What’s behind pro-poor growth? An investigation of its drivers and dynamics

Klasen, S ephan; Kneib, Thomas; Lo Bue, Ma ia C.; P e e, Vincenzo
A icle — Published Ve sion
Wha ’s behind p o-poo g ow h? An in es iga ion o i s
d i e s and dynamics
The Jou nal o Economic Inequali y
P o ided in Coope a ion wi h:
Sp inge Na u e
Sugges ed Ci a ion: Klasen, S ephan; Kneib, Thomas; Lo Bue, Ma ia C.; P e e, Vincenzo (2024) : Wha ’s
behind p o-poo g ow h? An in es iga ion o i s d i e s and dynamics, The Jou nal o Economic
Inequali y, ISSN 1573-8701, Sp inge US, New Yo k, NY, Vol. 23, Iss. 1, pp. 43-69,
h ps://doi.o g/10.1007/s10888-024-09628-7
This Ve sion is a ailable a :
h ps://hdl.handle.ne /10419/323387
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)
The Jou nal o Economic Inequali y (2025) 23:43–69
h ps://doi.o g/10.1007/s10888-024-09628-7
1 3
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s
d i e s anddynamics
S ephanKlasen1· ThomasKneib1· Ma iaC.LoBue2 · VincenzoP e e3
Recei ed: 23 Ma ch 2020 / Accep ed: 28 Ma ch 2024 / Published online: 30 May 2024
© The Au ho (s) 2024
Abs ac
S anda d g ow h incidence cu es desc ibe how g ow h episodes impac on he o e all
income dis ibu ion. Howe e , measu ing he p o-poo ness o he g ow h p ocess is com-
plex due o measu emen e o s, and o he e ec o shocks ha may hi he pe cen iles
o he income dis ibu ion in di e en ways. The e o e, s anda d g ow h incidence cu es
may mis ep esen he ue g ow h p ocess and i s dis ibu i e impac . Relying on a non-
anonymous app oach, we compa e ac ual g ow h episodes a each pe cen ile o he ini ial
pe sonalized dis ibu ion wi h coun e ac ual mobili y p o iles which ule ou he p es-
ence o shocks. We conside Indonesia in 2000–2007 and 2007–2014, wo g ow h spells in
which he e was subs an ial, signi ican upwa d mobili y among he ini ially poo e , a size-
able pa o which canno be explained by unobse ed indi idual endowmen s o s anda d
socio-economic a ibu es. The di e ence be ween ac ual and expec ed g ow h is ela ed, in
he ea ly 2000s, o he economy-wide ans o ma ions, which cha ac e ized he ea ly yea s
o he pos -Suha o e a. Howe e , in he mo e ecen yea s, i can be la gely a ibu ed o
indi idual eco e y om p e ious nega i e losses and high ulne abili y and eac i i y o
shocks o he poo .
Keywo ds Indonesia· Shocks· P o-poo ness· Mobili y
JEL Classi ica ion D31· I3· O12
1 In oduc ion
In he con ex o he Sus ainable De elopmen Goals, o p omo e coun ies’ de elopmen
and o aise he li ing s anda d o people a he bo om o he income dis ibu ion, he
Wo ld Bank G oup enewed i s s a egy by de ining wo goals: (i) ending he sha e o peo-
ple li ing in ex eme and ch onic po e y by 2030; and (ii) p omo ing sha ed p ospe i y
* Ma ia C. Lo Bue
ma iaca [email p o ec ed]
1 Uni e si y o Gö ingen, Gö ingen, Ge many
2 Uni e si y o T ies e, T ies e, I aly
3 Uni e si y o Pale mo, Pale mo, I aly
44
S.Klasen e al.
1 3
(Basu 2013). The i s goal deals wi h he educ ion o less han 3 pe cen o he sha e o
people li ing below he Wo ld Bank’s po e y line o US $1.25 pe day. Empi ical e i-
dence has documen ed ha economic g ow h ep esen s he main ool o achie e absolu e
po e y educ ion (Dolla and K aay 2002; Dolla e al. 2016).
Howe e , Basu (2013) has a gued ha he obse ed g ow h a es a e no enough o
e adica e po e y and a mo e equal dis ibu ion o g ow h bene i s is desi able. This is he
spi i o he “sha ed p ospe i y” goal, which calls o g ea e income g ow h o he poo -
es 40 pe cen o people. Wi h he de ini ion o hese wo goals, he Wo ld Bank G oup
ecognizes ha g ow h no only should be good o he poo bu also has o be “p o-poo ”.
The e o e, analysing he e ec s o income g ow h on po e y educ ion and assessing he
p o-poo ness o g ow h a e no only exe cises o academic esea che s bu also c ucial
challenges o policymake s.
P io o his, indeed, he e has been an in ense deba e among esea che s on he de i-
ni ion and measu emen o he p o-poo ness o g ow h, wi h wo al e na i e de ini ions
eme ging: absolu e e sus ela i e. The o me de ines g ow h as p o-poo when ei he he
absolu e income gain o he poo is la ge han he a e age income gain (s ong absolu e
p o-poo g ow h) o he poo expe ience a posi i e g ow h a e (weak absolu e p o-poo
g ow h). The ela i e de ini ion, ins ead, calls o he g ow h a e o he poo es pa o
he dis ibu ion o be la ge han he a e age g ow h a e. Klasen (2008) highligh s me i s
and weaknesses o each de ini ion, a guing ha he absolu e (weak) de ini ion is use ul o
measu e he “ a e” o p o-poo ness, while he ela i e de ini ion is pa icula ly sui able in
assessing he “s a e” o p o-poo ness.1
This conside a ion seems consis en wi h he wo k by Ra allion and Chen (2003), who
in oduced he g ow h incidence cu e (GIC). The GIC plo s he pe cen ile-speci ic income
g ow h a e be ween wo poin s in ime. By compa ing he a e age g ow h a e expe i-
enced by he indi iduals anked in he poo es pe cen iles wi h he a e age g ow h a e
o he o e all dis ibu ion, a g ow h p ocess can be de ined as p o-poo in absolu e ( ela-
i e) e ms i he o me is posi i e (la ge han he la e ). In his ega d, he e idence
documen ed by Dolla e al. (2016) sugges s p o-poo g ow h only in absolu e e ms, since
he poo es 40 pe cen ha e expe ienced a posi i e income g ow h a e, wi hou inc eas-
ing hei income sha e. Mo eo e , Ra allion and Chen (2003) and subsequen li e a u e
(Duclos 2009; Essama-Nssah 2005; K aay 2006; Son 2004) measu e he deg ee o p o-
poo ness o g ow h in an anonymous way, by ocusing only on he income change expe-
ienced by each pe cen ile o he dis ibu ion wi hou conside ing he iden i y o indi idu-
als loca ed on each pe cen ile. The e o e, wo al e na i e g ow h p ocesses gene a ing he
same income dis ibu ion as he p e ious pe iod a e conside ed equi alen , i espec i ely
o whe he indi iduals’ posi ions wi hin he income dis ibu ions a e unchanged o com-
ple ely eshu led. This coun e in ui i e esul makes he anonymous app oach unsa is-
ac o y when he in e empo al e alua ion o he g ow h p ocesses aims a assessing he
mobili y expe ienced by indi iduals. By emo ing he anonymi y assump ion, G imm
(2007) and Bou guignon (2011) p opose he “non-anonymous” e sion o he GIC, which
is ob ained by keeping cons an indi iduals’ posi ions in he ini ial income dis ibu ion.
Thus, he non-anonymous GIC plo s he income g ow h a e o all indi iduals as a unc ion
o hei quan ile in he ini ial dis ibu ion. A g owing s and o ecen li e a u e adop s his
1 See Essama-Nssah and Lambe (2009) o a e iew o he li e a u e de eloping indices o p o-poo ness
o g ow h. See also Duclos (2009) o a o mal cha ac e iza ion o absolu e and ela i e p o-poo ness.
45
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
non-anonymous app oach o e alua e p o-poo g ow h (Jenkins and Van Ke m 2016; Lo
Bue and Palmisano 2020; Palmisano 2018; Palmisano and Pe agine 2015).2
The indi idual income g ow h a e is also used by ano he ele an s and o li e a u e
