NRVA 2005 sampling scheme 23 May 05

NRVA 2005 Sampling Scheme
The analyt ical dom ain, t he unit at which we wish t he dat a
t o be st at ist ically represent at ive, is t o be t he 34 provinces.
I n addit ion t o t hese 34 provincial analyt ical dom ains t here
are 10 urban areas wit h populat ions larger t han 10,000
households. The survey will also collect dat a t o be
represent at ive of t hese 10 urban dom ains. There are four
in t ot al t here will be 44 analyt ical dom ains.
There are anot her 21 provinces t hat have urban
households as classified by t he CSO household list ing. The
urban household populat ions in t hese 21 provinces ranges
from j ust 64 households in Kapisa t o 7,746 in Parwan.
These urban populat ions are bot h sm all in num ber and not
t hought t o be represent at ive of a lifest yle t hat is so
significant ly different from t he rural households t hat t hese
households need to be sam pled separat ely from t he rural
populat ion. Therefore t hese urban households will be
j oined toget her wit h t he rural populat ion in single sam ple
fram e t o for m a m ixed rural- urban populat ion for t hese
provinces.
Experience from ot her surveys suggest s t hat for poverty

est im at es a sam ple num ber of households ranging from a
m inim um of 350 t o 500 produces a st able m ean wit h a
reasonably low st andard error. I t should also be
rem em bered t hat oft en t here are est im at es m ade on
indicat ors where not all households in t he sam ple provide
an observat ion ( e.g. households wit h school- age children)
hence reducing t he sam ple size furt her.
The rural provincial household populat ion varies
enorm ously according to CSO's lat est dat a for 34 provinces
which consist s of dat a for 28 provinces com plet ely
enum erat ed from t he CSO household list ing, and a furt her
t wo provinces part ially enum erat ed.
Collect ing represent at ive dat a wit h a proport ional sam ple a
province level creat es a challenge because of t he large
variat ion in provincial populat ion from t he sm allest
populat ion being in t he province of Nim roz wit h only
13,931 rural households t o Herat wit h 226,576 rural
households. I f we sam ple proport ionally for Nim roz wit h
350 households and use t he sam e sam pling fract ion for all
of a households in ot her provinces t he sam ple would be

prohibit ively large, m ore t han 80,000 households.
To com e wit hin t he budget , t he provinces Jawzj an wit h
50,900 r ur al households is t o be used as t he base
analyt ical dom ain for which we wish t o det erm ine t he
sam pling fract ion. Therefore if we wish t o sam ple 400
households in Jawzj an and use t he sam e sam pling int ensit y
for all ot her provinces of t he populat ion great er t han t hat
of Jawzj an.
For t hose provinces wit h populat ions less t han Jawzj an, if
t he num ber of household sam pled wit h t he st andard
sam pling fact ion is less t han 350 households, at furt her
clust ers are added when drawing t he sam ple to ensur e at
least 350 households for t hese provinces. The sam ple,
t herefore is no longer self weight ing. The sam e applies t o
seven out of t he 10 urban clust ers, where addit ional
households and clust ers have t o be drawn t o ensure 29
clust er s and 348 households.
For all of t hose analyt ical dom ains for which t he sam pling
int ensit y was st andard, wit h a sam ple int erval of 1519
households, t he sam ple was drawn from one spreadsheet .

Therefore when com paring t he predict ed an act ual num ber
of sam ples drawn from each province t here is a difference

of one clust er eit her way depending on t he random st art
wit hin t he order of households wit hin t hat province. This is
not im port ant as for all of t hese provinces t he num ber of
clust ers is great er t han t hose needed t o m eet t he m inim um
350 households. Also t he survey is no longer self- wait ing
because of t he addit ional households t hat have t o be
drawn in som e of t he urban and rural- t he urban provinces
wit h sm aller populat ions.
For t hose provinces or dist ricts wit hin provinces where t he
sam ple fram e is not yet available because t he CSO
household list ing exercise has not been conduct ed in t hose
areas, it m ay be possible t o collect dat a for t wo m ore
provinces during t he course of t he im plem ent at ion of t he
NRVA and draw random sam ples from t his fram e,
ot herwise, alt ernat ives will have to be sought which m ay
include using t he aim s village list and populat ion est im at es
from which t o draw a random sam ple. The wait ing for

t hese provinces can always be adj ust ed ex - post w hen t he
bet t er populat ion est im at es is produced by t he household
list ing exercise.
This sam ple does not include the m igrat ory kuchi
populat ion so far.
Village m aps wit h dwellings list ed are available from t he
recent household list ing exercise for t he 30 provinces t hat
have been enum er at ed so far, but for only villages great er
t han 250 households. A schem e for t hose villagers
select ed in t he random sam ple was less t han 250
households will have to be elaborat ed.
I t is proposed t hat on arrival at t he village, t he num ber of
households is det erm ined during t he m ale com m unit y
int erview. This num ber can be divided by 12, t o creat e a
sam pling int erval for households wit hin t he com m unit y,
and t hen enum erat ors select a household each t im e t hey
count t he sam pling int erval houses. Advant age of t his
m et hod as it should evenly spread t he sam pled households
t hroughout t he village.


