absolute” in the sense that it fixes amongst the poor. Liquiça 23.00 8.57 6.22 37.79 a given welfare level, or standard of living, over Manufahi 25.57 11.79 10.51 47.87 the domains of analysis. Three poverty indices Manatuto 24.16 11.15 8.51 43.81 within the Foster, Greer and Thorbecke (1984) POVERTY GAP INDEX 18 class of poverty measures are evaluated. This α The poverty gap index ( =1) is the average 19 family of measures can be written as: consumption shortfall of the population relative WEST to the poverty line. The shortfall is expressed as a Poverty in Timor-Leste 2014 Bobonaro 24.83 9.73 9.89 44.45 1 $ z − yi ' q α proportion of the poverty line and the non-poor are Covalima 25.66 11.68 12.11 49.45 Pα = ∑ & ) assigned a zero shortfall. The poverty gap index Oecussi 24.05 17.69 9.37 51.12 n i=1 % z ( takes into account the average consumption level of the poor and is often referred to as a measure Timor-Leste 25.01 11.50 9.86 46.37 where is some non-negative parameter, z is the of the depth of poverty. poverty line, y denotes consumption, i represents individuals, n is the total number of individuals in Note: At average prices of April 2014/April 2015 of each district. € the population, and q is the number of individuals SQUARED POVERTY GAP INDEX Source: TLSLS 2014 with consumption below the poverty line. Finally, the squared poverty gap index ( =2) α is sensitive not only to the mean consumption shortfall of the poor relative to poverty line, but HEADCOUNT INDEX also to the distribution of consumption amongst α The headcount index ( =0) measures the the poor. In contrast to the poverty gap index, percentage of the population whose consumption which gives equal weight to the consumption is below the poverty line. This index is the most shortfalls of all the poor, the squared poverty widely used poverty measure mainly because it gap index assigns higher weights to the larger is very simple and easy to interpret. However, poverty gaps, thus making the measure sensitive to transfers amongst the poor. The squared poverty gap index is also referred to as a measure of the severity TABLE 6: POVERTY INDICES, 2007 AND 2014 of poverty. 8 Headcount Poverty gap Squared poverty gap (Incidence) (Depth) (Severity) RESULTS 2007 2014 2007 2014 2007 2014 *** Timor-Leste 50.4 41.8 *** 13.8 10.4 *** 5.1 3.7 Average consumption and inequality Table 5 shows real consumption expenditure per person and the Gini index of inequality at the national level, and disaggregated for urban and rural areas. Nationally, average real consumption per person grew by a Rural *** 54.7 47.1 *** 15.5 12.2 *** 5.9 4.4 little over 10% during 2007-14 to about $2 per person per day in 2014. Though modest, this growth in mean Urban 38.3 28.3 *** *** 9.0 5.9 *** 3.0 1.8 consumption enabled the poverty reduction observed over this period. Also, there was no significant change in inequality (as measured by the Gini index) between 2007 and 2014. East 31.6 33.8 6.0 7.1 1.7 2.1 Poverty in Timor-Leste 2014 Centre *** 54.6 40.0 *** 15.4 9.8 *** 5.8 3.4 West 60.3 55.5 18.2 15.8 7.2 6.2 TABLE 5: REAL PER CAPITA CONSUMPTION AND INEQUALITY, 2007 AND 2014 East rural 32.2 36.0 6.1 7.7 1.7 2.3 East urban 25.4 21.0 5.1 3.8 1.5 1.0 CONSUMPTION PER PERSON GINI COEFFICIENT ($/person/month) *** Centre rural 64.4 48.3 *** 19.4 12.6 *** 7.6 4.5 Centre urban 39.1 26.4 *** *** 9.2 5.3 *** 3.1 1.6 2007 2014 2007 2014 West rural 62.7 57.6 19.2 16.8 7.7 6.7 Timor-Leste 53.9 60.1 0.28 0.29 20 West urban 42.5 46.0 10.6 11.5 3.7 3.8 21 Rural 50.3 55.5 0.26 0.27 Urban 64.2 71.7 0.29 0.29 Note: *, ** and *** indicate significantly different from 2007 at 10, 5 and 1 per cent, respectively. Poverty in Timor-Leste 2014 Source: TLSLS 2007 and TLSLS 2014. Note: Consumption per person at average national prices of April 2014/April 2015. Source: TLSLS 2007 and TLSLS 2014. The estimates in TABLE 6 indicate a significant District-level poverty estimates decline in poverty between 2007 and 2014. The TABLE 7 shows the district profile of poverty proportion of population below the poverty line together with average real consumption and Gini Poverty estimates: national, sectoral and regional declined from about 50% to 42%. There was also inequality indices for 2014. Table 6 reports the new estimates of poverty for 2014 at the national level, and disaggregated by urban and a significant fall in the poverty gap and squared rural areas, together with comparative estimates for 2007. poverty gap measures, indicating that welfare There is considerable variation in levels of poverty improvements are not just limited to those near rates across the 13 districts. For instance, the the poverty line. Poverty fell in both rural and headcount index ranges from 29% for Dili to urban areas, though the decline is larger in the 63% for Oecussi. More generally, with some urban sector. Across regions, the largest decline exceptions, poverty levels are lower in eastern 8 These measures satisfy some useful properties. For instance, all three poverty measures are subgroup decomposable in that the aggre in poverty is witnessed for the Central region. districts and higher in western districts, with the gate poverty measure equals the population-weighted average of the subgroup poverty measures. The poverty gap and the squared Poverty also declined in the West, while there central districts in the middle. (Annex D shows the poverty gap measures satisfy the monotonicity axiom, which requires that a welfare reduction for a poor household should cause mea sured poverty to increase. Finally, the squared poverty gap measure also satisfies with the transfer axiom, which requires that a re was a modest (though not statistically significant) 2014 poverty estimates along with their standard gressive transfer from a poor to a richer person should lead to an increase in measured poverty. Sen (1976) proposed the monotonicity increase in poverty in the East driven by the rise errors and confidence intervals.) and transfer axioms. For a discussion of these and other properties of poverty measures, also see Foster (2005). in rural poverty. Disparities in mean consumption levels largely both have relatively high and very similar Gini mirror those in the poverty rates. There is also indices of 0.30 and 0.31 respectively. In particular, some variation in inequality indices across the high level of poverty in Oecussi seems to be districts, although there does not seem to be a the product of low average consumption as well POVERTY REDUCTION: TIMOR-LESTE systematic relationship between average living as high levels of inequality. standards and level of inequality. For instance, IN THE INTERNATIONAL CONTEXT while Dili has the highest mean consumption (of $72) and Oecussi has the lowest (of $49), they TABLE 7: POVERTY INDICES AND REAL PER CAPITA CONSUMPTION BY DISTRICT, 2014 Poverty in Timor-Leste 2014 POVERTY INDICES Poverty Squared Consumption Table 8 juxtaposes the rate of poverty reduction line of $1.90 at 2011 Purchasing Power Parity Headcount gap poverty gap per person Gini (Incidence) (Depth) (Severity) ($/month) coefficient in Timor-Leste over 2007-2014 with rates of prices (PPP). poverty reduction observed in selected g7+ Timor-Leste 41.8 10.4 3.7 60.1 0.29 conflict-affected countries, CPLP countries as The 2011 PPP exchange rate for most countries well as some key Asian countries for which data is were estimated from International Comparison EAST available over a similar time period. Program (ICP) price surveys conducted in 2011. Baucau 32.6 6.8 2.0 63.0 0.25 As this survey was not undertaken in Timor Leste, Lautem 32.2 6.8 2.1 64.7 0.28 Because national poverty thresholds differ across PPP exchange rate for the country was “estimated” 22 23 Viqueque 36.9 7.8 2.4 61.7 0.26 countries, the comparison is made in terms of to be 0.56, implying that $0.56 in Timor Leste CENTRE achievements in poverty reduction using the had the same purchasing power as $1.00 in the Poverty in Timor-Leste 2014 Aileu 35.1 8.1 2.7 59.8 0.24 internationally comparable extreme poverty US in 2011. As a result, the PPP for Timor Leste Ainaro 43.2 9.4 3.0 58.6 0.26 Dili 29.1 6.0 1.9 72.0 0.30 Ermera 56.7 17.1 6.6 52.9 0.31 BOX 2 Liquiça 43.0 11.7 4.4 54.7 0.26 NATIONAL VERSUS INTERNATIONAL POVERTY LINE Manufahi 47.7 11.1 3.6 54.1 0.24 NATIONAL POVERTY LINE Manatuto 43.1 9.2 2.9 59.8 0.26 The national poverty line presented in this report is the most relevant measure of what it means WEST to be poor in Timor-Leste and is based on consumption patterns and prices prevailing in Timor- Bobonaro 51.7 12.6 4.4 53.0 0.26 Leste. The national poverty line is the most useful threshold for monitoring national poverty and for national policy making. Cova Lima 53.1 15.9 6.8 50.2 0.27 Oecussi 62.5 19.8 7.9 49.3 0.31 INTERNATIONAL EXTREME POVERTY LINE Given that the national poverty line varies from country to country, it is difficult to compare national poverty results internationally. Hence an international line which is based on the average national poverty lines of some of the poorest countries in the world and equivalent to $1.90 a day at 2011 Note: Consumption per person at average national prices of April 2014/April 2015. Purchasing Power Parity prices is used. This is equivalent to $40.45 per person per month at 2014 Source: TLSLS 2014. Timor-Leste prices, $5.92 less than the Timor Leste national poverty line. is less reliable than for other countries where Table 8 shows that poverty in Timor-Leste declined actual price surveys were conducted. Using the 2011 PPP of 0.56 and after adjusting for inflation by 16.9 percentage points, from 47.2% in 2007 to 30.3% in 2014. The rate of poverty reduction in DISTRICT-LEVEL POVERTY RATES USING THE INTERNATIONAL POVERTY LINE BOX 3 between 2011 and 2014, the international poverty Timor-Leste took place at a more rapid rate than line in 2014 Timor Leste prices is equivalent to in Haiti, Sierra Leone, and Togo (among the g7+ In order to estimate district level prevalence of poverty using the international $1.90 (2011 PPP) line, district level household expenditures were adjusted to national level prices using the ratio $1.33 per person or $40.45 per person per month. countries) as well as China and