aimed a measu ing income mobili y. No ably, he income mobili y p o iles p oposed by
Van Ke m (2009) ep esen an al e na i e o maliza ion o he non-anonymous GICs.3 How-
e e , while he analysis o mobili y is qui e de eloped (Fields 2008b; Fields and Ok 1999),
he in es iga ion o he e ec o mobili y on he p o-poo ness o g ow h is s ill limi ed.4
Bo h he non-anonymous GIC and mobili y p o iles may o e only a pa ial ep esen a-
ion o he indi idual income g ow h p ocess; his could be he esul o ei he shocks o
measu emen e o s. Fe ei a (2012) aised his issue in he con ex o he anonymous GIC,
p oposing an al e na i e in e p e a ion based on he li e a u e on coun e ac ual dis ibu-
ions (DiNa do e al. 1996; Juhn e al. 1993). Acco ding o Fe ei a (2012), he indi idual
income g ow h a e can be exp essed as he sum o a ious componen s, each measu ing
he impac o a speci ic de e minan , such as changes in wo ke cha ac e is ics o hei
co esponding e u ns. This app oach has been ecen ly applied by Fe ei a e al. (2019),
who es ima e a coun e ac ual GIC o ela e he dis ibu ional impac o economic g ow h
o changes o he s uc u e o he economy. E en Fields e al. (2015), in hei analysis o
ea nings mobili y in A gen ina, Mexico, and Venezuela, ecognize he con ounding ole o
measu emen e o s and ansi o y ea ning shocks o which indi iduals may be subjec ed
in he sho - un. The e o e, hey p opose a amewo k in which indi iduals’ ea nings a e
decomposed as he sum o wo componen s, one associa ed wi h obse able and pe manen
indi idual cha ac e is ics and ano he ha is ela ed o ansi o y ea ning componen s.5
By applying he non-anonymous GIC amewo k, in his pape we compa e ac ual g ow h
episodes a each pe cen ile o he ini ial pe sonalized dis ibu ion wi h a coun e ac ual pa -
e n o p edic ed income dynamics, which ules ou he p esence o shocks and measu emen
e o . Compa ison be ween he obse ed and he coun e ac ual non-anonymous GIC allows
an unde s anding o he ex en o which g ow h-shaped indi idual income ajec o ies ha e
esul ed om unexpec ed changes in he ma ginal e u n o indi idual socio-economic cha -
ac e is ics, which subs an ially changed indi idual ankings in he income dis ibu ion.
Using longi udinal su ey da a om Indonesia, we show ha g ow h has been gene ally
p o-poo o e he pe iod 2000–2014, wi h he incidence o g ow h in he ini ial poo es
quin ile being la ge han expec ed. We apply a double selec i i y model o s a e-depend-
ency o be e unde s and he na u e o hese unp edic ed pe cen ile-speci ic gains, which—
as we ind in his s udy—has e ol ed o e ime. Ou esul s, indeed, sugges s ha while he
economic ans o ma ions o he ea ly 2000s con ibu ed o imp o e he g ow h po en ial
2 Mo e speci ically, Jenkins and Van Ke m (2016) and Palmisano and Pe agine (2015) p opose a wel a e
analysis o he dis ibu i e impac o g ow h. Palmisano (2018) sugges s ha he iden i ica ion o indi idu-
als may be based on he anking in he inal dis ibu ion, and he e o e p o-poo ness is e alua ed by ocus-
ing on he income ajec o ies o indi iduals who become poo . Lo Bue and Palmisano (2020) p opose a
non-anonymous e sion o GIC o e alua e he pa e ns o mobili y expe ienced by he ch onic and ansi-
o y poo , whe e iden i ica ion is based ei he on he ini ial o he inal dis ibu ion.
3 The concep o mobili y is mul idimensional (Fields and Ok 1999), as i embodies ou di e en aspec s,
which a e desc ibed by Jan i and Jenkins (2014): e- anking wi hin he income dis ibu ion, income
g ow h, inequali y educ ion, and unce ain y.
4 In his ega d, excep ions a e he con ibu ions by Bá cena and Can ó (2018) and B esson e al. (2019).
5 By adop ing a wo-s age leas squa es p ocedu e, Fields e al. (2015) i s es ima e he pa o indi iduals’
ea nings associa ed wi h pe manen cha ac e is ics. Then, he p edic ed alues, which ep esen a p oxy o
he ini ial income o indi iduals, a e used in a second eg ession as he explana o y a iable o he indi idu-
als’ income changes.
46
S.Klasen e al.
1 3
a he bo om o he dis ibu ion, in he mo e ecen yea s he di e ence be ween ac ual and
expec ed g ow h me ely esul s om indi iduals’ abili y o eco e om p e ious nega i e
losses, a he han om pu e exogenous posi i e shocks.
The emainde o he pape is s uc u ed as ollows. In Sec ion2 we cha ac e ize he
coun e ac ual indi idual g ow h incidence cu e, in oduce he concep o p o-poo shock
wi hin he indi idual g ow h incidence cu e amewo k, and p esen he s a is ical in e -
ence p ocedu es applied. The empi ical illus a ion is p esen ed in Sec ion3, based on da a
om Indonesia o he pe iod 2000–2014. Sec ion3.2.2 concludes.
2 Se ing
2.1 The coun e ac ual indi idual g ow h incidence cu e
Le
F(
y
−1)
deno e he cumula i e dis ibu ion unc ion (
cd
) o he income obse ed in ime
−1
o a popula ion wi h bounded suppo
(0, ymax)
and ini e mean
𝜇
(F)=
∫ymax
0
ydF(y
)
. The
le in e se con inuous dis ibu ion unc ion o quan ile unc ion, showing he income o an
indi idual occupying posi ion
p −1∈(0,1)
in he dis ibu ion o incomes anked in inc eas-
ing o de , is de ined as
F−1(
p
−1)
∶= in
{
y
−1
∶F
(
y
−1)≥
p
−1}
To simpli y he exposi ion,
in he emainde o he pape we equi alen ly deno e he quan ile unc ion wi h
y −1(
p
−1)
.
Likewise,
F(
y
)
deno es he
cd
o income obse ed in pe iod
, while
y (
p
−1)
deno es he
income expe ienced in ime
by he indi idual anked
p −1
in pe iod
−1
. We ely on he
non-anonymous e sion o he g ow h incidence cu e (deno ed he ea e as indi idual GIC,
IGIC), whe e he iden i y o each indi idual is o malized by hei ank in he ini ial income
dis ibu ion. Following G imm (2007), in such a se ing, he income g ow h a e expe ienced
by he indi iduals loca ed a he
p h
pe cen ile in pe iod
−1
can be o malized as:
and, by in eg a ing he a ea below he IGIC up o he ini ial headcoun index
H −1
, one
ob ains he indi idual a e o p o-poo g ow h (IRPPG) ha is:
which de ines a non-anonymous pa e n o g ow h as p o-poo i i is posi i e (absolu e
de ini ion) o i exceeds he a e age g ow h a e measu ed o e he en i e dis ibu ion ( ela-
i e de ini ion).
A he gene ic ime
, he obse ed income
y
o each indi idual can be de ined as a unc-
ion o a ec o
C
o he cha ac e is ics (such as educa ion, employmen s a us, age, and
household demog aphic cha ac e is ics) and a measu emen e o deno ing he p opensi y
o mis epo income.6 Tha is, he pe cen ile-speci ic income dynamics can be decomposed
(1)
g
(p −1)=
y
(
p −1
)
y
−1(
p
−1)
−
1
(2)
IRPPG
=
1
H −1∫H
−1
0
g
(
p −1
)
dp −
1
6 As ecen ly in es iga ed by Angel e al. (2019), measu emen e o in epo ed income occu s, o exam-
ple, because o he p esence o a social desi abili y bias in su ey esponse o speci ic socio-demog aphic
cha ac e is ics o he esponden s. When pe capi a consump ion expendi u e is used as a p oxy o indi-
idual weal h (as in he empi ical applica ion o his pape ), i s mis epo ing is mos ly ela ed o socio-
demog aphic cha ac e is ics o he esponden s, he ecall bias and he su ey design.