Sampling for provinces and districts where CSO
household listing sample frame is not available.
The following provinces and dist rict s at t he t im e of
im plem ent at ion of t he NRVA 2005 did not have household
list ing dat a available.
Daikindi Day Kundi
Daikindi Gizab
Daikindi Kij ran
Daikindi Shahrist an
Ghazni Nawa
Ghazni Zana Khan
Hilm and Baghran
Hilm and Dishu
Hilm and Garm ser
Hilm and Gerishk
Hilm and Kaj aki
Hilm and Khanshen
Hilm and Lashk ar Gah
Hilm and Marj a
Hilm and Musa Qala

Hilm and Nad Ali
Hilm and Naw Zad
Hilm and Nawa
Hilm and Reg
Hilm and Sangin
Hilm and Washer

Kandahar Arghist an
Kandahar Ghorak
Kandahar Maruf
Kandahar Minasheen
Kandahar Nesh
Kandahar Reg
Kandahar Shah Wali Kot
Kandahar Shorabak
Pak t ik a Dela Khoshm and
Zabul Arghandab
Zabul At ghar
Zabul Day chopan
Zabul Khaki Afghan

Zabul Mizan
Zabul Nawbahar
Zabul Qalat
Zabul Seyouray
Zabul Shahj oy
Zabul Shahre Safa
Zabul Sham ulzayi
Zabul Shinkay

Having considered several options, t he use of FAO livestock
census village list wit h est im at ed num ber of households
was chosen t o construct a t em porary sam pling fram e from

which t o draw sam ple villages. This recognises t hat t he
FAO list is unlikely to include t he urban areas wit hin t hese
dist rict s. Once t he CSO household list ing has been
com plet ed for t hese provinces, appropriat e weight s can be
applied t o t he sam ple in each of t hese provinces and
dist rict s.


Kuchi sampling for NRVA 2005
The household list ing conducted by CSO has not effect ively
included t he m igratory kuchi populat ion to dat e, hence
t here is no effect ive sam pling fram e for t his populat ion.
Apparent ly, t his lack of enum erat ion of t he kuchi
populat ion includes t hose t hat have recent ly set t led. This
is exact ly t he sam e populat ion t hat was surveyed during
wint er/ spring 2004 by t he National Mult i- Sect orial
Assessm ent for Kuchi ( NAMAK) , I t he kuchi t hat is st ill
nom adic and t hose t hat have recent ly set t led since t he
onset of t he last drought period. This rem ains t he best
est im at e of t he current kuchi populat ion. The unit of
observat ion for t he survey was t he kuchi wint er locat ion,
where one or m ore kuchi com m unit ies m ay have been
locat ed.
The sam ple fram e for t his spring sum m er survey NRVA 05
was creat ed by const ruct ing the predict ed kuchi
populat ions t heir sum m er locat ion for which inform at ion
was collect ed from t he NAMAK survey. The geographic
locat ion of t hese sum m er kuchi com m unit y areas is not

precise and t he best we can do is construct a sam pling
fram e wit h kuchi com m unit ies ( donat ed by t ribe and sub
t ribe nam e) and t heir populat ion by dist rict .
We can apply t he sam e sam pling fract ion as appropriat e
for t he rural set t led and urban populat ions, and we can
effect ively select com m unit ies in t heir sum m er locat ion
dist rict s. This t hen gives us a sam ple fram e wit h a num ber
of kuchi com m unit ies to be select ed by sum m er dist r ict .
Pr act ically, t rying t o t rack down t he exact com m unit y
select ed in t he sam pling fram e is apparent ly close t o
im possible, and t herefore we will have to resor t t o
random ly select ing wit hin t he num ber of com m unit ies t hat
can be ident ified from inform at ion in t he dist rict .
The accom panying spreadsheet s indicat es t he full sam ple
fram e wit h com m unit y nam e and t he sum m ary of t he
num ber of kuchi com m unit ies by sum m er locat ion dist rict .
The lat t er will be used as t he planning t ool for dr awing t he
kuchi sam pling NRVA 2005.

The sum m er sam ple fram e produced households in for

general kuchi types in t he following t able.
Kuchi t ype
Ex- long range m igrat ory
Ex- short range m igrat ory
Long range m igrat ory
Short range m igrat ory
Tot al

HHs
16,116
21,282
123,792
78,669
239,859

The random sam ple drawn produced t he num ber following
num ber of prim ary sam ple unit s by kuchi t ype. 29 prim ary
sam pling unit s is t he m inim um needed t o produce
represent at ive dat a, t herefore it is clear t hat bot h long and
short range m igrat ory kuchi can confident ly be analysed as

analyt ical dom ains. Whereas t he ex- long and ex- short
range m igratory could form a single analyt ical dom ain, i.e.
t hose kuchi t hat a recent ly ex- m igrat ory but now set t led.
This recent ly set t led analyt ical dom ain would need 5 ext ra
prim ary sam ple unit s t o produce a sam ple for which
st at ist ics wit h a high level of confidence could be
calculat ed.
Because of t he indet erm inat e nat ure of ident ifying different
com m unit ies and t herefore different kuchi types, t he
figures in t he t able below m ay not be reflect ed in t he
act ual sam ple. Therefore t here will need to be som e
addit ional quest ions for t he kuchi com m unit ies to ident ify
t heir kuchi type st at us, and possible ad hoc ext ra prim ary
sam ple unit s collect ed, depending on t he frequency of t he
kuchi t ypes below being select ed as t he dist rict level.
Select ion and dist rict level by kuchi t ype will not be
possible, as it will not be possible t o reliably ident ify t he
kuchi t ype of com m unit ies from inform at ion at t he district
level.
Kuchi t ype sam ple
Ex long range m igratory
Ex short range m igrat ory
Long range m igrat ory
Short range m igrat ory
Margin

# PSUs
11
12
84
51
158

The t able above indicat es t he num ber of prim ary sam pling
unit s select ed classified by general kuchi t ype. This m ay
not be reflect ed in t he final sam ple, as ident ifying kuchi
com m unit ies and kuchi type at dist rict level will likely be
im possible. Quest ions will be included in t he m ail insurer
quest ionnaire to classify each kuchi prim ary sam pling unit
by general kuchi t ype.