Social Indicators Research

[Link]
02285-z
ORIGINAL

RESEARCH

Targeting Administrative Regions for

Multidimensional Poverty Alleviation: A Study on
Vietnam
Anh Thu Quang Pham1 · Pundarik Mukhopadhaya1 · Ha Vu1

Accepted: 3 February 2020

© Springer Nature B.V. 2020

Abstract
This study investigates seven dimensions of poverty in Vietnam (income, health, educa-
tion, housing, assets, basic services and economic status) using the Household Living
Standard Survey data of 2014. The Government of Vietnam disburses funds for poverty
alleviation to regions on the basis of incidence of household income poverty. Our study
shows that this method neither fully captures the complex regional diversity of poverty
nor does it accurately identify regions with a higher severity of poverty. For the first time
in poverty studies of Vietnam, we explore the role of multiple spatial levels on poverty in
multiple dimensions. Unlike the practice in the existing literature which classifies the poor
with an arbitrary poverty cut-off, we use a fuzzy method that allows the inclusion of
people who are in partial poverty. Furthermore, by utilizing random intercept multilevel
models to decompose the variation of poverty at the household, commune, district and
province levels, poverty maps for Vietnam are developed to visualize the spatial evidence
of the severity and incidence of poverty. We identify that the provinces that are relatively
less (more) poor in the income dimension are more (less) destitute in several other
dimensions, which clearly shows a need for special policy attention. Our method reveals
that the pov- erty ranking of provinces in regional Vietnam departs widely from those
obtained through traditional single-level analysis. This suggests that poverty in Vietnam
can be explained not only by characteristics at the household level, but also by contextual
factors at higher levels (commune/village, district, province). These empirical findings can
help Vietnamese policy makers determine suitable strategies to effectively target the most
deprived regions and to develop more appropriate poverty-alleviation programs.

Keywords Multilevel · Multidimensional · Fuzzy · Poverty · Province

JEL Classification I30 · I32 · I38

 Pundarik Mukhopadhaya
[Link]@[Link]
Anh Thu Quang Pham
arielpham2504@[Link]
Ha Vu
[Link]@[Link]

13
1
Department of Economics, Macquarie University, Sydney, Australia

13
Targeting Administrative Regions for Multidimensional
Poverty…
1 Introduction

Over recent decades, the outstanding achievements of Vietnam in economic growth and
reduction in poverty have transformed the nation from one of the poorest to a lower
middle- income country. With an average annual real GDP growth rate of 6.5% during the
period 2000–2017 (ADB 2018), Vietnam is now considered one of the most dynamic
emerging nations in the East Asian region. The poverty rate in Vietnam dropped from
28.9% in 2002 to 5.8% in 2016 (GSO 2017),1 and the provision of basic services has also
substantially improved (GSO 2015). For the next stage in the fight against poverty, the
Vietnamese gov- ernment announced the estimated budget for the national target
programme on poverty alleviation for 2016–2020 to be VND46.1 trillion (USD2.1
billion).
To date, the monitoring of poverty in Vietnam has utilised an approach based on
income or expenditure; a method that neglects non-income dimensions of poverty such as
health, education, living standards and so forth (Arouri et al. 2015; Lanjouw et al. 2017;
Mahade- van and Hoang 2016; Nguyen et al. 2017). However, there is no disagreement
about the necessity for identifying deprivation as a factor in a multidimensional approach,
and this has increasingly expanded the measurement of poverty in many countries. It is
also argued that using a multidimensional approach is not only a more efficient tool for
measuring pov- erty; it can also be employed as a tool for eliminating poverty (Alkire and
Santos 2010; Alkire et al. 2015; Ravallion 2011; Yang and Mukhopadhaya 2017; Yu
2013).
There is a small amount of literature available on multidimensional poverty in Vietnam
(for example, Asselin 2009; Baulch and Masset 2003; Roelen et al. 2010, 2012; Tran et al.
2015); however, these studies use an arbitrary poverty cut-off to divide the population in
a dichotomous group of poor and non-poor. Such a classification results in a huge loss
of information on those who are partially poor, especially when they are living in similar
conditions to those who are considered poor, but happen to just fall on the opposite side of
the poverty line (Makdissi and Wodon 2004). To overcome this shortcoming, this paper
employs a fuzzy method to measure poverty in Vietnam following Cerioli and Zani
(1990), Cheli and Lemmi (1995), and Betti and Verma (2008).2
Furthermore, given the constraints in the government’s budget, it is often desirable to
target the invariable scarce resources for those beneficiaries whose disparities in living
standards are the worst. Since inequality in economic growth and poverty in many devel-
oping countries, including Vietnam, is spatially concentrated (Huang and Magnoli Bocchi
2008; Klasen et al. 2007), public spending on poverty-alleviation programs is directed to
areas where the poverty rate is higher.3 To determine these target regions it is necessary
to evaluate poverty from the perspective of households’ socioeconomic and demographic
factors and also include some of the regional macro traits. In this regard, the traditional

1
Based on the national poverty line: for 2006–2010 the poverty lines are 360,000 and 450,000VND/
person/month for the rural and urban areas respectively; for the period 2011–2015, the poverty lines are
respectively 630,000 and 780,000 VND/person/month.
2
Because of the use of dichotomous variables to specify the poverty status of a household (for instance “1”
indicates poor and “0” not poor) in previous studies on Vietnam, the relative influence of household level
could not be calculated directly from the empirical model but was derived from the latent variable meth-
ods proposed by Browne et al. (2005) and Goldstein et al. (2002). This drawback is corrected through our
approach as we will show later. See also Pham and Mukhopadhaya (2018).
3
Empirical evidence also reveals that the impact on poverty is most powerful when poverty alleviation
efforts use spatial targeting for small administrative or geographic units; for instance, districts, villages or

13
 A. T. Q. Pham et
communes (Baker and Grosh 1994; Amarasinghe et al. 2005; Elbers et al. 2004). al.

13
Targeting Administrative Regions for Multidimensional
Poverty…
approach uses a single-level linear regression framework, which either aggregates up
to contextual level information or disaggregates down to household level information.
Moreover, the existing approach is prone to generating spurious outcomes, because poor
households in the same village/commune/district/province tend to be more similar to each
other than to those in different villages/communes/districts/provinces. Due to this within-
administrative area correlation, households within the same administrative region are not
independent. In technical terms, this group dependence can lead to a violation of the inde-
pendence assumption of linear regression. To avoid this problem, a multilevel model is an
alternative that partitions the error structure into components at the household and higher
levels.4 Estimation of separate error terms at each level of analysis avoids violation of the
assumption that originates in single-level regression models. The multilevel framework
has further advantages because it appropriately handles variables that are measured at
different levels, which allows the possibility of examining micro-level (household) and
macro-level (commune, district, province) determinants of poverty simultaneously. Using
a multilevel model, the current study identifies impoverished target regions in Vietnam
from the com- bined perspectives of regional macro traits and household micro
characteristics.
Beyond the household level, we also explore the relative effects of higher regional
administrative levels on different dimensions of poverty in Vietnam, after controlling for
household characteristics. This multilevel approach produces more efficient estimates than
those obtained from a standard single-level regression, and is therefore an effective meth-
odology to validly integrate the influences of household characteristics with contextual
fac- tors at the higher administrative levels to determine the poverty distribution of the
country. The approach allows us to decompose the error terms to evaluate the relative
impact of dif- ferent geographical levels that can be attributable to levels of deprivation
(Goldstein 2011; Raudenbush and Bryk 2002).
Our study contributes to the empirical literature in several ways. Firstly, this is the only
study to investigate the relative importance of each administrative level and hierarchical
clustering in explaining poverty in Vietnam beyond simply the income dimension. Sec-
ondly, although the Vietnam government has previously applied geographic targeting, it
has been criticized due to the limited effectiveness of targeting in benefiting those in need
since the government relied on a relatively arbitrary list of poor villages (Minot and
Baulch 2004; Nguyen et al. 2017; van de Walle 2002). By using a multilevel and
multidimensional approach, the present study derives an objective standard in locating the
impoverished administrative areas in the country, which will help in targeting resources
more efficiently. Thirdly, this paper brings out the complex characteristics of poverty in
Vietnam where the current income-based targeting of poverty alleviation has been
ineffective from the wider perspective of meeting the sustainable development goals. Our
examination identifies the regions that need specific support in the explicit dimensions of
poverty.
The paper is organized as follows. In the next section we describe the method used to
measure poverty and the empirical strategy of the multilevel model. Section 3 describes
the data, the choice of dimensions and the household-level variables. Section 4 presents
the empirical results and discussion, followed by conclusions and policy implications.

4
A limited number of studies is available in this respect. See, for example, Gräb and Grimm ( 2008) on
Burkina Faso, Kim et al. (2016) on India, and Arpino and Aassve (2014) on rural Vietnam among others.
However, all these studies utilise the uni-dimensional approach to measuring poverty, which cannot capture
13
 A. T. Q. Pham et
the multidimensional nature of deprivation. al.

13
Targeting Administrative Regions for Multidimensional
Poverty…
2 Methodology

2.1Measurement of Poverty—Income and Non-income Dimensions

The traditional approach to measuring poverty has been criticised for having two major
shortcomings. Firstly, it is unidimensional; that is, either income or expenditure level is
used as a proxy for poverty. Thus, the approach is limited in its capacity to capture the
complex nature of poverty and in informing policy makers on poverty-alleviation
strategies (Alkire and Santos 2013; Sen 1992). Secondly, the most widely used
multidimensional measures of poverty (for example, the Alkire-Foster measure) divide the
population into two groups: poor and non-poor, by using arbitrary poverty thresholds,
resulting in a sub- stantial loss of information from excluding those who are partially poor
(Makdissi and Wodon 2004). Our study applies a relative measure of poverty that does not
use the arbi- trary cut-off and includes people who are somewhat poor, following fuzzy
approach of Cerioli and Zani (1990). We consider that all households are subject to
poverty but of dif- ferent degrees, so that each has a certain propensity for poverty in the
continuum range of [0, 1] (Verma and Betti 2002). Unlike the Alkire-Foster and other
measures of poverty, this method allows for the possibility of determining the separate
magnitudes of income and non-income dimensions of household deprivation that we are
interested in. The following part of this section explains th e fuz zy computations of
poverty for income dimension (P P h)
and non-income dimensions (k . These measures are utilized to discuss the average
)
h
degrees of poverty in income and non-income dimensions ( APk,r ) for six economic
regions of the country in the Sect. 4.1, and in multilevel analyses discussed in the Sects.
4.2 and 4.3.
The income poverty measure for the household h is presented as:

Ph = 1 − Fh 1 − (1)
Lh

This measure varies between 0 (not poor) to 1 (maximum poor), where Lh is the value
of the Lorenz curve (which is the proportion of the total income earned by all households
who are less poor than household h), and Fh is the share of households less poor than
household h (which can be acquired from the cumulative distribution function, denoted as
F). This measure of poverty is highly sensitive because it combines the actual disparities
in income with the cumulative distribution function.
Similar to the income measure of poverty, we measure poverty for non-income dimen-
sions (see Betti et al. 2015). A non-income dimension may consist of more than one indi-
cator. First, a deprivation measure, dj,h for each indicator j in the dimension is determined,
and then by using a pre-assigned weight, all indicators are integrated into one index in a
dimension. The indicators are transformed into the interval 0 to 1 to determine the
depriva- tion score for each by the formula:

dj,h Z−
z=h 1 ≤ zh ≤ (2)
Z− Z
1
zh being the category to which household h belongs; Z is ordered categories of some depri-
vation indicator j (a higher Z indicates less deprivation).5
13
 A. T. Q. Pham et
al.

