Sie sind auf Seite 1von 42

GUIDELINES FOR SAMPLING ORPHANS AND OTHER VULNERABLE

CHILDREN
To Estimate the Size and Characteristics of OVC Populations
Anthony G. Turner, Sampling Consultant, May 2003
Introduction
National estimates of the population of orphans and other vulnerale !hildren "#$C%
in developing !ountries are imperfe!t. The methodology for those estimates often relies
upon demographi! models "&i!ego etal 2003% rather than !ensuses or surveys. 'here the
latter have een used, usually in !on(un!tion )ith household surveys * su!h as +,S
"+emographi! and ,ealth Survey% and M-CS "Multiple -ndi!ator Cluster Survey% *
!ondu!ted for roader purposes, the results re orphans have een in!onsistent and )idely
disparate, )hile there are no .no)n national efforts to measure the population si/e of
other vulnerale !hildren. #ne sour!e of and reason for poor estimation is that there has
not een a fo!used effort to develop spe!ifi! guidelines on ho) to survey #$C * and, in
parti!ular, ho) to sample them. This is partly due to the fa!t that they still !omprise a
tiny portion of the populations of most !ountries, not)ithstanding the rapid gro)th
o!!urring from the ,-$0A-+S epidemi!1 and for that reason there has not een, to date, a
strong push to study #$C in a !on!erted effort.
What the manual includes
This manual is intended to help !orre!t the gap in methodologi!al approa!hes for
surveying0sampling orphans and other vulnerale !hildren. -t provides detailed and
e2pli!it guidelines on sampling approa!hes "a% to estimate the si/e of the #$C
population, or its prevalen!e and "% to study the !hara!teristi!s of #$C, parti!ularly as a
target population of spe!ial interest in the !ampaign against ,-$0A-+S. The latter entails
gathering detailed information on #$C "demographi!, so!io3e!onomi!, and * for
,-$0A-+S * health status, ehavior, .no)ledge and0or attitudes%.
The manual is oriented to)ard developing !ountries ut the methods proposed are
e4ually appli!ale in developed nations.
The manual also emphasi/es sampling methods for surveys intended to study the
entire population of #$C, )hi!h may e !onsidered as !onsisting of three su3
populations5 "6% #$C in households, "2% #$C residing in institutions and other group
4uarters and "3% homeless #$C in!luding, espe!ially, street !hildren.
As ea!h of the three su3populations re4uires essentially independent sample surveys,
it is possile to use the guidelines for ea!h one as a separate endeavor in any !ountry that
)ishes to survey only one of the su3groups in a spe!iali/ed study. -n that !onte2t the
presentation distinguishes et)een linked surveys, )hen all three !omponents are
!ondu!ted in tandem, and stand-alone surveys restri!ted to one of the !omponents.
S!ientifi!, proaility sampling is ta.en as a given in these guidelines, e!ause
)ithout it there is no statisti!al (ustifi!ation for dra)ing inferen!es aout the population
that the sample is supposed to represent. 7roaility sampling is espe!ially !ru!ial for
national3level surveys intended to study the total population of #$C or their
!hara!teristi!s. -t is not only the theoreti!al asis for dra)ing s!ientifi! !on!lusions ut
the only te!hni4ue that a!tually provides a (ustifiale )ay of measuring the margin of
error in the results.
What the manual does not include
7artly e!ause of the need to emphasi/e formal, proaility sampling, informal
methods su!h as !onvenien!e, (udgmental or sno)3all sampling are not in!luded in the
guidelines. &ut su!h te!hni4ues are also e2!luded e!ause their appli!ation is generally
for small3s!ale monitoring and evaluation of intervention pro(e!ts in lo!al areas. The
manual does not provide guidelines for limited, spe!ial3purpose studies, )hi!h, )hile
important, are outside the s!ope of the manual.
Those te!hni4ues, along )ith rapid assessment methods su!h as fo!us group
dis!ussions and e2it intervie)s "from health fa!ilities% are useful as indi!ative, 8early3
)arning9 tools for rapid identifi!ation of oth prolems and solutions in lo!ali/ed pro(e!t
resear!h and design. Their use, ho)ever, is not intended to provide valid or reliale
statisti!al estimates of the various su3populations they are dra)n from. :i.e)ise, the
use of sentinel surveillan!e sites, )hi!h is a valuale te!hni4ue for the longitudinal
monitoring and evaluation of individuals and0or !ommunities ut not for statisti!al
estimation, is not !overed y these guidelines. ;esear!hers and others, therefore, )ho
have different o(e!tives from prevalen!e estimation and may e interested, instead, in
!ondu!ting more limited, fo!used, and lo!ali/ed studies aout #$C are en!ouraged to
!ontinue utili/ing the various rapid assessment and surveillan!e methods that are
availale.
Capture3re!apture methodology might e !onsidered a promising method y )hi!h to
!ondu!t a survey of street !hildren. -t is a te!hni4ue that entails sampling the target
su(e!ts on t)o o!!asions "8!apture9 and 8re!apture9%, mat!hing the t)o samples and
then estimating the total si/e of the population y applying the Chandrase.ar3+eming
formula, )hi!h is ased on the mat!h rate "Chandrase.ar 6<=<%. -t )as de!ided not to
promote or in!lude it in these guidelines, ho)ever, for several reasons. The theoreti!al
underpinnings of the te!hni4ue re4uire that the population eing studied e !losed, in the
sense that there must e virtually no su(e!ts either entering or e2iting the population
during the study period. This .ey !ondition )ould li.ely e drasti!ally violated in the
!ase of street !hildren. >a!h survey o!!asion must e a totally independent operation, a
!ondition almost impossile to attain )hen dealing )ith human populations. Moreover,
pra!ti!al implementation of !apture3re!apture methodology is rarely su!!essful e!ause
of the very real diffi!ulties in a!!urately mat!hing sample persons et)een the !apture
and re!apture phases. 7oor mat!hing essentially invalidates the method. Thus, it is felt
2
that !apture3re!apture !ould not provide a roust estimate sin!e three of the most
important !onditions for its appli!aility )ould e seriously violated.
