Can Weight of Evidence Method Improve Poverty Targeting?

Duong, Be Thanh; Sa iye , O khan; Zelle , Man ed
A icle — Published Ve sion
Can Weigh o E idence Me hod Imp o e Po e y
Ta ge ing?
Social Indica o s Resea ch
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Duong, Be Thanh; Sa iye , O khan; Zelle , Man ed (2025) : Can Weigh o
E idence Me hod Imp o e Po e y Ta ge ing?, Social Indica o s Resea ch, ISSN 1573-0921, Sp inge
Ne he lands, Do d ech , Vol. 178, Iss. 1, pp. 305-333,
h ps://doi.o g/10.1007/s11205-025-03576-z
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323668
S anda d-Nu zungsbedingungen:
Die Dokumen e au EconS o dü en zu eigenen wissenscha lichen
Zwecken und zum P i a geb auch gespeiche und kopie we den.
Sie dü en die Dokumen e nich ü ö en liche ode komme zielle
Zwecke e iel äl igen, ö en lich auss ellen, ö en lich zugänglich
machen, e eiben ode ande wei ig nu zen.
So e n die Ve asse die Dokumen e un e Open-Con en -Lizenzen
(insbesonde e CC-Lizenzen) zu Ve ügung ges ell haben soll en,
gel en abweichend on diesen Nu zungsbedingungen die in de do
genann en Lizenz gewäh en Nu zungs ech e.
Te ms o use:
Documen s in EconS o may be sa ed and copied o you pe sonal
and schola ly pu poses.
You a e no o copy documen s o public o comme cial pu poses, o
exhibi he documen s publicly, o make hem publicly a ailable on he
in e ne , o o dis ibu e o o he wise use he documen s in public.
I he documen s ha e been made a ailable unde an Open Con en
Licence (especially C ea i e Commons Licences), you may exe cise
u he usage igh s as speci ied in he indica ed licence.
h p://c ea i ecommons.o g/licenses/by/4.0/
Vol.:(0123456789)
Social Indica o s Resea ch (2025) 178:305–333
h ps://doi.o g/10.1007/s11205-025-03576-z
ORIGINAL RESEARCH
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
BeThanhDuong1,2 · O khanSa iye 1 · Man edZelle 1
Accep ed: 7 Ma ch 2025 / Published online: 6 Ap il 2025
© The Au ho (s) 2025
Abs ac
This s udy explo es he po en ial o adap ing a WOE (Logi ) me hod, a combina ion
o echniques such as weigh o e idence, in o ma ion alue, logi eg ession, and a
sco e-scaling app oach based on doubling he odds, o po e y a ge ing. To e i y he
e ec i eness and accu acy o his me hod, he s udy applies i o de elop a ge ing ools
based on in e na ional and Vie namese na ional po e y lines, and compa es hei accu acy
o commonly used po e y- a ge ing ools such as he Simple Po e y Sco eca d, P oxy
Means Tes , and Po e y Assessmen Tool. The s udy uses co e age a e as a c i e ion o
compa e he accu acy o a ge ing ools in iden i ying he poo a household and indi idual
le els. The esul s indica e ha a ge ing ools cons uc ed using he WOE (Logi )
me hod ou pe o m hose de eloped using o he me hods. Depending on he po e y
line, his me hod imp o es po e y iden i ica ion accu acy by 1.9–5.9 pe cen age poin s
o households and 1.5–3.3 pe cen age poin s o indi iduals compa ed o o he me hods.
The e o e, his s udy con ibu es an addi ional me hod o cons uc ing a ge ing ools wi h
high accu acy alongside exis ing app oaches.
Keywo ds Po e y a ge ing ool· P oxy means es · Simple po e y sco eca d· Po e y
assessmen ool· C edi isk sco eca d· Po e y a ge ing
1 In oduc ion
To iden i y impo e ished people eligible o ce ain po e y educ ion p og ams o ecei e
cash o in-kind ans e s aimed a imp o ing hei well-being, many coun ies p e e low-
cos bu e ec i e po e y assessmen and a ge ing ools such as P oxy Means Tes (PMT),
Po e y Assessmen Tool (PAT), o Simple Po e y Sco eca d (Sco eca d) (E ang e al.,
2022; Sch eine , 2023; Zelle e al., 2006). These a e ques ionnai es wi h speci ic answe s
* Be Thanh Duong
[email p o ec ed]; [email p o ec ed]
O khan Sa iye
o.sa iye[email p o ec ed]
Man ed Zelle
[email p o ec ed]
1 Hans-Ru henbe g-Ins i u e, Uni e si y o Hohenheim, S u ga , Ge many
2 Facul y o Ag icul u e andRu al De elopmen , Kien Giang Uni e si y, KienGiang, Vie nam
306
B.T.Duong e al.
and ela ed sco es o each ques ion, enabling au ho i ies o quickly p edic whe he
households a e abo e o below a gi en po e y line h ough s aigh o wa d and e i iable
ques ions. Fo b e i y, in his s udy, we e e o hese ools collec i ely as a ge ing ools.
When using a ge ing ools o sco ing, households wi h o al sco es below a ce ain
benchma k a e classi ied as poo , and ice e sa. This anspa ency makes use s and he
public mo e likely o accep and us he accu acy o a ge ing ools in iden i ying he poo
(A anasio e al., 2023; Lei e & Geo ge, 2014; Sch eine e al., 2014).