5
In the case of binary indicators, dj,h = 1 (maximally deprived) or dj,h = 0 (not deprived).

13
Targeting Administrative Regions for Multidimensional
Poverty…
Following Betti and Verma (1999, 2008) the weight of each indicator is calculated
within each dimension distinctly; as a product of the inverse of the average correlation
coefficients of all indicators in the dimension and the coefficient of variation. For conveni-
ence, the weights of the indicators are standardized to sum to 1 within each dimension.
Then a deprivation score is computed for poverty dimension k as follows:
,
Sk,h = (3)
k
wjdj,h
where wj is the weight of indicator j=1
j.
Like income poverty (as defined in Eq. (1)), we define a household’s degree of non-
income deprivations P(k)
h
as:

P(k)
h
= 1 − Fh(k) 1 − h (4)
L(k)
where F(k) is a distribution function of S and L(k) represents the value of the Lorenz curve
h k,h h
of Sk,h, calculated according to the form below:
� � ∑ S S >
∑
= Sk jSj >
h (k) k k
1− L (5)
k k k
Sh S
1
Having defined the deprivation of households as above, the poverty in dimension k
(income or non-income) of a region is computed as an average of all households’ poverty
in that region
∑ nr (k
Average poverty in region r = APk,r h=1
nr P)h,r (6)
=

where P(kh,r ) is the poverty measure of dimension k for the household h in region r, nr is
the number of households in region r. The average poverty in the region is computed for
the analysis of poverty profile in six administrative regions in Vietnam (see Sect. 4.1)

2.2Identification of Poverty-Prone Regions—Multilevel Models

To fight poverty with scarce resources, public spending on poverty-alleviation programs is

directed to areas where the poverty rate is higher than others. However, the unavailability
of dependable information on economic wellbeing at the local level is the main hindrance
to spatial targeting. Data on welfare levels are traditionally derived from household
surveys representative at the national level. These surveys are usually too small in size to
disaggre- gate reliable measures of poverty at local levels. On the other hand, the
population census data have a sufficient sample size but have limited welfare information
that is required to estimate poverty directly. To overcome this problem, Ghosh and Rao
(1994), Hentschel et al. (2000) and Elbers et al. (2002, 2003) proposed the approach of
small area estimation (SAE); combining census data with household survey data, and
employing estimations of households’ expenditure or income from the latter to the former.
The SAE method has been

13
 A. T. Q. Pham et
al.
utilized to develop poverty maps in some studies (Van der Weide et al. 2010; and
Lanjouw et al. 2017 for Vietnam6); however, they are prone to a number of limitations.
Firstly, such maps depend on individual-level analyses which do not take into account the
hierarchical or clustered structure of household or census data, where the lowest level
(households) are grouped within higher levels (communities, districts, provinces,
regions).7 As a result, the single-level studies violate the independence assumption of
observations which generate underestimation of standard errors of regression coefficients,
leading to an overstatement of statistical significance (Walsh 1947). Secondly, none of
these studies is able show the relative importance of each level in an explanation of
poverty. It may be difficult for policy makers to decide what sub-nation levels (for
example, whether at province or community level) should be given priority over the other
if the results from poverty maps show evi- dence supporting for spatial targeting of social
programs.
The available data on Vietnam comprise a four-level hierarchical structure with house-
holds at level 1, nested within communes/villages at level 2, districts at level 3, and prov-
inces at level 4 (Fig. 1 summarizes the hierarchical structure of the data). In this paper we
utilize a multilevel exploration of the spatial nature of poverty that overcomes the short-
comings of prevalent single-level analyses, and is more efficient in poverty-alleviation
strategies.8
Let Pkhcdp be the measurement of the poverty dimension k for household h nested within
communities c, which is further nested within districts d, and that is nested within province
p using poverty measures defined by the Eq. (1) and (4) presented in Sect. 2.1.9 Unlike
the traditional linear regression model, here we suppose that the means of P are differ-
ent in each hierarchical level. Moreover, we split the residual into 4 components that are
equivalent to the 4 levels in the data, to examine the nature of between-level differences,
which cannot be performed in the common linear regression models. We specify a series
of following four-level random intercept models for poverty dimension k of household h
in commune c, district d, province p:

Pk = β k + β k Xk + uk + vk + wk + ek
hcdp 0 hcdp p dp cdp hcdp (7)
p u cdp w hcdp e
uk ∼ N 0, σ2 , wk ∼ N 0, σ2 , ek ∼ N 0, σ2

Equation (7) estimates the outcome of dimension k, adjusted for a vector Xfkhcdp of
explanatory variables measured at the household level (see Sect. 3.3 for more details).
The parameter βk represents an overall mean of Yhcdp
k
(controlling for household charac-
0 k k k
across all households in the sample. The mean of Yhcdp for province p is β0 + up
teristics)
and so the random province effect ukp is the difference between province p’s mean and the
overall mean. Provinces with high values of upk tend to have households with high values of

6
For studies on other developing countries, see Amarasinghe et al. (2005), Carletto et al. (2007), Christi-
aensen et al. (2012), Demombynes et al. (2002) among others.
7
Therefore, households from the same community (say, the minorities) can have a poverty status more
similar to those in another area than to households from different communities, and the poverty status of
households in the same community in the same district are more like another than they are like those house-
holds of the same community in a different district.
8
We use STATA 15 software with the fully maximum likelihood (FML) method to obtain all parameter
estimations. To show the gross heterogeneity across provinces, districts, and communes, and to examine
which level(s) has the greatest impacts on poverty after controlling for household compositional effects,
this study employs multilevel random intercept models.

13
Targeting Administrative Regions for Multidimensional
Poverty…
9
It ranges from 0 (not at all poor) to 1 (totally poor).

13
 A. T. Q. Pham et
al.

Country p-Provinces (Level 4)

d-Districts (Level 3)
prov prov prov prov c-Communes/Villages
distr distr distr distr distr
distr distr distr (Level 2)
comm comm vill vill comm comm comm comm vill vill
comm comm
comm
comm
h-Households (Level 1)
comm
hh hh hh
hh hh hh
hh hh hh hh hh
hh
hh hh hh hh
hh hh hh
hh hh hh hh hh
hh hh hh
hh hh hh

Fig. 1 Nesting structure of households within communes/villages within districts within provinces

k
poverty in dimension k. The mean of Yhcdp for district d is β0k + upk + dp
vk and so the random
k
district effect vdp is the difference between district d’s mean and province p’s mean. Dis-
tricts with high (low) values of vkdp tend to have households with high (low) values of pov-
erty in dimension k, krelative to households from other districts belonging to the same prov-
ince. The mean of Y for commune c is βk + uk + vk + wk and so the random
commune hcdp 0 p dp cdp
effect wk is the difference between commune c’s mean and district d’s mean. Communes
cdpvalues of wk have a high value of dimension k or vice versa. The observed Yk
with high
for household h is cdp hcdp
βkk + uk + + wk + ek and so ek is the difference between
v
0 p dp cdp hcdp hcdp
household h’s observed measurements and commune c’s mean. The random effects and
residual errors in the random part of the model are assumed to be independent of the
vector of predictor variables and normally distributed with a mean of zero; the variances
between- province σ2 , between-district (σ2), between-commune/village σ2 , and
between-house-
u v w
hold (σe2), having adjusted for the predictor variables, describe the part of the outcome vari-
ation which is unexplained by the explanatory variables.

2.2.1 Variance Partitioning at Household, Commune, District, and Province

Levels

Based on the variance estimates of random effects, we calculate variance partition coef-
k
ficients (VPCs), which describe the percentage of variance in Yhcdp that is attributed to each
level (Goldstein et al. 2002). It may be noted that the total variance in a poverty dimension
for household h is the sum of the four separate variance components:

hcdp (8)
var Yk = σ2 + σ2 + σ2 +
σ2 u v w e

and the VPC for each level is calculated as the ratio of that level’s variance to the total vari-
ance. Thus, the province-level VPC is:
VPCu
σ2
13
Targeting Administrative Regions for Multidimensional
Poverty… u
= σ2 + σ2 + σ2 + σ2 (9)
u v w e

while the district-level VPC is:

13
 A. T. Q. Pham et
al.

σ2
VPCv = v (10)
σ2 + σ2 + σ2 +
σ2
u v w e

the commune-level VPC is:

σ2
w
VPCw = σ2 + σ2 + σ2 + (11)
σ 2
u v w e

and the household-level VPC is:

σ2
VPCe = σ2 e (12)
2
+ σ2 + σ2 +
σ
u v w e

VPCs allow us to establish the relative importance of provinces, districts, communes

and households as sources of variation in households’ poverty. VPCs evaluate the
percentage of outcome variation unexplained by the observed household variables that is
attributable to each level. The higher the VPC, the larger is the impact of that level (for
example the province) on the deprivation of households.
In the next section, based on the models of Eq. (7), the extent of poverty at various
admin- istrative levels in multiple dimensions is estimated by a maximum likelihood
method using following two steps:
Step 1: Models containing household-level variables, allowing for the varying
intercepts, are specified so as to inspect whether measurements of different dimensions of
poverty vary by higher administrative level. Then variance estimates and VPCs are obtained
from the mod- els by applying Eqs. (9)–(12).
Step 2: Maps are created that allow us to recognize areas of high poverty concentration
in the country. We apply the Empirical Bayes (EB) estimates (posterior estimates) to
obtain predictions of random errors at each level after fitting the multilevel models. We
then rank the residuals of each level together at 95% confidence intervals. These residuals
symbolise the administrative levels’ (for example, province) departures from the overall
mean
0
predicted by βk.
Hence a province whose confidence interval of residual does not overlap the mean
poverty value across all provinces is deemed significantly different from the average at the
5 percent level. The procedure proposed by Goldstein and Healy (1995) to assess the
province effect is:
(province-specific residual ± 1.96 × standard deviation of the
residual) (13)

Provinces with the upper bound calculated by Eq. (13) greater than zero were classified
as “good” provinces in the kth dimension. In other words, poverty levels in good
provinces, after controlling for household traits, are significantly below the overall level of
poverty for the whole country (which is the average for the provincial poverty). On the other
hand, provinces with the lower bound greater than zero were classified as “bad” provinces.
This means that in these provinces the levels of poverty are very high. Even the variation
of provincial macro traits in these provinces cannot significantly reduce poverty.
Provinces with residuals within the 95% coverage bounds of the overall province average

13
Targeting Administrative Regions for Multidimensional
Poverty…
were classified as “medium” prov- inces. In other words, poverty in the medium provinces
is not significantly different from the average poverty for the whole country after
considering both household traits and provincial variations.

13
 A. T. Q. Pham et
al.
3 Data and Choice of Dimensions

3.1Data and Sample Size

The Vietnam Housing Living Standard Survey (VHLSS) is conducted by the General Sta-
tistics Office (GSO) with technical support from the World Bank since 1993. This is a
com- prehensive survey that is representative of the whole country, the region, urban, rural
and province levels. The household survey covers detailed information on various aspects
of living standards; household demographics and household-level income, expenditure,
hous- ing conditions, ownership of durables, as well as the health, education and
employment of the household members, and participation in government programs.
Although the VHLSS is not the only large scale survey in the country, other surveys rarely
report the expendi- ture or income required to measure poverty directly. This paper uses
the current wave of VHLSS conducted nationwide in 2014. It covers 9399 households
nested within 3132 communes and wards, 698 districts, and 63 provinces. The largest city,
Hanoi, comprises 420 households, 140 communes, and 30 districts.

3.2Dimensions of Poverty Considered

Income and six non-income dimensions are considered in this study. To measure income
poverty, the equivalised household income is used by applying the modified OECD
scale.10 Total household income from all sources is included. 11 For non-income
dimensions, based on the available information in the 2014 VHLSS, we select 19
indicators grouped into six dimensions: education, health, housing, durable assets, basic
services, and economic sta- tus. A detailed list of the indicators is provided in Table 1.

3.2.1 Education

It has been determined in various welfare research studies that education level has signifi-
cant effects on poverty reduction (see, for example, Alkire 2007; Alkire and Santos 2014;
Santos et al. 2015). A number of previous studies in Vietnam report that there is high
over- lap between chronic income poverty and out-of-school primary school-age children
(Roe- len et al. 2010; Baulch and Masset 2003). Therefore, based on the Sustainable
Develop- ment Goals (SDGs; also called Global Goals for Sustainable Development) on
education, we include two indicators to identify the education dimension: average
schooling achieve- ment of adult members, and school attendance of the children.

3.2.2 Health

The SDGs argue that ensuring healthy lives is vital for sustainable development. Van
Doorslaer et al. (2006) and Van Minh et al. (2013) find that more than 60% of health care
payments in Vietnam are paid as out-of-pocket expenses by households, and this possi-
bly leads to impoverishment. Therefore, health is an important measure of poverty. To

10
To construct the equivalent scale, the first adult in the household is given a point 1, while each extra
member who is 15 years or above is assigned 0.5, and each member under the age of 15 is given 0.3.
11
Comprising wages, salary, and incomes from services, agricultural, fishery and forestry sectors.