Indonesia. between the district poverty lines and the national poverty line, reflective of the price difference This is considerably lower than the 2014 national across districts. District level international poverty rates are not computed for 2007 for the same poverty line of $46.37 per person per month. It reason district level national poverty rates are not computed for 2007: the TLSLS2 sample size does should therefore be noted that the international yield reliable estimates at the district level. line has less firm grounding than the national poverty line in the basic need requirements in The table and chart below provides poverty rates by districts using the $1.9 (2011 PPP) line. Note Timor Leste. In fact the minimum living standard that while poverty rates obviously fall when using international line, its distribution across districts Poverty in Timor-Leste 2014 remains almost the same (more apparent in the chart). is lowered when moving from the national poverty line to the international poverty line. Poverty rate using the Poverty rate using the District national poverty line international poverty line Aileu 35.1 23.7 TABLE 8: POVERTY REDUCTION IN SELECTED COUNTRIES Ainaro 43.2 28.9 Baucau 32.6 20.8 Bobonaro 51.7 36.2 Rate of poverty decline at $1.90 Cova Lima 53.1 41.8 (2011 PPP) poverty line Dili 29.1 18.9 24 Selected countries Period Average percentage points per year Ermera 56.7 46.8 25 Selected g7+ countries Lautem 32.2 20.8 Liquiça 43 32.9 Chad 2003-2011 3.1 Poverty in Timor-Leste 2014 Manatuto 43.1 34.8 Congo 2004-2012 1.9 Manufahi 47.7 31.1 Haiti 2001-2012 0.2 Oecussi 62.5 54.0 Viqueque 36.9 24.3 Sierra Leone 2003-2011 0.8 Timor Leste 41.8 30.3 Togo 2006-2011 0.3 Timor Leste 2007-2014 2.4 POVERTY RATES Selected CPLP countries 70 Angola 2000-2008 0.3 60 Mozambique 2002-2008 1.9 50 National poverty line Cabo Verde 2001-2007 1.8 40 Others Intenational poverty line 30 China 2002-2010 1.5 20 Indonesia 2005-2010 0.9 10 India 2004-2011 2.5 0 u ro bo u va ro a li a em an a M uto Oe hi Vi ssi ue m er iç le a Di a na na uc eq cu qu uf Li ut Ai at m Ai Ba qu an La Er Li Source: TLSLS 2014 and World Bank. Bo M Co Source: TLSLS 2014 E R SENSITIVITY OF POVERTY INCIDENCE Poverty in Timor-Leste 2014 This section investigates the robustness of the poverty measures (the poverty gap index and the poverty estimates presented above to different squared poverty gap index) are similar. methodological choices. The first subsection T I M O R deals with robustness with respect to correcting for household members’ age and sex in calorie CALORIE REQUIREMENTS requirements, the second subsection looks Poverty lines in 2007 and in 2014 were constructed at changes in the poverty estimates due to using a caloric norm of 2,100 calories per person the inclusion of expenses on festivities and per day. One may, however, want to allow for ceremonies in total household consumption, and differences in caloric requirements by age and 26 the third subsection assesses robustness against sex. This issue is investigated by replacing the 27 potential biases due to fieldwork team effects. For uniform per capita calorie requirement with a set brevity, this section presents results pertaining of age-gender-specific requirements, as shown Poverty in Timor-Leste 2014 to the measure of poverty incidence only (the in Table 9. L E S T headcount index); however results for the other TABLE 9: DAILY CALORIE REQUIREMENTS BY AGE AND SEX Age (years) Male Female Less than 1 584 584 1 to 3 1060 1060 4 to 6 1350 1350 7 to 9 1690 1690 10 to 12 2190 2010 13 to 15 2750 2330 16 to 17 3020 2440 18 or more 2730 2230 Source: TLSLS 2014. The calorie norms in Table 9 are drawn from the recommendations of the National Institute of Nutrition FESTIVITIES AND CEREMONIES their lump sum nature), which can distort the and the Indian Council of Medical Research (NIN-ICMR, 2010). Similar recommended caloric intakes are not The second piece of sensitivity analysis deals with “usual” consumption and poverty status of available for Timor-Leste, but the NIN-ICMR norms offer a reasonable benchmark to investigate potential spending on festivities and ceremonies, which households reporting such expenses. However, sensitivity to equivalence scales. includes expenses on marriages, births, funerals, since Timor-Leste is a country where festivities festivals and similar events. Conceptually, these and ceremonies are a more frequent part of the Table 10 (middle panel) reports the results of this exercise for poverty incidence (headcount index). As 9 expenses are part of consumption, but previous social and cultural tradition, it is worth exploring indicated in the table, the poverty lines underlying these estimates are based on a norm of 2,730 calories per poverty analyses excluded them largely for whether their inclusion in household consumption male adult equivalent for both 2007 and 2014. This translates into a requirement of 2,100 calories per person two related practical considerations: (a) such makes a difference. for 2007 and a slightly higher requirement of 2,168 calories per person for 2014, reflecting a change in the expenses are incurred infrequently, so even underlying age-sex distribution of the population. the longest recall period in the survey (the last Table 10 (right panel) reports the relevant results, 12 months) may not adequately reflect their which show that this makes little difference to occurrence for many households, and (b) these either the levels or changes in poverty incidence.10 Poverty in Timor-Leste 2014 TABLE 10: SENSITIVITY OF POVERTY INCIDENCE, 2007, 2014 expenditures are often rather large (reflecting Compared with the benchmark estimates of a Current estimates Equivalence scales Including festivities 2007 2014 2007 2014 2007 2014 TABLE 11: DISTRIBUTION OF 2014 SAMPLE BY DISTRICT AND TEAM Timor-Leste 50.4 41.8 50.4 44.6 50.1 41.6 Team District Total 1 2 3 4 5 6 7 8 Rural 54.7 47.1 54.7 50.5 54.3 46.8 Urban 38.3 28.3 38.3 29.8 38.5 28.2 Ainaro 30 14 0 30 0 60 120 120 374 Poverty lines * 25.1 46.4 25.1 47.7 26.5 50.1 Aileu 15 45 59 75 60 0 0 0 254 28 29 Food 15.4 25.0 15.4 26.0 15.5 25.5 Baucau 0 0 0 0 0 210 194 180 584 Rent 4.6 11.5 4.6 11.6 4.6 11.6 Bobonaro 45 120 119 72 149 0 0 0 505 Poverty in Timor-Leste 2014 Nonfood 5.2 9.9 5.2 10.2 6.4 12.9 Covalima 45 90 76 104 104 0 0 0 419 kcal/day/person 2,100 2,100 2,100 2,168 2,100 2,100 Dili 120 201 257 261 0 12 0 0 851 kcal/day/adult male - - 2,730 2,730 - - Ermera 45 60 73 59 177 60 0 0 474 Liquiça 30 45 45 29 248 0 0 0 397 * Poverty lines per person per month at average national prices of each year. Source: TLSLS 2007 and TLSLS 2014. Lautem 0 0 0 0 0 100 150 148 398 Manufahi 45 60 60 43 0 60 30 60 358 The resulting poverty lines for 2014 are slightly higher than the estimates based on the uniform caloric norm. Manatuto 0 15 37 60 0 90 97 88 387 The estimated poverty incidence for 2014 is about 3 percentage points higher, though the decline in poverty 375 Oecussi 90 0 0 0 0 0 0 465 relative to 2007 is maintained. Viqueque 0 0 0 0 0 150 150 150 450 Total 750 740 726 733 738 742 741 746 5,916 Source: TLSLS 2014. 10 Though not reported, the results are similar for the poverty gap and squared poverty gap indices. 9 The results for poverty gap and squared poverty gap measures are similar. decline from 50.4% in 2007 to 41.8% in 2014, FIELDWORK TEAM EFFECTS For a sufficiently large total sample size, the team’s subsample are not statistically significant. these estimates indicate a decline from 50.1% The TLSLS-3 deployed 8 teams to carry out the households surveyed by each team could be The only exception is in the case of team 3, for to 41.6%. The main reason for this small change entire fieldwork for the yearlong survey. All teams considered an independent random subsample of which the national poverty estimate is significantly is easy to appreciate. The inclusion of festivities received the same centralized training prior to the overall sample. However, in smaller samples, different at the 5% level and the estimate for and ceremonies certainly increases non-food and the launch of fieldwork. As noted in Annex A, this would only be approximately so, thus raising Bobonaro is significant at the 10% level. total consumption of households, but it also raises for quality assurance purposes, the sample was the possibility of team effects in finite samples. the allowance for the non-food component of the randomly distributed across quarters, districts Could this bias district-level poverty estimates? As there is just one exception involving a poverty line. Recall that the latter is estimated as and teams. Thus, the sample of each district, with We investigate this by conducting the following significant difference between the full sample the non-food (excluding rent) expenditure of the the exception of Oecussi, was randomly allocated experiment for each team. We consider ignoring and reweighted subsample estimates, the population whose food and rent expenditure is in to at least three different teams of fieldworkers the subsample surveyed by one particular team, benchmark poverty measures based on the full the neighbourhood of the food and rent poverty (interpenetrating sampling) whose fieldwork say team j. Thus, the households surveyed by sample can be considered to be robust overall lines. Thus, effectively, the methodology used for was spread over the four quarters. Oecussi was team j are assigned a sampling weight of zero, and to the presence of fieldwork team effects. As in Poverty in Timor-Leste 2014 the estimation of poverty lines offers an element surveyed by two teams, with a different team correspondingly, for each district team j was active the case of consumption including festivities and of built-in stability to the poverty estimates, visiting each quarter. The distribution of the final in, the sampling weights of households surveyed ceremonies (section 6.1), this is on account of the which is also reflected in the results reported in sample by team and district is shown in Table 11. by other teams are increased to achieve the same robustness of the underlying poverty estimation Table 10. total population for the district. Consumption, methodology, which recalibrates the