47
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
in o changes ela ed wi h indi idual cha ac e is ics and wi h he indi idual p opensi y o
mis epo income, and a ia ions in he income unc ion ha can be in e p e ed as a ia-
ions o he ma ginal e u ns associa ed wi h indi idual cha ac e is ics.
Le
yj
(
p
−1)
deno e he income o he indi idual anked in he
p h
posi ion in ime
−1
,
p edic ed acco ding o indi idual’s a ibu es a he beginning o pe iod
and he income
o he p e ious pe iod. Then, a coun e ac ual IGIC (CIGIC) can be de i ed o show he
income ha he indi idual loca ed a he gene ic
p h
pe cen ile would expe ience in pe iod
based on a linea p edic ion o he obse ed cha ac e is ics and uling ou he impac o
economic shocks and measu emen e o :
whe e he supe sc ip
j
indica es ha he p edic ed income o each indi idual esul s om
wo al e na i e eg ession models. Speci ically, when
j=FE
he i ed alues a e ex ac ed
om he ollowing panel wo-way eg ession model:
whe e he e m
τ
deno es he yea dummies, he pa ame e s
μi
and
ϑd
a e he indi idual
and he loca ion (e.g., p o ince o esidence) ixed e ec s espec i ely, and
ui,
ep esen s
he esidual e ms. By adding he p edic ion o he indi idual ixed e ec s o he s anda d
i ed alues,
yFE
cap u es he e ec s o changes in obse ed indi idual cha ac e is ics and
o unobse ed ime in a ian cha ac e is ics.7
Al e na i ely, when
j=QR
, we le he e ec o he p edic o s change acco ding o he
indi idual’s ank in he inal pe capi a (p.c. he ea e ) expendi u e dis ibu ion and ex ac
he p edic ed alues om a quan ile eg ession ha models he condi ional quan iles
q
o
he join dis ibu ion o p.c. expendi u e and i s p edic o s as
wi h he e ms
ϑd
and
ui,
deno ing he loca ion ixed e ec s and he e o e m espec i ely.8
To gauge he impac o economic shocks on he indi idual upwa d and downwa d
mobili y pa e ns, we need o compa e he IGIC in Eq.1 wi h he CIGIC in Eq.3. The di -
e en ial be ween hese wo cu es is de ined as
This esidual can be in e p e ed as a b oad measu e o he impac o he shock on he
pe cen ile-speci ic income g ow h a es. I includes, indeed, bo h he e ec o a ia ions o
(3)
g
j
(p −1)=
y
j
(
p −1
)
y
−1(
p
−1)
−
1
(4)
log(
y
i, )
=𝛽
0
+𝛽
1
C
i,
+𝛽
2
log
(
y
i, −1)
+𝜏
+𝜇
i
+𝜗
d
+u
i,
(5)
Qq
log
(
y
i, )
=𝛽
0
(q)+𝛽
1
(q)C
i,
+𝛽
2
(q)log
(
y
i, −1)
+𝜗
d
+u
i,
(6)
Δ
g =g (p −1)−gj
(p −1)=y
(
p −1
)
−y
j
(
p −1
)
y
−1(
p
−1)
7 I is o be no ed ha , as long as he ime dimension
doesn’ end o in ini y, a ixed e ec s es ima ion o
his dynamic linea panel equa ion esul s in a downwa d bias o he coe icien s o in e es (Nickell 1981).
Al e na i e es ima o s, such as he di e ence GMM o he sys em GMM ha e been p oposed o co ec o
his po en ial bias. Howe e , in he con ex o his s udy (which spans a pe iod o 14yea s co e ed in only
h ee wa es) he use o hese es ima o s en ailed he ejec ion o he null hypo hesis o he o e all alidi y
o he ins umen se such ha hey could no be applied.
8 P ecisely, we es ima e ou quan ile eg essions o he
q
=
.20, .40, .60, .80
condi ional quan iles o he
join dis ibu ion o income and i s p edic o s a ime
, ex ac om each model he p edic ed alues (
yQR
)
and assign hem o each indi idual depending on hei posi ion in he p.c. expendi u e dis ibu ion a ime
.
48
S.Klasen e al.
1 3
unobse ed cha ac e is ics and hei associa ed e u ns ha in luence indi idual incomes, and
he e ec o he e o componen , he ole o which is discussed and es ed in Sec ion3.2.
By using Eq.6, we de ine a shock as p o-poo in absolu e e ms i he a e age o he
Δg
up o he po e y line is posi i e, i.e., i he posi i e di e ences be ween he IGIC and he
CIGIC mo e han compensa e he nega i e ones o all pe cen iles up o he po e y line.
Tha is, an absolu e index o p o-poo ness o shocks can be o malized as
A ela i e de ini ion o p o-poo shock equi es ha he di e en ial de ined in Eq.6 is
on a e age la ge o he poo han o he ich. Tha is, le
𝛾
deno e he a e age di e ence
be ween he IGIC and he CIGIC o e he en i e dis ibu ion; hen a shock is p o-poo in
ela i e e ms i
PPS >𝛾
.
2.2 S a e‑dependency, sample o a ion, and eco e y ompas nega i e shocks
When examining he ole ha shocks ha e on he mobili y pa e ns o e subsequen spells
o g ow h, a complemen a y exe cise is o assess he na u e o he shocks hemsel es. Fo
example, one could ask whe he he posi i e shock implied in he se ing cha ac e ized by
IRPPG >0
and
PPS >0
is he ou come o a genuine posi i e shock expe ienced by he
ini ially poo e , o i i is a consequence o a eco e y om pas nega i e shocks.
To answe his ques ion, we need o assess, om an in e - empo al pe spec i e, whe he
he e is some o m o s a e-dependence, o cu en posi i e shocks a e exogenous o pas
nega i e shocks. Gi en he de ini ions in Eqs.3, 5 and 6, an indi idual posi i e shock
(
ps
i, )
can be de ined as a bina y indica o equal o
1
i
y
−
�
yQR
>
0
, and equal o
0
o he wise.
Le ’s s a by assuming ha each indi idual has a la en p opensi y o expe ience a posi i e
shock in ime
, and le ’s se he hypo hesis ha his is a unc ion o a ec o o indi idual and
place-o - esidence cha ac e is ics
(Xi, −1)
, he indi idual’s p opensi y o ha e expe ienced a
nega i e shock in he pas
(
ns∗
i, −1
)
, and o ha e been e ained
(
∗
i,
)
in he sample9:
Following he app oach p oposed by Cappella i and Jenkins (2004; 2008),
ns∗
i, −1
can be
de ined as:
whe e
Z
is a ec o o socio-economic a iables, including pa en al socio-economic back-
g ound. I his p opensi y exceeds some unobse ed alue (which can be se equal o