13
 Targeting Administrative Regions for Multidimensional
Poverty…
Table 1 Description of dimensions used for the computation of poverty
13

Dimensions Indicators Description Type of indicator Computed weight

Income Equivalised income Continuous
Non-monetary
Education Schooling achievement of Average level of education of Order 0.04
adult (15 years and above) household: No diploma (1)
members Primary school (2)
Lower secondary school
(3) Upper secondary
school (4) College (5)
Bachelor degree (6)
Higher education degree (7)
School attendance of children Households where there is at least one child or adolescent (6 to 15 years) Dichotomous 0.96
not attending school
Health Financial difficulties Household where there is at least one person who was sick in the last Dichotomous 0.82
12 months but the household could not afford to cover all health care
expenses for him/her
Health insurance Household where there is at least one person who does not have health Dichotomous 0.18
insurance or free health care certificate
Housing Condition roof Roof material: Order 0.48
Straw, canvas, tar paper (1)
Panels/Galvanized iron (2)
Tile (3)
Concrete, cement (4)
Condition wall Wall materials: Order 0.52
Branches/bamboo
(1)
Calcareous earth/straw (2)

al.
A. T. Q. Pham et
Wood/galvanized iron (3)
Fired brick, stone (4)
Concrete (5)
 Table 1 (continued)

Poverty…
Targeting Administrative Regions for Multidimensional
Dimensions Indicators Description Type of indicator Computed weight

Basic services Water Main water drinking

supply: Rainwater (1) Order 0.10
Unprotected spring sources (2)
Protected spring sources (3)
Hand-dug, non-reinforced and uncovered wells
(4) Hand-dug and covered wells (5)
Deep drill wells (6)
Public tap water (7)
Private tap water inside the house (8)
Sanitation Households with some of the
following: No toilet (1) Order 0.13
Toilet directly over the water (2)
Double vault compose latrine (3)
Squat toilet (4)
Flush toilet with septic tank/sewage pipes (5)
Energy Main source of lighting/cooking:
Candle/other (1) Order 0.76
Kerosene/gas (2)
Battery/diesel engine (3)
Electricity (4)
Durable assets Vehicle Household does not own any bike or motorbike Dichotomous 0.73
Telephone Household does not own any telephone, including mobile phone Dichotomous 0.17
TV Household does not own any black-and-white or colour television Dichotomous 0.11
13
 Table 1 (continued)
13

Dimensions Indicators Description Type of indicator Computed weight

Economic status Food Household whose own judgment on their consumption of food over
the last 30 days was insufficient to meet members’ needs Dichotomous 0.18
a
Foodstuff Household whose own judgment on their consumption of foodstuff
over the last 30 days was insufficient to meet members’ needs Dichotomous 0.09
Electricity Household whose own judgment on their consumption of electricity
over the last 30 days was insufficient to meet members’ needs Dichotomous 0.13
Water Household whose own judgment on their consumption of water over
the last 30 days was insufficient to meet members’ needs Dichotomous 0.26
Housing Household whose own judgment on their consumption of housing over
the last 30 days was insufficient to meet members’ needs Dichotomous 0.18
Clothes, footwear Household whose own judgment on their consumption of clothes or foot-
wear over the last 30 days was insufficient to meet members’ needs Dichotomous 0.14
Savings Household does not have any savings Dichotomous 0.02
a
While “food” includes agricultural products, “foodstuff” includes other food items

al.
A. T. Q. Pham et
maximize the available information in the VHLSS 2014, and to follow the targets of the
SDG on health,12 this study encompasses two indicators in the health dimension. The first
indicator determines a household as deprived in health if the family could not afford to
cover all necessary health care expenses for any household member who was sick in the
last 12 months. The second indicator determines a household as poor in the health dimen-
sion if there is at least one household member who does not participate in public health
insurance or have a free health care certificate.

3.2.3 Housing and Basic Services

Housing and accessibility to basic services are used in numerous poverty analyses and
comprise an important part of minimum cost-of-living (for example, Fiadzo et al. 2001;
Battiston et al. 2013; Santos et al. 2015). In Vietnam, a number of studies have concluded
that quality of housing and basic services has statistically significant effects on the prob-
ability that a household is poor (Minot 2000; Dasgupta et al. 2005; Baulch and Dat 2010).
We include conditions of roof and wall as two ordered indicators to define the housing
dimension. In basic services there are three indicators that are also recognized as important
in the SDGs: clean water, improved sanitation, and cooking fuel.

3.2.4 Durable Assets

Diverse studies consider ownership of a radio, television, telephone, and motorbike

or bicycle as indicators of poverty (see, among others, for Vietnam: Baulch and Masset
2003; Baulch and Dat 2010; Minot 2000; for Africa: Sahn and Stifel 2000; Heltberg and
Tarp 2002; for Mozambique: Stifel and Christiaensen 2007). In our study, a household is
deprived in durable assets when it does not have any one of the goods from the list.

3.2.5 Economic Status

Martinetti (2000) claims that the inclusion of the subjective indicator of well-being does
not substitute but complements our judgment and allows us to compare the findings devel-
oped from a combination of subjective and objective assessment criteria. The economic
status dimension, hence, is defined by seven indicators that take into account a subjective
perception on a household’s own condition or assessment by the household head, about
the level of contentment concerning sufficient consumption of essential goods such as
food, water, electricity, housing, clothes and footwear. In addition, we include the
households’ savings ability.

3.3Household Level Variables

The socio-economic characteristics of households are selected to connect the seven dimen-
sions of poverty. These are also the most used determinants in previous studies on poverty
in Vietnam (Arouri et al. 2015; Dasgupta et al. 2005; Imai et al. 2015; Lanjouw et al.
2017;

12
One of the targets of the SDG on health is “Achieve universal health coverage, including financial risk
protection, access to quality essential health-care services and access to safe, effective, quality and afford-
able essential medicines and vaccines for all”.
Targeting Administrative Regions for Multidimensional A. T. Q. Pham et al.
Poverty…
Paul et al. 2002, among others) as well as in studies on other developed countries (see
Azeem et al. 2016; Brady and Kall 2008; Brady et al. 2009; Chen and Wang 2015; Kim
et al. 2010). Table 2 provides a list and description of these variables.

4 Results of Empirical Analysis and Discussions

4.1Pattern of Average Regional Poverty in Vietnam in Different Dimensions

Table 3 shows the average degrees of poverty in seven dimensions for the whole country,
urban and rural, and for the six administrative regions, using the Eq. (1) for income dimen-
sion and the Eq. (4) for non-income dimensions. At the country level, no significant
dispar- ity is observed in the income and non-income dimensions except in housing and
education. It is perceived that while the lower degrees of deprivation in most non-income
dimensions move in line with the higher average levels of income, deprivation rates in
health move in the opposite direction. This suggests that the most deprived households in
the income dimension may not be the poorest in health. The government projects that
relate to financ- ing and health insurance, the Social Health Insurance and the Population
Coverage Rate (introduced in 1992) have made outstanding achievements in health
financing. By June 2010, 62% of the population had health insurance, with the objective
of 80% of Vietnamese being insured by 2020 (Van Tien et al. 2011).13 The estimation
presented in Table 3 shows the relative consistency of ranking between our income
poverty measure and the head- count poverty ratio based on official statistics in six regions
of the country. While the South East and the Red River Delta regions are least poor, the
Midlands and Northern Moun- tains, where the majority of Vietnam’s ethnic minorities
reside, have the highest income poverty.14
However, when looking more closely at the figures for each non-income dimension in
the last 6 columns of Table 3, it can be observed that differences across regions are sub-
stantial. The South East region experiences the lowest income poverty (0.24) but health
poverty (0.13) in this region is greater than that of the Central Highlands (0.12) where
income poverty (0.46) is nearly twice that of the South East. The region of the Red River
Delta has the lowest degrees of deprivation in the education, housing, and economic status
dimensions (0.17, 0.05, and 0.05, respectively). However, this region has the highest
popu- lation in the country (more than 20 million) and has the worst situation in health
poverty. It is worth noting that the capital of the country, Ha Noi City, is located in this
region, which has 50 thousand immigrants arriving every year (GSO, 2014). Since the
informal sector of the labour market is excluded from the benefit of health care insurance,
more than 37% of the Red River Delta’s citizens are not eligible for any health care
insurance or free health care programs (as calculated from VHLSS, 2014 data).
In contrast, the Midlands and Northern Mountains region experiences the highest pov-
erty values in health, basic services and economic status. However, compared to other
regions, health poverty here is moderate. The Midlands and Northern Mountains is the

13
In 2009 the Government of Vietnam passed the Health Insurance Law, which offers up to 100% subsi-
dies on health insurance premiums for the very poor, ethnic minorities, the elderly, and all children under
6 years of age.
14
The results in the previous studies show that poverty rates are lowest in the regions of Red River and
Mekong deltas; the Midlands and Northern Mountains region has the highest poverty rates in the income
dimension (see among others, Lanjouw et al. 2017; Cuong 2009; Baulch and Dat 2010).

13 13
Table 2 Various socio-economic variables used for the estimations of different dimensions of poverty in
multilevel models
Variable names Description Type of variable

Age Age of household head Continuous

Minority Ethnicity of household head 1: Minorities;
0: Kinh
Female Gender of household head 0: male;
1: female
Urban Residency of household 0: Rural
1: Urban
noMale No. of male in household Discrete
noFemale No. of female in household Discrete
noKid04 Number of kids below 5 in household Discrete
avgeduMale Average education of males in household Continuous
avgeduFemale Average education of females in household Continuous
Migrant household Households from which former members have migrated 1: Migrated household;
(moved to a different place from their place of 0: non-migrated
residence)
Percentage job 1 Percentage of wage/salary employment Proportion
Percentage job 2 Percentage of self-employment in agriculture, forestry, Proportion
aquaculture
Percentage job 3 Percentage of non-farm activities (self-engagement in Proportion
production, business, services outside agriculture,
forestry, aquaculture)

region that lags behind and cannot participate in the growth process of the country because
of its undeveloped infrastructure and disadvantaged physical environments (such as poor
soils, deforestation and environmental degradation).15 It can be noted that the Health Insur-
ance Law of the Government contributes significantly towards a reduction in the depriva-
tion level of health in the Midlands and Northern Mountains region due to that area having
the highest number of ethnic minorities.

4.2Multilevel Analyses: Magnitude of Various Dimensions of Poverty at

Different Administrative Levels

4.2.1 Influence of Household Characteristics

To determine the influence of household level variables prompting various dimensions of

poverty we use a multilevel regression model. We include household characteristics vari-
ables to the random intercept models, which allow intercepts to vary across higher levels
(as discussed in Sect. 2.2). Table 4 shows the results of separate random intercept models
for seven dimensions of poverty.16

15
See Glewwe et al. (2004), and Rambo and Lê (1996).
16
The results reported here are from the best models based on the Akaike Information criterion (AIC) (see
Goldstein, 2011). The likelihood ratio test was used to test whether a multilevel model is preferred to a sin-
gle-level model. We provide the results in Table 9 in the Appendix. Our tests show that the random effects
and residual errors in the random part of the model are independent of the vector of predictor variables and
approximately normally distributed. A sample of test results are provided in Appendix Table 8.
 Targeting Administrative Regions for Multidimensional
Poverty…
13

Table 3 Extent of income and non-income poverty in Vietnam—urban rural and 6 regions

Number house-
Average Poverty head- Income h
Non-income poverty (P(k))
income count ratioa poverty
holds (1,000VND) (%) (Ph) Education Health Housing Basic service Asset Economic status

Vietnam 9399 109,049.5 8.4 0.42 0.31 0.16 0.28 0.32 0.15 0.10
Urban 2781 153,865 3.0 0.24 0.21 0.15 0.21 0.10 0.20 0.06
Rural 6618 75,224.04 10.8 0.50 0.35 0.17 0.31 0.41 0.14 0.12
Red River delta 1992 123,947.8 4.0 0.34 0.17 0.21 0.05 0.22 0.15 0.05
Midlands and Northern Mountains 1662 82,312.01 18.4 0.56 0.34 0.18 0.41 0.51 0.12 0.20
Northern and Coastal Central 2067 96,892.3 11.8 0.46 0.28 0.15 0.17 0.29 0.14 0.10
Central Highlands 651 102,699.5 13.8 0.46 0.38 0.12 0.36 0.43 0.13 0.14
South East 1122 152,629.9 1.0 0.24 0.28 0.13 0.29 0.13 0.18 0.07
Mekong River delta 1905 106,491.1 7.9 0.43 0.45 0.15 0.50 0.35 0.20 0.08

a
The poverty head count ratio is identified according to the Government of Vietnam’s poverty line for 2011–2015 (1000VND/person/month): 630 for rural areas and 780 for
urban areas. Source for this column is GSO (2014)

al.
A. T. Q. Pham et
 Table 4 Estimates of the influences of household traits in different dimensions of poverty: multilevel models (VHLSS 2014)

Poverty…
Targeting Administrative Regions for Multidimensional
Variables Dimensions
Income Education Health Housing Basic services Durable asset Economic status

Age 0.000** 0.001*** – 0.001*** – 0.001*** – 0.001*** 0.004*** – 0.001***

(0.000) (0.000) (0.000) (0.000) (0.000) (0.000) (0.000)
Female 0.018*** 0.002 – 0.001 0.011* 0.014** – 0.008 0.007
(0.006) (0.004) (0.006) (0.006) (0.006) (0.008) (0.007)
Minority 0.158*** 0.066*** – 0.102*** 0.151*** 0.147*** 0.067*** 0.105***
(0.009) (0.006) (0.008) (0.009) (0.009) (0.010) (0.011)
Urban – 0.066*** 0.008* 0.001 – 0.042*** – 0.135*** 0.034*** – 0.022**
(0.007) (0.005) (0.007) (0.008) (0.008) (0.009) (0.009)
noMale – 0.010*** – 0.005** 0.021*** – 0.003 0.003 – 0.025*** 0.002
(0.003) (0.002) (0.003) (0.003) (0.003) (0.004) (0.004)
noFemale – 0.009*** 0.000 0.016*** – 0.009*** – 0.002 – 0.013*** – 0.005*
(0.003) (0.002) (0.003) (0.002) (0.002) (0.003) (0.003)
nokid04 0.017*** – 0.017*** 0.018*** 0.005 0.010** 0.000 0.017***
(0.005) (0.003) (0.004) (0.004) (0.004) (0.005) (0.005)
avgeduMale – 0.064*** – 0.137*** – 0.012*** – 0.021*** – 0.030*** 0.004 – 0.023***
(0.003) (0.002) (0.002) (0.002) (0.002) (0.003) (0.003)
avgeduFemale – 0.040*** – 0.074*** – 0.005* – 0.022*** – 0.028*** – 0.001 – 0.016***
(0.003) (0.002) (0.003) (0.003) (0.003) (0.003) (0.003)
Percentage job 1 – 0.103*** 0.052*** – 0.000 0.038*** 0.034*** – 0.062*** 0.020*
(0.009) (0.006) (0.009) (0.009) (0.009) (0.011) (0.011)
Percentage job 2 0.059*** – 0.010* 0.028*** 0.010 0.113*** – 0.076*** – 0.002
(0.009) (0.006) (0.008) (0.008) (0.008) (0.010) (0.010)
Percentage job 3 – 0.208*** – 0.036*** 0.054*** – 0.032*** – 0.070*** – 0.008 – 0.028**
(0.011) (0.007) (0.010) (0.010) (0.010) (0.013) (0.012)
Migration – 0.032*** – 0.004 – 0.001 – 0.018*** – 0.003 – 0.016** – 0.008
13

