Abstract

A multidimensional poverty index is constructed for the Philippines using the Alkire-Foster methodology and data from a 2011 annual poverty indicators survey. This is disaggregated into urban and rural population groups, as well as by dimension and administrative region. At the province level, the study finds a positive relation between poverty incidence and intensity, but the highest intensity levels are experienced in areas where incidence is not that high relative to other areas. Provinces with high incomes generally have low poverty indices and the relationship appears to be nonlinear. An examination of household poverty using mixed logit analysis shows that poverty risk rises with household size. A substantial reduction of the risk is observed for households with heads who were able to matriculate high school. The household head's health status has a negative impact on the household's risk of being poor. These are contrasted with the results using the income poverty definition. Policy implications are drawn from the calculations and the econometric results.

1.  Introduction

One of the major development objectives of nations is poverty reduction. In 2000, the United Nations set the elimination of extreme poverty and hunger as the first in their list of eight Millennium Development Goals (MDGs). This effectively committed member nations to coordinate and devote resources to reducing extreme poverty by 2015. Having achieved moderate success in the attainment of the MDGs after 15 years, the UN has resumed the battle against poverty with the 2030 Agenda for Sustainable Development.1

One important development in the fight against poverty is the realization that poverty is multifaceted and cannot be adequately characterized by a single type of indicator. This has led to a re-examination of the concept of poverty and the development of new measures to identify the poor and their characteristics. On the 20th anniversary of UNDP's Human Development Report, a radical revision of the Human Development Index (HDI) was introduced along with three new measures: the Multidimensional Poverty Index (MPI), the Inequality-adjusted Human Development Index, and the Gender Inequality Index (see UNDP 2010 and Klugman, Rodríguez, and Choi 2011).

The recent growth in the literature on the multidimensional approach to poverty analysis is partly due to the widely recognized limitations of the unidimensional approach. This paper introduces an index of multidimensional poverty for the Philippines constructed using a methodology developed by Alkire and Foster (2011a, 2011b). This was first used by Alkire and Santos (2014) in constructing a global MPI based on the HDI framework. This methodology has gained popularity, as seen by the number of publications in academic journals. The methodology has also been used by several countries in the formulation of their poverty reduction strategy, including Mexico, Bhutan, and Nepal.2

Although there are recent suggested improvements and extensions of the methodology (see, e.g., Cavapozzi, Han, and Miniaci 2015) there are also criticisms regarding the multidimensional approach to poverty analysis. Several articles in the special issue of the Journal of Economic Inequality in 2011 provide a view of the issues surrounding the methodological controversies. Among the contentious issues are the weights used, the absence of the role of prices, and the accounting of trade-offs between index components. The debate revolves around the issue of the best way to measure multidimensional poverty, because this will have implications on how useful the index is to policymakers. One side of the debate favors a scalar index that collapses information from various indicators into a single number as in the index of Alkire and Foster, whereas the other side prefers a “dashboard” approach, where several distinct indices may be used rather than a single index.

In this paper I do not intend to join this debate. I instead engage in two activities. First, I attempt to construct and evaluate a multidimensional poverty index for the Philippines based on Alkire and Foster's (AF) methodology. As in prior studies, the scalar index produced from a household data set is disaggregated by geographic area to allow for a ranking of poverty by regions or by provinces. Second, I perform a household-level analysis of poverty not usually conducted in previous multidimensional poverty studies. Logit analysis is used to understand the determinants of a household's state of well-being. Here, the state of well-being is defined by the poverty cutoff used in the construction of the index.

The multidimensional index of poverty constructed in this study analyzes the state of the poor in the Philippines using household data from the Annual Poverty Indicators Survey (APIS) collected in 2011 by the Philippine Statistical Authority (PSA). Section 2 discusses the methodology used in this study. Section 3 describes the data and Section 4 presents the MPI estimates. Section 5 presents an econometric analysis of the likelihood of a household falling into a state of poverty using the deprivation scores calculated in Section 4. Section 6 summarizes, discusses the policy implications, and proposes the direction for future research.

2.  Methodology

There are several studies on poverty in the Philippines including publications by the World Bank, the Asian Development Bank, and several unpublished working papers (see, e.g., World Bank 2001). Schelzig (2005) provides a comprehensive discussion of methodological problems observed in official poverty measures and trends in poverty analysis in the Philippines. These studies, which use income or expenditure as the main measure of poverty, provide insights into the effects of policies on the lives of the poor in society. They rely on official household census data, the Family Income and Expenditure Survey (FIES), collected every three years. Balisacan (2011) makes use of several household data sets aside from the FIES for different years to calculate aggregate and disaggregated multidimensional poverty measures. The academic literature on Philippine poverty is thin when compared with other countries (see Bayudan-Dacuycuy and Lim (2013) and the references cited therein). The present research contributes to the discussion on poverty in the Philippines, the measurement of multidimensional poverty, and the determinants of household poverty risk.

The capability approach to the measurement and analysis of poverty developed by Sen (1985) is the main foundation of the AF methodology. It relies on the notion that a person's capability to function, if left unrestricted, allows them to lead a life they can value. Living one's life may be thought of as performing a set of functions on entitlements (or goods) that directly affect their state of well-being. These can be events or activities, such as going to school or being adequately fed. Broadly speaking, restrictions that may be political, economic, social, or cultural in origin, when placed on these “functionings”, lead to diminished capabilities that hinder one's ability to continue living comfortably and reduce the quality of life. For example, an individual can go to school, but poor nutrition that leads to capability failure does not permit them to hurdle school work, get a job, and earn a living. The evaluation of these functionings and the capability to function is at the heart of the AF methodology.

The AF method attempts to capture the various facets of poverty that are not captured by unidimensional poverty measures and is a way of systematically constructing an index of poverty based on several indicators available from survey data. These may include several aspects of living conditions, constraints on resources or personal abilities that lead an individual to be multidimensionally poor. These indicators refer to a person's capabilities and/or possessions that enable a life that can in some sense be viewed as comfortable.

In the empirical implementation, the poverty measure is an aggregation of several dimensions. A dimension is in turn constructed using several deprivation indicators. The AF method is a counting approach to identifying the poor and is most useful when the indicators are binary ordinal. This study's measure belongs to a class of measures labeled Mα. A value of α = 0 yields the results from the counting approach. With cardinal data, values of α = 1 and α = 2 may be used to produce the normalized gap and the adjusted squared gap measures, respectively. Alkire and Foster (2011a, 2011b) and Alkire et al. (2015) provide more details on this.

From a set of m deprivation indicators (j=1,,m) and n individuals (i=1,,n), the MPI of this study is calculated in Section 4 using the dual cutoff method of Alkire and Foster as follows: First, a unique deprivation cutoff is set for each deprivation indicator to form m binary deprivation variables (with n observations). A fixed weight for each indicator is assigned and used to calculate a weighted sum of the m binary deprivation variables for individual i to arrive at their deprivation score and where the sum of the weights is equal to 1. Second, a single poverty cutoff score applied to all individual deprivation scores is set to segregate the poor from the non-poor. The scores of those classified by the poverty cutoff as non-poor are censored to zero. The average of the n deprivation scores (including those set to zero) is the MPI. Alkire et al. (2015) provide a detailed procedure to calculate the MPI, its decomposition into poverty incidence and intensity, as well as its decomposition by population subgroups and by dimension.

3.  The data, the dimensions, and the indicators

The data set used in this study is the 2011 APIS of the Philippines. This survey is conducted annually by the PSA and whenever the FIES is not conducted. The 2011 APIS has 43,833 household respondents covering the whole Philippines, 95.96 percent of whom were successfully interviewed. The survey contains detailed questions about the household such as individual household member characteristics, income and expenditures, government benefits received, hunger perceptions, and living conditions. The nationwide APIS of 2011 makes a vigorous effort to provide data on Philippine households that can be used to adequately describe the prevailing living standards, sources and uses of household funds, as well as the household members’ health status and educational attainment at the national and regional levels. The survey is designed to ensure that the data produced are representative of the state of the Philippine households for this period. To this end, the survey makes use of a three-stage sampling design in each stratum within a region. Primary sampling units (PSUs) are selected in stage 1, with probability proportional to the number of households derived from the 2000 Census of Population and Housing. In each PSU, enumeration areas (EA) are selected in stage 2, again with probability proportional to the number of households. In stage 3, housing units are selected in each sampled EA, using systematic sampling. Detailed information on the APIS data sets can be found in Ericta and Luis (2009).

The dimensions used by this study's MPI follow the main grouping of the global MPI first reported in the UNDP's 2010 Human Development Report (HDR). Like the HDI and global MPI, this study has three major dimensions—health, education, and living standards. Here, the dimensions are further disaggregated into 27 indicators. Some of the indicators of the study do not match those of the global MPI, however. For example, the latter has child mortality as a health indicator, which is not collected in the APIS. Instead, the APIS has information on who was sick in the previous month and who experienced hunger. Cooking fuel as one of the indicators of living standards is used in the global MPI but no information in the APIS data set can be used to substitute for this. Hence this study's standard of living dimension has one less indicator. Because of these differences, the MPI estimates of this study are not expected to match those reported in the HDR for the Philippines.

Table 1 shows the dimensions and their corresponding indicators, as well as the weights used in the calculation of the index. The final weight in the last column of Table 1 is derived as the product of the weights assigned to a dimension and its indicator. The weights assigned to elements in a group should sum to 1 (fractions in parentheses in the table). The three dimensions are each assigned a weight of one-third. A dimension has several indicators that are also given equal weights summing to 1. Thus, the health dimension has three indicators, each weighing one-third. The final weight of each health indicator is therefore 1/3 ⋅ 1/3 = 1/9. Some indicators are further grouped and given equal within-group weights. The assets indicator as shown in the table has four subgroups covering transportation, household equipment, communication, and entertainment, which are also assigned equal weights. As an example, the motorcycle asset's final weight is calculated as 1/3 ⋅ 1/5 ⋅ 1/4 ⋅ 1/2 = 1/120 (because the assets indicator belongs to the standard of living dimension and there are two transportation asset types). The sum of the final weights can be verified to equal 1.