0
), a
nega i e shock is obse ed:
(7)
PPS
=1
H −1∫H
−1
0
Δg
(
p −1
)
dp −
1
(8)
ps∗
i, =
(
ns∗
i, −1, ∗
i, ,Xi, −1
)
(9)
ns∗
i, −1=𝜂Z
i, −1
+𝜖i
(10)
ns
i, −1=1
[
ns∗
i, −1>0
]
9 A i ion is an issue ha in ou se ing can a ise om ei he sample a i ion o missing pe capi a expend-
i u e (in yea s
−
2,
−
1
and
) and/o in all he o he a iables used o ob ain p edic ed pe capi a expendi-
u e. I sample d opou s a e no andom and indi iduals wi h less a o able cha ac e is ics a e also less
likely o s ay in he sample, ou es ima ed ansi ion p obabili y o a posi i e shock expe ience in ime
will
be biased.
49
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
wi h
nsi, −1
being he obse able bina y indica o , equal o
1
i
y −1
−y
QR
≤0
and o
0
o h-
e wise. The indi idual’s chances o emaining in he sample a e cap u ed by
∗
i,
, he indi-
idual’s la en p opensi y o be e ained, which is a unc ion o a ec o
W
o indi idual
and household cha ac e is ics, including he a iables in
Z
and addi ional co a ia es on he
quali y o he in e iew:
whose obse ed coun e pa is:
Following he p ocedu es ecommended and adop ed in Sa ka e al. (2019), Tunali
(1986) and Vella (1998), we ocus on he eco e y case (i.e.,
nsi, −1=1
and
i, =1
), es i-
ma e Eqs.9 and 11 simul aneously wi h a bi a ia e p obi selec ion model, and ex ac he
ollowing wo selec ion co ec ion e ms:
and
whe e
Φ2(.)
is he bi a ia e s anda d no mal dis ibu ion unc ion,
Φ(.)
and
Φ(.)
a e
he s anda d no mal densi y and cumula i e dis ibu ion unc ions espec i ely, and
𝜌
=co
(
𝜖
i
,𝜀
i)
. To es o he ue exogenei y o posi i e shocks, we include he co ec-
ion e ms
λ�
i, −1
and
�
i, −1
in a linea p obabili y model o eco e y which es ima es he
p obabili y o expe iencing a posi i e shock in ime
, condi ional on nega i e shock expe-
ience in he pas and sample e en ion10:
I
𝛽=𝛾=0
, we can conclude ha i a posi i e shock expe ienced a ime
is obse ed,
his canno be iden i ied as a eco e y om nega i e shock in he pas , no can i be due o
sample e en ion.
3 Empi ical applica ion
This sec ion p esen s he empi ical applica ion o he app oach p oposed in he p e ious sec-
ion. By using he IGICs and he co esponding CIGICs, we i s illus a e he income g ow h
p ocess expe ienced in Indonesia o e he pe iod 2000–2014. We dis inguish wo sub-pe iods,
2000–2007 and 2007–2014. The pa e n o IGICs and CIGICs and he di e ences be ween hem
(11)
∗
i, =𝜁Wi, −1+𝜀i
(12)
i, =1
[
∗
i, >0
]
(13)
𝜆
�
i, −1=𝜙�𝜂Zi, −1�
Φ
�
𝜁Wi, −1−𝜌𝜂Zi, −1
√1−𝜌2
�
Φ
2�
𝜂Z
i, −1
,𝜁W
i, −1
;𝜌
�
(14)
𝜆
��
i, −1=𝜙�𝜁Wi, −1�
Φ
�𝜂
Zi, −1−
𝜌𝜁
Wi, −1
√1−𝜌2
�
Φ
2�
𝜂Z
i, −1
,𝜁W
i, −1
;𝜌
�
(15)
P ob(
ps
i
,
=1
|
ns
i
,
−1=1,
i
,
=1
)
=𝛼X
i
,
−1,+𝛽𝜆
�
i, −1
+𝛾𝜆
��
i, −1
+u
i
,
10 The applica ion o a linea p obabili y model in his con ex acili a es he inclusion o he co ec ion
e ms and he in e p e a ion o hei coe icien s.
50
S.Klasen e al.
1 3
a any pe cen ile a e in o ma i e o he impac o mobili y on he p o-poo ness o g ow h and on
he ole o shocks and measu emen e o in shaping he obse ed mobili y pa e ns. As a gued
abo e, ou in e p e a ion o he CIGICs essen ially hinges on he assump ion ha indi iduals’
p opensi y o unde -/o e - epo hei income is cons an o e ime, i.e., ha measu emen e o
is classical. To alida e his assump ion, we es whe he di e en con e gence pa ame e s p e-
dic ed unde he assump ion o classical measu emen e o a e su icien o yield consis en es i-
ma es, he eby suppo ing he alidi y o ou assump ion o a ime cons an e o e m. Las , we
assess he na u e o he shocks by implemen ing he p ocedu e desc ibed in Sec ion2.2.
3.1 Da a
The empi ical analysis elies on da a om he Indonesia Family Li e Su ey (IFLS), one o he
la ges longi udinal de eloping-coun y su ey da a-se s.11 We use h ee wa es (2000, 2007,
and 2014) and e alua e mobili y pa e ns in e ms o changes in mon hly household p.c. con-
sump ion expendi u e. This is a sui able p oxy o household wellbeing in de eloping coun-
ies, whe e he p ima y sou ce o income is om ag icul u e o in o mal sec o , and i can
also se e as an indica o o pe manen household income (Cu le e al. 1991; Meye and Sulli-
an 2003). To al household expendi u e includes household expendi u e on ood and non- ood
i ems. Da a on ood consump ion includes expendi u e on bo h sel -p oduced and pu chased
p oduc s. Households epo de ailed expenses on a ious ood i ems such as s aples, mea ,
d ied ui s, and ege ables on a weekly basis. Each ood expendi u e was hen mul iplied by
4.3 o ob ain he mon hly expendi u e. Household non- ood expendi u e is epo ed mon hly
and includes expenses on du ables, such as appliances and u ni u e, as well as non-du ables
(less equen ly pu chased i ems), housing cos s, and educa ion expenses.12 I also includes
ans e s in and ou o he household. He e ogenei ies in p ices ac oss ime and space a e
aken in o accoun by using empo al and spa ial de la o s wi h e e ence o Jaka a p ices in
2002.13 To cons uc a CIGIC we use obse ed p.c. expendi u e in yea
−1
and p edic ed p.c.
expendi u e in yea
, which is es ima ed using in o ma ion on p.c. expendi u e in he p e ious
wa e, household socio-demog aphic cha ac e is ics ( esidence and composi ion by age g oup),
and household head cha ac e is ics (gende , age, educa ion, and employmen s a us). Fo he
second pa o ou analysis (i.e., he p ocedu e illus a ed in Sec ion2.2) we also use IFLS2
om 1997 (F ankenbe g and Thomas 2000) o e ie e he a iables ha a e necessa y o es i-
ma e