poverty poverty lines and poverty measures are then lines in accordance with measured consumption. recalculated for such a reweighted sample, As shown in the detailed results in Annex E, any TABLE 12: SENSITIVITY OF POVERTY INCIDENCE TO EXCLUSION OF TEAMS and then we test for any statistically significant understatement or overstatement of nominal difference with respect to the benchmark poverty consumption due to potential team effects is All When excluding team estimates using the full sample for all teams. The compensated by a recalibration of the district 30 teams 1 2 3 4 5 6 7 8 31 presence of statistically significant differences poverty lines in the same direction. would be suggestive of finite sample biases due 41.8 41.2 42.2 46.2 ** 44.3 44.5 40.3 40.5 40.3 Timor-Leste Poverty in Timor-Leste 2014 to team effects. The experiment is repeated for Ainaro 43.2 44.6 44.3 50.0 44.6 45.3 45.3 45.9 37.9 each team j=1,2…8. The results are summarized Aileu 35.1 34.0 34.0 44.8 35.4 35.9 33.4 32.7 34.3 in Table 12.11 Baucau 32.6 34.0 34.0 34.0 34.6 35.1 34.2 33.8 29.7 Bobonaro 51.7 50.0 42.4 64.3 * 49.6 60.1 51.1 45.2 51.2 The first column of Table 12 reproduces the Covalima 53.1 52.3 55.6 53.3 55.0 52.9 51.8 51.5 51.9 poverty incidence estimates for the full sample Dili 29.1 23.6 30.4 32.7 34.8 30.5 24.7 26.4 27.8 as in Box 3. The remaining columns report the Ermera 56.7 56.4 58.4 60.3 61.2 57.2 57.6 55.5 55.5 poverty estimates when the subsample for one Liquiça 43.0 39.8 43.4 45.7 48.2 49.0 42.8 40.6 41.3 of the 8 teams is excluded with reweighting of the remaining sample. The results show that in Lautem 32.2 34.0 36.6 36.8 32.2 36.0 31.7 37.7 28.4 almost all cases the differences between the full Manufahi 47.7 46.2 49.4 52.5 47.6 50.3 47.3 49.6 45.7 sample poverty estimate (at both the national and Manatuto 43.1 45.1 44.0 46.0 47.3 46.2 34.6 40.7 41.4 district levels) and those excluding a particular Oecussi 62.5 72.4 60.4 64.5 62.8 64.3 62.0 60.3 61.6 Viqueque 36.9 37.8 41.0 42.9 38.7 41.4 32.8 35.0 39.2 Note: *, ** and *** indicate a significant difference from the full sample (“All teams”) estimate at 10, 5 and 1 per cent, respectively. Source: TLSLS 2014. 11 The full set of results of excluding teams is shown in Annex E. TABLE 13: DISTRIBUTION OF POOR BY AGE AND GENDER HOUSEHOLD CHARACTERISTICS AGE PERCENTAGE OF THE POOR POVERTY INCIDENCE AND POVERTY GROUPS National Male Female National Male Female Total 2007 100.0 51.0 49.0 50.4 50.6 50.2 ‹15 48.8 25.0 23.8 56.7 57.1 56.4 15-24 17.6 9.4 8.1 47.1 48.2 45.8 25-34 10.0 4.6 5.5 43.7 41.1 46.2 Poverty in Timor-Leste 2014 This section looks at several poverty correlates, 14 percentage point declines in poverty incidence 35-44 10.3 5.1 5.3 50.5 47.8 53.5 from household demographics and health status among female headed households in urban and of children to ownership of consumer durables rural areas respectively over 2007-2014. Overall, 45-60 10.0 5.3 4.7 44.2 47.5 41.0 and livestock, and assesses the extent to which comparing female-headed with male-headed 61+ 3.3 1.8 1.6 35.9 36.6 35.2 the change in observed in consumption poverty households, poverty incidence declined more matches changes in poverty correlates such as steeply among female-headed households over child malnutrition or patterns of consumption 2007-2014, by 17.4 percentage points relative to expenditure. It also assesses the extent to which a decline by about 8 percentage points amongst Total 2014 100.0 50.7 49.3 41.8 42.3 41.2 households characteristics differ between the male-headed households. At the national 32 poor and the non-poor. level, more than 90% of the poor live in male- ‹15 47.3 24.1 23.3 49.0 49.5 48.6 33 headed households. 15-24 17.2 9.0 8.2 39.5 40.3 38.6 Poverty in Timor-Leste 2014 25-34 10.6 4.8 5.8 37.3 35.6 38.7 DEMOGRAPHY AND POVERTY It is also worthwhile to point out that in 2007, Table 13 shows the distribution of the poor by female-headed households were less poor mainly 35-44 9.6 4.5 5.1 40.3 38.0 42.6 age and gender. As children under the age of 15 because of their small household size. In contrast, 45-60 9.9 5.6 4.3 36.4 39.2 33.5 account for more than 40% of the population in in 2014, poverty incidence among female-headed households is lower than male headed households 61+ 5.3 2.8 2.6 26.8 29.3 24.6 2014, they, as in 2007, also account for almost half of the poor population of the country. While even among similar household sizes (Table 15). poverty incidence has declined overall, except for a small rise in the proportion of poor that are elderly, there is very little change in the age and Source: TLSLS 2014, TLSLS 2007 sex distribution of the poor population 2007-2014 (Table 13). As in 2007, poverty incidence is lower among female-headed households than male-headed households, but the difference in the incidence rates between the two groups has more than doubled in 2014 (Table 14). This reflects the 22 and TABLE 14: POVERTY AMONGST FEMALE AND MALE-HEADED HOUSEHOLDS CONSUMPTION PATTERN AND POVERTY (weight lower than that of a reference child of a particular height) decreased significantly over Table 16 shows the share of major consumption 2007-2014. The incidence of wasting almost POVERTY INCIDENCE (%) PERCENTAGE OF THE POOR categories in total consumption. The share of halved while the incidence of underweight children National Rural Urban National Rural Urban food in total consumption expenditure declined declined by about 16 percentage points indicating from 66% to 54% over 2007-2014. Not only that, Total 2007 50.4 54.7 38.3 100.0 100.0 100.0 a substantial decline in acute malnutrition among the decline was larger for the poor (14 percentage Female-headed 45.0 47.2 39.2 9.0 8.6 10.9 children. However such rates of progress are points) than for the non-poor (11 percentage Male-headed 51.0 55.5 38.2 91.0 91.4 89.1 not observed for stunting (height lower than that points). However, the most significant change of a reference child of a particular age) where was in the share of rental expenditures, especially Total 2014 41.8 47.1 28.3 100.0 100.0 100.0 the decline was only about 5 percentage points. for the poor: rental share more than doubled from Female-headed 27.6 32.8 16.9 6.9 6.9 7.3 The stunting prevalence rate itself remains high 12.6% in 2007 to 27.2% in 2014. Share of non-food Male-headed 43.4 48.6 29.9 93.1 93.2 92.7 at 49.2% implying that about half the children Poverty in Timor-Leste 2014 expenditures increased marginally for both poor in the country continue to suffer from chronic and non-poor while shares of utilities, health, and malnutrition, most likely due to inadequate intake Source: TLSLS 2014 education declined. of essential micronutrients. TABLE 15. POVERTY AMONGST FEMALE AND MALE-HEADED HOUSEHOLDS BY HOUSEHOLD SIZE NUTRITIONAL STATUS OF CHILDREN % OWNERSHIP OF LIVESTOCK AND OTHER Table 17 provides information on the nutritional HOUSEHOLD POVERTY INCIDENCE DURABLE GOODS status of children based on anthropometric SIZE National Male Female measurements. It is clear that both the incidence There are some noticeable changes in ownership of consumer durables between 2007 and 2014. As 34 Total 2007 50.4 51.0 45.0 of underweight children (weight lower than that of 35 a reference child of a particular age) and wasting Table18 shows, ownership of mobile phones and 1 or 2 8.1 6.4 10.5 3 20.4 17.8 26.8 Poverty in Timor-Leste 2014 4 32.4 31.9 36.2 TABLE 16: SHARES OF MAJOR CONSUMPTION CATEGORIES IN TOTAL CONSUMPTION, BY POVERTY STATUS 5 43.8 42.5 53.1 6 53.6 53.0 61.4 (%) 7+ 61.6 61.4 64.9 2007 2014 National Non-poor Poor National Non-poor Poor Total 2014 41.8 43.4 27.6 Total 100.0 100.0 100.0 100.0 100.0 100.0 1 or 2 1.4 1.7 0.8 Food 66.2 64.3 70.6 54.0 53.1 57.0 3 11.5 13.2 6.1 Non-food 10.4 11.4 8.2 12.9 14.1 9.2 4 22.7 22.6 23.6 Utilities 8.7 9.4 7.0 8.0 8.6 5.9 5 27.5 27.4 28.0 Rent 12.7 12.8 12.6 24.2 23.2 27.2 6 43.8 44.2 39.8 Health 0.6 0.6 0.4 0.2 0.2 0.2 7+ 56.2 57.1 43.4 Education 1.4 1.5 1.1 0.7 0.7 0.6 Source: TLSLS 2014 Source: TLSLS 2014 TABLE 17. NUTRITIONAL STATUS OF CHILDREN UNDER 5 YEARS, 2007, 2014 TABLE 19: LIVESTOCK OWNERSHIP BY CONSUMPTION QUINTILE (%) 2007 2014 (% of the population living in households that own livestock) Male Female National Male Female National Buffalo Bali cow Cow Pig Goat Sheep Chicken Duck Underweight (weight for age) Total 52.5 44.5 48.6 36.2 28.5 32.4 Timor-Leste 2007 14 0 22 82 29 1 72 3 Severe 16.3 12.9 14.6 6.1 5.6 5.9 Quintile I 11 0 23 81 27 1 73 2 Moderate 36.2 31.6 33.9 30.1 22.8 26.5 Quintile II 11 0 23 84 28 1 72 1 Stunting (height for age) Poverty in Timor-Leste 2014 Total 56.3 51.5 53.9 53.9 44.3 49.2 Quintile III 14 0 22 84 32 2 73 2 Severe 24.6 22.9 23.8 22.7 12.7 17.8 Quintile IV 16 0 23 83 31 2 73 3 Moderate 31.7 28.6 30.2 31.2 31.5 31.4 Quintile V 16 0 19 79 29 2 67 4 Wasting (weight for height) Total 29.2 19.6 24.5 14.1 10.2 12.2 Severe 7.6 7.3 7.5 3.0 2.2 2.6 Timor-Leste 2014 11 1 25 79 21 1 70 1 Moderate 21.6 12.3 17.0 11.1 8.1 9.6 Quintile I 10 2 27 81 23 1 73 0 36 37 Note: Severe malnutrition refers to children with z-scores ‹ -3 and moderate malnutrition refers to children with z-scores Quintile II 10 1 29 81 24 1 73 1 between -3 and -2. Source: TLSLS 2014 Quintile III 13 1 23 84 22 2 75 2 Poverty in Timor-Leste 2014 TABLE 18: DURABLE GOODS OWNERSHIP BY URBAN AND RURAL AREAS Quintile IV 12 1 25 78 19 1 70 1 Quintile V 12 0 22 68 16 1 61 2 (% of the population living in households that own durable goods) Timor-Leste Rural Urban 2007 2014 2007 2014 2007 2014 Note: Each quintile comprises 20% of the population. Fans 5 14 0 4 19 38 Source: TLSLS 2014 Televisions 16 39 5 24 46 75 Video players 2 15 1 10 6 28 Tape players/CD players 10 10 5 6 25 19 Mobile phone 12 68 4 62 35 83 Motorcycles/scooters 7 24 3 17 18 44 Electric rice cooker 2 19 0 9 6 46 Source: TLSLS 2014 TABLE 20: AVERAGE NUMBER OF LIVESTOCK PER HOUSEHOLD (Among households that own livestock) MULTIDIMENSIONAL Buffalo Bali cow Cow Pig Goat Sheep Chicken Duck DEPRIVATION AND POVERTY Timor-Leste 2007 5.7 2.9 3.9 2.8 4.1 7.0 6.4 4.7 Rural 5.6 1.2 3.8 2.8 4.1 7.0 6.4 4.1 Urban 6.5 8.4 4.5 2.9 4.0 7.0 6.8 5.9 Timor-Leste 2014 6.5 3.4 5.1 3.1 4.5 9.0 7.3 5.3 Poverty in Timor-Leste 2014 The case for a separate