To c ea e anspa en a ge ing ools, a iables mus be ans o med in o ca ego ical
a iables, as each ques ion in a a ge ing ool equi es speci ic answe s wi h assigned
sco es. The e o e, ans o ming o iginal a iables in o ca ego ical ones is a undamen al
s ep in he cons uc ion p ocess o a ge ing ools. Values wi hin a a iable can be g ouped
in a ious ways, esul ing in nume ous new ca ego ical a iables om a single o iginal
a iable. Ca ego iza ion o a iables wi h wide- anging alues is e en mo e challenging,
as wi hou a speci ic c i e ion, i is di icul o de e mine he bes ca ego y g oupings. Each
new ca ego ical a iable can a ec he p edic abili y o a ge ing ools di e en ly. Al hough
he impac o a single a iable may be mino , ine ec i e g oupings ac oss many a iables
can p e en a ge ing ools om achie ing op imal p edic abili y. Addi ionally, he way
ca ego ies a e g ouped o a a iable can in luence s a is ical alues o p edic ion models,
such as c-s a is ic alues, adjus ed R-squa ed, o Akaike In o ma ion C i e ion. Typically,
he bes indica o s o a ge ing ools a e selec ed h ough a s epwise p ocess, in which
he ac ional change in hese s a is ical alues se es as one o he key selec ion c i e ia
(Ohlenbu g e al., 2022; Sch eine , 2023; Zelle e al., 2005). Consequen ly, how ca ego ies
a e g ouped wi hin a a iable can also a ec he indica o selec ion o a ge ing ools,
ul ima ely a ec ing hei p edic abili y. Howe e , hese aspec s emain unde explo ed, as
mos exis ing s udies p ima ily ocus on inding op imal p edic ion algo i hms o enhance
he accu acy o a ge ing ools. These s udies o en compa e a ious p edic i e modeling
echniques, anging om adi ional s a is ical me hods such as linea p obabili y models,
logis ic eg ession, o dina y leas squa es (OLS), and quan ile eg ession (Linh & Baulch,
2011; Sch eine , 2015; Skou ias e al., 2020; Zelle e al., 2006) o mo e ad anced machine
lea ning algo i hms (Alsha kawi e al., 2021; Kambuya, 2017; McB ide & Nichols, 2015).
Ne e heless, no adi ional eg ession model consis en ly ou pe o ms o he s ac oss
di e en con ex s (E ang e al., 2022; Sch eine , 2015). Fu he mo e, he accu acy gains
om using sophis ica ed machine lea ning algo i hms o e simple eg ession echniques
a e o en ma ginal (Hand, 2006).
Weigh o E idence (WOE) and In o ma ion Value (IV) echniques a e o en used
o add ess hese binning issues in he de elopmen o c edi isk sco eca ds (Ande son,
2007; Siddiqi, 2016). The WOE echnique helps assess he s eng h o he ela ionship
be ween g oups wi hin a a iable, known as bins, and a gi en dependen a iable (Lund,
2016). TheIV, on he o he hand, is a use ul measu e o e alua ing and compa ing he
p edic i e powe o a iables, and i can se e as a benchma k o selec op imal ca ego y
g oupings o indica o s (Siddiqi, 2006). Inco po a ing hese echniques acili a es he
a iable ans o ma ion p ocess, helping c ea e mono onic a iables wi h high p edic i e
powe . Mo eo e , applying he WOE echnique in a iable ans o ma ions has been shown
o imp o e he pe o mance o p edic i e modeling in c edi sco ing (Chen e al., 2020;
Pomazano , 2023; Sha ma, 2011).
Despi e he p o en e ec i eness o WOE and IV echniques in c edi isk assessmen ,
hei applica ion in po e y iden i ica ion emains unde explo ed. The e o e, his s udy
aims o examine he easibili y and e icacy o applying hese echniques o po e y
iden i ica ion. In c edi sco ing, logis ic eg ession and he sco e-scaling app oach
307
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
o doubling he odds a e o en used along wi h WOE and IV echniques o cons uc
sco eca ds ha p edic he de aul p obabili ies o cus ome s (Ande son, 2007; Siddiqi,
2016). Hence, we apply a combina ion o hese app oaches o iden i ying he poo . Fo
b e i y, we e e o his combina ion o app oaches as he WOE (Logi ) me hod in his
s udy, which includes WOE, IV, Logi , and he sco e-scaling app oach o doubling he
odds. Ta ge ing ools de eloped using his me hod a e called WOE ools. To e i y he
accu acy o WOE ools, we compa e hem o o he commonly used a ge ing ools such
as PMT, PAT, and Sco eca d. Fo cla i y and dis inc ion, we e e o he me hods used o
cons uc hese a ge ing ools as he PMT (OLS) me hod, PMT (Quan ile) me hod, and
Sco eca d (Logi ) me hod, espec i ely. The co e age a e, o po e y inclusion a e, is a
commonly used c i e ion o de ine he accu acy and benchma k sco e o a ge ing ools
(E ang e al., 2022; IRIS Cen e , 2005; Sch eine , 2023). The e o e, we apply his c i e ion
o measu e and compa e he accu acy o a ge ing ools while holding he p edic ed po e y
a e.
Based on an in ensi e li e a u e e iew, o he bes o ou knowledge, his s udy
is he i s o apply he WOE (Logi ) me hod o cons uc a ge ing ools. By doing so,
i in oduces ano he app oach o po e y iden i ica ion alongside exis ing me hods. This
me hod could help op imize inpu a iables o de elop high-accu acy a ge ing ools. In
his s udy, we apply he ou a o emen ioned me hods o de elop a ge ing ools based on
wo po e y lines: Vie nam’s na ional po e y line and he in e na ional po e y line. Ou
ocus is exclusi ely on building ools o iden i y he poo in he e hnic mino i ies, which
ha e he highes po e y a e in Vie nam. Al hough e hnic mino i ies1 comp ise only 14%
o Vie nam’s popula ion, hey cons i u e hal o he coun y’s poo . Thus, his g oup is o
special conce n o he Vie namese go e nmen (CEMA, 2020; CEMA e al., 2015). The
accu acy o a ge ing ools will be compa ed no only a he household le el bu also a
he indi idual le el, as e ec i e a ge ing ools should a ge households wi h mo e poo
membe s a he han hose wi h ewe membe s.