?inally, the manual does not provide sampling guidelines for trend measurements or
estimates of !hange. Again, those topi!s are most relevant for pro(e!t evaluation
purposes, )hereas this do!ument is !onfined to one3time, !ross3se!tional surveys *
usually on a large s!ale * to e2amine the prevalen!e and !hara!teristi!s of #$C.
Definitions of or!"n# $u%ner"&%e c!i%d
The internationally a!!epted definition
6
of an orphan is as follo)s5
AN #;7,AN -S A C,-:+ @N+>; 6A ',# ,AS :#ST #N> #; &#T, 7A;>NTS.
At the re!ent !onferen!e in Gaorone "@NA-+S0@N-C>? 2003%, the )or.ing
definition * not yet formally adopted 3 for #$C "in!luding orphan% is as follo)s5
Child below the age of ! who has lost one or both parents or li"es in a household with
an adult death #age !-$% &ears' in past ( months or is li"ing outside of famil& care
#homeless-street children and children in institutions')
The formal definitions are important e!ause they have impli!ations on the overall
survey purpose and data !olle!tion strategy in!luding the sampling methodology. 'hen
the survey o(e!tive is to estimate prevalen!e "the si/e of the #$C population% the entire
age group under 6A is oviously targeted. #n the other hand, a survey fo!used on
,-$0A-+S may, in some !ountry appli!ations, e !onfined to 6036B year3old !hildren,
espe!ially )hen se2ual issues are emphasi/ed. #r, the primary age group in another
survey might e #$C under the age of 60 )hen health status and so!io3e!onomi! impa!t
of ,-$0A-+S ma.e up the .ey !ontent of the 4uestionnaire. ;egarding sampling, the
target age group is an important issue e!ause sample si/es for household surveys are
larger, in terms of the numer of households that must e s!reened to lo!ate #$C, for
!hildren under 60 years old or 6036B year3olds !ompared to all #$C under 6A.
2
*n this manual it is assumed that all OVC under + are included in the sur"e& target
group. An impli!ation of this assumption, for the survey design as opposed to the sample
plan, is that different 4uestionnaire modules )ould seem to e in order * one fo!using on
health status and impa!t of ,-$0A-+S for #$C under 60 and the other emphasi/ing
se2ual ehavior, .no)ledge and attitudes for #$C 6036B.
6
See 8National A-+S 7rogrammes, a Guide to Monitoring and >valuation,9 @NA-+S0@SA-+, Geneva,
Cune 2000, p.630, )hi!h uses under 6D as the target age group1 the planned revision of this do!ument in
200= )ill sho) a !hange from under 6D to under 6A. -ndividual !ountries vary some)hat in defining
orphanhood )ith some using death of the mother only as the !riterion.
2
-f nh is the sample si/e * numer of households * to a!hieve a !ertain level of pre!ision for a survey of
#$C age 036B, the sample si/e for 6036B )ould e appro2imately 2.2Dnh and for 03< appro2imately 6.Anh.
3
?or use in household surveys, the standard survey 4uestions that have een developed
and )idely used to as!ertain orphan status in4uire aout )hether ea!h parent of the !hild
is still alive and living in the household. This )ould seem to e2!lude formally adopted
!hildren )ho have lost one or oth parents, ut it is not !lear ho) su!h !hildren are
treated in spe!ifi! survey appli!ations.
Sur$e' co$er"(e of t!e t"r(et ou%"tion
A!!ording to the Monitoring and >valuation Guide, page 63A, the indi!ator for survey
measurement, as distinguished from the definition, is 87er!ent of !hildren under 6D "no)
revised to under 6A% in a household survey )hose mother, father or oth parents have
died.9 -t is important to re!ogni/e that this indi!ator is not an estimate of the per!ent of
orphans in the population. The target population * orphans and0or #$C * is found in a
variety of residential settings. These in!lude traditional households, !ommunal group
!are 4uarters, orphanages and other institutions. -n addition there are homeless youth
living on the street and else)here in non3fi2ed pla!es of residen!e.
'hile it is thought that the great ma(ority of orphans !an e !aptured "in the parlan!e
of survey3ta.ing% in households "as opposed to other vulnerale !hildren )hi!h are more
li.ely to e living in group arrangements or on the street%, a household survey alone is a
ne!essary ut insuffi!ient !ondition to otain uniased estimates of either the prevalen!e
of #$C or their !hara!teristi!s. >stimates of prevalen!e )ould e under-stated through a
household survey alone and !hara!teristi!s )ould e biased to the degree that those
living in households are fundamentally different from those in institutions or on the
street.
Conse4uently, a survey intended to study the total population of #$C must in!lude
those living outside traditional households in!luding those that are homeless. >a!h of
these three su3populations must e co"ered y the survey pro!ess in order to a!hieve
uniased estimates of the numer, distriution and !hara!teristi!s of #$C.
Another aspe!t of survey !overage is geographi!. This manual emphasi/es national
level !overage in the survey appli!ation. Many !ountries, ho)ever, )ill e interested in
more limited geographi!al study of #$C su!h as those living in a parti!ular region,
provin!e or sele!ted !ities. ,ote that the guidelines in this manual appl& not onl& to
national sur"e&s of OVC but also to large- geographical sub-populations such as
regions- designated pro"inces or cities) -t is important to ear in mind the geographi!al
!overage area )hen !onsidering the sample si/e, sin!e sample si/es dis!ussed in the
guidelines apply irrespe!tively of )hether !overage is at the national level or at the level
of a su3national domain su!h as provin!e.