focus on multidimensional then a reduction in consumption poverty would Rural 6.4 3.3 5.1 3.2 4.2 8.5 7.4 5.1 poverty over and above poverty in terms of also result in multidimensional poverty reduction. Urban 7.2 8.6 4.9 2.7 6.8 11.7 6.7 5.6 consumption or income has both a conceptual But because such spontaneous “trickle down” to foundation as well as a basis in pragmatic other dimensions cannot be taken for granted, empirical considerations. monitoring of multidimensional poverty offers Source: TLSLS 2014 an important “dual check” on poverty reduction. The conceptual foundation of the term And indeed, in some cases a country may perform ‘multidimensional poverty’ is rooted in viewing better in the non-income than in the income televisions has increased, as also the ownership poverty as “capability failure” 12 and evaluating space. In addition, the focus on multidimensional 38 of motorcycles, and appliances such as fans and a range of specific capabilities including those poverty also has the potential to draw attention 39 electric rice cookers. Changes are large in both relating to health, education, shelter, and to “government failures” too, especially in rural and urban areas. access to basic amenities. While consumption relation to provision of public goods and how Poverty in Timor-Leste 2014 and income can themselves be considered markets function. However, the ownership of livestock, the principal multidimensional indicators of welfare, the case productive asset aside from land in Timor-Leste, for a distinctive focus on “multidimensional Table 21 presents results on a selection of shows a mixed picture. Over the period 2007- poverty” rests in the proposition that there non-income indicators of deprivation for 2007 2014, there was no apparent increase in the are many aspects of poverty that may not be and 2014. The selection of indicators in this proportion of the population owning different types adequately captured by consumption or income , 13 table is illustrative and is not intended to be of livestock in any of the expenditure quintiles and mainly due market failures of one sort or another. comprehensive. The 11 indicators shown in Table nationally (Table 19). However, there was some If consumption were indeed synonymous with 21 broadly encompass the three dimensions increase in the average numbers of different achievements in a range of multiple dimensions, of education, health and living standards that livestock amongst those who owned any livestock (Table 20). 12 Sen (1980, 1985, 1999) is the most notable exponent of this view. Sen’s writings on this subject are many; referenced here are only a few examples (including one of the early ones). 13 Even from the welfarist perspective of consumption expenditure as a money-metric of utility, consumption may be inadequate as there are other arguments in individual utility functions for which either markets (and hence prices) may either not exist or if they exist, they may be distorted. underpin the Multidimensional Poverty Index TABLE 21: MULTIDIMENSIONAL DEPRIVATION INDICATORS, 2007, 2014 (% OF POPULATION IN HOUSEHOLDS WITH THE FOLLOWING DEPRIVATIONS) (MPI) produced and published by the UNDP for over 100 countries since 2010. No No No safe Poor Poor Few Any child No member Any child Any child Any child electricity improved drinking quality cooking assets not attending with at least who is who is who is The estimates in Table 21 demonstrate sanitation water floor fuel school 5 yrs of school underweight stunted wasted substantial improvement between 2007 and 2014 for most of the 11 indicators. The reductions in 2014 deprivation incidence are large, often larger than the reduction in incidence of consumption poverty Timor-Leste 28 40 25 48 87 56 17 13 16 24 7 reported earlier. The improvement is notable Rural 37 48 29 60 91 68 18 17 17 25 6 across all (the East, Centre and West) regions of Urban 3 18 16 17 76 24 12 5 14 21 8 the country. Poverty in Timor-Leste 2014 Deprivation rates for individual dimensions cannot East 18 49 39 60 86 58 19 13 13 19 5 of course inform us of their joint distribution. Centre 29 32 20 43 84 51 16 12 15 24 6 Nonetheless, the estimates in Table 21 are West 37 50 23 47 96 67 15 19 21 28 10 certainly suggestive of sizeable reductions in multidimensional poverty over this period. The analysis undertaken for this paper however 2007 stops short of constructing an aggregate Timor-Leste 64 58 40 61 97 83 42 17 35 38 20 multidimensional poverty measure, which Rural 81 68 47 71 99 93 46 21 34 36 21 40 41 remains a topic for future analytical work. Key Urban 18 27 18 32 94 55 31 7 38 43 19 issues that need to be addressed in undertaking such multidimensional poverty analysis and some Poverty in Timor-Leste 2014 possible options that may be considered in the East 66 59 54 70 100 90 49 17 28 33 14 Timorese context are discussed in Annex F. Centre 57 51 30 56 96 78 41 15 38 43 20 West 81 74 49 63 99 91 36 24 33 30 2 Source and notes: TLSLS 2007 and TLSLS 2014. The indicators are Safe drinking water is water sourced from any of Few assets: defined as below. the following: Households not owning more than one of radio, television, piped water, public tap, borehole or pump, protected well, telephone, bike, motorbike or refrigerator and not owning a car No improved sanitation: protected spring or rainwater. or truck. The household’s sanitation facility is not improved or it is improved but shared with other households. A household is considered to have Poor quality floor: Any child not attending school: access to improved sanitation if it has some type of flush toilet or Refers to floors of dirt, sand or dung. Households with at least one child aged 6 to 11 currently not latrine, or ventilated improved pit or composting toilet, provided that attending school. they are not shared. Poor cooking fuel: Refers to dung, wood or charcoal. Any child who is underweight/stunted/wasted: No safe drinking water: Households with at least one child below 5 whose has z-score The household does not have access to safe drinking water or safe ‹-2 for weight-for-age/height-for-age/weight-for-height. drinking water is more than a 30-minute walk roundtrip from home. 42 Poverty in Timor-Leste 2014 and Health ANNEX A: Survey, the Multiple employment status of household members. FIGURE A 1: TIMOR-LESTE: DISTRICT MAP holds that were randomly selected and re-interviewed. TIMOR-LESTE SURVEY OF LIVING STANDARDS-3 specialized surveys such as the Demographic of topics covered is very broad, and encompasses measurements of children, and occupational and education status of households, anthropometric (TLSLS-3) is the third national survey of living The 2014 Timor-Leste Survey of Living Standards things, consumption expenditures, health and Information was collected on, among other Indicators Survey and a typical labor force survey. comprehensive multi-module survey and scope Cluster most of those that would be covered under more standards for the country. TLSLS-3 is also a Sample size Region 5: Oecussi); see Figure A 1. of delivering estimates for the urban and rural considerably expanded the sample size to a final 100 sucos covering one per cent of the population. one to three districts (Region 1: Baucau, Lautem which was spatially stratified with the intention About five and a half years later, the TLSLS 2007 cross-sectional sample of 4,477 households,14 representative, sample of 1,800 households from the first national living standards survey of its Region 4: Bobonaro, Cova Lima and Liquiçá; and 2007-January 2008. The 2001 TLSLS, being Manatuto; Region 3: Aileu, Dili and Ermera; and Viqueque; Region 2: Ainaro, Manufahi and segments of five regions, each comprising of August 1999, had a modest, though nationally kind following the independence referendum of in August-November 2001 and in January The first two national surveys were undertaken 14 The TLSLS 2007 also included a panel component of 900 households, which correspond to half of the 2001 TLSS sample of 1,800 house TABLE A 1: TIMOR-LESTE SURVEY OF LIVING STANDARDS 2014: SAMPLE DESIGN CENSUS 2010 LFS 2012 TLSLS 2014 (TLS-3) Domains Enumeration Areas Households Enumeration Areas Enumeration Areas Sample Size (HHs) Max Std Errors Urban Rural Total Urban Rural Total Urban Rural Total Urban Rural Total Urban Rural Total Urban Rural Total District 01 Aileu 15 84 99 1845 7819 9664 10 15 25 10 15 25 150 225 375 6.1% 4.8% 4.1% 02 Ainaru 3 75 78 444 6521 6965 3 14 17 3 14 17 45 210 255 10.6% 3.9% 3.7% 03 Baucau 20 185 205 3107 18148 21255 14 37 51 11 28 39 165 420 585 5.9% 3.5% 3.1% 04 Bobonaro 33 175 208 2721 14162 16883 17 35 52 11 23 34 165 345 510 5.8% 3.9% 3.4% 05 Cova Lima 13 98 111 1235 9870 11105 10 26 36 8 20 28 120 300 420 6.7% 3.9% 3.5% 06 Dili 228 69 297 28894 6330 35224 41 19 60 41 18 59 615 270 885 3.1% 4.7% 2.7% 07 Ermera 12 192 204 1148 18132 19280 9 23 32 9 23 32 135 345 480 6.3% 3.5% 3.1% 08 Liquica 6 103 109 755 9596 10351 6 27 33 5 22 27 75 330 405 8.5% 3.2% 3.0% 09 Lautem 16 107 123 2044 9403 11447 12 28 40 8 19 27 120 285 405 6.9% 4.3% 3.7% 10 Maufahi 14 71 85 1769 6087 7856 10 14 24 10 14 24 150 210 360 6.1% 5.0% 4.1% 11 Manatuto 12 63 75 1236 5688 6924 11 22 33 8 19 27 120 285 405 6.7% 3.5% 4.0% 12 Oecussi 10 110 120 1580 12310 13890 9 27 36 8 23 31 120 345 465 6.8% 3.7% 3.4% 13 VIqueque 7 88 95 945 12862 13807 7 26 33 6 24 30 90 360 450 7.8% 3.2% 3.0% Regions 01 Bau/Lau/Viq 43 380 423 6096 40413 46509 33 91 124 25 71 96 375 1065 1440 4.0% 2.0% 1.8% 02 Ain/Man/Man 29 209 238 3449 18296 21745 24 50 74 21 41 68 315 705 1020 4.4% 2.7% 2.4% 03 Ain/Dil/Erm 255 345 600 31887 32281 64168 60 57 117 60 56 116 900 840 1740 3.1% 2.7% 2.1% 04 Bob/Cov/Liq 52 376 428 4711 33648 38339 33 88 121 24 65 89 360 975 1335 3.7% 1.8% 2.0% 05 Oe cussi 10 110 120 1580 12310 13890 9 27 36 8 23 31 120 345 465 6.8% 3.7% 3.4% Total Timor-Leste 389 1420 1809 47723 136928 184651 159 313 472 138 262 400 2070 3930 6000 2.0% 1.1% 1.0% Note: The table presents the number of Enumeration Areas (EAs) and households from the 2010 Census (left-hand panel), the number of EAs visited by the 2012 Labor Force Survey (middle panel) and the number of EAs and households from the TLSLS 2014, the maximum absolute standard