The emaining sec ions o he pape a e o ganized as ollows: Sec .2 p esen s he da a
sou ce and me hodology; Sec .3 p esen s he s udy’s esul s and discussion; and Sec .4
summa izes he majo indings.
2 Da a andMe hodology
2.1 Da a
This s udy uses da a om Vie nam Household Li ing S anda d Su ey in 2018 (VHLSS
2018), conduc ed by he Gene al S a is ics O ice o Vie nam. This o icial da ase is
use ul o de eloping a ge ing ools because i includes necessa y in o ma ion such as
household demog aphics, homes ead cha ac e is ics (e.g., ype o house o oile ), du able
goods, educa ion, employmen , and income (Gene al S a is ics O ice, 2019). The p ima y
ocus o his s udy is he e hnic mino i y communi y, which has he highes po e y a e in
Vie nam. The e o e, we analyze da a om 7,863 e hnic mino i y households.
The VHLSS 2018 da ase con ains o e 600 indica o s co e ing a ious aspec s
o households, including demog aphics, housing, educa ion, employmen , and asse s.
1 Vie nam has 53 e hnic mino i ies and one e hnic majo i y (Kinh).
308
B.T.Duong e al.
Fi s , we emo e indica o s wi h mo e han 10% missing alues.2 We hen selec 129
s aigh o wa d and e i iableindica o s ha exhibi a s ong co ela ion wi h household
income, expendi u e, o po e y s a us. A e iden i ying candida e a iables, we spli he
da ase using s a i ied andom sampling in o a calib a ion sample (70%) and a alida ion
sample (30%). We use he calib a ion sample o cons uc a ge ing ools, and he alida ion
sample o e alua e hei accu acy.
2.2 Po e y line
The s udy e alua es he accu acy o ou me hods o cons uc ing a ge ing ools based on
na ional and in e na ional po e y lines. Howe e , he Vie namese go e nmen has di e en
na ional po e y lines o u al and u ban a eas. Since 90% o he e hnic mino i y popula ion
and 95% o poo e hnic mino i ies eside in u al a eas (CEMA, 2020; Gene al S a is ics O ice,
2019), his s udy uses only he u al po e y line o all e hnic mino i y communi ies (Table1).
2.3 Me hodology
2.3.1 Gene al S eps inDe eloping Po e y Ta ge ing Tools
To e i y he easibili y and e ec i eness o he WOE (Logi ) me hod, we compa e i wi h
h ee widely used me hods: Sco eca d (Logi ), PMT (OLS), and PMT (Quan ile). Each
me hod is used o de elop wo dis inc a ge ing ools based on na ional and in e na ional
po e y lines. To ensu e a ai compa ison ac oss all me hods, we limi he numbe o
indica o s in each a ge ing ool o en. Table2 p esen s a gene al s ep-by-s ep amewo k
o cons uc ing a ge ing ools.
S ep 1: G ouping esponses o indica o s
Each ques ion in a ge ing ools mus ha e a speci ic answe wi h a designa ed sco e.
This equi emen ensu es anspa ency and helps enume a o s accu a ely iden i y he
poo in he ield. To mee his equi emen , inpu a iables o cons uc ing a ge ing ools
should be ans o med o ca ego ical a iables. Nume ic a iables can be ans o med in o
many new ca ego ical a iables, as he e a e a ious ways o g oup ca ego ies. Mo eo e ,
c ea ing ca ego ies o a iables wi h a wide ange o alues is challenging, as he e a e
coun less ways o o m ca ego y g oupings. Howe e , exis ing s udies ha e no p o ided
de ailed guidance on how o e ec i ely c ea e hese ca ego ical a iables. The e o e, we
gene a e new ca ego ies o a iables based on hose in he exis ing ool o he Vie namese
go e nmen . To main ain p ac icali y, we limi he numbe o ca ego ies o each a iable
o se en, aligning wi h he Vie namese go e nmen ’s ool. Nex , we conduc a c oss-
abula ion analysis be ween hese new ca ego ical a iables and a gi en po e y s a us o
assess he ela ionship be ween he ca ego ies and po e y. I he ca ego ies do no align
wi h he gi en po e y s a us, we eg oup hem acco dingly o ensu e hey make sense and
ep esen a leas 5% o he sample.
Fo example, in Table3, we can ca ego ize household size in o six g oups: “1 membe ”,
“2 membe s”, “3 membe s”, “4 membe s”, “5 membe s”, “6 membe s”, and “7 membe s
o mo e”. By c ea ing a c oss- abula ion be ween his a iable and he na ional po e y
2 Rega ding a iables include missing alues wi h less han 10%, we eplace hei missing alues wi h hei
mean o mode alues.

309
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
Table 1 Po e y a es o he Vie namese e hnic mino i y communi y in e ms o income in 2018. Sou ce: Au ho s’ compu a ions based on he VHLSS 2018 da a
a The po e y s a us is de ined based on income o expendi u e, so he po e y a e in Table1 is di e en om he po e y a e published by he Vie namese go e nmen ,
which is based on a mul i-dimensional po e y s a us
b This is based on he PPP con e sion ac o o 7,891.2 om he Wo ld Bank in 2018 o con e Vie nam Dong (VND) o USD
Household po e y s a usa based on income and expendi u e Poo Non-poo All
Na ional po e y line (Income/pe son/day ≤ 2.92 USD in pu chasing powe pa i y)bN 1,293
(16.4%)
6,570
(83.6%)
7,863
(100%)
In e na ional po e y line (Expendi u e/pe son/day ≤ 1.9 USD in pu chasing powe pa i y) N 1,054
(13.4%)
6,809
(86.6%)
7,863
(100%)
310
B.T.Duong e al.