y
QR
and all he explana o y a iables used in columns 1 and 2 o Table6.
3.2 Resul s
3.2.1 Ac ual andcoun e ac ual indi idual g ow h incidence cu es
Figu e1(a) and Fig. 1(b) illus a e he g ow h p ocess expe ienced in Indonesia o e he
pe iod 2000–2007 and 2007–2014 espec i ely, while Table1 epo s some summa y s a is ics
abou hese p ocesses. The black dashed cu e co esponds o he IGIC, which desc ibes he
11 Fo de ails on he su eys see S auss e al. (2004) o 2000 (IFLS3); S auss e al. (2016) o 2007
(IFLS4); S auss e al. (2016) o 2014 (IFLS5).
12 Educa ion expenses a e epo ed o he pas yea . They include expendi u e on ui ion, uni o m, ans-
po a ion, and boa ding o child en li ing ou side he household.
13 Da a on bo h he consume p ice index (CPI) and egional po e y lines (u ban and u al) come om
Indonesia’s cen al s a is ics agency, Badan Pusa S a is ik (BPS).
57
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
should elimina e a smalle p opo ion
−𝛽(𝛽+1)
o he ini ial expendi u e gap be ween he
second and he hi d wa e. Hence, be ween he i s and he las wa e he o al p opo ional
con e gence pa ame e should be
−𝛽(𝛽+2)
. I , ins ead, we allow o measu emen e o ,
Eq.(18) yields ha
E(
𝜃
3|
𝛽,𝛼
)
=𝛼𝛽(𝛽+1
)
and
E(
𝜃
4|
𝛽,𝛼
)
=𝛼(𝛽+1)
2
−
1
.
Thi d, we ex end Model (23) as ollows:
Unde he hypo hesis o no measu emen e o we expec ha , once we con ol o
y2007
,
he e is a null ela ionship be ween consump ion expendi u e in 2000 and i s change
be ween 2007 and 2014
(
E
(𝜃5|𝛽
,
𝛼
=1
)
=0
)
while he expec ed alue o