errors for the estimation of prevalence, by geographic and urban/rural domains. Poverty in Timor-Leste 2014 43 With an expansion of statistical capacity within The right-hand panel of Table A1 above presents responsible for 50 EAs and visited one enumeration checking that the per capita calorie intake the country as well as a maturing of government the sample size and the maximum absolute area per week. Each enumerator within the team of each household (as reported by the food institutions and operations, greater demands were standard errors of the TLSLS 2014, by district and was responsible for interviewing 5 households, consumption sections of the questionnaire) placed on the third round of the living standards urban/rural domains. The left-hand panel shows with the enumeration of each household spread is within a reasonable range for the survey. The TLSLS 2014 faced the challenge of the number of EAs and households of the 2010 over four visits. household’s composition (the distribution of delivering estimates of comparable quality for Census, and the number of EAs visited by the LFS. household members by age and sex), (2) each of Timor-Leste’s 13 districts, leading to a The figures are broken down by district on top, The data for each PSU were transmitted on a checking for the consistency of each child’s further expansion of the sample size. To meet this and according to the regions used by the TLSLS continuous basis to the Directorate General of anthropometric measures (weight and height), challenge within the existing resource and time 2007 (TLSLS-2) at the bottom. Statistics (DGE) Office in Dili. The transfer was using the World Health Organization constraints, it was decided not to have a panel done through the internet using USB-modems. reference tables. component in the TLSLS 2014, but to augment The standard errors for most indicators of Data quality control 3. Monitoring selected indicators. As there is the total sample size to 6,000 households with 15 interest will be between 3 and 4 percentage perverse incentive for the interviewers to Poverty in Timor-Leste 2014 sample households in each of 400 enumeration points at the district level, but larger for the Three layers of supervision were utilized to assure modify some answers to reduce work, the third areas (EAs) stratified across rural and urban urban/rural subdivisions of each district. In other high data quality of the TLSLS 2014: layer of data quality control involved DGE Office sectors of the 13 districts. words, the survey can be expected to allow basic 1. Human supervision. This layer consists of in Dili monitoring the evolution of several comparisons between districts, but not for the (i) team supervisors re-visiting some randomly indicators using partial national databases, Sample design urban/rural portions of each district. Urban/rural selected households to ask selected questions periodically received from the field. Some comparisons would be possible at the national again; (ii) the core team from the DGE re-visiting examples of the indicators monitored were The TLSLS 2014 sample was selected in two level and marginally possible within the five some households to supervise the supervisors. average household size, average number stages: Census Enumeration Areas (EAs) as regions used by the TLSLS-2. The DGE prepared standard supervision forms of food items consumed, average number Primary Sampling Units (PSUs) and individual that were used for this purpose. of children under five, and average number of 44 households as secondary units. A cluster of 15 45 Like its predecessor, the TLSLS 2014 was fielded women between 15-49 years that can be households was visited in each PSU. 2. Computer-based quality controls. This over a yearlong period and the sample was expected to remain similar all four quarters of second layer of supervision involved the randomly distributed across the four quarters Poverty in Timor-Leste 2014 The sample frame for TLSLS 2014 makes use data entry program that complemented human the survey and across teams. of the year to capture seasonality. For quality of the recent 2010 Census, but indirectly, taking supervision, applying consistency checks assurance purposes, the sample of each district advantage of the household listings prepared for to each households’ data. The information was also randomly allocated to at least three ANNEX B: the Labor Force Survey (LFS) in 2012. The LFS registered on the questionnaires was different teams of fieldworkers (interpenetrating CONSUMPTION-BASED WELFARE INDICATOR distributed its sample of 472 EAs across the urban immediately entered into laptops, using an sampling). The two criteria (random allocation and rural segments of the 13 districts, with an “intelligent” data entry program that allows Two key elements of poverty measurement into quarters and teams) could be satisfied allocation that was considered adequate for the for both data entry and consistency checks. relate to (i) the construction of an individual simultaneously in all districts except Oecussi, TLSLS 2014. welfare indicator, and (ii) the construction of an which was visited by a different team each quarter. Possible inconsistencies were resolved appropriate poverty line against which the chosen The TLSLS 2014 only needed to visit a subsample by asking the household members the same welfare indicator will be compared in order to of 400 out of the 472 EAs for the LFS. The TLSLS Fieldwork organization and data management questions again, during the next visit. Teams classify individuals as poor or non-poor, and if 2014 subsample was allocated to the same strata The fieldwork organization for the TLSLS 2014 is did not move to another enumeration area poor, to measure their consumption shortfalls as the LFS, and the 400 PSUs were randomly similar to that used successfully for the TLSLS-2 until all inconsistencies were resolved in all of (relative to the poverty line) . This Annex describes selected with equal probabilities within each in 2007. The fieldworkers were organized into the 15 questionnaires. the procedures adopted in relation to the stratum. The detailed sample design is shown in 8 independent teams, each of them composed first element. Table A1. of one supervisor, three enumerators, one data Examples of the multiple consistency checks entry operator and one driver. Each team was included in the data entry program are: (1) Following well-established practice and consistent prepared and consumed outside the house’ and The general principle followed was to (a) to exclude months. Expenses over the last month entered with the approach in earlier poverty assessments `Foods and drinks prepared outside and brought items that do not directly contribute to household directly into the monthly consumption aggregate, for Timor-Leste, per capita household to be consumed at home’. consumption, (b) to exclude items that are lumpy whereas expenses recorded over the last 3 or 12 consumption is taken to be the measure of welfare and highly infrequent in nature. Following this months were converted to a monthly basis. at the individual level. 15 Thus, the first step in These food items are organized according to 14 principle, expenses related to taxes, festivities estimating poverty measures is to construct broad categories: cereals, tubers, fish, meat, and ceremonies, jewelry, furniture, household Rent a measure of total household consumption or eggs and milk products, vegetables, legumes equipment and other durables, donations, The last component of consumption relates to the “consumption aggregate” at the household and nuts, fruit, oil and fat, beverages and drinks, gambling/cash losses and bank deposits, were rent. The objective is to try to measure the value level, which could then be normalized by ingredients, miscellaneous, alcoholic drinks, and excluded: 21 non-food items in 2007 and 20 non- of the flow of “housing” services received by household size. tobacco and betel. food items in 2014 (deposits on savings accounts households from living in their dwelling. When a no longer needed to be excluded in 2014 because household rents its dwelling and rental markets For poverty measures to be comparable over The monetary value reported by the household this non-food item was not included in the Poverty in Timor-Leste 2014 are well-established, that value would be the time, the consumption aggregate also need to refers to the actual consumption of these items questionnaire). Altogether, 52 non-food items in actual rent paid by the household. However, in be constructed in a comparable manner over and includes all possible sources, including 2007 and 53 non-food items in 2014 were included Timor-Leste, only a handful of households rent time. As noted already, the consumption-related purchases, self-production and gifts or transfers. in the measure of household consumption. The their dwellings. Thus, paid rent cannot be used to modules of the TLSLS-2 in 2007 and the TLSLS-3 Food consumption is constructed by adding up the additional item in 2014 relating to mobile phone determine housing values for the vast majority of in 2014 were kept virtually identical to allow such consumption of all food items and is expressed on expenses was largely irrelevant for 2007 because non-renting households. comparability, and we follow the same procedures a monthly basis. of the very limited usage of mobile phones at for constructing consumption aggregates for that time. The survey asks households for estimates of the two rounds of the survey. The following Non-food how much their dwelling could be rented for, describes how the three main components of The non-food component introduces a couple of Reference period 46 so these “imputed” rents can be included in 47 consumption –food, rent and remaining non-food– practical issues: the choice of items to include As for the reference period, the TLSLS-2 captures the consumption aggregate. But self-reported were constructed. and the selection of the reference period. non-food consumption using two reference imputed rents are subject to measurement error. Poverty in Timor-Leste 2014 periods: the last month and the last 12 months. Hence, in order to minimize potential