Table 2 S eps o cons uc ing po e y a ge ing ools by he me hod
S ep Ac i i y Sco eca d (Logi ) me hod PMT (OLS) me hod PMT (Quan ile) me hod WOE (Logi ) me hod
1 G oup esponses o indica o s Manual ca ego iza ion based on c oss- abula ion wi h po e y s a us a iable WOE & IV echnique
2 Indica o selec ion “MaxC” s epwise “MaxR” s epwise PMT’s indica o s “MaxC” s epwise
3 Reg ession app oach Logi OLS Quan ile Logi
4 Gene a e sco eca ds Sco eca d (Logi ) me hod PMT (OLS) me hod PMT (OLS) me hod WOE (Logi ) me hod
5 Es ima e accu acy Co e age a e in iden i ying poo households and people based on na ional and in e na ional po e y lines
Sou ce o mo e de ails o each me hod Sch eine (2023) Nguyen & T an
(2018)
Zelle e al (2005) Siddiqi (2016)
311
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
s a us, we ind he ollowing po e y a es o each ca ego y: 16.2, 9.6, 8, 14.2, 19.9, 21.8,
and 32.1 pe cen age poin s, espec i ely. Since he i s ca ego y accoun s o only 3.1% o
he popula ion, we me ge i wi h he second ca ego y o o m a new ca ego y: “ < 3 mem-
be s”. This new ca ego y has a po e y a e o 11 pe cen age poin s. Howe e , he po e y
a es o hese ca ego ies a e inconsis en and con adic common sense, as he ca ego y
“ < 3 membe s” has a highe po e y a e han he ca ego y “3 membe s”. To add ess his,
we me ge he new ca ego y “ < 3 membe s” wi h he hi d ca ego y “3 membe s” esul ing
in he ca ego y “ < 4 membe s” wi h a po e y a e o 9.4 pe cen age poin s. Ul ima ely, he
household size a iable is ca ego ized in o i e g oups: “ < 4 membe s”, “4 membe s”, “5
membe s”, “6 membe s”, and “ ≥ 7 membe s”. The po e y a es o hese ca ego ies a e
now consis en and make sense: 9.4, 14.2, 19.9, 21.8, and 32.1 pe cen age poin s.
S ep 2: Selec ing indica o s o he p edic ion model
A e ob aining candida e a iables in S ep 1, his s ep selec s he en bes indica o s
o p edic ion models. Depending on he me hodology, hese indica o s can be selec ed
using he “MaxC” o “MaxR” s epwise app oaches. The “MaxC” app oach e e s o
a o wa d s epwise me hod pe o med manually o selec indica o s based on bo h
s a is ical and non-s a is ical c i e ia. The s a is ical c i e ion is he c-s a is ic, also known
as he concen a ion index (Ra allion, 2007) o he a ea unde he ecei e ope a ing
cha ac e is ic cu e (Wodon, 1997). A highe alue o his c i e ion indica es be e model
pe o mance (Baulch, 2002). Non-s a is ical c i e ia a e ou judgmen abou a ious
aspec s o he indica o s, such as use accep abili y, simplici y, e i iabili y, applicabili y,
a ie y, and so on. Fo mo e de ails abou he “MaxC” o wa d s epwise app oach, please
e e o Sch eine (2023). The “MaxR” app oach is simila o “MaxC” app oach, bu i
uses adjus ed R-squa ed as he s a is ical c i e ion. Fo mo e de ails abou he “MaxR”
app oach, please e e o Zelle e al. (2005).
S ep 3: T aining p edic ion model
In his s ep, we use selec ed indica o s o de elop p edic ion models and ob ain weigh s
o hose indica o s. Depending on he me hodology, he p edic ion models may use
Table 3 An example o manual ca ego iza ion by using c oss- abula ion analysis. Sou ce: Au ho s’ compu-
a ions based on he VHLSS 2018 da a
Ini ial ca ego iza ion
Ca ego y (Numbe o household
membe s)
1 2 3 4 5 6 ≥ 7
Numbe o poo and non-poo (Poo :
Non-poo )
28:145 58:548 78:893 215:1298 201:810 146:524 180:381
Poo a e o each ca ego y (%) 16.2 9.6 8.0 14.2 19.9 21.8 32.1
Second ca ego iza ion
Ca ego y (Numbe o household
membe s)
< 3 3 4 5 6 ≥ 7
Poo a e o each ca ego y (%) 11.0 8.0 14.2 19.9 21.8 32.1
Final ca ego iza ion
Ca ego y (Numbe o household
membe s)
< 4 4 56 ≥ 7
Poo a e o each ca ego y (%) 9.4 14.2 19.9 21.8 32.1
312
B.T.Duong e al.
Logi , OLS, o Quan ile eg ession algo i hms. De ailed in o ma ion abou he eg ession
algo i hms and dependen a iables used will be p o ided in he desc ip ion o each
me hod below.
S ep 4: T ans o ming he p edic ion model’s coe icien s o sco e
Based on he esul s o p edic ion models, he coe icien s o indica o s can be
ans o med in o sco es o o m a ge ing ools in h ee simple s eps. Fi s , we swi ch
e e ence ca ego ies o a iables i hei coe icien s a e nega i e. This ensu es all
coe icien s a e posi i e. Second, we mul iply he coe icien s o he a iables by 100.
Finally, we ound he sco es ob ained. The sum o hese sco es (excluding he in e cep ’s
coe icien ) is he o al sco e o p edic ing he po e y s a uso households. This common
and s aigh o wa d app oach is widely used o c ea e a ge ing ools and is also applied by
he Vie namese go e nmen o hei exis ing ools (Nguyen & Lo, 2016). The Sco eca d
and WOE (Logi ) me hods use di e en app oaches o ans o m coe icien s in o sco es,
which will be explained la e .