Table 1.
Dimensions, indicators, and weights
DIMENSION/indicator An individual in a household is deprived if: Final weight 
EDUCATION (1/3)   
Household head's education (1/3) Head had no schooling or did not graduate from elementary 1/9 
School participation (1/3) At least 1 non-head household member of school age was not able to go to school 1/9 
Schooling level (1/3) No non-head household member graduated from elementary 1/9 
HEALTH (1/3)   
Household member's health (1/3) More than 50% of household members (excluding head and spouse) were sick during the preceding month 1/9 
Head/spouse's health (1/3) Head or spouse was sick during the preceding month 1/9 
Hunger perception (1/3) At least 1 household member went hungry at least once in 3 months 1/9 
STANDARD OF LIVING (1/3)   
Electricity (1/5) No electricity in the dwelling place 1/15 
Sanitation (1/5) Open pit toilet or worse 1/15 
Water supply (1/5) Water supply is from undeveloped spring or worse 1/15 
Housing quality (1/5) Wall is made of light materials (1/3) 1/45 
 Roof is made of light materials (1/3) 1/45 
 Floor area of dwelling is 2 m2/person or less (1/3) 1/45 
Assets (1/5) Household does not possess: (1/4)  
 Car (1/2) 1/120 
 Motorcycle (1/2) 1/120 
 Air conditioner (1/4) 1/240 
 Washing machine (1/4) 1/240 
 Refrigerator (1/4) 1/240 
 Stove with oven (1/4) 1/240 
 Cellphone (1/5) 1/300 
 Phone [landline] (1/5) 1/300 
 Computer (1/5) 1/300 
 TV (1/5) 1/300 
 Radio (1/5) 1/300 
 Audio equipment (1/4) 1/240 
 Video equipment (1/4) 1/240 
 DVD player (1/4) 1/240 
 Karaoke (1/4) 1/240 
DIMENSION/indicator An individual in a household is deprived if: Final weight 
EDUCATION (1/3)   
Household head's education (1/3) Head had no schooling or did not graduate from elementary 1/9 
School participation (1/3) At least 1 non-head household member of school age was not able to go to school 1/9 
Schooling level (1/3) No non-head household member graduated from elementary 1/9 
HEALTH (1/3)   
Household member's health (1/3) More than 50% of household members (excluding head and spouse) were sick during the preceding month 1/9 
Head/spouse's health (1/3) Head or spouse was sick during the preceding month 1/9 
Hunger perception (1/3) At least 1 household member went hungry at least once in 3 months 1/9 
STANDARD OF LIVING (1/3)   
Electricity (1/5) No electricity in the dwelling place 1/15 
Sanitation (1/5) Open pit toilet or worse 1/15 
Water supply (1/5) Water supply is from undeveloped spring or worse 1/15 
Housing quality (1/5) Wall is made of light materials (1/3) 1/45 
 Roof is made of light materials (1/3) 1/45 
 Floor area of dwelling is 2 m2/person or less (1/3) 1/45 
Assets (1/5) Household does not possess: (1/4)  
 Car (1/2) 1/120 
 Motorcycle (1/2) 1/120 
 Air conditioner (1/4) 1/240 
 Washing machine (1/4) 1/240 
 Refrigerator (1/4) 1/240 
 Stove with oven (1/4) 1/240 
 Cellphone (1/5) 1/300 
 Phone [landline] (1/5) 1/300 
 Computer (1/5) 1/300 
 TV (1/5) 1/300 
 Radio (1/5) 1/300 
 Audio equipment (1/4) 1/240 
 Video equipment (1/4) 1/240 
 DVD player (1/4) 1/240 
 Karaoke (1/4) 1/240 

Source: Author's calculation.

The table shows the description of each indicator. The education dimension has three indicators, which separates the household head's education to give it more weight relative to other household members’ education. In the health dimension, the household head/spouse's health is also a separate indicator. The main health question in the APIS asks who was sick in the preceding month but does not ask the type or seriousness of the illness. With the available health information, the study assumes that the household is health-poor if more than 50 percent of household members were sick during the preceding month.

This indicator is designed to avoid capturing a situation where the household is not health poor because the sickness of, say, a few members (less than a majority) is due to some common illness that affects a healthy normal person. It is therefore assumed that the household is health poor regardless of the illness type if more than 50 percent are ill. The third health indicator makes use of the hunger perception data in the APIS. The indicators in the standard of living dimension are like those in the HDR except for housing quality. In this study, housing-poor households have dwelling places where walls or roofs are made of light materials or where the floor area per household member is 2 m2 or less.

The deprivation cutoffs are based on the deprivation definitions in Table 1, and the poverty cutoff is set to one-third, as in the global MPI in Alkire and Santos (2014). For the latter cutoff, several combinations of deprivations will classify individuals in a household as poor. For example, being deprived in all the indicators in each of the major dimensions alone classifies an individual to be multidimensionally poor. A combination of indicators, say, two health indicators plus one education indicator will do the same.

The AF methodology produces a measure of poverty based on several deprivations experienced simultaneously by the individual. Hence, an important part of the analysis is an assessment of the joint distribution of the indicators used in the calculation of the index.3 If the same individuals are shown to be deprived simultaneously in two indicators while the rest are not, then one of them is redundant and could be dropped from the index. Appendix A provides an assessment of redundancies in this paper.

4.  Poverty picture at disaggregated levels

The MPI estimates are presented in Tables 2 through 7. These are obtained from a sample of 42,061 households with an average family size of 4.6. Of the total sample population, 15.5 percent are identified to be multidimensionally poor as shown in the second column (H) of Table 2. The incidence of poverty is higher in rural compared with urban areas. This is also reflected in the huge urban–rural difference in the MPIs: 10.7 versus 2.9. Nonethless, as the table shows, there is a negligible difference in poverty intensity (defined as the average deprivation score of the poor population in the area). The average of the urban MPI and the rural MPI weighted by their population shares shown in the last column of Table 2 yields the national or aggregate MPI for the Philippines of 7.2.4

Table 2.
Multidimensional poverty index (by area)
  Incidence Intensity Sample population Percent 
 M0 H A Poor Total of total 
Aggregate 7.18 0.1552 46.29 29,965 193,092 100.00 
Urban 2.89 0.0645 44.86 5,631 87,315 45.22 
Rural 10.73 0.2301 46.62 24,334 105,777 54.78 
  Incidence Intensity Sample population Percent 
 M0 H A Poor Total of total 
Aggregate 7.18 0.1552 46.29 29,965 193,092 100.00 
Urban 2.89 0.0645 44.86 5,631 87,315 45.22 
Rural 10.73 0.2301 46.62 24,334 105,777 54.78 

Source: Basic data from APIS 2011.

Of the aggregate MPI, Table 3 shows that 18.2 percent is accounted for by the urban area and 81.8 percent is due to the rural population. The bottom panel of this table also shows the urban–rural dimensional breakdown as percent to total. The health dimension accounts for the lowest share, at 24 percent of the aggregate MPI, and education and standard of living dimensions come close to each other at 39 percent and 37 percent, respectively. These results are further disaggregated into indicators of the dimensions by subgroups in Table 4 as percent of aggregate MPI. Close to 30 percent of the aggregate MPI is due to the household head's health and education. Educational attainment and schooling level of household members who are not household heads receive a 21 percent share. In the living standards dimension, the collection of 15 assets account for 13 percent, followed by electricity at 8 percent. Sanitation, water supply, and housing quality together account for 16 percent of total aggregate MPI.

Table 3.
Multidimensional poverty index (by area and by dimension)
  Dimension 
 M0 Healtha Educationa Standard of livinga 
Aggregate 7.18 1.74 2.78 2.67 
Urban 2.89 0.94 1.07 0.88 
Rural 10.73 2.40 4.19 4.14 
  Percent distribution 
    Standard 
 M0 Health Education of living 
Aggregate 100.00 24.20 38.67 37.13 
Urban 18.21 5.93 6.74 5.54 
Rural 81.79 18.27 31.93 31.59 
  Dimension 
 M0 Healtha Educationa Standard of livinga 
Aggregate 7.18 1.74 2.78 2.67 
Urban 2.89 0.94 1.07 0.88 
Rural 10.73 2.40 4.19 4.14 
  Percent distribution 
    Standard 
 M0 Health Education of living 
Aggregate 100.00 24.20 38.67 37.13 
Urban 18.21 5.93 6.74 5.54 
Rural 81.79 18.27 31.93 31.59 

Source: Basic data from APIS 2011.

Note: a. weighted sum of censored headcount ratios of indicators in the dimension.

Table 4.
Multidimensional poverty index (by indicators, percent of total)
DIMENSION/indicator Aggregate Urban Rural 
[1] HEALTH 24.20 5.93 18.27 
Head/spouse's health 11.77 2.61 9.16 
Household member's health 5.79 1.52 4.27 
Hunger perception 6.64 1.80 4.84 
[2] EDUCATION 38.67 6.74 31.93 
School participation 12.24 2.25 9.99 
Schooling level 8.74 1.53 7.21 
Household head's education 17.70 2.97 14.73 
[3] STANDARD OF LIVING 37.13 5.54 31.59 
Sanitation 5.50 0.79 4.71 
Water supply 4.85 0.48 4.38 
Electricity 7.98 1.00 6.99 
Housing quality (roof) 2.17 0.33 1.84 
Housing quality (wall) 2.67 0.43 2.24 
Housing quality (floor area) 0.63 0.11 0.52 
Assets 13.33 2.41 10.92 
Total: [1] + [2] + [3] 100.00 18.21 81.79 
DIMENSION/indicator Aggregate Urban Rural 
[1] HEALTH 24.20 5.93 18.27 
Head/spouse's health 11.77 2.61 9.16 
Household member's health 5.79 1.52 4.27 
Hunger perception 6.64 1.80 4.84 
[2] EDUCATION 38.67 6.74 31.93 
School participation 12.24 2.25 9.99 
Schooling level 8.74 1.53 7.21 
Household head's education 17.70 2.97 14.73 
[3] STANDARD OF LIVING 37.13 5.54 31.59 
Sanitation 5.50 0.79 4.71 
Water supply 4.85 0.48 4.38 
Electricity 7.98 1.00 6.99 
Housing quality (roof) 2.17 0.33 1.84 
Housing quality (wall) 2.67 0.43 2.24 
Housing quality (floor area) 0.63 0.11 0.52 
Assets 13.33 2.41 10.92 
Total: [1] + [2] + [3] 100.00 18.21 81.79 

Source: Basic data from APIS 2011.