S")%e si*e
The sample si/e is a .ey parameter of sample design for #$C, )hether estimating the
si/e "prevalen!e% of the population or its distriution and !hara!teristi!s. The si/e of the
=
sample differs depending on )hether lin.ed or stand3alone surveys are planned. 'hen
the purpose of the survey is to loo. at the total population of #$C, then the 3 su3
!omponents "household residents, institutional #$C and homeless% must e !omined in
linked surveys. A study may also e designed for the purpose of e2amining only one of
the !omponents as a stand3alone survey.
The sample si/e to estimate prevalen!e is some)hat smaller than that needed to study
!hara!teristi!s, e!ause the latter re4uires detailed analysis of su3!ategories of #$C
"gender, age groups, sustantive groupings, et!.%. The fa!tors that affe!t sample si/e are
many and in!lude pre!ision re4uirements, !onfiden!e level needed, magnitudes of the
!hara!teristi!s eing estimated, sample design effe!t and others. Appendi2 A sho)s the
estimation details for !al!ulating the sample si/es, along )ith the !hoi!es for the various
parameters that ma.e up the mathemati!al formula.

'e )ill loo. at sample si/e along t)o dimensions. #ne is in terms of sample si/e
needed for prevalen!e estimation versus !hara!teristi!s. The other is sample allo!ation
among households, institutions and homeless. Chart 6 summari/es the various
possiilities, )hi!h are further elaorated in the remainder of this se!tion.
.ouseholds / pre"alence estimation
-t is useful, first, to as!ertain sample si/e re4uirements for the household !omponent
of #$C !overage, sin!e a large ma(ority of #$C are orphans that reside in households of
surviving parents or other relatives. 'e spe!ulate that the per!entage is BD and that is the
asis for the sample si/e figures that appear in Tale 6 elo), although a more pre!ise
sample si/e !an e !al!ulated "see Appendi2 A% for any !ountry that has more a!!urate
data on the per!entage of #$C in households. There are no .no)n estimates of #$C at
the national level ut 8Children on the &rin.9 "see e2a!t referen!e in Appendi2 +% does
provide estimates of orphans and those are used to help guide sample si/e !al!ulations.
D
C!"rt +, S")%in( Co$er"(e for V"rious T'es of OVC Sur$e's "nd Reco))ended
Minimum S")%e Si*es
T"r(et Pou%"tion T'e of
Esti)"te
Lin-ed or
St"nd."%one
Sur$e'
Sur$e'
T'e/s0
S")%e Si*e
1
All #$C 6B or
younger
7revalen!e
"si/e of #$C
population%
:in.ed ,ousehold 600032600
households
=
Group
4uarters
ECensusF
,omeless sites =003G00
homeless
persons "to
yield 60036D0
#$C%
All #$C 6B or
younger
Chara!teristi!s :in.ed ,ousehold 6D0033600
households
D
Group 4uarters 600 #$C
,omeless sites =00 homeless
persons "to
yield 600 #$C%
#$C in households Chara!teristi!s Stand3alone ,ousehold 6D0033600
households
#$C in group
4uarters, institutions
Chara!teristi!s Stand3alone Group
4uarters,
institutions
=00 #$C
,omeless #$C Chara!teristi!s Stand3alone ,omeless sites 6G00 homeless
persons "to
yield =00 #$C%
Tale note5 -n the last 2 ro)s the figures of =00 and 6G00 are regarded as are minimums1 )here udgets
permit it is re!ommended that A0036000 #$C e targeted.
A .ey fa!tor in the !onstru!tion of Tale 6 is that orphans "and #$C y e2tension%
!omprise a small per!entage of the total population of most developing !ountries. >ven
3
The sample si/e is douled )hen separate- e0uall& reliable data are )anted for oys, girls, in )hi!h !ase
the latter are regarded as t)o estimation domains.
=
See Tale 6.
D
See Tale 2.
G
in !ountries most affe!ted y A-+S, su!h as in su3Saharan Afri!a, the per!entage is only
aout =.=. Conse4uently, household sample surveys designed to estimate, )ith moderate
reliaility, the numer or proportion of #$C re4uire sample si/es of aout 600032600
households. The smallest sample si/e needed is in the su3Saharan Afri!an !ountries,
)here the average is aout 6000 households. -n the average :atin Ameri!an !ountry and
in China, )hi!h have, proportionately, mu!h smaller #$C populations aout t)i!e as
many households are needed for sampling "see Tale 6%.
T"&%e +, Esti)"ted Percent"(e# p# of OVC Li$in( in House!o%ds to Tot"% Pou%"tion
"nd Aro2i)"te S")%e Si*e /Nu)&er of House!o%ds0 Necess"r' to Me"sure p
G
Area
"6%
>stimated
per!entage, p, of
#$C 036B living in
households to total
population
"2%
Sample Si/e
",ouseholds%
"3%
>stimated
numer of #$C
in sample
"=%
Su3Saharan Afri!a =.3B 602B 2G<
Asia, e2!ept China 3.GD 62=6 2B6
China 2.6< 2600 2BG
:atin Ameri!a,
Cariean
2.3D 6<D2 2BD
The last !olumn of the tale is !omputed as "!olumn 2%0600 2 !olumn 3 2 G.0, )here G.0 is the e2pe!ted
household si/e on average.
-n Tale 6 the estimated per!entages of #$C )ere derived from 8Children on the
&rin.9 "@NA-+S0@N-C>?0@SA-+ 2002% figures, using various assumptions "refer to
Appendi2 A for details%. They are not intended to e a!!urate "and !ertainly )ill vary y
!ountry% ut merely serve as a starting point from )hi!h to !al!ulate the ne!essary
sample si/e.