𝜃6
will be simply
he con e gence pa ame e
𝛽
. I we allow o measu emen e o , Eq. (18) yields ha
E(
𝜃5
|
𝛽,𝛼
)
=(𝛽+1)
2
(𝛼−1)
𝛼
𝛼
2
(𝛽+1)
2
−1
and
E(
𝜃6
|
𝛽,𝛼
)
=1−𝛼(𝛽+1)+𝛼2𝛽(𝛽+1)
2
𝛼
2
(𝛽+1)
2
−1
implying ha measu emen
e o exace ba es he downwa d bias in he coe icien o
y2007
and causes an upwa d bias
on he coe icien o
y2000
.
Las ly, we eg ess he absolu e change in p.c. consump ion expendi u e be ween he las
wo wa es on i s change be ween he i s wo wa es:
In he absence o measu emen e o , we expec ha households ha expe ienced a
la ge g ow h in p.c. consump ion expendi u e be ween he i s wo wa es will expe ience
a slowe subsequen g ow h, i.e.,
E(
𝜃7
|
𝛽,𝛼=1
)
=
1
2
𝛽 . I he da a is measu ed wi h classi-
cal e o , Eq. (18) yields ha :
E(
𝜃7
|
𝛽,𝛼
)
=−1−𝛼+𝛼𝛽
2
2(1−𝛼−𝛼𝛽)
. Tha is, he nega i e co ela ion
be ween he wo subsequen changes in p.c. consump ion expendi u e should be la ge han
expec ed in he no-measu emen e o case. The es ima ed

𝜃k
coe icien s om Models
21–26 can be used in h ee dis inc app oaches ha complemen each o he o es o he
p esence o classical measu emen e o .
The i s app oach is o di ec ly es he hypo hesis ha consump ion expendi u e is
measu ed wi hou classical e o (i.e.,
𝛼=1
). Es ima es o
𝛼
can be p oduced by using
es ima es o

𝜃1
and

𝜃3
, as:
As epo ed in Table3, he eliabili y s a is ics
𝛼
in ou sample is 0.64, poin ing o he
p esence o classical measu emen e o .
A second app oach o es o he p esence o classical measu emen e o is o use all o
he es ima ed eg ession coe icien s om Models 21–26 and compa e hei ac ual alues
wi h hei expec ed alues unde he wo al e na i e scena ios ha consump ion expendi-
u e is measu ed wi hou o wi h classical measu emen e o . In he i s scena io, he
expec ed alues o hese coe icien s a e es ima ed assuming ha
𝛼=1
and ha he eg es-
sion coe icien

θ1
ep esen s he ue con e gence pa ame e
𝛽
. Al e na i ely, he second
scena io uses he eliabili y s a is ics p oduced by he da a, as in Eq. (27), and
E(

𝜃k
|
𝛽=

𝜃3

𝜃1+1)
.
(25)
Δyi
,2007
−
2014
=𝜇i+𝜃
5
yi
,2000
+𝜃
6
yi
,2007
+ui
(26)
Δyi,2007−2014 =
𝜇
i+
𝜃
7Δyi,2000−2007 +ui
(27)
𝛼
=
(

𝜃1+1
)2

𝜃
1
+
𝜃
3
+
1

58
S.Klasen e al.
1 3
I consump ion expendi u e da a is measu ed wi hou e o and he assumed condi ion
o a i s -o de au o eg essi e p ocess is alid, hen he es ima ed coe icien alues should
only di e due o sampling a ia ion om he p edic ed ones unde he hypo hesis o clas-
sical measu emen e o . Howe e , iola ion o hese assump ions may cause signi ican
di e ences be ween he es ima ed eg ession coe icien s and he p edic ed alues.21
As epo ed in Table3, apa om

𝜃2
, he es ima ed eg ession coe icien s a e e y di -
e en om he alues p edic ed unde he assump ion o no measu emen e o . Fo
ins ance, he e ec o
y2000
on
Δy2007−2014
(i.e., coe icien

𝜃3)
and
Δy2000−2014
(i.e., coe i-
cien

𝜃4)
is espec i ely one- hi d and app oxima ely 80 pe cen o he e ec ha we would
ha e expec ed unde he assump ion o no measu emen e o . Analogously, we obse e
ha measu emen e o exace ba es by app oxima ely 70 pe cen he nega i e co ela ion
(

𝜃7
)
be ween he wo subsequen changes in p.c. consump ion expendi u e and p oduces a
bias in he coe icien s

𝜃5
and

𝜃6
. On he o he hand, he es ima ed

𝜃k
coe icien s a e simila
o he alues epo ed in he hi d line o Table3, which a e p edic ed unde he assump ion
o classical measu emen e o .
A hi d app oach o p o ide e idence in suppo o agains he hypo hesis o classical
measu emen e o is o compa e he con e gence pa ame e s
𝛽
implied by each o he es i-
ma ed

𝜃k
coe icien s unde he wo al e na i e scena ios o no measu emen e o
(𝛼=1)
and ha da a is measu ed wi h e o , i.e.,
E(