The rural–urban disparity can again be seen in the censored headcount ratios in Table 5. Excluding the asset indicators from the comparison, the ratio is highest in the rural poor's education of the household head, where 17 percent of household heads had either no schooling or did not finish elementary schooling.

Table 5.
Censored headcount ratios: Percent of the (subgroup) population who are both poor and deprived in the indicator
 Health Education 
 Head and spouse's health Household member's health Hunger perception School participation Schooling level Household head's education 
Aggregate 7.61 3.74 4.29 7.91 5.65 11.44 
Urban 3.73 2.17 2.57 3.21 2.18 4.25 
Rural 10.81 5.04 5.71 11.79 8.51 17.38 
Standard of living 
  Water  Housing Housing Housing 
 Sanitation supply Electricity (roof) (wall) (floor area) 
Aggregate 5.93 5.23 8.60 7.00 8.62 2.02 
Urban 1.89 1.14 2.38 2.35 3.04 0.79 
Rural 9.26 8.61 13.75 10.84 13.23 3.04 
Assets 
   Air Washing   
 Car Motorcycle conditioner machine Stove  
Aggregate 15.41 14.60 15.43 15.10 15.20  
Urban 6.35 5.95 6.31 5.97 6.14  
Rural 22.88 21.74 22.96 22.63 22.69  
  Audio     
 Refrigerator equipment Video DVD Karaoke  
Aggregate 14.79 15.21 15.39 13.22 15.04  
Urban 5.79 6.16 6.33 4.98 6.11  
Rural 22.23 22.67 22.87 20.02 22.41  
 Cellphone Telephone Computer TV Radio  
Aggregate 9.35 15.37 15.37 11.37 11.56  
Urban 3.24 6.32 6.25 3.71 4.86  
Rural 14.39 22.85 22.90 17.68 17.08  
 Health Education 
 Head and spouse's health Household member's health Hunger perception School participation Schooling level Household head's education 
Aggregate 7.61 3.74 4.29 7.91 5.65 11.44 
Urban 3.73 2.17 2.57 3.21 2.18 4.25 
Rural 10.81 5.04 5.71 11.79 8.51 17.38 
Standard of living 
  Water  Housing Housing Housing 
 Sanitation supply Electricity (roof) (wall) (floor area) 
Aggregate 5.93 5.23 8.60 7.00 8.62 2.02 
Urban 1.89 1.14 2.38 2.35 3.04 0.79 
Rural 9.26 8.61 13.75 10.84 13.23 3.04 
Assets 
   Air Washing   
 Car Motorcycle conditioner machine Stove  
Aggregate 15.41 14.60 15.43 15.10 15.20  
Urban 6.35 5.95 6.31 5.97 6.14  
Rural 22.88 21.74 22.96 22.63 22.69  
  Audio     
 Refrigerator equipment Video DVD Karaoke  
Aggregate 14.79 15.21 15.39 13.22 15.04  
Urban 5.79 6.16 6.33 4.98 6.11  
Rural 22.23 22.67 22.87 20.02 22.41  
 Cellphone Telephone Computer TV Radio  
Aggregate 9.35 15.37 15.37 11.37 11.56  
Urban 3.24 6.32 6.25 3.71 4.86  
Rural 14.39 22.85 22.90 17.68 17.08  

Source: Basic data from APIS 2011.

Note: The aggregate value is the average of subgroup values weighted by population share. The overall MPI of 7.18 can be derived as the average of the aggregate values of the 27 indicators above using the final weights in Table 1.

The following are notable: 14 percent of the rural poor have no electricity; 14.4 percent of the rural poor are deprived of a cellphone (and is the lowest percentage in the assets category); 5.7 percent of the rural poor experienced hunger at least once in the 3 months prior to the survey.

A disaggregation by administrative region is presented in Table 6. There are three main island groupings in the Philippines: Luzon, Visayas, and Mindanao. Luzon, the biggest island, incorporates eight regions. Visayas, a group of islands in the middle of the Philippine archipelago, has three regions, and Mindanao, in the south, is composed of six regions. Each region is composed of provinces and cities. There are currently 81 provinces excluding the national capital region (Metro Manila), which is divided into four districts.

Table 6.
Multidimensional poverty index by administrative region
     Percent contribution 
REGION M0 H A % of n M0 Health Education Standard of living 
Aggregate MPI 7.18 0.1552 46.29      
National Capital Region 0.65 0.0158 41.20 10.95 1.00 0.33 0.42 0.24 
Cordillera Administrative Region 4.44 0.1068 41.52 4.07 2.51 0.53 1.06 0.93 
Region I - Ilocos Region 1.82 0.0434 41.90 5.29 1.34 0.45 0.52 0.37 
Region II - Cagayan Valley 3.94 0.0906 43.42 4.65 2.55 0.61 1.17 0.76 
Region III - Central Luzon 2.25 0.0539 41.72 7.82 2.45 0.99 0.87 0.58 
Region IVA - CALABARZON 2.02 0.0484 41.73 9.47 2.66 0.71 1.09 0.86 
Region IVB - MIMAROPA 11.02 0.2237 49.25 4.09 6.27 1.47 2.34 2.46 
Region V- Bicol 9.10 0.1968 46.23 5.82 7.38 2.18 2.14 3.05 
Region VI - Western Visayas 9.39 0.2063 45.53 6.70 8.76 2.57 2.77 3.42 
Region VII - Central Visayas 10.39 0.2203 47.15 6.65 9.61 2.51 3.41 3.69 
Region VIII - Eastern Visayas 11.34 0.2396 47.33 5.45 8.60 2.57 3.13 2.90 
Region IX - Zamboanga Peninsula 11.78 0.2510 46.94 4.25 6.96 1.35 2.79 2.83 
Region X - Northern Mindanao 8.74 0.1872 46.67 4.48 5.45 1.48 2.09 1.87 
Region XI - Davao 8.49 0.1796 47.29 5.46 6.46 1.30 2.83 2.33 
Region XII - SOCCSKSARGEN 12.54 0.2510 49.95 5.15 8.99 2.10 3.56 3.33 
Region XIII - Caraga 8.72 0.1901 45.89 4.38 5.32 1.80 1.74 1.78 
ARMM 18.52 0.4055 45.67 5.32 13.71 1.25 6.75 5.71 
Total    100.00 100.00 24.20 38.67 37.13 
     Percent contribution 
REGION M0 H A % of n M0 Health Education Standard of living 
Aggregate MPI 7.18 0.1552 46.29      
National Capital Region 0.65 0.0158 41.20 10.95 1.00 0.33 0.42 0.24 
Cordillera Administrative Region 4.44 0.1068 41.52 4.07 2.51 0.53 1.06 0.93 
Region I - Ilocos Region 1.82 0.0434 41.90 5.29 1.34 0.45 0.52 0.37 
Region II - Cagayan Valley 3.94 0.0906 43.42 4.65 2.55 0.61 1.17 0.76 
Region III - Central Luzon 2.25 0.0539 41.72 7.82 2.45 0.99 0.87 0.58 
Region IVA - CALABARZON 2.02 0.0484 41.73 9.47 2.66 0.71 1.09 0.86 
Region IVB - MIMAROPA 11.02 0.2237 49.25 4.09 6.27 1.47 2.34 2.46 
Region V- Bicol 9.10 0.1968 46.23 5.82 7.38 2.18 2.14 3.05 
Region VI - Western Visayas 9.39 0.2063 45.53 6.70 8.76 2.57 2.77 3.42 
Region VII - Central Visayas 10.39 0.2203 47.15 6.65 9.61 2.51 3.41 3.69 
Region VIII - Eastern Visayas 11.34 0.2396 47.33 5.45 8.60 2.57 3.13 2.90 
Region IX - Zamboanga Peninsula 11.78 0.2510 46.94 4.25 6.96 1.35 2.79 2.83 
Region X - Northern Mindanao 8.74 0.1872 46.67 4.48 5.45 1.48 2.09 1.87 
Region XI - Davao 8.49 0.1796 47.29 5.46 6.46 1.30 2.83 2.33 
Region XII - SOCCSKSARGEN 12.54 0.2510 49.95 5.15 8.99 2.10 3.56 3.33 
Region XIII - Caraga 8.72 0.1901 45.89 4.38 5.32 1.80 1.74 1.78 
ARMM 18.52 0.4055 45.67 5.32 13.71 1.25 6.75 5.71 
Total    100.00 100.00 24.20 38.67 37.13 

Source: Basic data from APIS 2011.

Note: MIMAROPA = Mindoro, Marinduque, Romblon, and Palawan; CALABARZON = Cavite, Laguna, Batangas, Rizal, and Quezon; SOCCSKSARGEN = South Cotabato, Cotabato, Sultan Kudarat, Sarangani, and General Santos City.

The results show that the Autonomous Region in Muslim Mindanao (ARMM) is the poorest region in the Philippines with an MPI of 18.5. A distant second, with an index of 12.5, is SOCCSKSARGEN (a region named with the acronym that stands for the region's four provinces and one major city), which is also in Mindanao. The MPI by region is shown in Figure 1. In column (5) of Table 6, we see that whereas ARMM accounts for the highest percent share of the aggregate MPI at 13.7 percent, SOCCSKSARGEN occupies the third rank with a share of 9.0 percent. Central Visayas, which has a lower MPI of 10.4 but has a larger population share, has the second highest share, with 9.6 percent.

Figure 1.

Multidimensional poverty index (by region)

Figure 1.