B
The sample si/e re4uirements for household surveys dis!ussed aove are in
the !onte2t of national surveys, ut the sample si/e figures in Tale 6 apply for su3
national areas "regions, sele!ted provin!es or !ities% as )ell, provided the estimated
per!entage of #$C in the designated su3national area is roughly those of !olumn 2. The
reason it is appli!ale to su3national areas is e!ause the reliaility of a sample estimate
depends on the size of the sample, not the si/e of the population. Nor does it depend on
the proportion of the population that the sample represents, unless the proportion is large
* say, over D per!ent "not generally the !ase for #$C%.
As noted in a footnote, the sample si/es in Tale 6 apply to total #$C. 'hen
separate estimates are )anted )ith e4ual reliaility for oys and girls, the figures of the
last t)o !olumns )ould e douled.
G
This is the estimate, p, of the proportion of #$C in the population measured )ith pre!ision of H0.6Dp and
at the <D per!ent level of !onfiden!e. See Appendi2 A for the !al!ulation formula and assumptions made.
B
-n the formula for !al!ulating sample si/e, it is ne!essary to use an estimate, ho)ever appro2imate, of the
per!entage, p, of the total population one is trying to as!ertain more pre!isely from the sample survey.
B
OVC in institutions and other group 0uarters / pre"alence estimation
To a!!ount for the institutional !omponent of #$C prevalen!e it is re!ommended that
a census of institutions e !ondu!ted, unless this su3population is very large. -t is
thought that fe) developing !ountries have large numers of institutions that house #$C.
A re!ent, !omprehensive study in ;)anda "?ern)ell 2002%, for e2ample, sho)ed there to
e a total of only 2= government3registered institutions "8una!!ompanied !hildren
!entres9% that house orphans "A6 per!ent% and other vulnerale !hildren "6< per!ent%.
#n!e the institutions have een identified, ho)ever, it is simple and ine2pensive, relative
to a sample survey, to enumerate all the #$C living in them. There may of !ourse e
pra!ti!al diffi!ulties )ith resistan!e y !enter dire!tors suspi!ious of the purpose of the
!ensus. Steps are ne!essary to gain )or.ing a!!ess to !enters, su!h as otaining a!tive
support y the ministry )ith responsiility for monitoring institutions for !hildren and0or
the organi/ation )hi!h funds ea!h institutionIs operations. More details on these points
are provided in se!tions on sample frame development and in step3y3step sampling
pro!edures later in this manual.
Thus, sampling and, y inferen!e, sample si/e play no role in !ountries. As su!h, the
estimated numer of #$C otained from the household survey !omponent, dis!ussed
aove, )ould simply e added to the !ount of #$C otained in an institution !ensus "plus
the homeless !omponent dis!ussed elo)% to otain the estimated si/e of the #$C
population.
-t is of !ourse !ru!ial * )hether for a !ensus or a sample survey 3 that all of the
institutions housing #$C e identified, to the e2tent possile, and this is dis!ussed later
in the se!tion on sampling frames.
Street children and other homeless OVC / pre"alence estimation
As previously noted, in the asen!e of etter information )e have assumed that BD
per!ent of #$C live in households.
A
&y that assumption, the remaining 2D per!ent live in
group 4uarters or are homeless. As )e )ould plan to !ondu!t a !ensus for #$C in
institutions it remains to estalish a plausile sample si/e for estimating the homeless.
The methodology for sampling and surveying homeless #$C is !ompli!ated, )ith the
main issues eing ho) to lo!ate them and to !ondu!t intervie)s. ?ortunately, sin!e the
numer of homeless #$C is !omparatively small, it is only ne!essary to sample aout
A
A good estimate of this per!entage is not availale and is only spe!ulated here. Also, )hatever the
average, the true figure is li.ely to vary !onsideraly !ountry3to3!ountry, )hi!h strongly suggests pilot3
testing in a fe) lo!ations to otain a etter figure efore designing and laun!hing a full survey.
A
60036D0 in order to round out the estimate of total prevalen!e. -n that regard )e might
spe!ulate that a properly designed and e2e!uted survey of homeless people )ould yield,
say, 2D33D per!ent that are #$C under 6A years old "the others )ould e adults%. Thus,
)e )ould attempt to design the #$C !omponent of a survey on prevalen!e y sampling
=00 to G00 homeless people and administering a s!reening intervie) to as!ertain age and
perhaps parental status "see Chart 6.%
-t must e re!ogni/ed that the ratio of total homeless persons to homeless #$C !an
vary drasti!ally from !ountry to !ountry or !ity to !ity. 'e have assumed throughout this
do!ument that the ratio is on the order of = to 6, an assumption )hi!h governs the sample
si/e !al!ulations and the illustrations that are given )ith respe!t to surveying street
!hildren. -n some pla!es there may e more homeless youth than adults. -n those !ases
the ratio is a!tually less than 6 and the numer of homeless that )ould have to e
s!reened in order to identify the #$C is redu!ed dramati!ally. Clearly, in the appli!ation
of the te!hni4ues des!ried in this manual in a parti!ular !ountry, it is ne!essary to
estimate, as !arefully as possile in advan!e of the survey, the proportion of homeless
persons that are li.ely to e #$C in order to !al!ulate the appropriate sample si/e.
Estimates of OVC characteristics / households- institutions- homeless
-n !ontrast to a prevalen!e survey )here only reliale estimates of the number of
#$C and their proportion of the total population are )anted, here the aim is to survey
#$C as the fo!al population to learn aout their demographi! and so!io3e!onomi!
!hara!teristi!s. 'hen the o(e!tive of the study is related to ,-$0A-+S, health status, at3
ris. ehavior, .no)ledge and attitudes are studied as )ell.