errors, a Food Items to include The chosen reference period is the last month. hedonic housing rental regression was estimated The food component is based on information that Regarding the first issue, the survey gathers However, households do not buy many non-food and the predicted imputed rent from this was obtained using a recall period over the last information on 73 non-food items. Two small items every month. In order to better capture the regression was included as part of consumption seven days. changes occurred between the two surveys: overall non-food consumption of the population, for all non-renting households. For those renting deposits on savings accounts was dropped in whenever households do not purchase a non-food their dwellings, the actual rent paid was included The survey collects information on 131 food items 2014 and expenses on mobile phones was added item in the last month but report its consumption in their consumption aggregate. in 2007 and 135 food items in 2014. The reason in 2014. These non-food items are organized in the last 12 months, the latter is converted to a for the small difference in the number of items in categories such as clothing and footwear, monthly basis and included as part of the non-food The sum of food, non-food and housing rentals is that (i) both local rice and imported rice have education, health, taxes, festivities, utilities and consumption. By contrast, the TLSLS-3 assigned (actual or imputed) gives the aggregate nominal been split into two subcategories each: subsidized household maintenance, etc. a single reference period to all non-food items consumption of the household. This nominal and non-subsidized, (ii) bottled water has been depending on how frequently they are purchased: consumption was further adjusted by a price index added as a new food item, and (iii) prepared food the last month, the last 3 months and the last 12 to reflect temporal differences in the cost of living. 16 and drinks has been split into `Foods and drinks 15 See Deaton (1997) and Ravallion (2016) for theoretical and practical arguments for using consumption as the measure of welfare for 16 Spatial cost of living differences are also taken into account, but this is done by way of estimating the poverty lines for six domains poverty estimation. (the rural and urban areas of the East, Centre and West regions) in 2007 or by district in 2014, as discussed before. Adjusting for temporal differences food and fuel consumption for urban and rural The final step in constructing the welfare indicator involves going from a measure of standard of living defined Temporal differences arise because households households, respectively. at the household level to one at the individual level. Following common practice, this adjustment requires are interviewed throughout the year, and on dividing the household consumption aggregate by the number of household members. account of inflation over the year, nominal This temporal price index is limited to food and consumption for, say, a household interviewed fuel (kerosene and firewood). This because the in March 2007 cannot be compared with another survey does not collect quantities of other non- ANNEX C: RENTAL MODEL interviewed in December 2007. food items because these quantities are not well defined and thus meaningful unit-values for these Annex C describes the rental model used to estimate the rental component of consumption and the rent Laspeyres price indices for urban and rural areas items cannot be constructed.17 The procedure poverty lines. were constructed based on urban and rural therefore has to assume that prices of other non- TABLE C 1: RENTAL MODELS, 2007, 2014 households’ respective reported consumption. food items changed proportionally with the prices In 2007, a quarterly food and fuel price index of food and fuel. Poverty in Timor-Leste 2014 was estimated having All-Urban January 2007/ 2007 2014 January 2008 as the base for the urban price A similar approach was taken in 2014, with Walls of concrete or brick 0.65 *** 0.35 *** index and All-Rural January 2007/January 2008 the base for the price indices being the entire Numbers of rooms 0.19 *** 0.13 *** as the base for the rural price index. The proxies Flush toilet of VIP latrine 0.24 *** 0.15 *** fieldwork period also: All-Urban April 2014/April Access to electricity 0.27 *** 0.06 *** for prices were the unit-values of food and fuel 2015 in urban areas and All-Rural April 2014/April Floor of ceramic, tiles, cement, concrete or brick 0.28 *** 0.20 *** consumed, and the weights were the average 2015 in rural areas. Table B1 shows the temporal shares of individual items in the total annual price index for urban and rural areas for 2007 East rural 1.30 *** - and 2014. East urban 1.27 *** - Center rural 1.55 *** - 48 TABLE B 1: THE TEMPORAL PRICE INDEX FOR RURAL AND URBAN AREAS, 2007, 2014 Center urban 1.90 *** - 49 West rural 1.50 *** - Rural Urban West urban 1.66 *** - Poverty in Timor-Leste 2014 TLSLS-2 * Ainaro - 3.35 *** January/March 2007 1.16 0.97 Aileu - 3.10 *** April/June 2007 1.06 1.00 Baucau - 3.35 *** Bobonaro - 3.06 *** July/September 2007 1.01 1.04 Covalima - 3.16 *** October 2007/January 2008 0.98 1.04 Dili - 3.52 *** TLSLS-3 ** Ermera - 3.05 *** April/June 2014 1.02 1.05 Liquiça - 2.87 *** July/September 2014 1.02 1.04 Lautem - 3.49 *** October/December 2014 1.00 0.97 Manufahi - 3.36 *** Manatuto - 3.36 *** January/April 2015 1.03 0.97 Oecussi - 3.50 *** * The base for the rural price index is All-Rural January 2007/January 2008 and for the urban price index is Viqueque - 3.45 *** All-Urban January 2007/January 2008. ** The base for the rural price index is All-Rural April 2014/April 2015 and for the urban price index is All- Urban April 2014/April 2015. N 4422 5768 Adjusted R-squared 0.92 0.97 Source: TLSLS 2007 and TLSLS 2014. Source and notes: Estimated with TLSLS 2007 and TLSLS 2014 data. The dependent variable is the actual or imputed rent. All 17 Only items that were consumed in all quarters and with at least 30 observations per quarter were included. independent variables except number of rooms are binary. *, ** and *** indicate significance at 10, 5 and 1 per cent, respectively. ANNEX D: STANDARD ERRORS AND CONFIDENCE INTERVALS Annex D demonstrates the standard errors and confidence intervals of the 2014 poverty estimates for the poverty incidence, poverty gap and the squared poverty gap measures. TABLE D 1: POVERTY INCIDENCE, 2014 TABLE D 2: POVERTY GAP, 2014 Poverty Standard (95% confidence interval) Poverty Standard (95% confidence interval) incidence error Lower bound Upper bound gap error Lower bound Upper bound Timor-Leste 41.8 1.4 39.0 44.5 Timor-Leste 10.4 0.5 9.4 11.4 Rural 47.1 1.8 43.6 50.5 Rural 12.2 0.7 10.8 13.5 Urban 28.3 2.0 24.3 32.2 Urban 5.9 0.6 4.8 7.1 Poverty in Timor-Leste 2014 East 33.8 2.4 29.1 38.4 East 7.1 0.7 5.7 8.5 Centre 40.0 2.0 36.1 43.9 Centre 9.8 0.7 8.3 11.3 West 55.5 2.9 49.9 61.2 West 15.8 1.3 13.3 18.3 East Rural 36.0 2.8 30.6 41.5 East Rural 7.7 0.8 6.0 9.3 East urban 21.0 3.6 13.9 28.1 East urban 3.8 0.8 2.2 5.4 Centre rural 48.3 2.7 43.0 53.7 Centre rural 12.6 1.1 10.4 14.7 Centre urban 26.4 2.4 21.6 31.3 Centre urban 5.3 0.7 3.9 6.7 50 51 West rural 57.6 3.4 51.0 64.2 West rural 16.8 1.5 13.8 19.7 West urban 46.0 4.3 37.5 54.5 West urban 11.5 1.5 8.6 14.4 Poverty in Timor-Leste 2014 Aileu 35.1 5.1 25.2 45.1 Aileu 8.1 2.1 3.9 12.3 Ainaro 43.2 5.2 33.1 53.4 Ainaro 9.4 1.6 6.2 12.5 Baucau 32.6 3.7 25.3 39.8 Baucau 6.8 1.0 4.9 8.7 Bobonaro 51.7 5.2 41.5 61.9 Bobonaro 12.6 1.8 9.1 16.2 Cova Lima 53.1 5.7 41.8 64.3 Cova Lima 15.9 3.1 9.8 22.0 Dili 29.1 2.5 24.2 34.0 Dili 6.0 0.7 4.6 7.5 Ermera 56.7 5.8 45.4 68.1 Ermera 17.1 2.7 11.8 22.4 Lautem 31.2 4.9 22.6 41.8 Lautem 6.8 1.6 3.8 9.9 Liquiça 43.0 8.1 27.1 59.0 Liquiça 11.7 2.4 7.0 16.5 Manufahi 47.7 5.4 37.0 58.4 Manufahi 11.1 1.6 8.0 14.2 Manatuto 43.1 6.1 31.1 55.0 Manatuto 9.2 2.3 4.8 13.7 Oecussi 62.5 3.5 55.6 69.3 Oecussi 19.8 0.1.7 16.5 23.1 Viqueque 36.9 3.7 29.6 44.2 Viqueque 7.8 1.3 5.2 10.4 Note: The estimation of standard errors and confidence intervals takes into account the three key features of survey design: strata, Note: The estimation of standard errors and confidence intervals takes into account the three key features of survey design: strata, primary sampling units and sampling weights. primary sampling units and sampling weights. Source: TLSLS 2014. Source: TLSLS 2014. TABLE D 3: SQUARED POVERTY GAP, 2014 TABLE E 1: THE IMPACT OF EXCLUDING TEAM 1, 2014 Squared Standard (95% confidence interval) ALL TEAMS EXCLUDING TEAM 1 poverty gap error Lower bound Upper bound Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex1) Consumption line Incidence Consumption line Incidence Timor-Leste 3.7 0.2 3.2 4.2 per person P per person P_ex1 Rural 4.4 0.3 3.8 5.1 Timor-Leste 60.8 44.9 41.2 0.3 61.4 46.4 41.8 Urban 1.8 0.2 1.4 2.3 Ainaro 56.8 44.0 44.6 -0.2 56.7 44.3 43.2 Aileu 54.6 40.0 34.0 0.1 54.2 41.1 35.1 59.7 42.9 34.0 -0.3 Baucau 59.6 43.3 32.6 East 2.1 0.3 1.6 2.7 Bobonaro 52.3 43.3 50.0 0.2 51.5 44.5 51.7 Centre 3.4 0.3 2.7 4.1 54.5 48.0 52.3 0.1 Covalima 54.4 49.4 53.1 West 6.2 0.7 4.9 7.5 Dili 89.1 53.5 23.6 1.5 87.8 56.2 29.1 Poverty in Timor-Leste 2014 41.7 36.3 56.4 0.0 Ermera 44.0 38.0 56.7 East Rural 2.3 0.3 1.7 3.0 45.5 36.1 39.8 0.3 Liquiça 45.4 37.8 43.0 East urban 1.0 0.3 0.4 1.6 59.6 41.4 34.0 -0.3 Lautem 59.3 41.6 32.2 Centre rural 4.5 0.5 3.5 5.5 56.8 47.1 46.2 0.2 Manufahi 56.3 47.9 47.7 57.8 43.7 45.1 -0.2 Manatuto 57.5 43.8 43.1 Centre urban 1.6 0.3 1.1 2.1 40.3 46.3 72.4 -0.9 Oecussi 54.8 51.1 62.5 West rural 6.7 0.8 5.1 8.2 60.4 44.1 37.8 -0.2 Viqueque 60.0 44.5 36.9 West urban 3.8 0.6 2.6 5.1 Source: TLSLS 2014. Aileu 2.7 1.0 0.8 4.6 52 Ainaro 3.0 0.6 1.8 4.2 TABLE E 2: THE IMPACT OF EXCLUDING TEAM 2, 2014 53 Baucau 2.0 0.4 1.3 2.7 Bobonaro 4.4 0.8 2.9 6.0 ALL TEAMS EXCLUDING TEAM 2 Poverty in Timor-Leste 2014 Cova Lima 6.8 1.8 3.3 10.2 Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex2) Dili 1.9 0.3 1.3 2.5 Consumption line Incidence Consumption line Incidence per person P per person P_ex2 Ermera 6.6 1.3 4.0 9.1 Timor-Leste 62.5 47.5 42.2 -0.2 61.4 46.4 41.8 Lautem 2.1 0.7 0.7 3.4 Ainaro 57.1 44.9 44.3 0.1 56.7 44.3 43.2 Liquiça 4.4 1.0 2.4 6.3 Aileu 55.9 41.7 34.0 0.2 54.2 41.1 35.1 Manufahi 3.6 0.6 2.3 4.8 59.7 43.6 34.0 -0.3 Baucau 59.6 43.3 32.6 Manatuto 2.9 1.0 0.9 4.8 Bobonaro 52.5 41.9 42.4 1.2 51.5 44.5 51.7 Oecussi 7.9 0.9 6.2 9.6 52.5 49.1 55.6 -0.3 Covalima 54.4 49.4 53.1 Viqueque 2.4 0.6 1.2 3.5 Dili 89.9 60.1 30.4 -0.3 87.8 56.2 29.1 44.9 38.5 58.4 -0.2 Ermera 44.0 38.0 56.7 Note: The estimation of standard errors and confidence intervals takes into account the three key features of survey design: strata, 44.2 37.8 43.4 0.0 Liquiça 45.4 37.8 43.0 primary sampling units and sampling weights. 