(0.006) (0.004) (0.006) (0.006) (0.006) (0.007) (0.007)

Constant 0.731*** 0.850*** 0.200*** 0.452*** 0.487*** 0.027 0.233***
(0.018) (0.012) (0.017) (0.025) (0.019) (0.020) (0.020)
 Table 4 (continued)
13

Variables Dimensions
Income Education Health Housing Basic services Durable asset Economic status

AIC 646.2753 – 9081.125 – 2502.607 – 2450.439 – 2389.851 1238.07 646.2753

0 hcdp p dp cdp hcdp u v w e

In Eq. (7), βk + βkXk is the fixed part of the model and uk + vk + wk + ek is the random part with random parameters σ2, σ2, σ2 , and σ2. The fixed part of the
model
denotes the overall mean relationship between the outcome and the explanatory variables; that is, the relationship that uses in the average province. The random part of the
model specifies how the province, district, and commune-specific relationships differ from this overall mean relationship. Table 4 presents the estimates of the fixed effects
Standard errors in parentheses
***p < 0.01, **p < 0.05, *p < 0.1

A. T. Q. Phametal.
 A. T. Q. Pham et
al.

In general, most explanatory variables are statistically significant and have expected
signs. For example, ethnic minority households have significantly higher rates of depriva-
tion in all dimensions compared to their non-minority counterparts, except in the health
dimension. The estimated coefficient for ethnic minority households has a 10.2% lower
health deprivation level than the ethnic majority citizens. This is possibly the effect of
gov- ernment policies and programs that have been in place to assist the ethnic minorities
in health insurance since 2005. According to a report from the Ministry of Finance, during
the period 2012–2014 an annual average of VND12500 billion (USD 545.8 million) was
spent to support ethnic minorities, students and children under 6 years of age by providing
free health insurance cards.
The results also reveal that, as expected, higher average education levels (for both
males and females) have significant effect in lowering deprivation rates in most
dimensions. These results are consistent with numerous previous studies on the positive
effects of edu- cation on reducing poverty. 17 Furthermore, the rates of deprivation in
almost all dimen- sions are significantly lower for households with higher proportion of
members working in non-farm activities. Households with older heads have a lower
deprivation level in the health, housing, basic services, and economic status dimensions,
but suffer from higher deprivation rates in income, education, and durable assets. The
findings shown in Table 4 affirm the complex nature of poverty in Vietnam and highlight
the necessity of targeting poverty beyond the current approach based on income
deprivation alone.

4.2.2 Partitioning Variation at Higher Administrative Levels

Figure 2 presents VPCs in seven dimensions of poverty attributable to each administrative

level after controlling for specific household characteristics. The VPCs in these models
enable us to establish the relative importance of provinces, districts, communes/villages
and households in explaining the variation in households’ deprivation in different dimen-
sions of poverty. In general, as observed in Fig. 2, household level accounts for the highest
per cent of the total variation in deprivation rates. This suggests that household
characteris- tics in Vietnam play a crucial role in explaining the income and non-income
poverty status of households. However, nearly 15% VPCs in the higher levels indicate that
households from the same higher administrative area are significantly more alike than
housholds from different areas. Therefore, the impact of province, district,
commune/village levels is also an important clustering factor on the level of deprivation in
all dimensions.
Specifically, while the VPCs of the housing and basic services dimensions report the
highest values at higher administrative levels, the VPCs of the health, education and asset
dimensions have the lowest values at these levels. For example, after controlling for
house- hold characteristics we observe that more than 50% of the total unexplained
variation in housing deprivation is above the household level. Especially in this dimension
the prov- ince level has a VPC value of 0.346, 18 showing that the province level plays an
important role in explaining the variance of housing poverty. As far as the asset dimension
is con- cerned, 13.4% of the variation in deprivation is explained by factors at commune,
district,

17
See, for example, in Romania (Mihai et al. 2015), in South Sudan (Shimeles and Verdier-Chouchane
2016), and in Vietnam (Klasen et al. 2015; Gloede et al. 2015).
2

13
Targeting Administrative Regions for Multidimensional
Poverty…
0.025
u
(0.025+0.007+0.006+0.034 = 0.364 . Table 10 in the Appendix provides the crude values of
)
v
= 2 +σ2
σ2 +σ2 +σ
u w e

13
18
A. T. Q. Pham et
al.
σ
all variations. These are the ingredients of the computation of VPCs presented in Fig. 2.

13
Targeting Administrative Regions for Multidimensional
Poverty…

and province levels. These high values of VPCs confirm the necessity of using a
multilevel approach in analysing poverty in Vietnam. Our results indicate that the
contextual factors of province, district, and commune/village should be taken into account
in future poverty studies in Vietnam. Detailed analysis of contextual factors is beyond the
scope of this study; however, our findings show the presence of gross significant
heterogeneity across administration levels. This highlights the drawbacks of the current
poverty-alleviation pro- grams by the government which rely solely on targeting
administrative areas on the basis of a single-level observation (i.e. head count of
household income poverty).

4.2.3 Maps of Poverty in Vietnam Based on Multilevel Analysis

Having observed the significant variation of poverty at higher levels (particularly at the
province level), this section aims to classify the provinces of Vietnam on the basis of vary-
ing intensity of poverty. Using the predictions of the random effects (errors of province,
district, commune, and household levels) of our model, we examine the contribution of
unobserved contextual effects of the poverty status of the household. Unlike previous
well- being studies in Vietnam, our empirical findings are derived from a multilevel model
which simultaneously takes into account the effects of both the household level factors and
the between administrative area (mainly province) variability of poverty. We present in
Fig. 3 the maps for each poverty dimension, clearly distinguishing the bad, medium, and
good provinces (following the method we discussed in Sect. 2.2) as obtained from the
multilevel models (see right panel of Fig. 3). The numbers within the maps represent
ranking of the provinces in seven dimensions of poverty. The province ranking is further
discussed in the Sect. 4.3.1.
These maps reveal some interesting facts. Firstly, there is no province which is classi-
fied as good across all dimensions. Secondly, provinces identified as bad in the income
dimension are not necessarily determined as bad in non-income dimensions. This dem-
onstrates that current poverty-alleviation programs in Vietnam, which are based on a uni-
dimensional approach, may lead to inappropriate decision-making and mal-utilization of
scarce resources. Our results from a multilevel analysis reveal significant implications that
should encourage province authorities to focus on certain other aspects of poverty. For
example, quite a few provinces in the south are good in terms of the income dimension,
but bad in terms of education or housing. Thus, the local government in these provinces,
while designing policies and programs to reduce poverty, need to improve the deprivation
levels in education and housing.
The Ministry of Labour, Invalids and Social Affairs (MOLISA) has prime accounta-
bility for the redistribution of funds for social support programs to the poor in Vietnam.
MOLISA has applied a “bottom-up” approach that produces a list of poor households,
which is aggregated up to village/commune level and then to district and province levels.
It is challenging to ensure consistent exercise of those criteria in the field when applying
different guidelines and standards in collecting data by different provinces (Conway
2001). Moreover, it has also been claimed that the list of poor households based on
MOLISA’s approach is relatively arbitrary and fails to properly determine the majority of
the poor (Baulch and Minot 2002; Nguyen and Tran 2014). This results in limited
efficiency in pov- erty-alleviation programs based on the list from MOLISA (Walle 2002;
Minot and Baulch 2004). Our identification of provinces according to incidence and
severity of household poverty provides useful indications to enhance the effectiveness of
dissemination of pov- erty-alleviation funds.
13
 A. T. Q. Pham et
al.
1.000

0.866
0.852

0.844
0.820
0.900

0.736
0.800

0.700

0.567
0.600

0.474
0.500

0.346
0.400

0.300

0.182
0.136

0.125
0.116

0.113
0.200
0.098
0.093

0.082
0.072

0.069
0.068

0.058

0.057
0.053

0.052
0.039

0.026
0.025
0.021

0.010

0.100

-
INCOME EDUCATION HEALTH HOUSING BASIC ASSET ECONOMIC

Province District Commune Household

Fig. 2 Variance partitioning in various levels in dimensions of poverty

Figure 3 shows a clear difference in distribution of bad and good provinces in income
and non-income dimensions between Northern and Southern Vietnam. In the income and
health dimensions, while bad provinces are only located in Northern Vietnam (Red River
Delta, Midlands and Northern Mountains, and Northern and Coastal Central regions), the
majority of good provinces are located in South Vietnam (Central Highlands, South East,
and Mekong River Delta regions). However, the scenario is reversed in the education and
housing dimensions where most good provinces belong to North Vietnam and a greater
number of bad provinces are situated in South Vietnam.
According to GSO (2014), the highest poverty rates 19 in the country prevail particu-
larly in three regions: Midlands and Northern Mountains (18.4), Central Highlands (13.8),
and the Northern and Coastal Central region (11.8) (see Table 3). It is also interesting
to observe that while the official poverty rates in the Midlands and Northern Mountains
region are nearly twice the rates experienced in Northern and Coastal Central, the latter
region having the greater number of bad provinces in the income dimension. Also, in the
Central Highlands region, which is the second poorest in the country (according to GSO
2014), not one province shows as bad in our income measure of poverty. This discrepancy
in the results is a manifestation of the between-province variation that has been mostly
neglected in previous studies.
The Vietnam Government’s perspective towards poverty reduction policies has expe-
rienced transmutations during the process of socio-economic development. In the begin-
ning of “Doi Moi” (1986–2000) the government prioritized hunger eradication and
poverty

19
The Vietnam Government’s poverty lines for 2014 were updated by CPI as 605 and 750 thousand VNDs
per capita per month for the rural and urban areas respectively (GSO 2016).
13
Targeting Administrative Regions for Multidimensional
Poverty…

Fig. 3 Maps of provincial poverty (various dimensions)—single-level versus multilevel analyses

13
 A. T. Q. Pham et
al.

Fig. 3 (continued)

13
Targeting Administrative Regions for Multidimensional
Poverty…

Fig. 3 (continued)

13
 A. T. Q. Pham et
al.

Fig. 3 (continued)

reduction, in 2000–2015 poverty policies were extended to support access to health ser-
vices, education, housing, and clean water for the poor. Since MOLISA’s list of poor
households that are eligible for specific social support programs (such as education or
health insurance) is based on the income approach, the majority of the beneficiaries reside
in those regions where income poverty rates are highest. Thus, many households that are
detected as non-income poor in our analysis are not considered by the Department as eligi-
ble for social support programs.20,21
While at the household level, there is a low likelihood of a “minority” household
having high levels of health poverty (as seen in Table 4), the multilevel analyses maps
in Fig. 3

20
Public social support programs include: health insurance support; exemption and reduction of healthcare
and tuition fees for the poor; scholarships; vocational training; housing support for the poor; provision of
clean and clear water; and food support (GSO 2014).
21
In this context, it is worth noting that the largest ethnic minority populations who are considered the
poorest in the country reside in the poorest regions of the country (GSO 2014). Since 2005, the government
has implemented policies to support poor households and ethnic minorities in sustainable poverty reduc-
tion. According to the VHLSS 2014, nearly 70 % of the beneficiaries of social support programs are in
the three poorest regions. According to MOLISA (2016), on an annual average, the total state budget in
the period 2012–2014 allocated for social support programs reached nearly VND 25,500 billion, includ-
ing health care for the poor (VND12,500 billion); education, training and vocational training for the poor
(nearly VND12,000 billion); accommodation support, clean water and electricity provision (VND,000 bil-
lion). Although these programs contribute to notable achievements in poverty reduction of the country, our
findings indicate obvious drawbacks in monitoring and evaluating poverty.