-t is generally a!!epted that the minimum numer of survey su(e!ts needed for fairly
reliale detailed analysis * that is, )here !ross3taulations are needed y gender, age
groupings and sustantive !ategories 3 is aout =00. #viously, reliaility improves the
larger that numer e!omes, and it )ould e etter to sample G00, A00 or 6000 #$C
)hen resour!es permit.
Tale 2 sho)s the re4uired numer of households to e sampled to otain various
numers of #$C under 6A years old.
T"&%e 3, Aro2i)"te Nu)&er of House!o%ds to S")%e in Order to Loc"te OVC
A(e 4.+5 for Det"i%ed Stud' of C!"r"cteristics
<
Area =00 #$C G00 #$C A00 #$C 6000 #$C
Su3Saharan
Afri!a
6D2D 22BD 30D0 3A00
Asia, e2!ept
China
6A2D 2BD0 3GD0 =DBD
China 30D0 =DBD G600 BG2D
<
?or !ountries that might design an #$C survey !onfined to a different age group, the appro2imate numer
of households needed for #$C age 03< or 6036B are, respe!tively, 6.A times and 2.2D times the numers
sho)n in Tale 2.
<
:atin Ameri!a,
Cariean
2AD0 =2D0 DGBD B600
The figures are rounded to the nearest 2D1 see Appendi2 A for !al!ulation formulas.
The figures in Tale 2, as in Tale 6, apply to the BD per!ent of the #$C population
that )ould e e2pe!ted to live in households. Note that the sample si/es sho)n )ould
also e appli!ale in !ases )here the survey team )as studying #$C in households as a
stand-alone target population, that is, )ithout intending to !omine the results )ith other
data from institutions or the homeless.
'hen lin.ed data are )anted, that is, !omined results from household, institutional
and homeless surveys, the sample si/e for #$C in institutions and homeless need e only
aout 600 from ea!h sour!e. That implies sele!ting 600 #$C from institutions ut aout
=00 homeless persons, the latter of )hi!h )ould have to e s!reened to as!ertain their age
group and, possily, parental status. Note that the !omparatively small sample si/es for
the institutional and homeless !omponents pertains e!ause the o(e!tive is to otain
enough !ases to round out the estimates for total #$C "not (ust those in households%. -f
the study o(e!tives in!luded ma.ing reliale comparisons among the three su3
!omponents and analy/ing their distriutions in various )ays, then )e )ould e dealing
)ith 8!hara!teristi!s9 "as opposed to prevalen!e% in the same !onte2t as if ea!h
!omponent )as a stand3alone survey. ;eferen!e to Chart 6 sho)s ho) the sample si/es
)ould e larger a!!ordingly.
?or surveys intended to e2amine the institutional or homeless population as a stand3
alone target in its o)n right, the sample si/e should e =00 in ea!h !ase * that is, =00
#$C from institutions or 6G00 homeless persons to e s!reened for #$C status.
Note that from Chart 6 a !ensus is !ited as the methodology to use for prevalen!e
estimation for !hildren in institutions "unless the numer of institutions is very large%,
)hile samples are implied for !hara!teristi!s. +epending on !ir!umstan!es in a given
!ountry, this may imply either "a% sampling oth the institutions and their inhaitants in
t)o stages or "% sampling the inhaitants in all of the institutions. This point is
elaorated further in the se!tion, Sampling design and approaches.
'ith respe!t to oy3girl domain estimates, the numers in Tale 2 and those
dis!ussed in the previous paragraphs of this suse!tion )ould e douled for separate,
e4ually reliale data.
S")%e fr")e construction
As implied previously, tapping #$C for survey3ta.ing re4uires that a typology of
#$C e arti!ulated. The typology is in terms of the lo!ations )here #$C !an e found
for purposes of !ondu!ting survey intervie)s. 'e have identified those venues in the
previous se!tions as traditional households, group living 4uarters in!luding institutions
and various pla!es )here homeless !hildren sleep "or !ongregate%.
60
A separate and independent sample frame is re4uired for ea!h of these three venues.
?or the household !omponent, development of a sample frame is not ne!essary, as )ill e
seen elo). ?or institutions, frame !onstru!tion is straight3for)ard in !on!ept ut not
ne!essarily in implementation. ?or homeless !hildren, development of a feasile sample
frame is !ompli!ated oth in !on!ept and implementation. -n addition, a list frame of
.no)n #$C might also e usefully employed in !ountries )here lists of good 4uality
e2ist. The list frame, ho)ever, )ould e used in !on(un!tion )ith the other three,
e!ause all its memers )ould e e2pe!ted to e in!luded, y definition, in at least one of
the three.
The sample plan for an #$C survey intended to study all #$C )ould thus e ased
on a multi-frame design. -n !ases )here a resear!h effort is fo!used on only one of the
!omponent parts "for e2ample, #$C living on the street%, a single frame approa!h
suffi!es. >a!h of the different frames is dis!ussed elo).
.ouseholds
Consistent )ith our !on(e!ture that, on average, BD per!ent of a !ountryIs #$C are
housed in traditional households, the principal frame for sampling that !omponent is
oviously a household frame. The features of household sampling frames are )ell
.no)n. Moreover, it is re!ommended that )hen household surveys are used for #$C
measurement they e done as so3!alled riders to e2isting household surveys rather than
mounting an independent survey from s!rat!h. The !ost of a separate survey dedi!ated
e2!lusively to #$C is li.ely to e eyond the means of most !ountries. ?urthermore,
)hen a rider module is used the household frame and other parameters of the sample
design are pre3determined y the designers of the host survey to )hi!h the rider is
atta!hed. ?or these reasons, )e )ill not dis!uss the general parameters of sample frame
!onstru!tion )ith respe!t to households.