59.4 42.8 36.6 -0.7 Lautem 59.3 41.6 32.2 Source: TLSLS 2014. 58.4 49.8 49.4 -0.2 Manufahi 56.3 47.9 47.7 58.6 44.9 44.0 -0.1 Manatuto 57.5 43.8 43.1 58.9 52.2 60.4 0.4 Oecussi 54.8 51.1 62.5 ANNEX E: 60.2 45.6 41.0 -0.8 Viqueque 60.0 44.5 36.9 SENSITIVITY ANALYSIS: FIELDWORK TEAMS Annex E shows the results of the analysis examining sensitivity of poverty incidence with respect to Source: TLSLS 2014. fieldwork teams. TABLE E 3: THE IMPACT OF EXCLUDING TEAM 3, 2014 TABLE E 5: THE IMPACT OF EXCLUDING TEAM 5, 2014 ALL TEAMS EXCLUDING TEAM 3 ALL TEAMS EXCLUDING TEAM 5 Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex3) Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex5) Consumption line Incidence Consumption line Incidence Consumption line Incidence Consumption line Incidence per person P per person P_ex3 per person P per person P_ex5 Timor-Leste 61.4 46.4 41.8 60.5 47.1 46.2 -2.2** Timor-Leste 64.3 49.6 44.5 -1.4 61.4 46.4 41.8 Ainaro 56.7 44.3 43.2 56.6 47.1 50.0 -0.9 Ainaro 56.4 44.2 45.3 -0.3 56.7 44.3 43.2 Aileu 54.2 41.1 35.1 47.7 40.7 44.8 -1.2 Aileu 59.1 44.7 35.9 -0.1 54.2 41.1 35.1 Baucau 59.6 43.3 32.6 59.4 43.4 34.0 -0.3 59.2 44.0 35.1 -0.5 Baucau 59.6 43.3 32.6 Bobonaro 51.5 44.5 51.7 44.7 44.4 64.3 -1.8* Bobonaro 57.7 52.6 60.1 -1.2 51.5 44.5 51.7 Covalima 54.4 49.4 53.1 51.2 46.9 53.3 0.0 62.5 56.9 52.9 0.0 Covalima 54.4 49.4 53.1 Dili 87.8 56.2 29.1 91.3 59.5 32.7 -0.9 Dili 87.4 56.5 30.5 -0.4 87.8 56.2 29.1 Poverty in Timor-Leste 2014 Ermera 44.0 38.0 56.7 41.1 36.8 60.3 -0.4 56.1 47.4 57.2 -0.1 Ermera 44.0 38.0 56.7 Liquiça 45.4 37.8 43.0 43.4 36.7 45.7 -0.2 57.6 48.8 49.0 -0.6 Liquiça 45.4 37.8 43.0 Lautem 59.3 41.6 32.2 59.2 43.1 36.8 -0.7 59.0 42.3 36.0 -0.6 Lautem 59.3 41.6 32.2 Manufahi 56.3 47.9 47.7 53.8 47.7 52.5 -0.6 56.0 48.6 50.3 -0.4 Manufahi 56.3 47.9 47.7 Manatuto 57.5 43.8 43.1 57.0 43.9 46.0 -0.4 57.3 44.3 46.2 -0.4 Manatuto 57.5 43.8 43.1 Oecussi 54.8 51.1 62.5 54.7 52.3 64.5 -0.4 54.4 51.7 64.3 -0.4 Oecussi 54.8 51.1 62.5 Viqueque 60.0 44.5 36.9 59.9 45.8 42.9 -1.2 59.7 45.6 41.4 -0.8 Viqueque 60.0 44.5 36.9 Source: TLSLS 2014. Source: TLSLS 2014. 54 TABLE E 4: THE IMPACT OF EXCLUDING TEAM 4, 2014 TABLE E 6: THE IMPACT OF EXCLUDING TEAM 6, 2014 55 ALL TEAMS EXCLUDING TEAM 4 ALL TEAMS EXCLUDING TEAM 6 Poverty in Timor-Leste 2014 Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex4) Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex6) Consumption line Incidence Consumption line Incidence Consumption line Incidence Consumption line Incidence per person P per person P_ex4 per person P per person P_ex6 Timor-Leste 58.6 45.1 44.3 -1.2 61.4 46.4 41.8 Timor-Leste 62.4 45.1 40.3 0.7 61.4 46.4 41.8 Ainaro 55.9 43.7 44.6 -0.2 56.7 44.3 43.2 Ainaro 52.7 43.7 45.3 -0.3 56.7 44.3 43.2 Aileu 51.8 38.0 35.4 0.0 54.2 41.1 35.1 Aileu 54.3 39.7 33.4 0.2 54.2 41.1 35.1 59.4 43.5 34.6 -0.4 Baucau 59.6 43.3 32.6 60.4 44.1 34.2 -0.3 Baucau 59.6 43.3 32.6 Bobonaro 50.2 42.6 49.6 0.3 51.5 44.5 51.7 Bobonaro 51.6 44.0 51.1 0.1 51.5 44.5 51.7 51.2 45.9 55.0 -0.2 Covalima 54.4 49.4 53.1 54.5 48.9 51.8 0.2 Covalima 54.4 49.4 53.1 Dili 79.9 54.1 34.8 -1.3 87.8 56.2 29.1 Dili 88.7 54.0 24.7 1.3 87.8 56.2 29.1 41.1 36.9 61.2 -0.6 Ermera 44.0 38.0 56.7 43.9 36.0 57.6 -0.1 Ermera 44.0 38.0 56.7 43.2 36.0 48.2 -0.5 Liquiça 45.4 37.8 43.0 45.5 37.6 42.8 0.0 Liquiça 45.4 37.8 43.0 59.1 41.5 32.2 0.0 Lautem 59.3 41.6 32.2 59.8 40.2 31.7 0.1 Lautem 59.3 41.6 32.2 55.3 47.9 47.6 0.0 Manufahi 56.3 47.9 47.7 57.1 47.1 47.3 0.0 Manufahi 56.3 47.9 47.7 54.8 43.4 47.3 -0.5 Manatuto 57.5 43.8 43.1 57.9 41.8 34.6 0.9 Manatuto 57.5 43.8 43.1 54.4 50.5 62.8 -0.1 Oecussi 54.8 51.1 62.5 55.0 50.5 62.0 0.1 Oecussi 54.8 51.1 62.5 59.8 44.3 38.7 -0.3 Viqueque 60.0 44.5 36.9 57.4 40.5 32.8 0.7 Viqueque 60.0 44.5 36.9 Source: TLSLS 2014. Source: TLSLS 2014. TABLE E 7: THE IMPACT OF EXCLUDING TEAM 7, 2014 ANNEX F: THE MULTIDIMENSIONAL POVERTY INDEX ISSUE IN DEVELOPING A MULTIDIMENSIONAL The MPI is defined following the general ALL TEAMS EXCLUDING TEAM 7 POVERTY INDEX FOR TIMOR-LESTE methodology of Alkire and Foster (2011) as: Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex7) Consumption line Incidence Consumption line Incidence The UNDP currently estimates a Global per person P per person P_ex7 Multidimensional Poverty Index for over 100 MPI=M_0=H. A Timor-Leste 61.4 46.4 41.8 61.4 45.7 40.5 0.6 countries –including Timor-Leste. However, some Ainaro 56.7 44.3 43.2 55.2 43.8 45.9 -0.3 countries, such as Mexico have chosen to develop where M_0 is the multidimensional poverty Aileu 54.1 39.5 32.7 0.3 54.2 41.1 35.1 a national MPI in order to develop a methodology, index or what is also referred to as the “adjusted 59.4 44.3 33.8 -0.2 Baucau 59.6 43.3 32.6 headcount index”. It can be expressed as the that may be more relevant for policymaker’s Bobonaro 51.6 43.0 45.2 0.9 51.5 44.5 51.7 needs, and to have greater ownership and control product of H, the headcount index or the proportion 54.3 48.0 51.5 0.2 Covalima 54.4 49.4 53.1 over the process. of the population that is multidimensionally-poor, Dili 87.8 54.6 26.4 0.8 87.8 56.2 29.1 Poverty in Timor-Leste 2014 Ermera 44.0 38.0 56.7 44.0 36.8 55.5 0.2 and A, the average intensity of deprivation of the Liquiça 45.4 37.8 43.0 45.4 36.8 40.6 0.2 Annex F briefly explains the Multidimensional poor. Thus, M_0 is interpretable as either (i) the Lautem 59.3 41.6 32.2 58.6 43.0 37.7 -0.7 poverty measure, explains the methodology used average intensity of deprivation for the population 56.7 48.8 49.6 -0.2 Manufahi 56.3 47.9 47.7 by the UNDP for estimating multidimensional or (ii) the headcount index adjusted by the average 56.1 41.5 40.7 0.3 Manatuto 57.5 43.8 43.1 intensity of deprivation of the poor. poverty, and then discusses a set of key issues 54.9 49.6 60.3 0.4 Oecussi 54.8 51.1 62.5 that would need to be addressed in developing a 62.6 45.5 35.0 0.4 Viqueque 60.0 44.5 36.9 multidimensional poverty index (MPI) for Timor- UNDP’S GLOBAL MULTIDIMENSIONAL POVERTY Leste. Each issue involves methodological INDEX (MPI) Source: TLSLS 2014. choices, and in light of these, the Annex also As shown in Figure F1, UNDP’s global MPI is 56 TABLE E 8: THE IMPACT OF EXCLUDING TEAM 8, 2014 presents some options that could be considered based on ten indicators covering the three broad 57 in the Timorese context. ALL TEAMS EXCLUDING TEAM 8 Poverty in Timor-Leste 2014 Nominal Poverty Poverty Nominal Poverty Poverty t (P-P_ex8) Consumption line Incidence Consumption line Incidence FIGURE F 1: GLOBAL MPI (UNDP) – TEN INDICATORS per person P per person P_ex8 Timor-Leste 62.1 45.6 40.3 0.8 61.4 46.4 41.8 10 Indicators Ainaro 64.6 46.2 37.9 0.7 56.7 44.3 43.2 Aileu 54.3 39.9 34.3 0.1 54.2 41.1 35.1 Nutrition Baucau 59.6 43.3 32.6 59.2 40.8 29.7 0.5 Health Bobonaro 51.7 43.5 51.2 0.1 51.5 44.5 51.7 Child Mortality 54.5 48.4 51.9 0.2 Covalima 54.4 49.4 53.1 Dili 87.8 55.6 27.8 0.4 87.8 56.2 29.1 THREE 44.1 36.5 55.5 0.2 Ermera 44.0 38.0 56.7 DIMENSIONS Education Years of Schooling Liquiça 45.4 37.8 43.0 45.5 36.3 41.3 0.2 OF POVERTY School Attendance 60.3 39.9 28.4 0.4 Lautem 59.3 41.6 32.2 56.3 46.5 45.7 0.3 Manufahi 56.3 47.9 47.7 Electricity 62.2 46.7 41.4 0.2 Manatuto 57.5 43.8 43.1 Living Sanitations Water Oecussi 54.8 51.1 62.5 54.9 49.8 61.6 0.2 Standard Floor 60.2 45.7 39.2 -0.4 Viqueque 60.0 44.5 36.9 Cooking Fuel Assests Source: TLSLS 2014. Source: UNDP (2014). TABLE F 1: GLOBAL MPI: INDICATORS, DIMENSIONAL CUT-OFFS AND WEIGHTS The deprivation thresholds for each of the ten sets of issues to be considered, regarding: (i) The indicators are shown in Table F1. The Table also dimensions and weights to include in the index; (ii) Dimension Indicator Deprived if... Relative Weight shows the relative importance of each indicator in the way in which multidimensional poverty will be EDUCATION Years of Schooling No household member has completed 1/6 the overall index, indicated by its weight and the defined based on the dimensions and weights and; five years of schooling weights sum up to one. A person is considered (iii) the method of aggregation. Child School No child is attending school up to the age 1/6 multidimensionally-poor if deprived in at least Attendance at which they should finish class 6 1/3rd of the weighted dimensions. In particular, the following questions need to HEALTH Child Mortality Any child has died in the family 1/6 be answered: Nutrition Any adult or child for whom there is 1/6 With this basic set-up, UNDP/ OPHI constructed 1. Which indicators (dimensions) should be information is malnourished estimates of multidimensional poverty for Timor- included in an MPI? Leste based on the Demographic and Health 2. How should deprivation be defined in the LIVING Electricity The household has no electricity 1/18 Survey (DHS) for 2009-10, as reported in Table F2. respective dimensions? Poverty in Timor-Leste 2014 STANDARD Improved The household’s sanitation facility is 1/18 3. How should the different dimensions Sanitations not improved (according to MDG guidelines), or it is improved but shared MEASUREMENT CONSIDERATIONS be weighted? other households.** How 4. should dimensional deprivations The approach underlying Tables 1 and 2 (the Safe Drinking The household does not have access to 1/18 be aggregated into an overall measure of “global MPI” measure) represents one clear drinking water (accoding to MDG guidelines) multidimensional poverty? option for the construction of an MPI for Timor- or safe drinking water is no more than a 5. How should the multidimensionally-poor 30-minute walk from home, roundtrip.