13
Targeting Administrative Regions for Multidimensional
Poverty…

show that the regions of Red River Delta and Midlands and Northern Mountains are the
worst performers in terms of the health dimension. The latter region has the highest pop-
ulation of ethnic minorities. This implies that ethnic minorities living in the Red River
Delta or Midlands and Northern Mountains regions experience higher levels of depriva-
tion in health than households with same background living in the other regions. Even
if the households in this region are eligible for free health care certificates according to
MOLISA’s approach, the provision is insufficient. Nearly 70% of ethnic minority house-
holds who, based on the VHLSS 2014, are eligible for health insurance or free health care
certificates were unable to cover their health care expenses. In these regions the average
distance from communes where minorities live to the nearest health care facilities is great-
est compared to the other regions in the country. This necessitates an additional amount
in total health care expenses for the poor in the Midlands and Northern Mountains region,
which points to an important implication that not only household characteristics but also
contextual factors are vital in evaluating health poverty. 22
We note in Fig. 2 that variability in some dimensions at the commune level is substan-
tial. Since maps of district and commune levels could not be created due to a lack of infor-
mation, Table 5 presents a cross-tabulation of commune/village- versus province-level
pov- erty in seven dimensions to analyse the clustering of good, medium, and bad
communes. We observe that it is rare for a bad commune to be nested within a good
province as well as a good commune to be nested within a bad province. It is often the
case that a bad com- mune is nested within a bad province, and a good commune in good
province. Of seven poverty dimensions, only the housing and basic services dimensions
have bad communes nested in good provinces: in the case of housing, there are 73 good
communes clustered in bad provinces and 66 bad communes clustered in good provinces;
while in the case of basic services, there are 48 communes clustered in bad provinces and
10 bad communes clustered in good provinces. This finding clearly indicates the
complexities in housing and basic services poverty in Vietnam (a similar observation is
made in Fig. 2). We note also that income, education, durable assets and economic factors
of poverty are less complexly clustered at the hierarchical levels.

22
We also note that the government does provide ad hoc support on the basis of regional characteristics.
For example, households in the Northern and Coastal Central regions are more likely to be affected by
natural disasters than the households in other regions. The proportion of households affected by storms,
typhoons, floods and flash floods is highest in these regions (Bui et al. 2014, and Chaudhry and
Ruysschaert 2007). During the period 2012–2014, more than 500,000 poor households of this region were
supported in housing and building houses to respond to floods and flash floods. The highest percentage of
households benefiting from the housing support programs to this region might be one reason that the
provinces of this region present significantly lower levels of deprivation in the housing dimension than the
overall average. Although ethnic minorities have higher levels of housing deprivation than the ethnic
“majorities”, the for- mer living in the Northern and Coastal Central region may be better off in the
housing dimension than households living in other regions.

13
 A. T. Q. Pham et
al.
4.3 Targeting Provinces for Ameliorating Poverty—Single
Versus Multilevel Analysis

4.3.1 Traditional Single-Level Analysis Versus the Multilevel Analysis

—Variations in Poverty Ranking of Provinces

As we observe above, the target provinces for controlling poverty as determined by our
multilevel analysis are different from those obtained from the traditional single-level
analy- sis. In this section we further refine our identification of provinces that need prime
atten- tion for poverty alleviation.
In Appendix Table 11, computed values of poverty are presented without considering
the effects of variation at the higher (commune/district/province) levels. These poverty
val- ues depend solely on household traits, which is the traditional method utilized in
Vietnam (on income poverty only) for targeting areas for financial support in eradicating
poverty. We rank the provinces from best (rank 1 for least poverty) to worst (province
with the greatest poverty ranks 63) and present them in the left panel of Fig. 3 and in
Table 12 in the Appendix. In the right panel of Fig. 3, the provinces are marked by rank
according to the predictions of the random effects (province errors) of multilevel models,
which constitute the contribution of unobserved contextual factors to the level of
deprivation, controlled for differences in the characteristics of households. Thus, Fig. 3
(and Table 12) presents the poverty ranking from both methods in seven poverty
dimensions. For each dimension,
p
the province level residual, uk of dimension k for
province p is predicted after the models are fitted. The predictions and their standard errors
are yielded by using the Empirical Bayes (EB) method. We then rank the predicted errors
(or residuals) of each level together with 95% confidence intervals. These residuals
symbolise the province level departure
0
from the overall mean predicted by the βk. A
province with negative (positive) predictions of resid- uals is likely to have households
whose measurements of deprivation are below (above) the population mean. A detailed
diagram of the estimation for ranking is presented in the Appendix (Fig. 5).
A clear difference can be seen in the ranking of provinces between the two approaches,
especially in the income, health, and basic services dimensions. In the right panel of Fig. 3
(using a multilevel analysis), all the “bad” provinces with the highest values in income
poverty are located in the Midlands and Northern Mountains region. However, the distri-
bution of the worst provinces in terms of a single-level study are less concentrated in the
Midlands and Northern Mountains and the Northern and Coastal Central regions. In the
health dimension, provinces in the Midlands and Northern Mountains region depart sub-
stantially in ranking between the two models. All provinces except two in this region are
in the middle (those ranked 40 and above) of the distribution from the single-level
analysis, while accounting for the between-province variation, we find 71 provinces are in
a worse situation. In the basic services dimension, while the majority of lower-ranked
provinces are located in the Midlands and Northern Mountain region in the single-level
analysis, the dis- tribution of “bad” provinces in the multilevel model is concentrated in
both the Northern and Coastal Central region and the Midlands and Northern Mountains
region.
Table 6 presents the list of provinces and the values of the differences in ranking
between the single-level and multilevel analyses. A positive (negative) value indicates that
the province is in a better (worse) position in the poverty rankings in terms of a multilevel
analysis. Tables 6, 11 and Fig. 3 reveal that the changes in the average absolute ranking in
the income dimension for bad and good provinces are greatest among the seven poverty

13
Targeting Administrative Regions for Multidimensional
Poverty…
Table 5 Cross-tabulation of commune versus province level poverty—various dimensions

N, % of total commune

Good Medium Bad

Commune-level Income Poverty

Province-level Income Poverty Good 6 0.19% 856 27.33% 0 0.00%
N (% of total commune) Medium 0 0.00% 1681 53.67% 0 0.00%
Bad 6 0.19% 561 17.91% 22 0.70%
Commune-level Education Poverty
Province-level Education Poverty Good 0 0% 884 28.22% 0 0.00%
N (% of total commune) Medium 0 0% 1312 41.89% 0 0.00%
Bad 0 0% 914 29.18% 22 0.70%
Commune-level Health Poverty
Province-level Health Poverty Good 0 0% 777 24.81% 0 0.00%
N (% of total commune) Medium 0 0% 1428 45.59% 0 0.00%
Bad 0 0% 914 29.18% 13 0.42%
Commune-level Housing Poverty
Province-level Housing Poverty Good 3 0.10% 1370 43.74% 66 2.11%
N (% of total commune) Medium 18 0.57% 673 21.49% 21 0.67%
Bad 73 2.33% 845 26.98% 63 2.01%
Commune-level Basic Services Poverty
Province-level Basic Services Poverty Good 39 1.25% 930 29.69% 10 0.32%
N (% of total commune) Medium 22 0.70% 1401 44.73% 74 2.36%
Bad 48 1.53% 570 18.20% 38 1.21%
Commune-level Durable Assets Poverty
Province-level Durable Assets Poverty Good 0 0.00% 88 2.81% 0 0.00%
N (% of total commune) Medium 18 0.57% 2502 79.89% 37 1.18%
Bad 0 0.00% 437 13.95% 50 1.60%
Commune-level Economic status Poverty
Province-level Economic status Poverty Good 0 0.00% 0 0.00% 0 0.00%
N (% of total commune) Medium 0 0.00% 2826 90.23% 136 4.34%
Bad 11 0.35% 106 3.38% 53 1.69%

dimensions. This means that variation in the higher-level contextual factors in regard to
the income dimension in determining poverty in Vietnam is substantially high. Thus,
disburse- ment of poverty-alleviation funds in provinces on the basis of household traits
alone is not the most efficient way of reducing income poverty in Vietnam. Consideration
of regional variations must be included in any strategy for attacking the problems of
poverty.

4.3.2 Large Changes in Ranking of Provinces: Single Versus Multilevel

Analysis

The poverty scenario in Vietnam can be studied in more detail by looking at large changes
in ranking between single and multilevel analyses. For this purpose, let’s consider a
change in ranking of 5 points as a large change. We note from results presented in
Table 7 that in the income dimension, 21 provinces show an increase in ranking, while for
23 prov- inces the ranking decreases. A 5-point change in the ranking of 70% of provinces

13
 A. T. Q. Pham et
al.
is defi- nitely a significant variation; as the utilization of funds from poverty-alleviation
programs

13
 Table 6 Difference in poverty rankings of provinces—multilevel versus single-level analysis

Poverty…
Targeting Administrative Regions for Multidimensional
Province Income Education Health Housing Basic Asset Economic

Red River Delta Ha Noi City –9 – 42 2 –4 –7 42 – 26

Quang Ninh –2 –8 –4 –2 –9 26 – 15
Vinh Phuc – 26 –7 10 –1 –9 – 26 – 16
Bac Ninh 1 –4 5 1 3 8 –7
Hai Duong – 17 2 –1 0 2 4 – 18
Hai Phong City –2 –2 –3 –1 –6 30 – 21
Hung Yen – 18 –4 2 1 –3 2 – 20
Thai Binh –5 –1 –6 1 2 23 – 12
Ha Nam – 13 0 8 –1 –8 – 30 –6
Nam Dinh – 14 4 2 2 7 2 –9
Ninh Binh – 16 7 3 –1 2 –6 – 17
Midlands and Northern Mountainous Areas Ha Giang 10 13 – 17 7 1 –5 0
Cao Bang 10 18 – 20 7 –1 – 33 7
Bac Can 4 20 – 15 10 4 – 32 5
Tuyen Quang 6 16 1 7 11 – 39 14
Lao Cai 3 1 – 13 3 1 – 21 30
Dien Bien 19 5 – 18 4 0 – 33 0
Lai Chau 10 6 3 9 0 –5 0
Son La 5 4 – 16 6 –1 – 15 23
Yen Bai 13 12 –2 1 7 – 21 –1
Hoa Binh 16 22 –7 9 9 – 45 3
Thai Nguyen – 17 –1 5 0 – 10 – 34 1
Lang Son 20 19 – 14 12 3 – 40 5
Bac Giang 1 9 6 4 8 – 10 1
13

Phu Tho – 12 6 –2 –1 2 – 20 1
 Table 6 (continued)
13

Province Income Education Health Housing Basic Asset Economic

Northern and Coastal Central Region Thanh Hoa –9 3 –2 –2 2 – 24 –5

Nghe An – 12 4 –1 –1 –2 –8 –3
Ha Tinh – 13 –7 5 –1 –9 – 44 1
Quang Binh – 16 –5 10 –2 – 11 – 35 – 11
Quang Tri – 17 – 16 0 –2 2 2 –2
Thua Thien - Hue – 21 – 14 2 1 1 32 –4
Da Nang City – 22 – 27 –1 – 10 –9 49 – 25
Quang Nam –7 4 1 –4 2 24 1
Quang Ngai –6 3 3 2 –1 – 27 5
Binh Dinh 2 –5 4 3 –4 –2 –9
Phu Yen –2 0 5 1 3 – 22 –2
Khanh Hoa 3 – 36 – 21 – 36 – 19 23 – 10
Ninh Thuan 12 10 1 5 6 45 18
Binh Thuan 6 4 3 –2 0 22 5
Central Highlands Kon Tum 2 7 –3 8 1 –8 32
Gia Lai 21 –5 –3 3 4 – 29 –1
Dac Lac 0 –6 –1 3 –8 – 31 –4
Dac Nong 20 9 –1 –1 11 – 33 21
Lam Dong 1 –2 –4 –1 –6 15 –2

A. T. Q. Phametal.
 Table 6 (continued)

Poverty…
Targeting Administrative Regions for Multidimensional
Province Income Education Health Housing Basic Asset Economic

Southeastern Area Binh Phuoc 12 9 –6 0 0 – 25 –4

Tay Ninh 4 –7 4 –3 5 8 0
Binh Duong 1 4 5 –4 1 18 0
Dong Nai 4 –4 5 –3 –1 41 –9
Ba Ria - Vung Tau 3 –8 4 –4 0 44 –2
Ho Chi Minh City –2 – 29 –8 –7 –4 54 – 25
Mekong Delta Long An 6 7 12 –5 –6 –2 –4
Tien Giang 5 –1 8 –4 6 27 –9
Ben Tre 9 –5 –3 –5 –7 –8 –3
Tra Vinh 22 0 –1 1 5 –9 26
Vinh Long –2 –1 10 –5 –3 –8 12
Dong Thap –8 –5 10 –3 7 13 2
An Giang 5 –1 11 –7 5 31 17
Kien Giang 16 –1 1 –4 –1 6 22
Can Tho –7 –4 3 –4 –8 29 –6
Hau Giang –2 –5 16 –7 –2 9 8
Soc Trang 19 5 –8 –2 7 30 21
Bac Lieu 14 –1 6 –1 6 27 – 16
Ca Mau 7 –5 5 –2 0 25 6
13
 A. T. Q. Pham et
al.
Table 7 Number of provinces where ranking changed by more than 5 units