S ep 5: Es ima ing he ool’s accu acy
We use co e age a e as a key c i e ion o compa e he accu acy o a ge ing ools. The
co e age a e, also known as he inclusion o he poo o po e y accu acy, is cu en ly used
by he Vie namese go e nmen o assess he accu acy o hei a ge ing ools (IRIS Cen e ,
2005; Nguyen & Lo, 2016). The co e age a e is he pe cen age o households co ec ly
p edic ed as poo by a gi en a ge ing ool, exp essed as a pe cen age o he o al numbe
o obse ed poo households. In addi ion, we also measu e he co e age a e o a ge ing
ools a he pe son le el o compa ison. This means we also calcula e he co e age a e o
a ge ing ools when iden i ying poo indi iduals. This is impo an because iden i ying
poo households wi h la ge amily sizes is mo e bene icial han iden i ying hose wi h
ewe membe s.
We measu e he co e age a e o a ge ing ools when hey pe o m in alida ion
samples. This helps us a oid he o e i ing issue because a ge ing ools ce ainly pe o m
well in calib a ion samples, which a e used o cons uc hem. I also allows us o gauge
he accu acy o a ge ing ools in iden i ying he poo wi hin an unobse ed popula ion.
Addi ionally, we apply a boo s ap echnique o es ima e he s anda d e o o he co e age
and a ge ed a es by calcula ing hose measu es and hei di e ences o he 1,000
epe i ions om he boo s ap sample (Chen & Sch eine , 2009; E on & Tibshi ani, 1994).
This app oach p o ides a eliable assessmen o ools’ pe o mance.
Fo a ai compa ison, we compa e he co e age a es a a cu -o wi h he same
p edic ed po e y a e, which is he a e o he a ge ed popula ion in alida ion samples.
Due o he g anula i y o he sco es, i is challenging o ob ain cu -o sco es wi h he same
p edic ed po e y a e ac oss all a ge ing ools. The e o e, we ix he po e y a e p edic ed
by a speci ic a ge ing ool and use linea in e pola ion o es ima e he co e age a es o he
emaining a ge ing ools.
2.3.2 The Sco eca d (Logi ) Me hod
Sco eca ds a e alsode eloped based on he p inciples o he PMT (OLS) me hod. Sco e-
ca d (Logi ) me hod uses he sco es o p oxy indica o s o cons uc Sco eca ds o iden i-
ying he poo . Sch eine (2010) applies logis ic eg ession o es ima e he weigh s o p oxy
indica o s, and uses his own me hod o con e he indica o s’ weigh s o non-nega i e
319
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
o he na ional WOE ool is 1.9, 5.9, and 4.3 pe cen age poin s highe han hose o he
na ionalSco eca d, PMT, and PAT, espec i ely.
Rega ding he in e na ional po e y line, Table 4 shows ha he in e na ional WOE
ool achie es he highes accu acy among in e na ional a ge ing ools, wi h a co e age
a e o 54.3 pe cen age poin s. In compa ison, he in e na ional Sco eca d and PMT ha e
lowe accu acy, wi h co e age a es o 51.6 and 51.3 pe cen age poin s, espec i ely, while
he in e na ional PAT has he lowes accu acy, wi h a co e age a e o 50.9 pe cen age
poin s. This means ha he co e age a e o he in e na ional WOE ool is 2.7, 3.0, and
3.4 pe cen age poin s highe han hose o he in e na ional Sco eca d, PMT, and PAT,
espec i ely.
To es ima e s anda d e o s o he co e age a es and p edic ed po e y a es o na ional
and in e na ional a ge ing ools, he boo s ap sampling me hod equi es he ac ual cu -
o sco es o hese ools. Howe e , as men ioned ea lie , na ional a ge ing ools do no
ha e exac cu -o sco es ha achie e he same p edic ed po e y a e, and nei he do
in e na ional a ge ing ools. The e o e, o es ima e s anda d e o s, we use cu -o sco es
ha yield p edic ed po e y a es closes o he obse ed po e y a es. The esul s in
Table5 show ha he s anda d e o s o he na ional WOE ool di e only sligh ly om
hose o o he na ional a ge ing ools, and i s p- alue is iden ical o hose o he o he
na ional ools. Simila ly, o he in e na ional po e y line, Table 5 indica es ha he
s anda d e o s o he in e na ional WOE ool a e compa able o hose o o he in e na ional
a ge ing ools, and he p- alues o all ou in e na ional a ge ing ools a e he same.
3.2 The Me hods’ Accu acy inIden i ying Poo People
The accu acy compa ison o he ou me hods in iden i ying poo indi iduals ollows he
same app oach used o iden i ying poo households. We apply he p e iously cons uc ed
a ge ing ools o he co esponding po e y lines and calcula e he co e age a e o
each ool. The key di e ence is ha he na ional and in e na ional po e y lines a e now
measu ed a he indi idual le el (household membe ). As a esul , hese po e y lines a e
highe han hose used in he household-le el analysis.
Table 6 p esen s he co e age and p edic ed po e y a es o he na ional Sco eca d,
which a e 64.1 and 20.8 pe cen age poin s, espec i ely. This means ha when a ge ing
20.8% o he popula ion a he indi idual le el, he na ional Sco eca d co ec ly iden i-
ies 64.1% o hem as poo people. Fo he same a ge ed popula ion sha e, he co e -
age a es o he na ional PMT, PAT, and WOE ools a e 62.2, 62.7, and 65.5 pe cen age
Table 6 The accu acy o a ge ing ools in iden i ying poo people by he po e y line
Uni : (%)
Sco eca d PMT PAT WOE ool
Na ional po e y line (in pe son)
Co e age a e 64.1 62.2 62.7 65.5
P edic ed po e y a e 20.8 20.8 20.8 20.8
In e na ional po e y line (in pe son)
Co e age a e 55.5 56.9 56.2 58.8
P edic ed po e y a e 17.2 17.2 17.2 17.2

320
B.T.Duong e al.