-t must e pointed out that there are signifi!ant methodologi!al !onsiderations, )hi!h
go eyond the sampling issues, on the sustan!e of )hat might e added to a roader
household survey. ?or e2ample, guidan!e is needed on ho) to avoid pitfalls that !ould
yield ina!!urate information su!h as misrepresentation of orphans as non3orphans and
vi!e versa. As.ing the )rong 4uestions or the right 4uestions in the )rong )ay generates
distorted results. 7ra!ti!al advi!e informed y field e2perien!e gathering this type of data
is needed of ho) to avoid the signifi!ant ris.s of over and under3!ounting orphans.
These roader methodologi!al issues are eyond the s!ope of this manual.
There are, ho)ever, t)o sampling3related aspe!ts spe!ifi! to the measurement of
#$C that surveys )hi!h use household frames are li.ely to en!ounter.
60
They are
important for survey !overage and steps must e ta.en to avoid ias from under3
estimation.
60
These are not so mu!h prolems )ith the household frame as su!h or even )ith sampling ut rather )ith
the sur"e& methods generally used in implementing household surveys in developing !ountries.
66
The first !on!erns !hild3headed households, )hi!h are often ruled out3of3s!ope in
traditional household surveys. -t is perhaps apparent that this survey pra!ti!e )ould have
to e avoided in a survey intended to estimate #$C prevalen!e or !hara!teristi!s. Thus,
)hen using an #$C module as a rider, it is important for the survey team to revie) the
pro!edures for the host survey to ensure that !hild3headed households are not e2!luded.
The se!ond aspe!t !on!erns the !overage of !hildren in !ommunal !are at the village
level, a !ultural pra!ti!e often found in Afri!an villages "and perhaps else)here% and, in
parti!ular, for !hildren orphaned y A-+S. ,ousehold surveys are not designed to !over
su!h !hildren, as the survey instruments see. to identify and in!lude only members of
households )ho usuall& eat or sleep there. Any !hild under general, (urisdi!tional !are of
the village as a )hole and not atta!hed to a specific household )ould e missed in a
traditional household survey. This t&pe of under-co"erage ma& in fact account for a
significant part of the under-estimation of orphans that has plagued some sur"e&s)
There is a fairly simple solution to !orre!t for the under3!overage of orphans in
parti!ular and perhaps other vulnerale !hildren as )ell, des!ried in the previous
paragraph, in household3ased surveys in general, and espe!ially, )hen the household
survey is used as a rider for #$C measurement and follo)3up. Most household surveys
in developing !ountries !onsist of a stage of sele!tion in the rural areas involving villages
or parts of villages. $illages are often defined as the first or se!ond stage sampling units.
?or #$C measurement, it )ould e re!ommended that the !hief or head of ea!h sample
village e !onta!ted and as.ed to supply spe!ifi! survey information aout the #$C
under !ommunal !are y the village ut not memers of parti!ular households. An
e2ample of su!h !ommunal arrangements might e 8drop3in9 !enters for #$C of the type
.no)n to e2ist in &ots)ana.
>a!h #$C identified, or a su3sample of them, )ould e eligile for the survey
intervie) * in person or y adult pro2y, depending upon age. The survey )eight that is
appropriate to produ!e the #$C estimates for that portion of the sample pi!.ed up y
8!ommunal9 intervie)s is the inverse of the proaility of sele!ting the village "generally
the )eight asso!iated )ith the first3stage unit of sele!tion%, modified, if ne!essary, y a
)eight for su3sampling )henever a sample of #$C so identified is intervie)ed.
-t is important to note that this type of survey in4uiry at the village level has most
li.ely not een done. -t is re!ommended therefore that a !ommunity level 4uestionnaire
to administer to the village head e developed and pre3tested.

Orphan lists
This su3se!tion applies mainly in !ases )here a !ountry "or provin!e, !ity, et!.%
)ould plan to !ondu!t a dedicated #$C survey, )ithout enefit of the rider approa!h. -t
!an also e applied in rider surveys, ut to a lesser e2tent.
:ist sampling depends on )hether a suitale list or set of lists of .no)n #$C "most
li.ely orphans only% e2ists in a given !ountry or (urisdi!tions )ithin !ountries. 'here
62
lists do e2ist, they should e e2plored for their potential use as a list frame to supplement
the household frame. -f it is thought, for e2ample, that a signifi!ant portion of .no)n
orphans is listed on an availale set of rosters, sampling from those lists )ould redu!e the
si/e of the sample needed from the household !omponent. -n !ountries )here ministries
of so!ial )elfare or human resour!es maintain administrative rosters of orphans, those
rosters should e revie)ed for use as a sampling frame. The revie) )ould as!ertain
)hether they are suffi!iently !urrent, a!!urate and )hether names and lo!atale addresses
of the registrants are availale. Su!h rosters need not e !omplete to e useful, e!ause a
household3ased frame )ould also have to e used to !over #$C not listed on offi!ial
registers, espe!ially non3orphaned, other vulnerale !hildren.
A survey !ompli!ation in surveying #$C sampled from t)o different, overlapping
frames "list frame and household frame% is that un3dupli!ation pro!edures have to e
implemented to avoid over3estimation y doule !ounting. The methodology for un3
dupli!ation re4uires a time3!onsuming and error3prone mat!hing operation
66
to find out
)hi!h intervie)ees sele!ted from the household frame also appear on the list frame.
Su!h !ases must either e eliminated or 8do)n3)eighted.9 'hat is sometimes not
!ompletely understood is that this is a re4uirement irrespe!tive of )hether the !ase in
4uestion happened to e sele!ted on the list frame. The mathemati!al proaility and
!on!omitant )eight depend on the chance of eing sele!ted, not )hether a!tually
sele!ted.