Number of Provinces Monetary Education Health Housing Basic Asset Economic

Total number
Decrease of provinces21= 63
in ranking 14 12 7 15 15 19
Increase in ranking 23 19 18 10 14 13 18
Red River delta (total number of provinces = 11)
Decrease in ranking 8 3 1 0 5 2 11
Increase in ranking 0 1 3 0 1 2 0
Midlands and Northern Mountainous Areas (total number of provinces = 14)
Decrease in ranking 2 1 8 0 2 4 1
Increase in ranking 9 11 2 8 4 2 5
Northern and Coastal Central Region (total number of provinces = 14)
Decrease in ranking 9 6 0 1 3 5 3
Increase in ranking 2 2 3 1 1 5 4
Central highlands (total number of provinces = 5)
Decrease in ranking 0 1 0 0 2 2 0
Increase in ranking 2 2 0 1 1 1 2
Southeast area (total number of provinces = 6)
Decrease in ranking 0 3 2 1 0 0 2
Increase in ranking 1 1 2 0 1 1 0
Mekong River delta (total number of provinces = 13)
Decrease in ranking 2 0 1 5 3 2 2
Increase in ranking 9 2 8 0 6 2 7

in Vietnam is based only on MOLISA’s list of income-poor households, these significant

departures in the ranking of provinces in the two approaches could have a serious implica-
tion for the correct targeting of the current poverty reduction efforts in the country.
In the non-income dimensions, variation in the poverty rankings of provinces is also
evidenced in the education, health, and economics dimensions, with 32 (51%), 30 (48%),
and 37 (59%) provinces respectively, showing a substantial change in ranking (Table 7).
In particular, the region of Midlands and Northern Mountains reports the largest number
of provinces changing their rank in the health and education dimensions. This region is
ranked as the best in the health and second worst in the education dimension in the tra-
ditional single-level poverty analysis (Tables 3 and 10). However, the results of a multi-
level analysis indicate that mostly all provinces in this region have a higher rank
(indicating higher levels of poverty) in the education dimension, with a lower rank (less
poverty) in the health dimension. Given the fact that the Midlands and Northern
Mountains is the poorest region in terms of income poverty, the remarkable departures in
ranking of provinces in the region strongly indicate that both income and non-income
dimensions should be consid- ered in a multilevel analysis for designing more effective
poverty-reduction strategies for Vietnam in the future.

13
Targeting Administrative Regions for Multidimensional
Poverty…
5 Conclusions

This study investigates income and non-income dimensions of poverty in Vietnam in

light of regional variations (due to contextual effects) and household traits. Seven dimen-
sions of household poverty are measured using a fuzzy method. Four administrative lev-
els (comprising provinces, districts, communes/villages and households) are partitioned
in seven dimensions of deprivation utilizing VHLSS 2014 data. Our results show that the
amount of variation in non-income and income poverty that is attributable to provinces
and to communes/villages is considerable even after controlling for the recognized factors
at the household level. For example, of the total variation in poverty in the basic services
and housing dimensions, 13.6% and 34.6%, respectively, are attributable to provinces and
18.2% and 8.2%, respectively, are attributable to communes/villages (Fig. 2). These sig-
nificant variations at province and commune/village levels imply that a definition of pov-
erty-prone areas based on household deprivation alone cannot fully expound the method of
regional poverty targeting in Vietnam. This reveals a need for multilevel analysis of
poverty for the efficient utilization of poverty-alleviation funds. However, studies in this
respect are limited and those that are available are based on an income approach to poverty
alone.23
Poverty alleviation has always played a critical role in the sustainable development
strategies of the Vietnamese government. The government has approved a series of pov-
erty reduction and social support programs with a total budget of over USD 2 billion and
has taken a multidimensional approach for the period 2016–2020; however, the methodol-
ogy used for measuring poverty and in identification of the poor are based on an income
approach only. Our empirical results prove that in Vietnam there are many regions in
which people are better off in the income dimension of poverty but are worse off in some
non- income dimensions. For example, the region of Mekong River Delta is a relatively
good region in the income dimension with three good and no bad provinces. However,
of the 13 provinces in this region, all 13 are classified as bad provinces in education, and
10 are bad in the housing dimension. Especially, Kien Giang province was identified
as good in the income dimension but is found to be bad in most of the non-income
dimensions (see Tables 5 and 6). The proportion of poor people who have benefited from
social sup- port programs in the region of Mekong River Delta is always lower than in the
region of Midlands and Northern Mountains, and lower than in the region of Northern
and Coastal

23
To date, there are only two studies in which multilevel models are applied to explore the role of location-
specific contributions to households’ welfare in Vietnam. Using panel data from two waves of the Vietnam
Living Standards Measurement Surveys (VLSMS) 1993 and 1997, Arpino and Aassve (2014) investigate
the contribution of villages in households’ exit from poverty. Haughton and Nguyen (2010) focus on inves-
tigating geographical variations in the inequality gap in expenditure levels between urban and rural areas.
While both studies apply the income approach to measure the welfare levels of households, we examine
multiple dimensions of poverty for the whole of the country. Arpino and Aassve (2014) employed EB pre-
dictions of village-level random effects to find the good and bad villages and regions in rural Vietnam.
Since the authors do not include the higher levels in their multilevel model, the random errors accumu-
late all the unobserved contextual factors at village and higher geographic levels. Therefore, the variance
component at the village level is inflated because the specified model disregards higher hierarchical lev-
els (Tranmer and Steel 2001). There are four administrative levels involved in our empirical models that
allow us not only to distinguish the random effects of each higher level, but also to gain more efficient
estimations. Haughton and Nguyen (2010) show that spatial effects play an important role in inequality
gaps between urban and rural areas. Although the authors take into account all four administrative levels in
their empirical model, they do not examine the relative importance of these levels on household welfare nor

13
 A. T. Q. Pham et
al.
which members in each level report higher effects on household welfare.

13
Targeting Administrative Regions for Multidimensional
Poverty…
Central in every program. This disbursement of poverty-alleviation funds is justifiable
from the perspective of income poverty, but our findings indicate that the region is worse
off in terms of education, housing and durable asset poverty. Thus, targeting the poor by
the current income poverty approach as used by MOLISA results in widening the disparity
in deprivation of non-income dimensions in the Mekong River Delta. We observe that the
poverty-alleviation approach in Vietnam leads to financial overlapping activities and inef-
ficiencies in the use of resources and in implementation of policies, and poor monitoring.
Our results show that the variance partitioning at the household level is at the
maximum for all dimensions. However, the consistent and significant variations at higher
levels of administration indicate that even after controlling for predictors of poverty at the
house- hold level, considerable variability in poverty is left unexplained for the province,
district, and commune/village levels. In particular, the amount of variation in most
dimensions of poverty attributed to provinces was greater than at district or
commune/village levels. Therefore, this study suggests that in future analyses of poverty
in Vietnam in particular, and developing countries in general, attention should also be paid
to examining provincial characteristics.

Appendix

Since the major analyses of the study are based on province level, we provide here tests of
normal distribution at province level. We utilize the Skewness-Kurtosis (Jarque–Bera) test
in Stata to test against the null hypothesis that random effects and residual errors are nor-
mally distributed. Therefore, in Table 8, a value of adj chi2 greater than 0.05 (Prob > chi2)
implies its significance at 5% level. Consequently, the null hypothesis cannot be rejected
and random effects and residual errors show normal distribution. In Table 8, the normality
assumption of deprivation is satisfied in all dimensions (Figs. 4, 5).
Based on the result of the likelihood ratio test in Table 9, the estimated multilevel mod-
els (Models in Table 4) is preferred to OLS models (Tables 10, 11, 12).

13
 A. T. Q. Pham et
al.
Table 8 Tests for normal distribution of random effects (province level)

Obs Pr(Skewness) Pr(Kurtosis) adj chi2 Prob > chi2

Income 63 0.15 0.37 3.03 0.22

Education 63 0.63 0.02 5.28 0.07
Health 63 0.29 0.03 5.61 0.06
Housing 63 0.15 0.37 3.03 0.22
Basic Services 63 0.90 0.20 1.72 0.42
Durable Asset 63 0.20 0.64 1.94 0.38
Economic Status 63 0.13 0.12 4.70 0.10

Fig. 4 Map of Vietnam

by regions and provinces

13
Targeting Administrative Regions for Multidimensional
Poverty…
Income Education

Health Housing

Basic Services Durable Assets

Economic Status

Fig. 5 EB predictions of province-level random effects for multiple dimensions of poverty with 95% confidence
intervals (Since residuals are assumed to follow a normal distribution with zero mean, the zero line in Fig. 1 repre-
sents the estimated overall mean,βˆk. Black dots represent the estimated province-level residuals, uˆpk, which are the
differences between province j 0 βˆk.)
’s mean of the kth dimension of poverty, and the estimated overall mean, 0

13
 A. T. Q. Pham et
al.
Table 9 Likelihood ratio tests
for multilevel models versus Income LR 33 2 = 579.03 Pr > 3 2
=
single- 0.0000
level models Non-income Education LR 3 3 2 = 460.78 Pr > 3 2 =
3 0.0000
Health LR 3 3 2 = 368.38 Pr > 3 2 =
3
0.0000
Housing LR 3 32 = 3566.86 Pr > 3 2 =
0.0000
32
Basic services LR 3 = 1587.89 Pr > 3 2 =
0.0000

Table 10 Variances from multilevel models

Dimensions Levels
Household Commune District Province

Income 0.044 0.004 0.002 0.004

Non-income Education 0.018 0.001 0.000 0.001
Health 0.038 0.004 0.000 0.002
Housing 0.034 0.006 0.007 0.025
Basic 0.032 0.010 0.007 0.008
Asset 0.061 0.004 0.004 0.002
Economic 0.051 0.008 0.009 0.002

13
 Targeting Administrative Regions for Multidimensional
Poverty…
Table 11 Measures of poverty in Vietnam by provinces
13

Provinces No. of obser- Income Education Health Housing Basic Asset Economic
vations

Red River delta Ha Noi City 420 0.227 0.149 0.219 0.038 0.123 0.213 0.031
Quang Ninh 147 0.277 0.225 0.129 0.085 0.190 0.183 0.074
Vinh Phuc 135 0.369 0.190 0.239 0.066 0.263 0.065 0.060
Bac Ninh 138 0.242 0.207 0.271 0.045 0.184 0.110 0.078
Hai Duong 183 0.369 0.154 0.267 0.027 0.205 0.112 0.050
Hai Phong City 186 0.272 0.161 0.212 0.076 0.196 0.188 0.054
Hung Yen 147 0.425 0.137 0.240 0.056 0.285 0.138 0.056
Thai Binh 189 0.478 0.137 0.098 0.026 0.284 0.172 0.017
Ha Nam 123 0.413 0.163 0.218 0.022 0.417 0.118 0.104
Nam Dinh 195 0.361 0.199 0.233 0.039 0.198 0.087 0.039
Ninh Binh 129 0.454 0.194 0.188 0.033 0.304 0.117 0.047
Midlands and northern mountainous areas Ha Giang 105 0.665 0.501 0.138 0.657 0.681 0.218 0.433
Cao Bang 102 0.631 0.351 0.178 0.394 0.668 0.140 0.211
Bac Can 102 0.684 0.340 0.186 0.462 0.601 0.123 0.249
Tuyen Quang 114 0.539 0.302 0.214 0.441 0.490 0.066 0.156
Lao Cai 102 0.578 0.368 0.118 0.555 0.489 0.150 0.104
Dien Bien 102 0.621 0.490 0.120 0.634 0.635 0.134 0.323
Lai Chau 102 0.707 0.575 0.196 0.691 0.728 0.235 0.370
Son La 126 0.635 0.366 0.143 0.503 0.597 0.175 0.151
Yen Bai 114 0.555 0.348 0.196 0.622 0.488 0.125 0.243
Hoa Binh 114 0.535 0.268 0.184 0.246 0.512 0.060 0.247
Thai Nguyen 147 0.407 0.241 0.199 0.196 0.346 0.076 0.130

al.
A. T. Q. Pham et
Lang Son 108 0.600 0.311 0.195 0.319 0.582 0.105 0.200
Bac Giang 168 0.406 0.263 0.222 0.102 0.302 0.079 0.114
Phu Tho 156 0.482 0.194 0.195 0.222 0.356 0.090 0.065
 Table 11 (continued)

Poverty…
Targeting Administrative Regions for Multidimensional
Provinces No. of obser- Income Education Health Housing Basic Asset Economic
vations