𝜃k
|
𝛽,𝛼=𝛼
)
. We obse e, in he ou h and
i h aw o Table3, ha he de i ed es ima es o
𝛽
unde his la e hypo hesis lie wi hin a
ela i ely na ow ange, whe eas hose ob ained unde he assump ion o no measu emen
e o do no . This esul indica es ha ou es ima es ob ained unde he assump ion o clas-
sical measu emen e o a e enough o p oduce consis en es ima es, p o iding he e o e
also suppo o he absence o non-classical measu emen e o .
In e es ingly, his inding aligns wi h ou esul s om a alida ion exe cise p oposed
by Fields e  al. (2003) o es i he ac ual expendi u e dynamics simply esul om
mean- e e ing non-classical measu emen e o , gene a ing a spu ious ela ion be ween
he base-yea epo ed expendi u e and he associa ed change. The es conside s he
a io o he minimum amoun o a iance o s ochas ic measu emen e o ela i e o a -
iance o ue income ha would be equi ed o o e u n he obse ed pa e n o con e -
gence. I his a io is la ge enough o exceed a c i ical h eshold, he downwa d pa e n
o ou es ima ed IGICs can be e alua ed as obus agains he hypo hesis o eg ession o
he mean. The es , which is conduc ed o di e en combina ions o he se ial co ela-
ion coe icien s and o he co ela ion be ween base-yea expendi u e and measu emen
e o , is epo ed in Table4. Resul s sugges ha he es ima ed nega i e slope o he
IGIC is obus agains non-classical measu emen e o in bo h pe iods, wi h he a ios
la gely exceeding he minimum c i ical h eshold o 0.3 ac oss mos o he combina ions
o he se ial co ela ion coe icien and o he co ela ion be ween base-yea expendi u e
and measu emen e o .22
21 Such disc epancies can also occu i he o he unde lying assump ions a e no alid. Fo ins ance, he
assump ion o a cons an slope migh be p oblema ic o e a ela i ely long ime span as he one conside ed
in his analysis. Ye , as epo ed in TableA1 in he Appendix, he coe icien s on ini ial p.c. consump ion
expendi u e ha e a simila magni ude (i.e., -0.46 and -0.45) in he wo eg essions wi h dependen a iables
de ined by changes in p.c. consump ion expendi u e be ween he i s and second wa e and be ween he
second and hi d wa e.
22 By elying on wo alida ion s udies based on U.S. da a, Fields e al. (2003) assume ha a c edible ange
o he minimum c i ical h eshold o his a io is equal o abou 0.1 o 0.3.
59
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
Table 3 Reg ession coe icien s and implied pa ame e alues
The ull eg essions esul s on he es ima ion o he coe icien s in he i s aw a e shown in Table A1 in he Appendix. Signi icance le els:
∗p<0.10;
∗∗ p<0.05;
∗∗∗ p<0.01.
+ No a ailable gi en ha
E(𝜃
5
|𝛽
,
𝛼
=1
)
=
0
.
Sou ce: Au ho s’ es ima ions based on IFLS da a
𝜃k
𝜃1
𝜃2
𝜃3
𝜃4
𝜃5
𝜃6
𝜃7
Es ima ed alues o 
𝜃k
-0.459***
(0.007)
-0.448***
(0.007)
-0.086***
(0.008)
-0.546***
(0.007)
-0.221***
(0.008)
0.570***
(0.008)
-0.392***
(0.007)
P edic ed alues o 
𝜃k
unde he
assump ion o no measu emen
e o 
(𝛽=−0.459, 𝛼=1)
-0.459 -0.459 -0.248 -0.707 0.000 -0.459 -0.229
P edic ed alues o 
𝜃k
unde he assump-
ion o classical measu emen
e o 
(𝛽=−0.158, 𝛼=0.643)
-0.459 -0.459 -0.086 -0.545 0.229 -0.583 -0.406
Values o 
𝛽
implied by
𝜃k
(i 𝛼=1)
-0.459 -0.448 -0.095 -0.326 na+-0.570 -0.784
Values o 
𝛽
implied by
𝜃k
(i 𝛼=0.643)
-0.159 -0.142 -0.159 -0.160 -0.170 CN -0.208
60
S.Klasen e al.
1 3
3.2.3 The na u e o  heshocks
As implied in ou esul s so a , in bo h pe iods expendi u e g ow h was gene ally p o-poo .
The nega i ely sloped IGIC ma ches wi h he expec a ions on wha he ela i e gains a
each pe cen ile should be, gi en indi idual socio-economic a ibu es and he e u ns asso-
cia ed wi h hem. Ne e heless, a sizeable po ion o his p og essi e pa e n canno be
ully accoun ed o by his, as he ac ual g ow h a es o he poo a e signi ican ly la ge
han he p edic ed ones. We need, he e o e, o unde s and i his “unexpec ed” posi i e
g ow h o he poo esul ed om e en s ha do no ela e o indi idual exposu e o nega-
i e shocks in he pas (e.g., changes in he labou ma ke ha inc eased he e u ns o edu-
ca ion), o i he unp edic ed income dynamics simply e lec indi iduals’ eco e y om
pas nega i e shocks, due o example o imp o emen s in hei abili y o cope wi h nega-
i e shocks in he pas , o simply he dissipa ion o a pas nega i e shock. To shed ligh on
his ques ion, we conside o each o he wo g ow h spells (2000–2007 and 2007–2014)
he p opo ion o indi iduals ha , a he end o each pe iod expe ienced a posi i e shock
(
y
−�yQR
>0
)
, condi ional on e en ion and on obse ing, a he beginning o he pe iod, a
nega i e income shock
(
y −1−yQR
−1
≤
0
)
. These indi iduals amoun o abou 24 pe cen o
he obse a ions e ained in he panel and o abou 13 pe cen o he en i e sample (see
Table5).
A i ion a ime
a ises om ei he sample a i ion o missing pe capi a expendi u e
and/o in all he o he a iables used o ob ain p edic ed pe capi a expendi u e. Because
indi idual shock expe ience is measu ed based on he household-le el expendi u e a i-
able, he co a ia es used in he double selec i i y eg ession model a e also measu ed
a he household le el. P ecisely, he co a ia es e e o he household head and his/he
spouse (age, sex, employmen s a us, educa ion), and o he household i sel (se e al a i-
ables summa izing household composi ion and pa en al socio-economic backg ound). The
s anda d e o s a e boo s apped and es ima ed o be obus o he e oskedas ici y and a bi-
a y se ial co ela ion among obse a ions in he same p o ince.
Table 4 Ra io o measu emen e o a iance o ue expendi u e a iance implying ze o co ela ion
be ween ue ini ial expendi u e and ue change in expendi u e
Sou ce: Au ho s’ es ima ions based on IFLS da a
Co ela ion be ween base-yea expendi u e
and measu emen e o
Se ial co ela ion
coe icien
2000–2007
𝛽=−0.459
2007–2014
𝛽=−0.448
0 0 0.848 0.812
0 0.1 1.041 0.991
0 0.2 1.346 1.273
-0.1 0 0.687 0.657
-0.1 0.1 0.843 0.803
-0.1 0.2 1.090 1.031
-0.2 0 0.543 0.519
-0.2 0.1 0.666 0.634
-0.2 0.2 0.861 0.815
-0.4 0 0.305 0.292
-0.4 0.1 0.375 0.357
-0.4 0.2 0.485 0.458
61
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
As implied by he es ima ed co ela ion e m be ween nega i e shock expe ience a he
baseline and e en ion (Table6), hose e ained in he sample a e mo e likely o expe ience
a nega i e shock a he beginning o he i s pe iod. Howe e , o he nex pe iod we do no
ind s a is ically signi ican e idence o ini ial-condi ions selec i i y o sample a i ion. We
also obse e ha in bo h pe iods, indi iduals wi h a lowe socio-economic backg ound a e
mo e likely o be e ained in he sample. These a e indi iduals om households in which
he household head’s spouse is an unpaid wo ke and has low educa ion le els, wi h lowe
socio-economic backg ound associa ed wi h hei amily o o igin.
Howe e , we see also ha – apa om his common end – he household demog aphic
d i e s o sample e en ion change subs an ially om one pe iod o he o he . Speci ically,
smalle households and households wi h amily membe s abo e he age o 16 a e mo e
likely o d op ou o he sample in he i s pe iod bu mo e likely o be e ained in he sec-
ond pe iod.
When looking a he d i e s o nega i e shock expe ience a he baseline, ou esul s
sugges ha lowe socio-economic backg ound o he amily o o igin (as p oxied by yea s
o educa ion o he a he o he household head) inc eases he likelihood o a nega i e
shock. Howe e , cu en socio-economic cha ac e is ics o he household head and o his
spouse, such as he le el o educa ion and a job as go e nmen wo ke a e, especially in he
Table 5 S a e dependency and ini ial shock expe ience wi h and wi hou non- e ained sample
Panel A: 2000-2007
S a us a ime
S a us a ime −1 >  ≤  no e ained
Sample e ained
−1 >  −1 28.41 24.26
−1 ≤  −1 24.56 22.77
All 52.97 47.03
all indi iduals
−1 >  −1 15.99 13.65 22.75
−1 ≤  −1 13.82 12.81 20.98
All 29.81 26.46 43.73
Panel B: 2007-2014
S a us a ime
S a us a ime −1 >  ≤  no e ained
Sample e ained
−1 >  −1 28.10 24.87
−1 ≤  −1 24.33 22.70
All 52.43 47.57
all indi iduals
−1 >  −1 14.91 13.20 24.35
12.91 12.05 22.58
All 27.83 25.25 46.93
Pooled ansi ions om IFLS, wa es 2–5. Sample size ( e ained) = 15,960. Re ained indi iduals a e ol-
lowed in 1997–2000-2007–2014 and wi h non-missing a iables on pe capi a expendi u e and i s p edic-
o s in each yea . To al sample size in Panel (A): 28,364. To al sample size in Panel (B): 30,073. Panel
(A) includes indi iduals e ained plus indi iduals wi h non-missing pe capi a expendi u e in 1997 and
2000 and comple e in o ma ion on he p edic o s o pe capi a expendi u e. Panel (B) includes indi iduals
e ained plus indi iduals wi h non-missing pe capi a expendi u e in 2000 and 2007 and comple e in o ma-
ion on he p edic o s o pe capi a expendi u e.
Sou ce: Au ho s’ es ima ions based on IFLS da a
62
S.Klasen e al.
1 3
Table 6 P obabili y o e en ion and ini ial nega i e shock expe ience – ma ginal e ec s o explana o y a iables
Re en ion Ini ial s a us:
nsi,2000 =1
Re en ion Ini ial s a us:
nsi,2007 =1
(1) (2) (3) (4)
Age (yea s) o HH head 0.008***
(0.002)
0.009***
(0.002)
0.008***
(0.002)
0.011***
(0.002)
Age squa ed o HH head -0.000***
(0.000)
-0.000***
(0.000)
-0.000***
(0.000)
-0.000***
(0.000)
Female headed HH (dummy) -0.016
(0.012)
-0.046***
(0.017)
-0.000
(0.011)
0.009
(0.013)
Yea s o schooling HH head -0.000
(0.001)
0.010***
(0.001)
-0.002*
(0.001)
0.028***
(0.001)
Yea s o schooling HH spouse -0.003*
(0.002)
-0.003
(0.002)
-0.007***
(0.002)
0.001
(0.001)
HH size 0.009
(0.009)
-0.098***
(0.012)
0.015*
(0.008)
-0.081***
(0.013)
HH size squa ed -0.001
(0.001)
0.005***
(0.001)
-0.001**
(0.001)
0.003***
(0.001)
Ra io o amily membe s aged 19 + -0.166***
(0.058)
0.263***
(0.065)
0.088**
(0.041)
0.317***
(0.038)
Ra io o amily membe s aged 16–18 -0.166***
(0.044)
0.225***
(0.065)
0.235***
(0.047)
0.342***
(0.056)
Ra io o amily membe s aged 13–15 -0.018
(0.069)
0.175***
(0.066)
0.275***
(0.052)
0.269***
(0.036)
Ra io o amily membe s aged 6–12 -0.069
(0.072)
0.050
(0.055)
-0.298***
(0.045)
0.085**
(0.036)
HH head is go e nmen wo ke (dummy) 0.005
(0.017)
0.158***
(0.018)
-0.040
(0.069)
0.127
(0.083)
HH head is p i a e wo ke (dummy) 0.039***
(0.012)
-0.049***
(0.017)
0.012
(0.038)
-0.024
(0.060)
HH head is unpaid wo ke (dummy) 0.000
(0.029)
0.024
(0.054)
0.131
(0.155)
-0.153
(0.173)
HH spouse is go e nmen wo ke (dummy) -0.017
(0.029)
0.151***
(0.036)
-0.317**
(0.129)
1.765***
(0.053)