Multidimensional poverty index (by region)

The MPI, and its components the incidence and intensity of poverty, are shown in Figures 1, 2, and 3, respectively. It is quite clear in Figure 1 that only 6 out of the 17 regions, all located in Luzon, have MPIs that are below the national MPI of 7.2. Only two Luzon regions, Bicol and MIMAROPA, have MPIs above the national average. A wide difference exists between the regional groupings above and below the national MPI. The regional ranking in terms of poverty incidence in Figure 2 is no different from the MPI ranking and the same six regions have an incidence below the national average. Poverty intensity is above the national average for 7 out of 17 regions, as shown by the first seven bars of Figure 3. Four are in Mindanao, two are in the Visayas, and one in Luzon. Although the incidence of poverty is highest in ARMM, it is in SOCCSKSARGEN where poverty is most intense.

Figure 2.

Headcount ratio (H by region)

Figure 2.

Headcount ratio (H by region)

Figure 3.

Poverty intensity (A by region).

Figure 3.

Poverty intensity (A by region).

Table 7 has three panels that show the top 10 provinces sorted by MPI, by poverty incidence, and by poverty intensity.5 The first panel of the table shows that Saranggani in SOCCSKSARGEN is the poorest province in the Philippines, with an MPI of 23.5. Given the regional MPI ranking, it is not surprising that four out of the five provinces of ARMM are in the top 10 list of poorest provinces. Masbate in the Bicol region has the third highest MPI. The second and third panels of Table 7 show 10 provinces with the highest incidence and intensity of poverty, respectively. Tawi-Tawi has the highest poverty incidence, followed by Masbate and Sulu. At first glance, it seems that high incidence goes together with high intensity of poverty (see the second panel of the table). This is not the case, however, when provinces are sorted according to intensity as seen in the bottom panel of the table.

Table 7.
Multidimensional poverty index: 10 poorest provinces by MPI, by incidence and by intensity
Region Province M0 H A Mediana income Meana income 
 (1) Sorted by M0 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
ARMM Tawi-Tawi 22.93 0.4980 46.05 8,355 10,389 
Region V- Bicol Masbate 22.31 0.4586 48.64 9,740 15,263 
ARMM Maguindanao 20.59 0.4422 46.57 6,967 8,774 
ARMM Sulu 20.33 0.4502 45.16 11,580 12,950 
Region VIII - Eastern Visayas Eastern Samar 18.49 0.3816 48.45 9,072 16,238 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
Region VIII - Eastern Visayas Biliran 15.12 0.3110 48.60 11,972 25,076 
Region IX - Zamboanga Peninsula Zamboanga del Norte 14.74 0.3105 47.48 7,988 12,911 
 (2) Sorted by incidence (H) 
ARMM Tawi-Tawi 22.93 0.4980 46.05 8,355 10,389 
Region V- Bicol Masbate 22.31 0.4586 48.64 9,740 15,263 
ARMM Sulu 20.33 0.4502 45.16 11,580 12,950 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
ARMM Maguindanao 20.59 0.4422 46.57 6,967 8,774 
Region VIII - Eastern Visayas Eastern Samar 18.49 0.3816 48.45 9,072 16,238 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region VII - Central Visayas Siquijor 14.72 0.3349 43.96 17,262 31,631 
Region VIII - Eastern Visayas Samar (Western) 14.63 0.3121 46.89 9,283 17,817 
 (3) Sorted by intensity (A) 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
Region IVB - MIMAROPA Palawan 13.86 0.2668 51.93 11,252 19,119 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region VI - Western Visayas Antique 9.75 0.1940 50.28 9,652 18,034 
Region X - Northern Mindanao Bukidnon 14.31 0.2848 50.25 11,300 17,468 
Region IVB - MIMAROPA Oriental Mindoro 9.30 0.1857 50.09 12,126 17,748 
Region XII - SOCCSKSARGEN South Cotabato 8.65 0.1731 50.00 14,120 22,476 
Region VII - Central Visayas Bohol 9.13 0.1850 49.37 10,414 18,919 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
Region XII - SOCCSKSARGEN Cotabato 13.26 0.2711 48.92 11,087 18,707 
Aggregate MPI  7.18 0.1551 46.29   
Region Province M0 H A Mediana income Meana income 
 (1) Sorted by M0 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
ARMM Tawi-Tawi 22.93 0.4980 46.05 8,355 10,389 
Region V- Bicol Masbate 22.31 0.4586 48.64 9,740 15,263 
ARMM Maguindanao 20.59 0.4422 46.57 6,967 8,774 
ARMM Sulu 20.33 0.4502 45.16 11,580 12,950 
Region VIII - Eastern Visayas Eastern Samar 18.49 0.3816 48.45 9,072 16,238 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
Region VIII - Eastern Visayas Biliran 15.12 0.3110 48.60 11,972 25,076 
Region IX - Zamboanga Peninsula Zamboanga del Norte 14.74 0.3105 47.48 7,988 12,911 
 (2) Sorted by incidence (H) 
ARMM Tawi-Tawi 22.93 0.4980 46.05 8,355 10,389 
Region V- Bicol Masbate 22.31 0.4586 48.64 9,740 15,263 
ARMM Sulu 20.33 0.4502 45.16 11,580 12,950 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
ARMM Maguindanao 20.59 0.4422 46.57 6,967 8,774 
Region VIII - Eastern Visayas Eastern Samar 18.49 0.3816 48.45 9,072 16,238 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region VII - Central Visayas Siquijor 14.72 0.3349 43.96 17,262 31,631 
Region VIII - Eastern Visayas Samar (Western) 14.63 0.3121 46.89 9,283 17,817 
 (3) Sorted by intensity (A) 
Region XII - SOCCSKSARGEN Saranggani 23.47 0.4453 52.70 8,510 14,139 
Region IVB - MIMAROPA Palawan 13.86 0.2668 51.93 11,252 19,119 
ARMM Basilan 17.66 0.3437 51.39 11,408 14,365 
Region VI - Western Visayas Antique 9.75 0.1940 50.28 9,652 18,034 
Region X - Northern Mindanao Bukidnon 14.31 0.2848 50.25 11,300 17,468 
Region IVB - MIMAROPA Oriental Mindoro 9.30 0.1857 50.09 12,126 17,748 
Region XII - SOCCSKSARGEN South Cotabato 8.65 0.1731 50.00 14,120 22,476 
Region VII - Central Visayas Bohol 9.13 0.1850 49.37 10,414 18,919 
Region XI - Davao Davao Oriental 16.99 0.3456 49.17 9,561 13,651 
Region XII - SOCCSKSARGEN Cotabato 13.26 0.2711 48.92 11,087 18,707 
Aggregate MPI  7.18 0.1551 46.29   

Source: Basic data from APIS 2011.

Note: a. Nationwide per capita income poverty threshold = 9,465.00 Philippine pesos (see footnote 6 for more details).

As we can see, Saranggani, which has one of the highest levels of poverty incidence, tops the list in terms of poverty intensity. The next six provinces, which have intensity indices of 50 or higher, have a poverty incidence that is not as pronounced. Nonetheless, there is a positive relation between incidence and intensity. This can be observed in Figure 4, which plots the two sub-indices against each other. The area with the lowest MPI, Metro Manila's second district, is identified in the graph. To avoid clutter, only provinces with the highest incidences (H ≥ 0.40) and those with the highest intensities (A ≥ 50.0) are labeled in the graph. Saranggani earns the distinction as the province that belongs to both sets. These results can help policymakers decide on what type of poverty reduction policy should be pursued on a per province basis (i.e., determine whether a province should be subjected to policies designed to minimize intensity or to lessen incidence).

Figure 4.

Head count ratio and poverty intensity by province (2011)

Figure 4.

Head count ratio and poverty intensity by province (2011)

How does the MPI and its components relate to the traditional income measure of poverty? A partial answer is most conveniently demonstrated visually. Figures 5 and 6 plot the MPI against mean and median per capita income, respectively. In both diagrams, provinces with high per capita incomes generally have low MPIs. The relationship appears to be nonlinear. This is reflected in both the level and dispersion of per capita income, which are higher for the top five low-MPI provinces when compared with the top five high-MPI provinces. The horizontal dashed line in the diagrams segregates provinces above and below the per capita income poverty threshold. The threshold of 9,465.00 Philippine pesos is the mean of the per capita income of households in the third income decile in the APIS 2011 data set used by this study.6 As can be seen, only Maguindanao falls below the income poverty line when the mean is used whereas eight provinces—Maguindanao, Zamboanga del Norte, Sultan Kudarat, Lanao del Sur, Tawi-Tawi, Saranggani, Eastern Samar, Western Samar—are below the line when the median is used. All these provinces have MPIs above the national aggregate MPI of 7.18, along with 35 other provinces that are above the income poverty line.

Figure 5.

MPI and mean per capita income by province (2011)

Figure 5.

MPI and mean per capita income by province (2011)

Figure 6.

MPI and median per capita income by province (2011)

Figure 6.

MPI and median per capita income by province (2011)

It is clear from the discussion that the MPI can be a very useful guide to policy because it can be disaggregated by dimension and by geographic area. With different levels of disaggregation, the areas can be ranked and areas facing severe budget constraints can be given attention and aid. More than this, these calculations indicate a clear direction for policy at the national and regional levels. This is further discussed in Section 6.

5.  Analysis at the household level

The index is derived by aggregating deprivation indicators from household survey data. Ironically, however, the AF methodology does not provide for analysis at the household level. It is therefore up to the researcher to lay out in detail the characteristics of the poor household that sets it apart from the non-poor household. That is, more statistical analysis needs to be done after the setting of the household deprivation thresholds (first cutoff), the calculation of the individual deprivation scores and the identification of the poor population through the poverty (second) cutoff. Data sets that can produce an MPI are rich in household information that can help validate the results of the MPI calculations.

The variable of interest in the household level analysis of this study is a binary variable constructed using a poverty cutoff of one-third. This variable segregates the poor from the non-poor and is used in generating the individual's censored deprivation scores to arrive at the MPI in the preceding section.7 Table 8 presents the cross-tabulation of household size against this variable. Of the 42,061 households, 15.6 percent are identified as poor. Households with three to four members account for 35 percent of the total. This household group also has the lowest within-group proportion of the poor (12 percent) and the highest share of the total non-poor households (36.5 percent).