Table 7 The s anda d e o and p- alue o a ge ing ools in iden i ying poo people by he po e y line
Na ional po e y line (in pe son) Uni (%) S d. e o p- alue In e na ional po e y line (in pe son) Uni (%) S d. e o p- alue
Sco eca d Co e age a e 64.1 1.078 0.001 Sco eca d Co e age a e 55.5 1.173 0.001
P edic ed po e y a e 20.8 0.408 0.001 P edic ed po e y a e 17.2 0.354 0.001
PMT Co e age a e 61.2 1.119 0.001 PMT Co e age a e 56.7 1.200 0.001
P edic ed po e y a e 20.3 0.407 0.001 P edic ed po e y a e 17.0 0.356 0.001
PAT Co e age a e 61.2 1.108 0.001 PAT Co e age a e 55.1 1.194 0.001
P edic ed po e y a e 20.1 0.407 0.001 P edic ed po e y a e 16.9 0.359 0.001
WOE ool Co e age a e 64.2 1.082 0.001 WOE ool Co e age a e 58.5 1.168 0.001
P edic ed po e y a e 20.2 0.391 0.001 P edic ed po e y a e 17.0 0.352 0.001
321
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
poin s, espec i ely, as shown in Table6. These esul s indica e ha he na ional WOE ool
achie es he highes accu acy in iden i ying poo indi iduals based on he na ional po e y
line. Speci ically, he co e age a e o he na ional WOE ool is 1.5, 3.3, and 2.8 pe cen age
poin s highe han hose o he na ionalSco eca d, PMT, and PAT, espec i ely.
Rega ding he iden i ica ion o poo indi iduals based on he in e na ional po e y line,
he in e na ional WOE ool also demons a es he highes accu acy among all a ge ing
ools. Speci ically, i s co e age a e exceeds hose o he Sco eca d, PMT, and PAT by 3.2,
1.9, and 2.6 pe cen age poin s, espec i ely.
In iden i ying poo indi iduals, bo h na ional and in e na ional WOE ools exhibi only
mino di e ences in s anda d e o s compa ed o hose o o he a ge ing ools, as shown
in Table7. Addi ionally, he p- alues o he WOE ools a e he same as hose o he o he
a ge ing ools.
3.3 Discussion
The compa isons abo e show ha he WOE (Logi ) me hod ou pe o ms he o he h ee
me hods. Howe e , due o di e ences among he ou me hods and he lack o con olled
condi ions, pinpoin ing he exac easons o his ou pe o mance is challenging.
None heless, we will highligh key di e ences be ween he me hods o examine how hese
a ia ions po en ially in luence accu acy and indica o selec ion in a ge ing ools.
The i s di e ence lies in how esponses o each indica o a e binned. O he me hods
c ea e ca ego ies based on he esul s o c oss- abula ion, aiming o c ea e g oups ha
consis en ly align wi h po e y s a us. In he WOE (Logi ) me hod, binning wi h he WOE
echnique does no s op a ha equi emen , bu aims o maximize IV, he p edic i e powe
o an indica o . This app oach o en esul s in di e en binning pa e ns o a iables wi h
a wide ange o alues, such as “elec ici y bill,” “numbe o poul y,” “dependen a io,”
and so on. This is because wi hou he c i e ion o IV, i is di icul o de e mine he bes
ca ego y combina ions o hese a iables.
Va iables wi h di e en ways o binning can lead o di e ences in selec ing indica o s
because, wi hin he same a iable, di e en ca ego y combina ions will esul in di e en
alues o he c-s a is ic o adjus ed R squa e. These a e essen ial s a is ical c i e ia in he
“MaxC” o “MaxR” o wa d s epwise app oaches, which selec indica o s based on small
imp o emen s in hese c i e ia. As a esul , binning wi h di e en me hods can lead o he
selec ion o di e en indica o s in a ge ing ools, which ul ima ely a ec s accu acy. This
can be seen in compa ing he WOE (Logi ) and Sco eca d (Logi ) me hods: despi e bo h
applying he same “MaxC” s epwise app oach and using Logi models, hei a ge ing
ools di e in one indica o , and wo o h ee indica o s ha e di e en binning pa e ns.
Consequen ly, hei accu acy di e s.
Addi ionally, o he Sco eca d (Logi ) and WOE (Logi ) me hods, he c-s a is ic alue
is also an impo an c i e ion o assessing he accu acy o a ge ing ools. We obse e
ha a iables wi h highe IVs o en4 end o achie e highe c-s a is ic alues. Since he
WOE (Logi ) me hod maximizes he IV o indica o s du ing hebinning p ocess, nume ic
a iables wi h a wide ange o alues binned in his way o en ha e highe c-s a is ic alues
compa ed o hose binned using c oss- abula ion. While his dispa i y is mino , i can
accumula e when a ge ing ools con ain many such indica o s. As a esul , his di e ence
may con ibu e o he accu acy gap be ween Sco eca ds and WOE ools.
4 This end weakens and no always co ec when di e ences in IVs a e e y minimal.
322
B.T.Duong e al.
Ano he ac o ha may con ibu e o di e ences in indica o selec ion and a ge ing
accu acy is he na u e o he a iables used in he eg essions. The WOE (Logi ) me hod
employs nume ic alues ans o med using he WOE echnique, whe eas o he me hods
ely on ca ego ized a iables. Howe e , we canno de ini i ely conclude ha using WOE-
ans o med a iables di ec ly impac s accu acy o indica o selec ion, as he indica o s
used in he ou a ge ing ools di e , po en ially in oducing bias.