*** Leste. There are some obvious advantages of be identified? following this option. First, estimates based on Flooring The household has a dirt, sand and 1/18 6. How should the multidimensional deprivation dung floor this particular implementation of an MPI already of the poor be aggregated? 58 exist for Timor-Leste, and thus set a notable 59 Cooking Fuel The household cooks with dung, wood, 1/18 or charcoal precedent. Second, since UNDP has followed this approach globally, comparable estimates of Additional dimensions Assets The household does not own more than 1/18 Poverty in Timor-Leste 2014 one radio, TV, telephone, bike motorbike MPI exist for more than 100 countries which helps Beyond the global MPI measure, there may or refrigerator and does not own a car put the MPI for Timor-Leste in a comparative be several other options for dimensions and or a truck international perspective. weights that could be considered, such as adding in a consumption poverty dimension, or an Source: UNDP (2014). This particular implementation of the MPI is employment dimension. feasible with the TLSLS data with the possible exception of the child mortality indicator. Consumption poverty dimension TABLE F 2: MPI FOR TIMOR-LESTE, 2009-10 However, that indicator may be replaced by one The global MPI could be augmented by introducing of reproductive health for women, for instance, the additional dimension of consumption poverty Survey Year Multidimensional Percentage of Average deprivation pre- or post-natal care, professionally assisted itself, which could be assigned a weight of one-third, Poverty Index poor people Intensity amongst the poor deliveries. Such information on women’s while the weights of the other dimensions could be (MPI = H x A) (H) (A) reproductive health is readily available from scaled down proportionately (such that the weights DHS 2009-10 National 0.360 68.1% 52.9% the TLSLS. sum up to one). If the cross-dimensional cut-off Urban 0.176 47.3% 37.3% of 1/3rd is used for identifying multidimensional Rural 0.414 53.7% 77.0% Should the government of Timor-Leste, however, poverty, then consumption poverty’s weight of wish to develop its own approach to measuring 1/3rd will ensure that a consumption-poor person multidimensional poverty, there are three main is also multidimensionally poor. However, this still Source: OPHI (2014). leaves open the possibility that someone who is working-age household members with “gainful” Defining multidimensional poverty that it ensures regressive transfers from a more not poor in terms of consumption is nonetheless employment, or an ordinal representation An additional consideration for Timor-Leste is the to a less deprived person are poverty-increasing – multidimensionally-poor if deprived in at least of that. 18 Since employment is at least one definition of multidimensionally poor. There are a a property that is not always satisfied by measures one-third of the other weighted dimensions. significant determinant of living standards, the number of questions here that need answering. that use cross-dimensional cut-offs.19 The introduction of consumption poverty also weights for the standard-of-living indicators Should the definition of multidimensionally poor has the attraction that it would allow (for a given could be adjusted downward to accommodate this be based on (i) one cross-dimensional cut-off Aggregation set of weights and cross-dimensional cut-off) a additional dimension. value, (ii) more than one value, or (iii) no value? A further consideration relates to the choice of decomposition of how much of multidimensional which aggregate measures to use when monitoring poverty is attributable to consumption poverty The above illustrate just a couple of possibilities. multidimensional poverty. The headcount index One cross-dimensional cut-off value against other non-consumption dimensions. There are potentially many other options for measures incidence, and the adjusted headcount The chosen value of the cross-dimensional indicators and how they should be weighted. index takes into account intensity. Neither of these cut-off can make a big difference to measured Employment dimension Which variant is appropriate for Timor-Leste are sensitive to inequality, however, and so can be Poverty in Timor-Leste 2014 multidimensional poverty. Figure F2 shows One could consider introducing employment warrants a thoughtful consideration of the many augmented with additional measures. how sensitive the multidimensional headcount as an additional dimension, in keeping with alternatives by the country’s policymaking and index (H) can be to the cut-off for Timor-Leste. the significant policy concern of employment statistical agencies, and should ideally also be The multidimensional headcount index (H), Using a higher cut-off than one-third (the cut- generation in Timor-Leste. The variable could open to wider public discussion as a means of though very easy to understand, is not a terribly off deployed by the global MPI) dramatically be specified, for instance, as the proportion of consensus-building. good measure of multidimensional poverty. lowers the proportion of multidimensionally- poor; for instance, a cut-off of half (implying, to M_0 (=H× A), the adjusted headcount index, is a be mutidimensionally-poor a person must be better measure than H as it takes into account FIGURE F 2: MULTIDIMENSIONAL HEADCOUNT (H) FOR TIMOR-LESTE (2009-10) FOR deprived in at least 50% of weighted dimensions) the average intensity of deprivations amongst the DIFFERENT VALUES OF THE CROSS-DIMENSIONAL OR POVERTY CUT-OFF lowers H from 68% to 39%. poor; it thus rules out such incorrect inferences 60 61 as those of no change in multidimensional poverty % when, for instance, while H may not have changed, More than one cross-dimensional cut-off value Poverty in Timor-Leste 2014 80 or no value the average intensity of deprivation amongst the 65% multidimensionally-poor may have declined 70 There is also the possibility of using more than one value for the cross-dimensional cut-off, or (or increased). 60 50% of not using the cross-dimensional cut-off at 50 However, even M_0 may be inadequate as a 39% all. The former is in the spirit of investigating 40 robustness of comparisons of multidimensional measure of multidimensional poverty as it is Percentage of people 30 poverty (changes over time or differences across insensitive to dispersion or inequality in the 23% distribution of deprivations amongst the poor. In Severe 20 sub-populations) to the choice of this cut-off. The Poverty 9% latter takes us to the so-called “union approach” This can be illustrated with an example. Consider 10 4% 4 dimensions and a population of 100 persons. 1% to identification, where a person is considered H 0 Suppose 40 of them are multidimensionally- 33.3%+ 40%+ 50%+ 60%+ 70%+ 80%+ 90%+ multidimensionally-poor if deprived in any dimension. The union approach also has the merit poor, i.e. H = 0.4. Further, suppose on average Intensity of Poverty a multidimensionally-poor person is deprived in Source: OPHI (2014). 18 The notion of “gainful” employment can be made more precise based on information available from the employment module of the TLSLS. 19 See Datt (2014) for a formal statement of this property as well as further discussion of why the union approach may be worth considering. 50% of dimensions (or two dimensions), i.e. A = 0.5. Then, the MPI measure M_0 = 0.4 x 0.5 = 0.2. But note that the measure M_0 can also be written as M_0 = (total number of deprived dimensions of the poor / total number of possible deprivations) = REFERENCES (80 / 400) = 0.2. M_0 depends on the total number of deprivations but not their dispersion or concentration, which is a limitation if one wants the measure of multidimensional poverty to be sensitive to the degree of inequality in the distribution of Poverty in Timor-Leste 2014 deprivations. Measures with such sensitivity do exist and may be worth considering (Datt, 2014). Alkire, S., Foster, J.E. (2011): Counting and multidimensional poverty measurement. Journal of Public Economics, And again, one should not underrate the possibility 95(7-8): 476-487. of using more than one aggregate measure of multidimensional poverty. Datt, Gaurav (2014): Making every deprivation count: multidimensional poverty without the “dual cut-off”. Working Paper, Monash University. Deaton, Angus (1997): The Analysis of Household Surveys. World Bank and Johns Hopkins University Press, Baltimore. 62 63 Foster, J. (2005): Poverty indices. In: A. de Janvry and R. Kanbur (eds.) Poverty, Inequality and Development: Essays Poverty in Timor-Leste 2014 in Honor of Erik Thorbecke, Kluwer Academic Publishers. Foster, J., Greer, J. and E. Thorbecke (1984). A Class of Decomposable Poverty Measures, Econometrica, 52: 761-765. OPHI (2014). “Timor-Leste Country Briefing”, Multidimensional Poverty Index Data Bank. Oxford Poverty and Human Development Initiative (OPHI), University of Oxford. Available at: www.ophi.org.uk/multidimensional- poverty-index/mpi-country-briefings National Institute of Nutrition and Indian Council of Medical Research (NIN-ICMR) (2011): Dietary Guidelines for Indians – A Manual, Second Edition, Hyderabad. Ravallion, Martin. (2008). “Poverty Lines” in The New Palgrave Dictionary of Economics. 2nd Edition, Larry Blume and Steven Derlauf (eds), Palgrave Macmillan, London. _____ (2016): Economics of Poverty: History, Measurement and Policy. Oxford University Press. Sen, Amartya (1976). Poverty: An Ordinal Approach to Measurement, Econometrica, 46: 437-446. _____ (1980): “Equality of what?” (Lecture delivered at Stanford University, 22 May 1979), in MacMurrin, Sterling M., The Tanner lectures on human values 1, Salt Lake City, Utah: University of Utah Press. _____ (1985): Commodities and Capabilities. New York, NY: North-Holland. _____ (1999): Development as Freedom. New York: Oxford University Press. UNDP (2014): Human Development Report 2014: Sustaining Human Progress: Reducing Vulnerability and Building Resilience. Macmillan, New York. World Bank (2003). Timor-Leste: Poverty in a New Nation: Analysis for Action, Washington DC. World Bank (2008). Timor-Leste: Poverty in a Young Nation. Dili and Washington DC. 64 Poverty in Timor-Leste 2014