Table 8.
Household size
 Number of households Row percent Column percent 
 Poor Non-poor Total Poor Non-poor Total Poor Non-poor Total 
Smallest (1/2) 1,615 5,696 7,311 22.09 77.91 100.00 24.60 16.05 17.38 
Small (3/4) 1,773 12,968 14,741 12.03 87.97 100.00 27.01 36.53 35.05 
Medium (5/6) 1,780 10,768 12,548 14.19 85.81 100.00 27.11 30.34 29.83 
Large (7/9) 1,179 5,202 6,381 18.48 81.52 100.00 17.96 14.66 15.17 
Extra large (≥10) 218 862 1,080 20.19 79.81 100.00 3.32 2.42 2.57 
Total 6,565 35,496 42,061 15.61 84.39 100.00 100.00 100.00 100.00 
 Number of households Row percent Column percent 
 Poor Non-poor Total Poor Non-poor Total Poor Non-poor Total 
Smallest (1/2) 1,615 5,696 7,311 22.09 77.91 100.00 24.60 16.05 17.38 
Small (3/4) 1,773 12,968 14,741 12.03 87.97 100.00 27.01 36.53 35.05 
Medium (5/6) 1,780 10,768 12,548 14.19 85.81 100.00 27.11 30.34 29.83 
Large (7/9) 1,179 5,202 6,381 18.48 81.52 100.00 17.96 14.66 15.17 
Extra large (≥10) 218 862 1,080 20.19 79.81 100.00 3.32 2.42 2.57 
Total 6,565 35,496 42,061 15.61 84.39 100.00 100.00 100.00 100.00 

Source: Basic data from APIS 2011.

Table 9 provides information on the educational attainment of the head of the poor and non-poor households. Seventy-five percent of household heads of the poor population did not finish elementary schooling. Add those who did not finish high school (labeled Elementary + in Table 9) and this goes up to 92.3 percent. Household heads with no schooling account for only 3.2 percent of the total but 70.9 percent of them belong to the poor households. It is clear from the table that the proportion of the poor within groups goes down as the head's education rises.

Table 9.
Household head's education
 Number of households Row percent Column percent 
 Poor Non-poor Total Poor Non-poor Total Poor Non-poor Total 
No schooling 961 395 1,356 70.87 29.13 100.00 14.64 1.11 3.22 
Some elementary 3,967 4,807 8,774 45.21 54.79 100.00 60.43 13.54 20.86 
Elementary + 1,131 11,675 12,806 8.83 91.17 100.00 17.23 32.89 30.45 
High school + 455 13,777 14,232 3.20 96.80 100.00 6.93 38.81 33.84 
College + 51 4,842 4,893 1.04 98.96 100.00 0.77 13.65 11.63 
Total 6,565 35,496 42,061 15.61 84.39 100.00 100.00 100.00 100.00 
 Number of households Row percent Column percent 
 Poor Non-poor Total Poor Non-poor Total Poor Non-poor Total 
No schooling 961 395 1,356 70.87 29.13 100.00 14.64 1.11 3.22 
Some elementary 3,967 4,807 8,774 45.21 54.79 100.00 60.43 13.54 20.86 
Elementary + 1,131 11,675 12,806 8.83 91.17 100.00 17.23 32.89 30.45 
High school + 455 13,777 14,232 3.20 96.80 100.00 6.93 38.81 33.84 
College + 51 4,842 4,893 1.04 98.96 100.00 0.77 13.65 11.63 
Total 6,565 35,496 42,061 15.61 84.39 100.00 100.00 100.00 100.00 

Source: Basic data from APIS 2011.

Data describing household characteristics can be used to analyze the state of well-being of households using econometric methods. Desai and Shah (1988) were one of the first groups to use econometric techniques to analyze relative deprivation in Great Britain. More recently, Whelan, Nolan, and Maître (2014) explains deprivation scores calculated for several European countries using household characteristics.

Column (1) of Table 10 presents the standard logit estimate of the state of the household where the dependent variable takes the value of 1 if the household is poor and zero otherwise. The household head's education described in Table 9 is represented by categorical dummies. Sex, health of the household head, and overseas Filipino worker (OFW) transfers to households are dummy variables. The rest are continuous variables in logarithms except household size and household health, which is defined by the percentage of non-head household members who were sick at the time the survey was conducted. To account for nonlinearities in household size visible from Table 8, its square is included as a separate variable. As seen in Table 10, all variables except squared household size are significant and signed in accordance with expectations.

Table 10.
Logit estimation results I
 Multidimensional poverty Income poverty 
 (1) Standard logit (2) Margins (dy/dx(3) Mixed logit (4) Margins (dy/dx(5) Standard logit (6) Margins (dy/dx(7) Mixed logit (8) Margins (dy/dx
MPI poor     0.716 0.107 0.642 0.090 
     (0.047)** (0.007)** (0.085)** (0.013)** 
Household income −1.155 −0.065 −1.120 −0.061     
 (0.044)** (0.002)** (0.058)** (0.003)**     
Household size 0.201 0.011 0.138 0.008 0.417 0.057 0.482 0.064 
 (0.042)** (0.002)** (0.064)* (0.003)* (0.031)** (0.004)** (0.049)** (0.007)** 
(Household size)2 −0.001 0.000 0.002 0.000 −0.018 −0.002 −0.020 −0.003 
 (0.003) (0.000) (0.004) (0.000) (0.002)** (0.000)** (0.004)** (0.000)** 
Household head's education: (Base category: No schooling)     
Some elementary −0.774 −0.092 −0.511 −0.057 0.099 0.017 0.017 0.003 
 (0.092)** (0.012)** (0.104)** (0.013)** (0.089) (0.015) (0.207) (0.033) 
Elementary + −3.200 −0.277 −2.957 −0.235 −0.159 −0.027 −0.200 −0.031 
 (0.100)** (0.012)** (0.088)** (0.017)** (0.090)+ (0.015)+ (0.208) (0.033) 
High school + −3.768 −0.300 −3.434 −0.253 −1.014 −0.152 −0.909 −0.131 
 (0.114)** (0.012)** (0.102)** (0.018)** (0.092)** (0.015)** (0.228)** (0.036)** 
College + −4.144 −0.312 −3.925 −0.268 −2.538 −0.288 −2.642 −0.295 
 (0.217)** (0.013)** (0.250)** (0.021)** (0.128)** (0.016)** (0.276)** (0.042)** 
Household head's age −0.819 −0.046 −0.628 −0.034 0.876 0.120 0.634 0.084 
 (0.144)** (0.008)** (0.157)** (0.008)** (0.098)** (0.013)** (0.113)** (0.014)** 
Household head's sex −0.258 −0.014 −0.214 −0.011 −0.374 −0.050 −0.236 −0.031 
(female = 1) (0.066)** (0.004)** (0.057)** (0.003)** (0.042)** (0.005)** (0.053)** (0.007)** 
Household head's health 1.908 0.127 1.928 0.124 −0.091 −0.012 −0.190 −0.025 
(sick = 1) (0.051)** (0.004)** (0.083)** (0.005)** (0.038)* (0.005)* (0.039)** (0.005)** 
Percent of household 2.383 0.135 2.439 0.133 0.072 0.010 −0.141 −0.019 
members who were sick (0.069)** (0.004)** (0.095)** (0.004)** (0.051) (0.007) (0.096) (0.013) 
Average years of schooling −2.737 −0.155 −2.754 −0.151 −1.294 −0.177 −1.298 −0.171 
of household members (0.060)** (0.003)** (0.085)** (0.006)** (0.040)** (0.005)** (0.078)** (0.011)** 
Average age of 0.598 0.034 0.561 0.031 −1.486 −0.203 −1.448 −0.191 
household members (0.118)** (0.007)** (0.157)** (0.008)** (0.084)** (0.011)** (0.115)** (0.014)** 
Remittances −0.203 −0.011 −0.197 −0.011 −1.821 −0.198 −1.840 −0.202 
(receive = 1) (0.100)* (0.005)* (0.121) (0.006) (0.074)** (0.006)** (0.095)** (0.016)** 
σintercept2 (Region)   0.132    0.437  
   (0.066)*    (0.269)  
σintercept2 (Province)   0.218    0.406  
   (0.046)**    (0.124)**  
Pseudo-R2 0.542    0.292    
LR test vs. logit model (χ2)   494.99    3275.43  
p value   (0.000)    (0.000)  
 Multidimensional poverty Income poverty 
 (1) Standard logit (2) Margins (dy/dx(3) Mixed logit (4) Margins (dy/dx(5) Standard logit (6) Margins (dy/dx(7) Mixed logit (8) Margins (dy/dx
MPI poor     0.716 0.107 0.642 0.090 
     (0.047)** (0.007)** (0.085)** (0.013)** 
Household income −1.155 −0.065 −1.120 −0.061     
 (0.044)** (0.002)** (0.058)** (0.003)**     
Household size 0.201 0.011 0.138 0.008 0.417 0.057 0.482 0.064 
 (0.042)** (0.002)** (0.064)* (0.003)* (0.031)** (0.004)** (0.049)** (0.007)** 
(Household size)2 −0.001 0.000 0.002 0.000 −0.018 −0.002 −0.020 −0.003 
 (0.003) (0.000) (0.004) (0.000) (0.002)** (0.000)** (0.004)** (0.000)** 
Household head's education: (Base category: No schooling)     
Some elementary −0.774 −0.092 −0.511 −0.057 0.099 0.017 0.017 0.003 
 (0.092)** (0.012)** (0.104)** (0.013)** (0.089) (0.015) (0.207) (0.033) 
Elementary + −3.200 −0.277 −2.957 −0.235 −0.159 −0.027 −0.200 −0.031 
 (0.100)** (0.012)** (0.088)** (0.017)** (0.090)+ (0.015)+ (0.208) (0.033) 
High school + −3.768 −0.300 −3.434 −0.253 −1.014 −0.152 −0.909 −0.131 
 (0.114)** (0.012)** (0.102)** (0.018)** (0.092)** (0.015)** (0.228)** (0.036)** 
College + −4.144 −0.312 −3.925 −0.268 −2.538 −0.288 −2.642 −0.295 
 (0.217)** (0.013)** (0.250)** (0.021)** (0.128)** (0.016)** (0.276)** (0.042)** 
Household head's age −0.819 −0.046 −0.628 −0.034 0.876 0.120 0.634 0.084 
 (0.144)** (0.008)** (0.157)** (0.008)** (0.098)** (0.013)** (0.113)** (0.014)** 
Household head's sex −0.258 −0.014 −0.214 −0.011 −0.374 −0.050 −0.236 −0.031 
(female = 1) (0.066)** (0.004)** (0.057)** (0.003)** (0.042)** (0.005)** (0.053)** (0.007)** 
Household head's health 1.908 0.127 1.928 0.124 −0.091 −0.012 −0.190 −0.025 
(sick = 1) (0.051)** (0.004)** (0.083)** (0.005)** (0.038)* (0.005)* (0.039)** (0.005)** 
Percent of household 2.383 0.135 2.439 0.133 0.072 0.010 −0.141 −0.019 
members who were sick (0.069)** (0.004)** (0.095)** (0.004)** (0.051) (0.007) (0.096) (0.013) 
Average years of schooling −2.737 −0.155 −2.754 −0.151 −1.294 −0.177 −1.298 −0.171 
of household members (0.060)** (0.003)** (0.085)** (0.006)** (0.040)** (0.005)** (0.078)** (0.011)** 
Average age of 0.598 0.034 0.561 0.031 −1.486 −0.203 −1.448 −0.191 
household members (0.118)** (0.007)** (0.157)** (0.008)** (0.084)** (0.011)** (0.115)** (0.014)** 
Remittances −0.203 −0.011 −0.197 −0.011 −1.821 −0.198 −1.840 −0.202 
(receive = 1) (0.100)* (0.005)* (0.121) (0.006) (0.074)** (0.006)** (0.095)** (0.016)** 
σintercept2 (Region)   0.132    0.437  
   (0.066)*    (0.269)  
σintercept2 (Province)   0.218    0.406  
   (0.046)**    (0.124)**  
Pseudo-R2 0.542    0.292    
LR test vs. logit model (χ2)   494.99    3275.43  
p value   (0.000)    (0.000)  