Another important disadvantage of using a list frame of .no)n #$C, )hether alone
or in !omination )ith a household frame, is that the !hildren sele!ted are li.ely to e
randomly s!attered aout in the population rather than !lustered "e2!ept )hen the list
in!ludes #$C living in institutions%. -ntervie)ing !osts on a per !ase asis are mu!h
higher for #$C sele!ted from the list as opposed to those sele!ted in the household
frame, the latter of )hi!h is !lustered.
Although list sampling may have some advantages, espe!ially if a very high
per!entage of #$C is listed on offi!ial registers, the disadvantages of !umersome un3
dupli!ation re4uirements and higher per !ase !osts lead us not to re!ommend list
sampling in general. Note that list sampling is not stri!tly ne!essary any)ay, sin!e the
household and institutional frames )ill !over all the registered #$C )ith .no)n
addresses. This do!ument does not in!lude list sampling as an option in the step)ise
pro!edures given further on. -t is re!ommended, ho)ever, that the sample design
spe!ialist in any !ountry !ontemplating an #$C survey e2plore the feasiility for list
sampling in terms of !osts, !omple2ity and appli!aility.
*nstitutions- other group 0uarters
-n this se!tion )e spea. of the sampling frame of institutions and other group living
4uarters, although in general it is re!ommended that a !ensus of su!h pla!es e !ondu!ted
for #$C prevalen!e estimation "in )hi!h !ase no sampling is involved%. The
66
The !hief diffi!ulty in mat!hing operations of this type is in !omparing names and de!iding )hen a
mat!h o!!urs.
63
re4uirements for !ompiling a list of non3household residential estalishments for #$C
are the same, ho)ever, irrespe!tive of )hether a !ensus or a sample is underta.en.
#f the three venues )here !hildren may e found, those living in institutions that
house #$C should e the easiest to identify. :ists of institutions that are run y the
government, religious or non3governmental organi/ations must e !ompiled. This,
ho)ever, should not e espe!ially diffi!ult, as su!h institutions are li.ely to e )ell3
.no)n and !ompiling a list of them easily a!!omplished. -n order for the list to e
a!!eptale as a frame it should e, as mu!h as possile, !omplete, a!!urate and !urrent.
The !onta!t persons for identifying institutions in!lude offi!ials at ministries of so!ial
)elfare or human resour!es, religious leaders, NG#s that fo!us on youth and any
organi/ation that )or.s on ,-$0A-+S issues. Su!h offi!ials should e sought out not
only at the national level ut also at the levels of provin!e, distri!t and large !ity, in order
to ensure !omplete, or ma2imum, !overage.
#rphanages and temporary foster !are fa!ilities oviously elong on the list, ut also
any other institutions that house !hildren for one reason or another. Cuvenile detention
!enters, (ails and even adult !orre!tional fa!ilities that also have young inmates should e
in!luded. &y the same to.en, military arra!.s04uarters ought to e in!luded, even
though the ma(ority of their inhaitants )ould li.ely e older than 6B.
An important !onsideration is that most institutions )ould li.ely e separated y
gender. -n that !ase, sampling )ould e done independently in ea!h set, thus resulting in
a separate set of sample institutions for oys and for girls.
,omeless shelters re4uire spe!ial treatment, even though they may e regarded as
8group living 4uarters.9 -t is re!ommended, ho)ever, that homeless shelters not e
in!luded on the institutional frame e!ause of the transient nature of their o!!upants.
-nstead, shelters )ould e part of the sample frame for dealing )ith homeless !hildren,
dis!ussed in the ne2t su3se!tion.
-nformation to e otained from ea!h residential estalishment for !onstru!ting the
list in!ludes only its name, lo!ation and the numer of resident #$C. The latter, )hi!h is
the so3!alled 8measure of si/e,9 is needed for three reasons. ?irst, if the overall numer
of institutions is small enough that a !ensus of #$C is to e !ondu!ted, the measure of
si/e informs the survey team aout the )or.load for intervie)ing in ea!h institution.
Se!ond, )here the numer of institutions is so large that a sample of them must e
sele!ted, the measure of si/e is used to administer the sampling pro!edure * systemati!
proaility3proportionate3to3si/e "pps% sele!tion pro!ess. Third, it is used to estalish the
sample proailities and )eights.
-t is important to e a)are that in those institutions that house other persons as )ell as
#$C the measure of si/e should in!lude only the #$C su3population. -t is also
important to note that the #$C !ount need not e perfe!t in order to serve its purpose as a
measure of si/e for sampling "though it should e a fairly !lose appro2imation%. The
6=
e2!eption is that the measure of si/e should e as a!!urate as possile )hen a !ensus of
institutions is to e !ondu!ted.
Street children and other homeless OVC
-t is perhaps ironi! that the !omponent of #$C populations that is most li.ely the
smallest, at least in many !ountries, is the most diffi!ult to survey. The prin!ipal
!hallenge for sampling #$C that live on the street is in lo!ating them. Meeting that
!hallenge is largely a matter of frame development. ?irst, it should e pointed out that
any organi/ation planning a survey effort intended to !over street !hildren )ill have to
!arefully define, operationally, )hat is meant y a street !hild. -n some !onte2ts, for
e2ample, it might in!lude !hildren )ho spend most of their time on the street as truants
ut nevertheless sleep in the home of their parents ea!h night. The tas. of defining street
!hildren for survey3ta.ing is eyond the s!ope of this manual, )hi!h is fo!used on the
sampling methods, and elongs instead to the survey manager )or.ing in !ollaoration
)ith government authorities and professionals )ho are developing its overall !on!epts,
o(e!tives, design. The pro!edures for sampling in!luding development of the frame,
ho)ever, are the same no matter )hat definition of street !hildren is used.