Northern and coastal central region Thanh Hoa 246 0.538 0.240 0.171 0.148 0.381 0.104 0.123
Nghe An 225 0.520 0.232 0.169 0.162 0.425 0.159 0.136
Ha Tinh 150 0.531 0.180 0.196 0.130 0.396 0.074 0.115
Quang Binh 120 0.500 0.227 0.214 0.135 0.396 0.117 0.096
Quang Tri 102 0.494 0.262 0.295 0.206 0.294 0.128 0.155
Thua Thien - Hue 135 0.415 0.347 0.072 0.179 0.138 0.159 0.089
Da Nang City 123 0.240 0.157 0.067 0.167 0.032 0.247 0.051
Quang Nam 159 0.529 0.285 0.082 0.208 0.241 0.169 0.044
Quang Ngai 147 0.534 0.324 0.127 0.151 0.375 0.102 0.123
Binh Dinh 162 0.395 0.265 0.108 0.150 0.305 0.142 0.056
Phu Yen 123 0.453 0.333 0.159 0.118 0.317 0.071 0.103
Khanh Hoa 138 0.375 0.281 0.125 0.175 0.157 0.200 0.075
Ninh Thuan 102 0.461 0.447 0.145 0.285 0.139 0.177 0.114
Binh Thuan 135 0.385 0.450 0.163 0.260 0.236 0.167 0.093
Central highlands Kon Tum 102 0.508 0.420 0.071 0.266 0.482 0.170 0.103
Gia Lai 141 0.480 0.406 0.106 0.419 0.457 0.107 0.227
Dac Lac 165 0.450 0.371 0.131 0.312 0.479 0.132 0.214
Dac Nong 102 0.523 0.391 0.132 0.453 0.427 0.032 0.071
Lam Dong 141 0.389 0.305 0.145 0.353 0.294 0.177 0.056
13
 Table 11 (continued)
13

Provinces No. of Income Education Health Housing Basic Asset Economic

obser-
vations
Southeastern area Binh Phuoc 120 0.390 0.369 0.138 0.408 0.362 0.090 0.166
Tay Ninh 135 0.367 0.415 0.179 0.364 0.245 0.123 0.060
Binh Duong 177 0.168 0.312 0.121 0.275 0.076 0.117 0.033
Dong Nai 207 0.246 0.274 0.121 0.309 0.179 0.210 0.055
Ba Ria - Vung Tau 132 0.297 0.323 0.152 0.283 0.095 0.184 0.068
Ho Chi Minh City 351 0.143 0.177 0.120 0.229 0.032 0.256 0.079
Mekong delta Long An 156 0.395 0.408 0.139 0.403 0.354 0.158 0.066
Tien Giang 171 0.348 0.391 0.161 0.379 0.267 0.170 0.109
Ben Tre 153 0.463 0.398 0.156 0.398 0.483 0.194 0.038
Tra Vinh 129 0.489 0.477 0.086 0.536 0.448 0.151 0.086
Vinh Long 135 0.428 0.387 0.142 0.403 0.368 0.150 0.072
Dong Thap 168 0.473 0.461 0.129 0.501 0.308 0.156 0.113
An Giang 186 0.394 0.521 0.152 0.540 0.155 0.155 0.073
Kien Giang 162 0.429 0.522 0.152 0.604 0.497 0.265 0.090
Can Tho 138 0.370 0.425 0.163 0.437 0.224 0.212 0.086
Hau Giang 111 0.449 0.465 0.156 0.546 0.440 0.223 0.061
Soc Trang 144 0.496 0.488 0.117 0.602 0.395 0.274 0.071
Bac Lieu 114 0.417 0.501 0.182 0.567 0.360 0.214 0.070
Ca Mau 138 0.485 0.459 0.175 0.635 0.364 0.263 0.099
Overall mean 0.420 0.309 0.163 0.281 0.318 0.155 0.103

al.
A. T. Q. Pham et
 Table 12 Ranking of provinces based on poverty measures (ascending from least to most deprived) – single and multilevel analyses

Poverty…
Targeting Administrative Regions for Multidimensional
Name of provinces Income Education Health Housing Basic Asset Economic

SLA a
MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA

Red River delta Ha Noi City 3 12 3 44 56 53 5 9 5 12 54 54 2 27

Quang Ninh 8 10 15 23 18 22 11 13 12 21 47 40 26 40
Vinh Phuc 14 40 10 17 59 48 9 10 20 29 3 12 16 31
Bac Ninh 5 4 14 18 62 56 7 6 11 8 16 19 28 34
Hai Duong 13 30 4 2 61 61 3 3 15 13 17 15 8 26
Hai Phong City 7 9 6 8 52 54 10 11 13 19 49 45 10 30
Hung Yen 28 46 1 5 60 57 8 7 23 26 28 24 13 32
Thai Binh 38 43 2 3 6 12 2 1 22 20 43 5 1 13
Ha Nam 25 38 7 7 55 46 1 2 43 51 21 27 39 44
Nam Dinh 10 24 13 9 58 55 6 4 14 7 9 8 5 14
Ninh Binh 34 50 11 4 45 41 4 5 27 25 19 17 7 24
Midlands and northern mountainous areas Ha Giang 61 51 59 45 22 39 62 55 62 60 56 57 63 62
Cao Bang 59 49 37 19 40 59 40 33 61 61 29 26 55 47
Bac Can 62 58 34 14 44 58 49 39 59 55 23 23 60 54
Tuyen Quang 54 48 27 11 54 52 47 40 54 43 4 20 52 37
Lao Cai 57 54 39 37 10 23 55 52 53 52 31 42 40 10
Dien Bien 58 39 58 52 12 30 60 56 60 59 27 34 61 60
Lai Chau 63 53 63 56 48 44 63 54 63 62 58 58 62 61
Son La 60 55 38 63 25 63 51 45 58 63 44 63 50 63
Yen Bai 55 42 36 24 50 51 59 58 52 45 24 16 58 58
Hoa Binh 52 36 23 1 43 49 28 19 56 47 2 2 59 55
Thai Nguyen 24 41 19 20 51 45 23 23 31 41 7 10 48 46
Lang Son 57 37 29 10 47 60 36 24 57 54 14 4 54 48
13

Bac Giang 22 21 21 12 57 50 12 8 26 18 8 11 44 42
Phu Tho 40 52 12 6 46 47 26 27 33 31 11 9 18 17
 Table 12 (continued)
13

Name of provinces Income Education Health Housing Basic Asset Economic

SLAa MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA

Northern and coastal central region Thanh Hoa 53 62 18 15 38 40 16 18 39 37 13 14 46 50

Nghe An 47 59 17 13 37 38 19 20 44 46 38 32 49 51
Ha Tinh 50 63 9 16 49 43 14 15 41 50 6 1 45 43
Quang Binh 45 61 16 21 53 42 15 17 42 53 18 25 35 45
Quang Tri 43 60 20 35 63 62 24 26 25 23 25 35 51 52
Thua Thien - Hue 26 47 35 48 3 1 22 21 6 5 37 31 32 35
Da Nang City 4 26 5 32 1 2 20 30 1 10 59 53 9 33
Quang Nam 49 56 26 22 4 3 25 29 18 16 40 18 6 5
Quang Ngai 51 57 32 29 16 13 18 16 38 39 12 36 47 41
Binh Dinh 21 19 22 27 8 4 17 14 28 32 30 29 12 21
Phu Yen 33 35 33 33 33 28 13 12 30 27 5 13 37 38
Khanh Hoa 16 13 25 25 15 16 21 25 9 9 51 47 27 19
Ninh Thuan 35 23 51 40 26 25 33 28 7 1 46 51 43 25
Binh Thuan 17 11 52 47 36 33 29 31 17 17 39 39 34 28
Central highlands Kon Tum 46 44 49 41 2 5 30 22 50 49 41 49 38 6
Gia Lai 39 18 46 50 7 10 45 42 48 44 15 6 57 57
Dac Lac 32 32 41 46 19 20 35 32 49 57 26 38 56 59
Dac Nong 48 28 44 34 20 21 48 49 45 34 1 3 23 2

al.
A. T. Q. Pham et
Lam Dong 18 17 28 30 27 31 37 38 24 30 45 41 14 16
 Table 12 (continued)

Poverty…
Targeting Administrative Regions for Multidimensional
Name of provinces Income Education Health Housing Basic Asset Economic

SLAa MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA SLA MLA

Southeastern area Binh Phuoc 19 7 40 31 21 27 44 44 35 35 10 7 53 56

Tay Ninh 12 8 48 54 41 37 38 41 19 14 22 22 15 15
Binh Duong 2 1 30 26 13 8 31 35 3 2 20 21 3 3
Dong Nai 6 2 24 28 14 9 34 37 10 11 52 46 11 20
Ba Ria-Vung Tau 9 6 31 38 28 24 32 36 4 4 48 44 20 22
Ho Chi Minh City 1 3 8 36 11 19 27 34 2 6 60 60 29 53
Mekong delta Long An 22 16 47 39 23 11 42 47 32 38 36 48 19 23
Tien Giang 10 5 43 43 34 26 39 43 21 15 42 50 41 49
Ben Tre 36 27 45 49 32 35 41 46 51 58 50 43 4 7
Tra Vinh 42 20 56 55 5 6 52 51 47 42 33 33 30 4
Vinh Long 29 31 42 42 24 14 43 48 37 40 32 30 24 12
Dong Thap 37 45 54 58 17 7 50 53 29 22 35 37 42 39
An Giang 20 15 61 61 29 18 53 60 8 3 34 28 25 8
Kien Giang 30 14 62 62 30 29 58 62 55 56 62 61 33 11
Can Tho 15 22 50 53 35 32 46 50 16 24 53 52 31 36
Hau Giang 31 33 55 59 31 15 54 61 46 48 57 56 17 9
Soc Trang 44 25 57 51 9 17 57 59 40 33 63 59 22 1
Bac Lieu 27 13 60 60 42 36 56 57 34 28 55 55 21 18
Ca Mau 41 34 53 57 39 34 61 63 36 36 61 62 36 29
a
SLA: Ranking based on single-level analysis of poverty. MLA: Ranking based on multilevel analysis of poverty
13
Targeting Administrative Regions for Multidimensional
Poverty…
Reference
s
ADB (Asian Development Bank). (2018). Key Indicators for Asia and the Pacific 2018 available at www.
[Link]/statistics.
Alkire, S. (2007). Choosing dimensions: The capability approach and multidimensional poverty. Chronic
Poverty Research Centre Working Paper, (88).
Alkire, S., Roche, J. M., Ballon, P., Foster, J., Santos, M. E., & Seth, S. (2015). Multidimensional poverty
measurement and analysis. Oxford : Oxford University Press.
Alkire, S., & Santos, M. E. (2010). Acute multidimensional poverty: A new index for developing coun-
tries (No. HDRP-2010-11). Human Development Report Office (HDRO), United Nations Develop-
ment Programme (UNDP).
Alkire, S., & Santos, M. E. (2013). A multidimensional approach: Poverty measurement & beyond. Social
Indicators Research, 112(2), 239–257.
Alkire, S., & Santos, M. E. (2014). Measuring acute poverty in the developing world: Robustness and scope
of the multidimensional poverty index. World Development, 59, 251–274.
Amarasinghe, U., Samad, M., & Anputhas, M. (2005). Locating the poor: Spatially disaggregated poverty
maps for Sri Lanka (Vol. 96). Colombo: IWMI.
Arouri, M., Nguyen, C., & Youssef, A. B. (2015). Natural disasters, household welfare, and resilience: Evi-
dence from rural Vietnam. World Development, 70, 59–77.
Arpino, B., & Aassve, A. (2014). The role of villages in households’ poverty exit: Evidence from a multi-
level model for rural Vietnam. Quality & Quantity, 48(4), 2175–2189.
Asselin, L. M. (2009). Analysis of multidimensional poverty: Theory and case studies (Vol. 7). Berlin:
Springer.
Azeem, M. M., Mugera, A. W., & Schilizzi, S. (2016). Poverty and vulnerability in the Punjab, Pakistan: A
multilevel analysis. Journal of Asian Economics, 44, 57–72.
Baker, J. L., & Grosh, M. E. (1994). Poverty reduction through geographic targeting: How well does it
work? World Development, 22(7), 983–995.
Battiston, D., Cruces, G., Lopez-Calva, L. F., Lugo, M. A., & Santos, M. E. (2013). Income and beyond:
Multidimensional poverty in six Latin American countries. Social Indicators Research, 112(2), 291–
314.
Baulch, B., & Dat, V. H. (2010). Poverty dynamics in Vietnam, 2002–2006. World Bank. [Link]
[Link]/curated/en/309531468267611416/pdf/642780WP0P107600Box0361535B0PUBLIC0
.pdf.
Baulch, B., & Masset, E. (2003). Do monetary and nonmonetary indicators tell the same story about
chronic poverty? A study of Vietnam in the 1990s. World Development, 31(3), 441–453.
Baulch, B., & Minot, N. (2002). The spatial distribution of poverty in Vietnam and the potential for target-
ing. Washington DC: The World Bank.
Betti, G., Gagliardi, F., Lemmi, A., & Verma, V. (2015). Comparative measures of multidimensional depri-
vation in the European Union. Empirical Economics, 49(3), 1071–1100.
Betti, G., & Verma, V. (1999). Measuring the degree of poverty in a dynamic and comparative context: A
multi-dimensional approach using fuzzy set theory. In Proceedings, ICCS-VI (Vol. 11, p. 289).
Betti, G., & Verma, V. (2008). Fuzzy measures of the incidence of relative poverty and deprivation: A
multi-dimensional perspective. Statistical Methods and Applications, 17(2), 225–250.
Brady, D., Fullerton, A. S., & Cross, J. M. (2009). Putting poverty in political context: A multi-level analy-
sis of adult poverty across 18 affluent democracies. Social Forces, 88(1), 271–299.
Brady, D., & Kall, D. (2008). Nearly universal, but somewhat distinct: The feminization of poverty in afflu-
ent Western democracies, 1969–2000. Social Science Research, 37(3), 976–1007.
Browne, W. J., Subramanian, S. V., Jones, K., & Goldstein, H. (2005). Variance partitioning in multilevel
logistic models that exhibit overdispersion. Journal of the Royal Statistical Society: Series A
(Statistics in Society), 168(3), 599–613.
Bui, A. T., Dungey, M., Nguyen, C. V., & Pham, T. P. (2014). The impact of natural disasters on household
income, expenditure, poverty and inequality: Evidence from Vietnam. Applied Economics, 46(15),
1751–1766.
Carletto, C., Dabalen, A., & Moubayed, A. (2007). Constructing and using poverty maps for policy
making: The experience in Albania. In More than a pretty picture: Using poverty maps to design
better policies and interventions. Washington DC: The World Bank.
Cerioli, A., & Zani, S. (1990). A fuzzy approach to the measurement of poverty. In C. Dagum and M.
Zenga (Eds.) Income and wealth distribution, inequality and poverty: Proceedings of the second
international conference on income distribution by size: Generation, distribution, measurement
and applications.