Note: Standard errors in parentheses; +statistically significant at the 10 percent level; *statistically significant at the 5 percent level; **statistically significant at the 1 percent level.

Column (3) of Table 10 shows the estimation results from a mixed logit specification. This is a three-level logit model with a random intercept for regions and a random intercept for provinces. The latter is nested within the former. The specification takes advantage of the clustered structure of the data and accounts for unobserved inter-group differences as well as intra-group correlations that is not possible with standard logit.8 The variances of the intercepts and not the intercepts themselves are estimated and are reported in the table to be statistically significant in general. The likelihood ratio test indicates that the mixed logit estimate differs significantly from the standard logit.

The estimated parameters of the models are difficult to interpret because they are on a logit scale. One should instead calculate the average marginal effects of the explanatory variables on the (predicted) probabilities. The columns beside the parameter estimates show the average marginal effects of the explanatory variables on the predicted probability of being in the state of poverty. Note that the average marginal effects in the mixed model are consistently lower than the standard logit model in absolute value terms. As expected, higher incomes cut the probability of being poor. Thus, in column 4, an increase in household income of approximately 10 percent reduces the probability by 0.6 percentage points.9 Although household size is significant, its square is not and is nearly zero. Here, an additional household member raises poverty risk by 0.8 percentage points. All other variables are significant except the remittances dummy.

Note that some education and health variables used as predictors are related to some components of the MPI from which the binary dependent variable is derived. For example, one of the education indicators in the MPI is a binary variable that takes a value of 1 when the household head has no schooling or did not graduate from elementary and zero otherwise, and a variable used as a predictor in the logit model is a categorical variable that classifies the household head's education into five categories enumerated in Table 9. The extent and direction of the bias of this overlap on the results are unknown. To check the robustness of the results, the same predictor variables except income are used in the same model but using an income poverty variable as the dependent variable. This also serves to evaluate the income measure of poverty in relation to this study's measure. It should be evident from the numbers in Table 11 that persons identified as income poor are not necessarily multidimensionally poor.

Table 11.
Comparison of definition of poverty by MPI and by income
  Income  
  Non-poor Poor Total 
MPI Non-poor 66.3 18.1 84.4 
 Poor 6.6 9.0 15.6 
 Total 72.9 27.1 100.0 
 Total households: 42,061 
  Income  
  Non-poor Poor Total 
MPI Non-poor 66.3 18.1 84.4 
 Poor 6.6 9.0 15.6 
 Total 72.9 27.1 100.0 
 Total households: 42,061 

Note: See footnote 6 for income threshold explanation.

Close to 75 percent of the total number of households are either poor or non-poor according to both definitions of poverty, whereas 25 percent are poor or non-poor in one poverty definition but not in the other. Given this discrepancy in classification, there is a compelling reason to view the two poverty measures as complementing each other rather than as competing measures.

The logit results shown in columns (5)–(8) of Table 10 are, to some extent, comparable with the results shown in columns (1)–(4). The marginal effects of the size variables on income poverty probability are all significant and much larger. Note, however, that the coefficient of the household head's health variable is significant but of the wrong sign. The other health variable is statistically insignificant. The MPI is also used as an explanatory variable and is statistically significant. Results indicate that a discrete change in MPI from 0 to 1 raises income poverty risk by 10 percentage points.

The results in Table 10 imply that an additional household member has the same effect on poverty risk regardless of household size. To determine the extent of the effects of different sizes, the model is re-estimated where the household size variable is grouped into five categories (see Table 8). A second set of estimates is shown in Table 12. As can be seen in column 1 of this table, all variables are significant except for the small household size dummy. In the mixed logit estimate shown in column 3, the coefficient of the medium household size dummy becomes insignificant along with the small household size dummy and the remittances dummy's statistically significance is reduced to 10 percent.

Table 12.
Logit estimation results II
 Multidimensional poverty Income poverty 
 (1) Standard logit (2) Margins (dy/dx(3) Mixed logit (4) Margins(dy/dx(5) Standard logit (6) Margins (dy/dx(7) Mixed logit (8) Margins (dy/dx
Household income −1.114 −0.063 −1.075 −0.059     
 (0.043)** (0.002)** (0.058)** (0.003)**     
Household size: (Base category: 1 to 2)     
Small (3 to 4) −0.020 −0.001 −0.110 −0.006 0.301 0.038 0.389 0.047 
 (0.084) (0.004) (0.107) (0.006) (0.068)** (0.008)** (0.092)** (0.012)** 
Medium (5 to 6) 0.312 0.017 0.158 0.008 0.802 0.109 0.971 0.127 
 (0.101)** (0.005)** (0.135) (0.007) (0.074)** (0.009)** (0.099)** (0.015)** 
Large (7 to 9) 0.779 0.045 0.609 0.035 1.183 0.169 1.454 0.200 
 (0.117)** (0.007)** (0.188)** (0.010)** (0.082)** (0.011)** (0.102)** (0.016)** 
Extra large (≥10) 1.319 0.082 1.134 0.069 1.236 0.178 1.543 0.214 
 (0.162)** (0.011)** (0.191)** (0.011)** (0.108)** (0.016)** (0.140)** (0.022)** 
Household head's education: (Base category: No schooling)     
Some elementary −0.788 −0.094 −0.523 −0.058 −0.029 −0.005 −0.066 −0.011 
 (0.093)** (0.012)** (0.102)** (0.013)** (0.084) (0.015) (0.204) (0.033) 
Elementary + −3.221 −0.281 −2.975 −0.237 −0.483 −0.083 −0.452 −0.071 
 (0.100)** (0.012)** (0.088)** (0.017)** (0.083)** (0.015)** (0.223)* (0.037)+ 
High school + −3.803 −0.304 −3.463 −0.256 −1.35 −0.210 −1.171 −0.173 
 (0.115)** (0.012)** (0.103)** (0.018)** (0.086)** (0.015)** (0.243)** (0.040)** 
College + −4.22 −0.317 −3.992 −0.272 −2.888 −0.344 −2.925 −0.336 
 (0.219)** (0.013)** (0.248)** (0.021)** (0.123)** (0.015)** (0.294)** (0.046)** 
Household head's age −0.645 −0.036 −0.461 −0.025 0.963 0.133 0.771 0.103 
 (0.143)** (0.008)** (0.153)** (0.008)** (0.096)** (0.013)** (0.132)** (0.017)** 
Household head's sex −0.321 −0.018 −0.274 −0.015 −0.432 −0.058 −0.300 −0.039 
(female = 1) (0.066)** (0.004)** (0.056)** (0.003)** (0.042)** (0.005)** (0.057)** (0.007)** 
Household head's health 1.902 0.126 1.920 0.124 0.051 0.007 −0.066 −0.009 
(sick = 1) (0.051)** (0.004)** (0.084)** (0.005)** (0.036) (0.005) (0.047) (0.006) 
Percent of household 2.393 0.135 2.453 0.134 0.232 0.032 0.002 0.000 
members who were sick (0.069)** (0.004)** (0.097)** (0.004)** (0.049)** (0.007)** (0.093) (0.012) 
Average years of schooling −2.707 −0.153 −2.726 −0.149 −1.462 −0.202 −1.437 −0.191 
of household members (0.060)** (0.003)** (0.086)** (0.006)** (0.039)** (0.005)** (0.091)** (0.014)** 
Average age of 0.352 0.020 0.322 0.018 −1.595 −0.221 −1.596 −0.212 
household members (0.116)** (0.007)** (0.151)* (0.008)* (0.082)** (0.011)** (0.128)** (0.016)** 
Remittances −0.221 −0.012 −0.216 −0.012 −1.859 −0.204 −1.878 −0.206 
(receive = 1) (0.099)* (0.005)* (0.125)+ (0.007)+ (0.074)** (0.006)** (0.099)** (0.017)** 
σintercept2 (Region)   0.138    0.460  
   (0.068)*    (0.275)+  
σintercept2 (Province)   0.219    0.391  
   (0.045)**    (0.119)**  
Pseudo-R2 0.542    0.285    
LR test vs. logit model (χ2)   505.76    3336.7  
p value   0.000    0.000  
 Multidimensional poverty Income poverty 
 (1) Standard logit (2) Margins (dy/dx(3) Mixed logit (4) Margins(dy/dx(5) Standard logit (6) Margins (dy/dx(7) Mixed logit (8) Margins (dy/dx
Household income −1.114 −0.063 −1.075 −0.059     
 (0.043)** (0.002)** (0.058)** (0.003)**     
Household size: (Base category: 1 to 2)     
Small (3 to 4) −0.020 −0.001 −0.110 −0.006 0.301 0.038 0.389 0.047 
 (0.084) (0.004) (0.107) (0.006) (0.068)** (0.008)** (0.092)** (0.012)** 
Medium (5 to 6) 0.312 0.017 0.158 0.008 0.802 0.109 0.971 0.127 
 (0.101)** (0.005)** (0.135) (0.007) (0.074)** (0.009)** (0.099)** (0.015)** 
Large (7 to 9) 0.779 0.045 0.609 0.035 1.183 0.169 1.454 0.200 
 (0.117)** (0.007)** (0.188)** (0.010)** (0.082)** (0.011)** (0.102)** (0.016)** 
Extra large (≥10) 1.319 0.082 1.134 0.069 1.236 0.178 1.543 0.214 
 (0.162)** (0.011)** (0.191)** (0.011)** (0.108)** (0.016)** (0.140)** (0.022)** 
Household head's education: (Base category: No schooling)     
Some elementary −0.788 −0.094 −0.523 −0.058 −0.029 −0.005 −0.066 −0.011 
 (0.093)** (0.012)** (0.102)** (0.013)** (0.084) (0.015) (0.204) (0.033) 
Elementary + −3.221 −0.281 −2.975 −0.237 −0.483 −0.083 −0.452 −0.071 
 (0.100)** (0.012)** (0.088)** (0.017)** (0.083)** (0.015)** (0.223)* (0.037)+ 
High school + −3.803 −0.304 −3.463 −0.256 −1.35 −0.210 −1.171 −0.173 
 (0.115)** (0.012)** (0.103)** (0.018)** (0.086)** (0.015)** (0.243)** (0.040)** 
College + −4.22 −0.317 −3.992 −0.272 −2.888 −0.344 −2.925 −0.336 
 (0.219)** (0.013)** (0.248)** (0.021)** (0.123)** (0.015)** (0.294)** (0.046)** 
Household head's age −0.645 −0.036 −0.461 −0.025 0.963 0.133 0.771 0.103 
 (0.143)** (0.008)** (0.153)** (0.008)** (0.096)** (0.013)** (0.132)** (0.017)** 
Household head's sex −0.321 −0.018 −0.274 −0.015 −0.432 −0.058 −0.300 −0.039 
(female = 1) (0.066)** (0.004)** (0.056)** (0.003)** (0.042)** (0.005)** (0.057)** (0.007)** 
Household head's health 1.902 0.126 1.920 0.124 0.051 0.007 −0.066 −0.009 
(sick = 1) (0.051)** (0.004)** (0.084)** (0.005)** (0.036) (0.005) (0.047) (0.006) 
Percent of household 2.393 0.135 2.453 0.134 0.232 0.032 0.002 0.000 
members who were sick (0.069)** (0.004)** (0.097)** (0.004)** (0.049)** (0.007)** (0.093) (0.012) 
Average years of schooling −2.707 −0.153 −2.726 −0.149 −1.462 −0.202 −1.437 −0.191 
of household members (0.060)** (0.003)** (0.086)** (0.006)** (0.039)** (0.005)** (0.091)** (0.014)** 
Average age of 0.352 0.020 0.322 0.018 −1.595 −0.221 −1.596 −0.212 
household members (0.116)** (0.007)** (0.151)* (0.008)* (0.082)** (0.011)** (0.128)** (0.016)** 
Remittances −0.221 −0.012 −0.216 −0.012 −1.859 −0.204 −1.878 −0.206 
(receive = 1) (0.099)* (0.005)* (0.125)+ (0.007)+ (0.074)** (0.006)** (0.099)** (0.017)** 
σintercept2 (Region)   0.138    0.460  
   (0.068)*    (0.275)+  
σintercept2 (Province)   0.219    0.391  
   (0.045)**    (0.119)**  
Pseudo-R2 0.542    0.285    
LR test vs. logit model (χ2)   505.76    3336.7  
p value   0.000    0.000  