63
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
Table 6 (con inued)
Re en ion Ini ial s a us:
nsi,2000 =1
Re en ion Ini ial s a us:
nsi,2007 =1
(1) (2) (3) (4)
HH spouse is p i a e wo ke (dummy) 0.035**
(0.014)
-0.104***
(0.027)
0.156
(0.136)
-0.021
(0.262)
HH spouse is unpaid wo ke (dummy) 0.048***
(0.014)
-0.065***
(0.018)
0.174***
(0.068)
0.267**
(0.105)
Pa en al SES: Mo he ’s educa ion -0.008***
(0.002)
0.001
(0.001)
-0.006***
(0.001)
-0.004**
(0.002)
Pa en al SES: Fa he ’s educa ion 0.004**
(0.002)
-0.006***
(0.002)
-0.002
(0.002)
-0.003**
(0.001)
Pa en al SES: Mo he is e i ed (dummy) -0.018
(0.013)
-0.002
(0.014)
0.016
(0.011)
-0.001
(0.014)
Pa en al SES: Fa he is e i ed (dummy) 0.005
(0.013)
-0.010
(0.011)
-0.013
(0.010)
-0.016
(0.010)
Pa en al SES: Mo he is unemployed (dummy) -0.046***
(0.010)
-0.004
(0.009)
0.003
(0.010)
-0.003
(0.009)
Pa en al SES: Fa he is unemployed (dummy) 0.003
(0.013)
-0.005
(0.010)
-0.010
(0.010)
-0.010
(0.009)
Accu acy o he in e iew (dummy) 0.047
(0.030)
0.121
(0.103)
Ra ing o he in e iew missing (dummy) 0.018
(0.051)
-0.108***
(0.028)
Se iousness o he in e iew (dummy) 0.070*
(0.038)
0.028
(0.091)
Co ela ion be ween unobse able ac o a ec ing
i,
and
nsi, −1
0.037*** -0.014
Obse a ions 28,345 28,345 30,026 30,026
P o ince ixed e ec s yes yes yes Yes
HH = household; SES = socio-economic s a us. Robus s anda d e o s clus e ed a he p o ince le el in pa en hesis; Signi icance le els:
∗p<0.10;
∗∗ p<0.05;
∗∗∗ p
<
0.01
Fo each sample, he es ima ed coe icien s a e ob ained om a bi a ia e p obi model ha join ly es ima es he p obabili y o ini ial s a us and e en ion, ollowing he double
selec i i y model. Omi ed ca ego y o he employmen s a us o HH head and HH head’s spouse is sel -employmen
Sou ce: Au ho s’ es ima ions based on IFLS da a
64
S.Klasen e al.
1 3
Table 7 P obabili y o expe iencing a posi i e shock, condi ional on pas nega i e shock and e en ion
Boo s apped s anda d e o in pa en hesis. Signi icance le els:
∗p<0.10;
∗∗ p<0.05;
∗∗∗ p<0.01.
Omi -
ed ca ego y o he employmen s a us o HH head and HH head’s spouse is sel -employmen . Cons an no
epo ed.
Sou ce: Au ho s’ es ima ions based on IFLS da a
Reco e y in 2007 Reco e y in 2014
(1) (2)
Age (yea s) o HH head -0.010***
(0.003)
-0.010***
(0.003)
Age squa ed o HH head 0.000***
(0.000)
0.000***
(0.000)
Female headed HH (dummy) -0.024*
(0.015)
-0.017
(0.013)
Yea s o schooling HH head -0.012***
(0.002)
-0.011**
(0.006)
Yea s o schooling HH spouse -0.008***
(0.001)
-0.004***
(0.002)
HH size 0.104***
(0.016)
0.119***
(0.017)
HH size squa ed -0.005***
(0.001)
-0.006***
(0.001)
Ra io o amily membe s aged 19 + -0.241***
(0.057)
-0.285***
(0.068)
Ra io o amily membe s aged 16–18 -0.192***
(0.056)
-0.276***
(0.047)
Ra io o amily membe s aged 13–15 -0.226***
(0.051)
-0.097
(0.064)
Ra io o amily membe s aged 6–12 -0.256***
(0.042)
-0.213***
(0.059)
HH head is go e nmen wo ke (dummy) -0.050*
(0.027)
-0.124
(0.106)
HH head is p i a e wo ke (dummy) -0.032**
(0.013)
0.019
(0.085)
HH head is unpaid wo ke (dummy) 0.030
(0.029)
-0.338**
(0.152)
HH spouse is go e nmen wo ke (dummy) -0.004
(0.028)
-0.358***
(0.072)
HH spouse is p i a e wo ke (dummy) -0.034
(0.022)
0.029
(0.159)
HH spouse is unpaid wo ke (dummy) -0.030**
(0.015)
-0.002
(0.171)
Selec ion – Re en ion -0.051
(0.052)
-0.070***
(0.024)
Selec ion – Nega i e shock a ime
−1
0.109
(0.089)
0.016
(0.094)
Wald Tes 
𝛽=𝛾=0
2.35 14.04***
p- alue 0.309 0.001
Obse a ions 15,955 15,960
R-squa ed 0.100 0.097
P o ince ixed e ec s yes yes
65
Wha ’s behindp o‑poo g ow h? Anin es iga ion o i s d i e s…
1 3
i s pe iod, signi ican ly and posi i ely ela ed o he expe ience o a nega i e shock a he
baseline. Tu ning o he indi idual p obabili y o eco e y (i.e., posi i e shock expe ience,
condi ional on p e ious nega i e shock expe ience), we obse e in Table7 ha his ends
o be ins ead highe o indi iduals wi h a lowe socio-economic backg ound. The coe i-
cien s on he selec ion co ec ion e m on e en ion
(
𝜆��
i, −1
)
is nega i e in bo h pe iods bu
only s a is ically signi ican o he p obabili y o eco e y in 2014. In his la e pe iod, he
exogenei y es o ini ial condi ions
(𝛽=𝛾=0)
s ongly ejec s he hypo hesis ha
𝜆�
i, −1
and
𝜆��
i, −1
a e join ly ze o, sugges ing ha he posi i e shocks obse ed be ween 2007 and
2014 a e d i en by sample e en ion and a e likely o be iden i ied as a simple dissipa ion
o p e ious nega i e shocks.
On he o he hand, he coe icien s o he selec ion co ec ion e ms on e en ion and
on ini ial nega i e shock expe ience a e join ly and indi idually no signi ican in he i s
pe iod. The lack o signi icance o he sign o he selec ion e ms on ini ial nega i e shock
expe ience implies ha he unobse ed ac o s ha aised he p obabili y o expe iencing
a nega i e shock a he baseline did no play a ole in in luencing he chances o a posi i e
shock a he end o he pe iod. This esul can be in e p e ed in ligh o he a ious eco-
nomic ans o ma ions expe ienced in Indonesia du ing hose yea s, such as he pe pe u-
a ion o he e ec s o he 1997/98 economic c ises23 o he ea ly 2000s oil p ice shocks
( esul ing in ou model as un-p edic ed educ ions in consump ion expendi u e) and he
economic changes expe ienced by he coun y in he Re o masi e a. In he 2000s, pa ly
as a consequence o global ma ke ends and a na u al- esou ce expo boom, Indonesia
expe ienced high a es o economic g ow h and apid s uc u al change which a ec ed he
composi ion o labou demand. No ably, employmen ose mainly in he se ice and in
low-skill sec o s (Coxhead and Sh es ha 2016). Whe eas eal labou ea nings s agna ed,
he new employmen oppo uni ies abso bed a la ge sha e o low-skilled wo ke s and, as
ound in se e al s udies (e.g., Su yahadi e al. 2012; Su yada ma e al. 2013), con ibu ed
subs an ially o po e y educ ion. As implied by ou indings, his ype o exogenous eco-
nomic shocks, indeed, played a ole in gene a ing g ow h oppo uni ies a he bo om o he
dis ibu ion.
4 Concluding ema ks
G ow h incidence cu es a e he main ool p oposed o assess he dis ibu i e impac o
g ow h. Howe e , his ool is unsa is ac o y o a deepe in es iga ion o he na u e o he
obse ed g ow h pa e n, which can mask ei he measu emen e o s o he p esence o
shocks a ec ing pe cen iles in di e en ways.
This pape o e s a guide o co ec ly in e p e ing he p o-poo ness and mobili y impli-
ca ions o g ow h p ocesses wi hin he con ex o he IGIC amewo k. As a i s s ep, we
compa e he ac ual g ow h episodes a each pe cen ile o he ini ial pe sonalized dis ibu-
ion wi h a coun e ac ual pa e n o income g ow h p edic ed on he basis o indi idual
a ibu es. As a second s ep, we examine he di e ence be ween ac ual and coun e ac ual
indi idual g ow h a es. This allows us o unde s and whe he unp edic ed posi i e g ow h
o he ini ially poo is he esul o genuine posi i e shocks, a ou ing upwa d mobili y,
23 As shown in Ra allion and Lokshin (2007), he 1997/98 c isis had an app eciable long- e m impac on
mean consump ion and on he incidence o po e y. P ecisely, almos a one-qua e d op in consump ion
and a leas hal o he obse ed po e y coun in 2002 was a ibu able o he c isis.
66
S.Klasen e al.
1 3
o whe he i can be a ibu ed o p ocesses o s a e-dependence and so o indi idual abil-
i y o eco e om p e ious nega i e shocks.
The me hodological amewo k is applied in he con ex o a sample o 15,960
indi iduals om Indonesia ollowed o e wo se en-yea pe iods, 2000–2007 and
2007–2014. Ou esul s documen ha he e has been subs an ial and signi ican upwa d
mobili y among he ini ially poo e . Howe e , a signi ican pa o his p og essi e
g ow h canno be econciled wi h ei he unobse ed indi idual endowmen s o changes
in ce ain socio-economic a ibu es. The main ac o d i ing he di e ence be ween
ac ual and coun e ac ual g ow h a e is he eco e y om p e ious nega i e shocks in
ecen yea s, as well as mo e genuine economic shocks in he ea ly 2000s. Fo Indone-
sia, he en i e pe iod conside ed in his pape has been one o apid and sha p changes
in he economy and in socie y. The yea 2000 ma ks he ansi ion om he au oc a ic
ule o Suha o, he eco e y om he Asian inancial c isis, he beginning o a p ocess
o decen aliza ion, and, subsequen ly, he commodi y boom – ou di e en economic,
poli ical, and social e en s ha a guably had an impac on people’s li es and so on hei
income ajec o ies. Se e al s udies (e.g., B esson e al. 2017; G imm 2007; Lo Bue and
Palmisano 2020), including he p esen one, ha e shown ha he e has been g ow h in
his pe iod and ha he incidence o g ow h has been la ge among he ini ially poo .
Bu why do he poo exhibi highe g ow h a es han hose indi iduals ini ially belong-
ing o iche pe cen iles? The indings o his s udy sugges ha he apid economic
ans o ma ions o he ea ly 2000s played a ole in shaping he g ow h po en ial a he
bo om o he dis ibu ion. Con e sely, in line wi h he snapsho o ising inequali y and
alling po e y depic ed by he Wo ld Bank (2016), ou esul s also imply ha wha is
obse ed in he mo e ecen yea s is he p oduc o he coexis ence o high ulne abil-
i y and eac i i y o shocks o he poo and o economic secu i y o he middle and
uppe -middle class ha con inued o g ow acco ding o expec a ions. We do obse e
high mobili y among he bo om 30 pe cen , bu his has o be in e p e ed simply as
esilience and abili y o escape ch onic po e y, a he han as a signal o inc eased
oppo uni ies o climb he socio-economic ladde .
Supplemen a y In o ma ion The online e sion con ains supplemen a y ma e ial a ailable a h ps:// doi.
o g/ 10. 1007/ s10888- 024- 09628-7.
Acknowledgemen s S ephan Klasen sadly passed away be o e he e ision o his a icle; we a e g a e ul
o all his con ibu ions o his p ojec . The au ho s acknowledge he inancial suppo o he Minis y o
Science and Cul u e o Lowe Saxony (MWK) and om he F ench S a e in he amewo k o he In es -
men s o he Fu u e p og amme IdEx Uni e si é de Bo deaux / GPR HOPE. A p e ious e sion o his
s udy has been p epa ed wi hin he UNU-WIDER p ojec Social mobili y in he Global Sou h – concep ,
measu es and de e minan s. The au ho s a e g a e ul o he Edi o , wo anonymous e iewe s, F ancisco
Fe ei a, Se gio Fi po, An onio Gal ao, Ga y Fields, Michael G imm, Fabio Clemen i and he pa icipan s
in he 8 h ECINEQ (Socie y o he S udy o Economic Inequali y) Mee ing o e y use ul commen s and
sugges ions.
CRediT au ho s a emen : S ephan Klasen: Concep ualiza ion, Me hodology. Thomas Kneib: Me hodol-
ogy, W i ing—Re iew & Edi ing. Ma ia C. Lo Bue: Concep ualiza ion, Fo mal Analysis, W i ing – O igi-
nal D a , W i ing—Re iew & Edi ing. Vincenzo P e e: Fo mal Analysis, Visualiza ion. The da ase s gene -
a ed du ing and/o analysed in his s udy a e a ailable om he co esponding au ho on eques . The usual
disclaime applies.
Funding Open access unding p o ided by Uni e si à degli S udi di T ies e wi hin he CRUI-CARE
Ag eemen .