13
 A. T. Q. Pham et
al.
Held at the University of Pavia, Italy, September 28–30, 1989 (pp. 272–284). Berlin, Heidelberg:
Springer.
Chaudhry, P., & Ruysschaert, G. (2007). Climate change and human development in Viet Nam: A case study
for the human development report 2007/2008. Hanoi: UNDP.
Cheli, B., & Lemmi, A. (1995). A “Totally” fuzzy and relative approach to the multidimensional analysis of
poverty. Economic Notes-SIENA, 24, 115–134.
Chen, K. M., & Wang, T. M. (2015). Determinants of poverty status in Taiwan: A multilevel approach.
Social Indicators Research, 123(2), 371–389.
Christiaensen, L., Lanjouw, P., Luoto, J., & Stifel, D. (2012). Small area estimation-based prediction meth-
ods to track poverty: Validation and applications. The Journal of Economic Inequality, 10(2), 267–297.
Conway, T. (2001). Using government data to target activities to poor communes and monitor poverty
reduction: A review of options for the Cao Bang-Bac Kan Rural Development Project. Hanoi: Com-
mission of the European Communities.
Cuong, N. V. (2009). Updating poverty maps without panel data: Evidence from Vietnam. Asian Economic
Journal, 23(4), 397–418.
Dasgupta, S., Deichmann, U., Meisner, C., & Wheeler, D. (2005). Where is the poverty–environment nexus?
Evidence from Cambodia, Lao PDR, and Vietnam. World Development, 33(4), 617–638.
Demombynes, G., Elbers, C., Lanjouw, J., Lanjouw, P., Mistiaen, J., & Özler, B. (2002). Producing an
improved geographic profile of poverty: Methodology and evidence from three developing coun-
tries (No. 2002/39). WIDER Discussion Papers//World Institute for Development Economics
(UNU-WIDER).
Elbers, C., Fujii, T., Lanjouw, P., Yin, W., & Özler, B. (2004). Poverty alleviation through geographic
targeting: How much does disaggregation help?. Washington DC: The World Bank.
Elbers, C., Lanjouw, J. O., & Lanjouw, P. (2002). Micro-level estimation of welfare (Vol. 2911). Wash-
ington DC: World Bank Publications.
Elbers, C., Lanjouw, J. O., & Lanjouw, P. (2003). Micro-level estimation of poverty and inequality.
Econometrica, 71(1), 355–364.
Fiadzo, E. D., Houston, J. E., & Godwin, D. D. (2001). Estimating housing quality for poverty and
development policy analysis: CWIQ in Ghana. Social Indicators Research, 53(2), 137–162.
Ghosh, M., & Rao, J. N. K. (1994). Small area estimation: An appraisal. Statistical Science, 9(1), 55–76.
Glewwe, P., Agrawal, N., & Dollar, D. (Eds.). (2004). Economic growth, poverty, and household welfare in
Vietnam. Washington DC: World Bank Publications.
Gloede, O., Menkhoff, L., & Waibel, H. (2015). Shocks, individual risk attitude, and vulnerability to pov-
erty among rural households in Thailand and Vietnam. World Development, 71, 54–78.
Goldstein, H. (2011). Multilevel statistical models (Vol. 922). Hoboken: Wiley.
Goldstein, H., Browne, W., & Rasbash, J. (2002). Partitioning variation in multilevel models. Understand-
ing Statistics: Statistical Issues in Psychology, Education, and the Social Sciences, 1(4), 223–231.
Goldstein, H., & Healy, M. J. (1995). The graphical presentation of a collection of means. Journal of the
Royal Statistical Society. Series A (Statistics in Society), 158, 175–177.
Gräb, J., & Grimm, M. (2008). Spatial inequalities explained: Evidence from Burkina Faso.
GSO (General Statistics Office of Vietnam). (2014). Statistical Yearbook of Vietnam 2014. Statistical Pub-
lishing House.
GSO (General Statistics Office of Vietnam). (2015). Statistical Yearbook of Vietnam 2015. Statistical Pub-
lishing House.
GSO (General Statistics Office of Vietnam). (2017). Statistical Yearbook of Vietnam 2017. Statistical Pub-
lishing House.
Haughton, D., & Nguyen, P. (2010). Multilevel models and inequality in Vietnam. Journal of Data Science,
8(2), 289–306.
Heltberg, R., & Tarp, F. (2002). Agricultural supply response and poverty in Mozambique. Food policy,
27(2), 103–124.
Hentschel, J., Lanjouw, J. O., Lanjouw, P., & Poggi, J. (2000). Combining census and survey data to trace
the spatial dimensions of poverty: A case study of Ecuador. The World Bank Economic Review, 14(1),
147–165.
Huang, Y., & Magnoli Bocchi, A. (Eds.). (2008). Reshaping economic geography in East Asia. Washington
DC: The World Bank.
Imai, K. S., Gaiha, R., & Thapa, G. (2015). Does non-farm sector employment reduce rural poverty and vul-
nerability? Evidence from Vietnam and India. Journal of Asian Economics, 36, 47–61.
Kim, K. S., Lee, Y., & Lee, Y. J. (2010). A multilevel analysis of factors related to poverty in welfare states.
Social Indicators Research, 99(3), 391–404.

13
Targeting Administrative Regions for Multidimensional
Poverty…
Kim, R., Mohanty, S. K., & Subramanian, S. V. (2016). Multilevel geographies of poverty in India.
World Development, 87, 349–359.
Klasen, S., Grimm, M., & McKay, A. (2007). Determinants of pro-poor growth: Analytical issues and
findings from country cases. Houndmills: Palgrave Macmillan.
Klasen, S., Lechtenfeld, T., & Povel, F. (2015). A feminization of vulnerability? Female headship, pov-
erty, and vulnerability in Thailand and Vietnam. World Development, 71, 36–53.
Lanjouw, P., Marra, M., & Nguyen, C. (2017). Vietnam’s evolving poverty index map: patterns and
implications for policy. Social Indicators Research, 133(1), 93–118.
Mahadevan, R., & Hoang, V. N. (2016). The nexus between poverty and deprivation in Vietnam. Journal
of Policy Modeling, 38(2), 290–303.
Makdissi, P., & Wodon, Q. (2004). Fuzzy targeting indices and orderings. Bulletin of Economic
Research, 56(1), 41–51.
Martinetti, E. C. (2000). A multidimensional assessment of well-being based on Sen’s functioning
approach. Rivista internazionale di scienze sociali, 108, 207–239.
Mihai, M., Ţiţan, E., & Manea, D. (2015). Education and poverty. Procedia Economics and Finance, 32,
855–860.
Minot, N. (2000). Generating disaggregated poverty maps: An application to Vietnam. World Develop-
ment, 28(2), 319–331.
Minot, N., & Baulch, B. (2004). The spatial distribution of poverty in Vietnam and the potential for tar-
geting (Vol. 842, p. 229). Washington DC: World Bank Publications.
Nguyen, C. V., & Tran, A. (2014). Poverty identification: Practice and policy implications in Vietnam.
Asian-Pacific Economic Literature, 28(1), 116–136.
Nguyen, C. V., Tran, T. Q., & Van VU, H. (2017). Ethnic Minorities in Northern Mountains of Vietnam:
Employment. Poverty and Income. Social Indicators Research, 134(1), 93–115.
Paul, G., Gragnolati, M., & Zaman, H. (2002). Who gained from Vietnam’s boom in the 1990s. Eco-
nomic Development and Cultural Change, 50(4), 773–792.
Pham, A. T. Q., & Mukhopadhaya, P. (2018). Measurement of poverty in multiple dimensions: The case
of Vietnam. Social Indicators Research, 138(3), 953–990.
Rambo, A. T., & Lê, T. C. (1996). Development trends in Vietnam’s northern mountain region. Hono-
lulu: East-West Center, Program on Environment.
Raudenbush, S. W., & Bryk, A. S. (2002). Hierarchical linear models: Applications and data analysis
methods (Vol. 1). Thousand Oaks: Sage.
Ravallion, M. (2011). On multidimensional indices of poverty. Washington DC: The World Bank.
Roelen, K., Gassmann, F., & de Neubourg, C. (2010). Child poverty in Vietnam: Providing insights
using a country-specific and multidimensional model. Social Indicators Research, 98(1), 129–145.
Roelen, K., Gassmann, F., & de Neubourg, C. (2012). False positives or hidden dimensions: What can
monetary and multidimensional measurement tell us about child poverty in Vietnam? International
Journal of Social Welfare, 21(4), 393–407.
Sahn, D. E., & Stifel, D. C. (2000). Poverty comparisons over time and across countries in Africa. World
Development, 28(12), 2123–2155.
Santos, M. E., Villatoro, P., Mancero, X., & Gerstenfeld, P. (2015). A multidimensional poverty index for
Latin America (No. ophiwp079). Queen Elizabeth House, University of Oxford.
Sen, A. K. (1992). Inequality reexamined. New York, Cambridge: Harvard University Press.
Shimeles, A., & Verdier-Chouchane, A. (2016). The key role of education in reducing poverty in South
Sudan. African Development Review, 28(S2), 162–176.
Stifel, D., & Christiaensen, L. (2007). Tracking poverty over time in the absence of comparable consump-
tion data. The World Bank Economic Review, 21(2), 317–341.
Tran, V. Q., Alkire, S., & Klasen, S. (2015). Static and dynamic disparities between monetary and mul-
tidimensional poverty measurement: Evidence from Vietnam. In Measurement of poverty, depriva-
tion, and economic mobility, pp. 249–281. Emerald Group Publishing Limited.
Tranmer, M., & Steel, D. G. (2001). Ignoring a level in a multilevel model: Evidence from UK census
data. Environment and Planning A, 33(5), 941–948.
Van der Weide, R., Cuong, N. V., & Truong, T. N. (2010). Poverty and inequality maps for rural Vietnam:
An application of small area estimation. Washington DC: The World Bank.
Van Doorslaer, E., O’Donnell, O., Rannan-Eliya, R. P., Somanathan, A., Adhikari, S. R., Garg, C. C., et al.
(2006). Effect of payments for health care on poverty estimates in 11 countries in Asia: An analysis of
household survey data. The Lancet, 368(9544), 1357–1364.
Van Minh, H., Phuong, N. T. K., Saksena, P., James, C. D., & Xu, K. (2013). Financial burden of
household out-of-pocket health expenditure in Viet Nam: Findings from the National Living Standard
Survey 2002–2010. Social Science and Medicine, 96, 258–263.

13
 A. T. Q. Pham et
al.
Van Tien, T., Phuong, H. T., Mathauer, I., & Phuong, N. T. K. (2011). A health financing review of Viet-
nam. World Health Organization. [Link]
[Link].
Verma, V., & Betti, G. (2002). Longitudinal measures of income poverty and life-style deprivation. Diparti-
mento di scienze statistiche: Università di Padova.
Walle, D. V. D. (2002). The Static and Dynamic Incidence of Vietnam’s Public Safety Net, World Bank
Policy Research, Working Paper, no. WPS 2791. The World Bank.
Walsh, J. E. (1947). Concerning the effect of intra class correlation on certain significance tests. The Annals
of Mathematical Statistics, 18(1), 88–96.
Yang, J., & Mukhopadhaya, P. (2017). Disparities in the level of poverty in China: evidence from China
family panel studies 2010. Social Indicators Research, 132(1), 411–450.
Yu, J. (2013). Multidimensional poverty in China: Findings based on the CHNS. Social Indicators
Research, 112(2), 315–336.

Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps
and institutional affiliations.

13