An obse a ion we can con i m is ha eg essions using ca ego ized a iables
some imes encoun e issues wi h coe icien mono onici y. The Sco eca d (Logi ) and PMT
(OLS) me hods use ca ego ical a iables, so each ca ego y has i s own coe icien . This can
lead o si ua ions whe e a ca ego ical a iable migh help he model achie e he highes
c-s a is ic o highes adjus ed R-squa e bu canno be selec ed because hei coe icien s a e
non-mono onic.5 In such cases, we do no d op hese a iables; ins ead, we eca ego ize
hem o achie e mono onic coe icien s and e un he s epwise p ocess. Whe he he
Fig. 1 The associa ion be ween he o al sco e and p edic ed ou comes a pe son le el by in e na ional a -
ge ing ools
5 A a iable wi h non-mono onic coe icien s leads o inconsis encies in he sco ing o esponses wi hin a
ques ion.
323
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
e-selec ed indica o emains he same o no , his change o en esul s in a model wi h a
lowe c-s a is ic o adjus ed R-squa e. This issue a ises when selec ing indica o s such as
he numbe o dependen s and numbe o ag icul u al labo e s o he na ional Sco eca d,
and he highes educa ion le el o he in e na ional Sco eca d. The PMT (Quan ile)
me hod aces simila challenges, so o main ain consis en coe icien s ac oss ca ego ies,
we need o selec quan iles ha esul in lowe c-s a is ic alues. In con as , he WOE
(Logi ) me hod does no ace his issue because WOE a iables a e nume ic, and each
a iable has only one coe icien .
The hi d di e ence be ween me hods is he eg ession algo i hms used. Howe e , we do no
conside his ac o as a main eason o he accu acy imp o emen seen in he WOE (Logi )
me hod. Exis ing s udies show ha he Logi model some imes ou pe o ms OLS o Quan ile
eg essions, bu a o he imes, Logi models a e less e ec i e (Linh & Baulch, 2011; Sch eine ,
2015; Zelle e al., 2006). This pa e n is also obse ed in his s udy. Fo ins ance, in iden i y-
ing poo people, he na ional Sco eca d achie es highe accu acy han he na ional PMT and
PAT, while he in e na ional Sco eca d pe o ms wo se han he in e na ional PMT and PAT. In
some cases, bo h he Logi and OLS eg essions yield simila a ge ing accu acy (E ang e al.,
2022). Addi ionally, al hough bo h he WOE (Logi ) and Sco eca d (Logi ) me hods use logis ic
eg ession, WOE ools s ill ou pe o m Sco eca ds.
The inal di e ence be ween he WOE (Logi ) me hod and he o he me hods is he
app oach o ans o ming coe icien s in o sco es. Figu e1 illus a es ha he o al sco es
con e ed by he WOE (Logi ) me hod a e mo e consis en wi h he p edic ed po e y
p obabili ies han hose con e ed by he Sco eca d (Logi ) me hod. This sugges s ha
he Sco eca d (Logi ) me hod may inad e en ly lowe i s accu acy, as many indi iduals
wi h simila o e en highe p edic ed po e y a es a he selec ed cu -o sco e may s ill
be excluded. The magni ude o he o al sco e could be a con ibu ing ac o o his issue.
Howe e , he sco e ans o ma ion app oaches o he PMT (OLS) and PMT (Quan ile)
me hods a e simila , wi h la ge o al sco e magni udes. While he o al sco es in he PMT
(Quan ile) me hod align wi h indi iduals’ expendi u es, he o al sco es in he PMT (OLS)
me hod do no . As a esul , bo h he PMT (OLS) and PMT (Quan ile) me hods gene ally
pe o m wo se han he WOE (Logi ) me hod. Ne e heless, concluding ha he sco e
ans o ma ion app oach o he WOE (Logi ) me hod is supe io o o he s could be biased,
as he a ge ing ools cons uc ed by he ou me hods use di e en indica o s, and he mag-
ni ude o he o al sco es a ies ac oss he me hods.
This discussion highligh s he di e ences be ween me hods and he po en ial ou comes
esul ing om hese di e ences. Based on hese ou comes, we can only conclude ha using
bo h he WOE binning echnique and WOE a iables leads o di e ences in binned and
selec ed a iables compa ed o o he me hods. We canno p o ide p ecise quan i a i e
assessmen s o accu acy imp o emen s o indi idual echniques wi hin he WOE (Logi )
me hod because he necessa y condi ions o a ai assessmen a e no me , and his is
beyond he scope o he s udy. Howe e , his p esen s an in e es ing esea ch gap o explo e
in he u u e.
4 Conclusion
This s udy explo es he easibili y and e icacy o using he WOE (Logi ) me hod o
de elop a ge ing ools. To achie e his, we compa e he accu acy o he WOE (Logi )
me hod wi h ha o o he powe ul and common me hods, such as he Sco eca d (Logi ),
324
B.T.Duong e al.
PMT (OLS), and PMT (Quan ile) me hods. Fo a ai compa ison, we use he co e age a e
o compa e he accu acy be ween a ge ing ools while holding he p edic ed po e y a e.
The esul s indica e ha a ge ing ools cons uc ed using he WOE (Logi ) me hod
achie e highe accu acy han hose de eloped by o he me hods. Depending on he na ional
o in e na ional po e y line, he WOE (Logi ) me hod can imp o e accu acy in iden i ying
poo households by app oxima ely 1.9–5.9 pe cen age poin s o co e age a e compa ed
o o he me hods, while his accu acy imp o emen anges om 1.5 o 3.3 pe cen age
poin s in iden i ying he poo a he pe son le el. This small imp o emen is meaning ul
on a na ional scale. Fo ins ance, acco ding o he upda ed epo by he Commi ee o
E hnic Mino i y A ai s o Vie nam (CEMA, 2020), Vie nam has 745,441 poo households
among e hnic mino i ies. I he Vie namese go e nmen aims o a ge all o hese poo
households, applying he WOE (Logi ) me hod could po en ially inc ease he numbe o
co ec ly iden i ied poo households by 14,163 o 43,981 compa ed o o he me hods.