Note: Standard errors in parentheses; +statistically significant at the 10 percent level; *statistically significant at the 5 percent level; **statistically significant at the 1 percent level.

The estimated marginal effects of small and medium household sizes are likewise insignificant in the mixed logit model, implying that they are unrelated to the household's state of poverty. However, household sizes larger than six members have statistically significant marginal effects and the larger the household size is, the larger is the effect on the probabilities. This is shown in Figure 7.

Figure 7.

Multidimensional poverty: Average marginal effects of household size

Figure 7.

Multidimensional poverty: Average marginal effects of household size

The educational attainment of the household head at any level alleviates the poverty situation of the household in general. This is seen in the results in Table 12 and in Figure 8. It is interesting to note that a discrete change from the base category of no schooling to some elementary education reduces the probability by 5.8 percentage points. The household does much better however, if the household head graduates from elementary plus some high school, as this lowers the probability by as much as 23.7 percentage points. Higher education levels, High School+ and College+, further shrink this probability by less than 2 percentage points over their previous levels.

Figure 8.

Multidimensional poverty: Average marginal effects of education of household head

Figure 8.

Multidimensional poverty: Average marginal effects of education of household head

The health of the members and the head of the household raise the probability of being classified into a state of poverty. It increases by 1.34 percentage points as the proportion of sick household members rises by 10 percentage points. The probability rises by 12.4 percentage points with a sick household head. The schooling of household members, however, reduces the probability by 1.49 percentage points for every 10 percent rise in average years. Small but statistically significant effects, less than 2 percentage points, are observed with respect to rest of the variables. For an approximately 10 percent increase in its level value, poverty likelihood rises with average household age (0.18 percentage points [pp]) but declines with the head's age (0.25 pp). The likelihood declines with the presence of OFW remittances (1.20 pp). The household is less likely to be in poverty with a female head (1.50 pp).

The logit results for income poverty shown in columns (5)–(8) of Table 12 can be compared with the results shown in columns (1)–(4). The marginal effects of the size variables on income poverty probability are all significant and much larger. These effects also rise with size itself. The difference is that in the multidimensional case, only the large and the extra-large households matter for poverty risk.

The direction and magnitude of the marginal effects of the education variable are not very different from the previous results. Nevertheless, although the marginal effects decline more uniformly in the case of income poverty, a huge drop is seen from the “Some Elementary” education level to the next level in the case of multidimensional poverty. Succeeding declines as shown in Figure 8 are less than 2 percentage points from preceding levels. At the college level, the decline for poverty income probability is 33.6 points, compared with 27.2 points for multidimensional poverty. Figures 9 and 10 showing these effects on the probability of being in an income poverty state can be compared with Figures 7 and 8.

Figure 9.

Income poverty: Average marginal effects of household size

Figure 9.

Income poverty: Average marginal effects of household size

Figure 10.

Income poverty: Average marginal effects of education of household head

Figure 10.

Income poverty: Average marginal effects of education of household head

As in the previous table, the health variables are not significant when the income poverty definition is used. The problem with this is that a policymaker confronted with this result will find it hard to justify putting priority on health care for the poor. This also reflects the difficulty of relying on a narrowly defined unidimensional measure that cannot account for other dimensions representing quality of life.

There are no a priori expected signs for the direction of effects of household head age and the average age of the households. It is noted that these variables are significant in both estimates but have opposite signs. The OFW remittances variable is significant in both estimates. Its marginal effect of 20.6 percentage points on income poverty likelihood dwarfs the 1.2 points obtained for multidimensional poverty. The large marginal effect can be explained from the income poverty perspective as an income supplement that is used for consumption rather than for investments in health and education.

6.  Summary, policy implications, and future research direction

This study constructs a multidimensional index of poverty for the Philippines using 2011 household data. This index is disaggregated into urban and rural population groups. The breakdown of the index by dimension, by administrative region, and by province is also reported. At the provincial level, the study finds a generally positive relation between the incidence and intensity of poverty, but the highest intensity levels are experienced in areas where incidence is not that high relative to other areas. Provinces with high per capita income generally have low poverty indices. The relationship appears to be nonlinear. This is reflected in both the level and dispersion of per capita income, which are higher for the top five low-index provinces when compared with the top five high-index provinces.

An examination of multidimensional household poverty using mixed logit analysis shows that large-sized households have higher poverty risk. A substantial reduction of poverty risk is observed for households with heads who at a minimum were able to enter high school. The household head's health status has a negative impact on the household's risk of being poor. Small but statistically significant effects of less than 2 percentage points in the probability of being in the state of poverty are observed with respect to rest of the variables. Poverty likelihood rises with average household age but declines with the head's age. The likelihood declines with the presence of OFW remittances. The household is less likely to be in poverty with a female head. These are contrasted with results of estimates using the income poverty definition.

The calculation and disaggregation of the MPI in this study provides some policy direction with respect to poverty reduction. The wide urban–rural disparity in the MPI offers a strong case for policies aimed at developing rural areas. Urban development strategies such as infrastructure development would serve to reduce the large differences in MPIs across regions by allowing greater mobility of goods and labor which helps raise the level of regional economic activity.

Overall, the results affirm the importance of health and education policies to address the non-monetary aspects of poverty. The health dimension covers 24 percent of the MPI and education accounts for 38 percent. A recent study shows that health outcomes such as maternal death, infant mortality, and the incidence of childhood stunting and malnutrition have changed negligibly over the past 25 years (Solon et al. 2017). Furthermore, health expenditures by the household are mainly out-of-pocket expenses. Thus, the poor are unable to adequately access quality health care. The study's results suggest that more public investment in education—especially at the primary level to raise the number of elementary graduates—would contribute to the success of any long-term poverty reduction strategy. As noted earlier, the net decline in poverty risk can go as high as 18 percentage points as the household head moves from the “some primary schooling” category to the “elementary graduate plus” category. A family planning program is of course another important part of this strategy if household size is to be constrained. The econometric results imply that policymakers can set a threshold family size no larger than six because going beyond this number increases the likelihood of being in a state of poverty.