This s udy in oduces an addi ional app oach o go e nmen s seeking o de elop
accu a e a ge ing ools. While he WOE (Logi ) me hod may seem complex in explana ion,
i is nei he new no di icul o implemen , as i is commonly used in c edi sco ing.
Va ious accessible esou ces, guidelines, and so wa e packages a e a ailable o popula
and powe ul ools such as R, SAS, and Py hon. These packages allowuse s o cons uc
a ge ing ools using he WOE (Logi ) me hod in jus a ew simple s eps.
The i s limi a ion o his s udy is ha i ocuses solely on es ing whe he applying he
WOE (Logi ) me hod—a comp ehensi e app oach o cons uc ing c edi isk sco eca ds,
including echniques such as binning ca ego ies by WOE and IV, using WOE a iables,
applying logis ic eg ession, and using he sco e-scaling app oach o doubling he po e y
odds—imp o es he accu acy o a ge ing ools. The e o e, he s udy can only conclude
ha he applica ion o all hese echniques oge he imp o es he accu acy o WOE ools.
The indi idual con ibu ions o each componen echnique o he accu acy imp o emen
emain unclea , as many condi ions a e no con olled o a ai compa ison, making any
conclusion po en ially biased. Ne e heless, his is a esea ch gap ha should be explo ed
in he u u e i any esea che in ends o apply any single echnique in heWOE (Logi )
me hod o he PMTs.
A second limi a ion is ha he WOE (Logi ) me hod has only been applied o e hnic
mino i y communi ies in Vie nam. Howe e , a ge ing ools de eloped using his me hod
ha e been e ec i e in iden i ying he poo a bo h he household and pe son le els among
he 53 e hnic mino i y g oups, which a y in cha ac e is ics, li ing egions, po e y
a es, and le els o po e y (poo , mode a ely poo , o ex emely poo ) (CEMA, 2020).
The e o e, he WOE (Logi ) me hod should also be applicable o cons uc ing a ge ing
ools o iden i y he poo among he e hnic majo i y, which includes only he Kinh e hnic
g oup, in Vie nam. We ecommend ha he compa a i e analysis o he WOE (Logi )
me hod p esen ed in his pape be epea ed o o he coun ies and da a se s. I esul s a e
simila o hose p esen ed he e, he WOE (Logi ) me hod can be applied o po e y sco ing
in hose coun ies.
Appendix
See Tables8, 9 and 10.

325
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
Table 8 Na ional Sco eca d cons uc ed by he Sco eca d (Logi ) me hod
Indica o Value Poin s Sco e
1. How much did you pay o you elec ici y bill las mon h? A. Less han 50,000 VND 0
B. 50,000 VND—99,000 VND 3
C. 100,000 VND—199,000 VND 7
D. 200,000 VND—349,000 VND 11
E. 350,000 VND o mo e 19
2. How many membe s o he household wo k as non-ag icul u al labo e s (includes side jobs)? A. One o none 0
B. Two 7
C. Th ee 12
D. Fou o mo e 17
3. How many membe s does he household ha e? A. Less han ou 9
B. Fou 5
Fi e 3
Six 2
Se en o mo e 0
4. Was he household included in he an i-poo p og am in he p e ious yea (2017)? A. Yes 0
B. No 5
5. Wha is he o al land a ea owned by he household? A. Less han wo hec a es 0
B. Two hec a es o mo e 6
6. Is anyone in he household wo king away om home? A. Yes 10
B. No 0
7. How many membe s o he household wo k in he ag icul u al sec o as hei main job? A. None 12
B. One 9
A. Two, Th ee o Fou 4
B. Fi e o mo e 0
326
B.T.Duong e al.
Table 8 (con inued)
Indica o Value Poin s Sco e
8. How many membe s o he households a e dependen s? A. None 8
B. One o wo 4
C. Th ee 1
D. Fou o mo e 0
9. Does he household own a cow, bu alo, o ho se? A. No 0
B. Yes 6
10. How many mo o bikes does he household own? A. None 0
B. One 3
C. Two 6
D. Th ee o mo e 9
To al sco e
327
Can Weigh o E idence Me hod Imp o e Po e y Ta ge ing?
Table 9 Na ional WOE ool cons uc ed by he WOE (Logi ) me hod
Indica o Value Poin s Sco e
1. How much did you pay o you elec ici y bill las mon h? A. Less han 40,000 VND 17
B. 40,000 VND—89,000 VND 30
C. 90,000 VND—159,000 VND 43
D. 160,000 VND—239,000 VND 57
E. 240,000 VND o mo e 74
2. How many membe s o he household wo k as non-ag icul u al labo e s (includes side jobs)? A. One o none 12
B. Two 40
C. Th ee 57
D. Fou o mo e 65
3. How many membe s does he household ha e? A. Less han ou 44
B. Fou 37
C. Fi e 31
D. Six 30
E. Se en o mo e 22
4. Was he household included in he an i-poo p og am in he p e ious yea (2017)? A. Yes 24
B. No 45
5. How many poul ies does he household own? A. Less han 30 32
B. 30 o 39 35
C. 40 o 59 53
D. 60 o mo e 74
6. Is anyone in he household wo king away om home? A. Yes 71
B. No 32
7. How many mo o bikes does he household own? A. None 26
B. One 32
C. Two 51
D. Th ee o mo e 64
328
B.T.Duong e al.
Table 9 (con inued)
Indica o Value Poin s Sco e
8. Does he household own a cow, bu alo, o ho se? A. No 33
B. Yes 58
9. How many membe s o he household wo k in he ag icul u al sec o as hei main job? A. None 68
B. One 50
C. Two o h ee 31
D. Fou 24
E. Fi e o mo e 15
10. How many membe s o he households a e dependen s? A. None 53
B. One 40
C. Two 34
D. Th ee 20
E. Fou o mo e 12
To al sco e