Although these policy implications are drawn based on a detailed examination of multidimensional poverty without an income dimension, these do not imply that the monetary dimension should be set aside. As mentioned previously, there are clear reasons to view the two poverty measures as complementary because one measure's weakness is the other one's strength.

The method used allows researchers to produce a rich set of results that would not normally be obtained using other methods. Indeed, as one goes to a higher level of disaggregation, interesting results are obtained. As shown in this study, disaggregation at the regional level yields results that are useful in policymaking at the national level. Disaggregation at the provincial level is useful not only to high-level policymakers but also to local government officials, especially the provincial governors, as it relates directly to their constituents. Unfortunately, the sample sizes in the succeeding APIS surveys of 2013 and 2014 were significantly trimmed down to around 10,000 households, or a drop of approximately 75 percent from the 2011 sample. Consequently, with limited data, classification by province became impossible to do. On a more positive note, the APIS series started in 1998 and is conducted only when the FIES is not conducted. With the regularity of the surveys and the ease with which the study's methodology is implemented, the monitoring of multidimensional poverty at the regional/provincial level over time has become feasible. Thus, one can proceed with an analysis over several data sets using pseudo-panel methods. This is an area for further research.

References

Alkire
,
Sabina
, and
James
Foster
.
2011a
.
Counting and Multidimensional Poverty Measurement
.
Journal of Public Economics
95
(
7–8
):
476
487
.
Alkire
,
Sabina
, and
James
Foster
.
2011b
.
Understandings and Misunderstandings of Multidimensional Poverty Measurement
.
Journal of Economic Inequality
9
:
289
314
.
Alkire
,
Sabina
,
James
Foster
,
Suman
Seth
,
Maria Emma
Santos
,
Jose Manuel
Roche
, and
Paola
Ballón
.
2015
.
Multidimensional Poverty Measurement and Analysis
.
New York
:
Oxford University Press
.
Alkire
,
Sabina
, and
Maria Emma
Santos
.
2014
.
Measuring Acute Poverty in the Developing World: Robustness and Scope of the Multidimensional Poverty Index
.
World Development
59
:
251
274
.
Alkire
,
Sabina
, and
Suman
Seth
.
2015
.
Multidimensional Poverty Reduction in India Between 1999 and 2006: Where and How
?
World Development
72
:
93
108
.
Balisacan
,
Arsenio
.
2011
.
What Has Really Happened to Poverty in the Philippines? New Measures, Evidence, and Policy Implications
.
UPSE Discussion Paper No. 2011-14
.
Quezon City
:
University of the Philippines School of Economics.
Bayudan-Dacuycuy
,
Connie
, and
Joseph Anthony
Lim
.
2013
.
Family Size, Household Shocks and Chronic and Transient Poverty in the Philippines
.
Journal of Asian Economics
29
:
101
112
.
Cavapozzi
,
Danilo
,
Wei
Han
, and
Raffaele
Miniaci
.
2015
.
Alternative Weighting Structures for Multidimensional Poverty Assessment
.
Journal of Economic Inequality
13
:
425
447
.
Desai
,
Meghnad
, and
Anup
Shah
.
1988
.
An Econometric Approach to the Measurement of Poverty
.
Oxford Economic Papers
40
(
3
):
505
522
.
Ericta
,
Carmelita
, and
Jeremias
Luis
.
2009
.
A Documentation of the Annual Poverty Indicators Survey
.
PIDS Discussion Paper No. 2009-20
.
Manila
:
Philippine Institute for Development Studies.
Goldstein
,
Harvey
.
2011
.
Multilevel Statistical Models
.
West Sussex, UK
:
John Wiley & Sons
.
Klugman
,
Jeni
,
Francisco
Rodríguez
, and
Hyung-Jin
Choi
.
2011
.
The HDI 2010: New Controversies, Old Critiques
.
Journal of Economic Inequality
9
:
249
288
.
National Statistical Coordinating Board
.
2013
.
2012 Full Year Official Poverty Statistics
.
Makati
:
National Statistical Coordinating Board.
Philippine Statistical Authority
.
2013
.
Results of the 2011 Annual Poverty Indicators Survey
.
APIS Technical Notes
.
Manila
:
Philippine Statistical Authority.
Schelzig
,
Karin
.
2005
.
Poverty in the Philippines: Income, Assets and Access
.
Manila
:
Asian Development Bank.
Sen
,
Amartya
.
1985
.
Commodities and Capabilities
.
Amsterdam
:
North-Holland/Elsevier.
Solon
,
Orville
,
Carlo Irwin
Panelo
,
Rebecca
Ramos
, and
Alejandro
Herrin
.
2017
.
The Challenge of Reaching the Poor with a Continuum of Care: A 25-year Assessment of Philippine Health Sector Performance. Project Report
.
Washington, DC
:
United States Agency for International Development.
United Nations Development Programme
.
2010
.
Human Development Report 2010, The Real Wealth of Nations: Pathways to Human Development
.
New York
:
Palgrave Macmillan
.
Whelan
,
Christopher
,
Brian
Nolan
, and
Bertrand
Maître
.
2014
.
Multidimensional Poverty Measurement in Europe: An Application of the Adjusted Headcount Approach
.
Journal of European Social Policy
24
(
2
):
183
197
.
World Bank
.
2001
.
Philippines Poverty Assessment
,
volumes I and II
.
Washington, DC
:
The World Bank.
Yu
,
Jiantuo
.
2013
.
Multidimensional Poverty in China: Findings Based on the CHNS
.
Social Indicators Research
112
:
315
336
.

Notes

1 

The objectives are embodied in 17 time-bound targets known as the Sustainable Development Goals (SDGs) which build on the eight MDGs (see www.un.org/sustainabledevelopment/sustainable-development-goals/).

2 

See https://ophi.org.uk/. Examples of country papers on the MPI are Yu (2013), Whelan, Nolan, and Maître (2014), and Alkire and Seth (2015).

3 

This contrasts with the “dashboard” approach where indicators (or sub-indices) are used independently to assess the multidimensional aspects of poverty and hence only the marginal distribution matters.

4 

The Philippine MPI in the 2010 HDR is 6.70 (H = 0.126; A = 0.535). As mentioned previously, this is difficult to compare with the present result because the data sources are different.

5 

Only 10 out of a total of 81 provinces and four districts of the national capital region are reported in the Table to conserve space. The full set of provinces can be found at http://cba.upd.edu.ph/bautista/MPI.

6 

The median per capita income of the third decile group is 9,466.00 Philippine pesos. The threshold used in this study is defined in the APIS 2011 technical notes (PSA 2013). A National Statistical Coordinating Board (2013) document puts the official monthly poverty threshold in 2012 for a household with five members at 7,890.00 Philippine pesos. Dividing this by 5 and multiplying by 6 months (the coverage period of the survey) yields 9,468.00 Philippine pesos, which is close to the figure above.

7 

Note that, in theory, the methodology's smallest unit of analysis is the individual. Individual intra-household comparisons and analyses are not attempted here because some data are collected only at the household level.

8 

For more details on mixed models, see Goldstein (2011).

9 

As mentioned earlier, all continuous predictor variables are in natural logarithms except the household members’ health and household size. An increase (decrease) in the log value of a variable by 0.1 is approximately an increase (decrease) in the level value by 10.5 percent (9.5 percent). To approximate the effects of a 10 percent change in the predictor in levels, the corresponding marginal parameter must be multiplied by 0.1.

Appendix A.  Assessment of redundancies

In the construction of the index, it is important to examine the relation among indicators to avoid possible redundancies. The correlation between two indicators can be calculated to determine the redundancy of an indicator. Because the indicators are binary variables, the relevant measures of relationship are the tetrachoric correlation and Cramer's V. Table A.1 presents both measures and a redundancy ratio proposed by Alkire et al. (2015). The redundancy ratio uses the percentage of people deprived in indicators i and j, (di,dj), and the percentage of people deprived in both indicators simultaneously, dij, to form the ratio: R=dij/min(di,dj).

This measure lies between 0 and 1 (inclusive) like Cramer's V, and the tetrachoric correlation lies between –1 and +1 (inclusive). A high value in any of the three measures may indicate redundancy.

The values in Table A.1 are the average of the measures of each of the 12 non-asset indicators with the 11 other indicators. The 15 asset indicators are excluded from this group; instead, their measures are averaged and are shown in the last row of the table. As can be seen, there appears to be no strong indication of redundancy among the 12 indicators. For the asset indicators, however, the redundancy ratio registers a high value of 0.929. Here, the redundancy ratio appears to be more appropriate because it is a direct measure of the duplication of deprivation. The issue of whether to retain or remove a redundant indicator is not a settled one and there can be various reasons to follow either one of these alternative courses of action. These are further discussed in Alkire et al. (2015).

Table A.1
Summary of relationship among indicators
Indicator Redundancy ratio Tetrachoric correlation Cramer's V 
Sanitation 0.386 0.397 0.193 
Water supply 0.311 0.294 0.135 
Housing (roof) 0.406 0.369 0.201 
Housing (wall) 0.480 0.380 0.209 
Housing (floor area) 0.296 0.223 0.076 
Electricity 0.411 0.412 0.219 
School participation 0.282 0.220 0.099 
Schooling level 0.285 0.213 0.106 
Household head's education 0.437 0.321 0.166 
Head/spouse's health 0.313 0.080 0.040 
Household member's health 0.197 0.081 0.036 
Hunger perception 0.248 0.227 0.089 
Assets 0.929 0.492 0.246 
Indicator Redundancy ratio Tetrachoric correlation Cramer's V 
Sanitation 0.386 0.397 0.193 
Water supply 0.311 0.294 0.135 
Housing (roof) 0.406 0.369 0.201 
Housing (wall) 0.480 0.380 0.209 
Housing (floor area) 0.296 0.223 0.076 
Electricity 0.411 0.412 0.219 
School participation 0.282 0.220 0.099 
Schooling level 0.285 0.213 0.106 
Household head's education 0.437 0.321 0.166 
Head/spouse's health 0.313 0.080 0.040 
Household member's health 0.197 0.081 0.036 
Hunger perception 0.248 0.227 0.089 
Assets 0.929 0.492 0.246