>2amples of 8floating9 populations "no fi2ed residen!e% in!lude orphaned or
aandoned !hildren, runa)ays, persons involved in prostitution and0or drugs. The latter
t)o groups in!lude persons of all ages, though a signifi!ant proportion )ould e #$C
and other young people.
To lo!ate #$C for survey3ta.ing re4uires going to the sites )here they !ongregate,
visit fre4uently or sleep. The numer of persons to e found at a parti!ular site varies
from time to time and this variation must e ta.en into a!!ount in sample design. These
floating populations of youth must e sampled using the !on!ept of time-location sites, a
method of !overing moile youth populations that is uniased and adheres to the tenets of
proaility sampling. >a!h time3lo!ation site identified is defined as a separate primary
sampling unit "PS1%, all of )hi!h ta.en together then !omprise the sample frame. PS1s
are thus defined and !reated as fre4uented sites in !omination )ith time intervals.
A!!ordingly, the same ph&sical site is usually in!luded in the sampling frame more than
on!e, depending on different times of the day, or )ee., that sampling is to ta.e pla!e.
#n the grounds that everyone sleeps some)here, it is re!ommended that the sampling
of street children be confined to places where &outh are known to sleep su!h as rail)ay
stations and other transportation terminals, under ridges, aandoned uildings, va!ant
lots and any others that !ommunity leaders .no) aout and !an pinpoint.
Constru!tion of the frame of PS1s, using the time3lo!ation definition, must egin
)ith !ommunity e2perts and .ey informants )ho are .no)ledgeale aout the ehavior
of homeless or street youth. These e2perts are as.ed to identify "so3!alled ethnographi!
mapping% all the .no)n lo!ations )here the youth may sleep. ?or the sites identified,
!alendar segments of dis!rete time intervals are then asso!iated )ith ea!h one and the
site3time intervals therey e!ome separate PS1s.
6D
@se of the sleep3site !on!ept seems to e the most feasile for estalishing a !losed,
non3overlapping frame of time3lo!ation PS1s. Su!h an approa!h )ould diminish
!onsideraly the estimation prolems that )ould arise from dupli!ate !overage * the
statisti!al !han!e of in!luding the same street !hild from oth her venue for sleeping and
a site )here she performs various )a.ing a!tivities. To apply the approa!h a!!urately,
ho)ever, re4uires that the sleep sites e identified )ith near 6003per!ent perfe!tion. And
it is re!ommended that all time intervals e used in the PS1 !onstru!tion, sin!e sleeping
)ould ta.e pla!e during daylight hours as )ell as night.
?urther details on PS1 !onstru!tion for homeless !hildren are provided in the se!tion
on sampling approa!hes and in the step3y3step pro!edures of Appendi2 C.
S")%in( desi(n "nd "ro"c!es
.ouseholds
?or the household !omponent the most promising methodology in terms of !ost
effe!tiveness is the so3!alled double sample approa!h * also referred to as post3
stratifi!ation. An initial large sample
62
of households is used to !ondu!t simple s!reening
intervie)s to identify and lo!ate households )ith #$C. The s!reening 4uestion )ould e
of the type mentioned in the @NA-+S Monitoring and >valuation Guide, that is, )hether
8the household is !urrently !aring for any !hildren under the age of E6AF )hose father,
mother or oth parents have died.9
-n pra!ti!e, the s!reener should e applied through use of the rider approa!h, that is,
y appending it to another household survey that has a suffi!iently large sample. Timing
of !ourse is !ru!ial and the rider approa!h must e dis!ussed and approved for the host
survey during the early planning stages of the latter. -n !ountries that intend to !ondu!t a
+,S in a time frame !ommensurate )ith planning for an #$C survey, the former 3 )ith
its rigorous sampling methods and typi!al sample si/e of aout G000 households 3 )ould
e a suitale vehi!le in )hi!h to use the rider for initial s!reening to identify households
)ith #$C.
63
Note that e!ause the host survey )ill have een designed y others, )e )ill not
dis!uss su!h matters as geographi! stratifi!ation, !luster si/es, stages of sele!tion for the
#$C module. -t is in!ument upon the sampling te!hni!ian, ho)ever, to ensure that the
treatment of those matters in the host sample design meets the needs for the #$C survey.
62
The initial s!reening may also e done in a de!ennial !ensus provided the timing is right and agreement
!an e rea!hed to add to the !ensus a minimum set of 4uestions that identify orphans, although
identifi!ation of other vulnerale !hildren may re4uire more 4uestions than a !ensus !an easily
a!!ommodate.
63
The sample design for a host survey is generally pre3determined ut the sampling pra!titioner in !harge of
the #$C survey should ensure herself that the former is suitale )ith respe!t to su!h points as the use of
proaility methods, geographi!al !overage, stratifi!ation !riteria, numer and si/e of !lusters and, of
!ourse, sample si/e.
6G
,e0she must also ensure the sample si/e for the host survey is ig enough to !omply )ith
the re4uirements for #$C measurement as depi!ted in Tales 6 and 2. -f the host survey
has a sample si/e that is too ig, it is reasonale to sele!t only a su3sample for the #$C
module, )hi!h may e done in one of t)o )ays.
To illustrate, suppose 3000 households are needed from the host survey in order
to find =00 #$C "see Tale 2%, ut the host survey uses a sample of G000. -n that
!ase, A00 #$C are li.ely to e identified if the rider is appended to every host
survey 4uestionnaire. The survey team may de!ide either "6% to apply the
s!reening rider to only 6 in every 2 of the host survey households in order to .eep
to the original =00 !ases or "2% to apply it in all G000 of the host survey
households. -n option 2, the de!ision !ould then e ta.en later )hether to !ondu!t
the se!ond3phase, detailed 4uestionnaire on #$C in all A00 households "udget
permitting%, or su3sample those households y half and thus revert to the
originally intended =00.
To estimate OVC pre"alence it is a simple matter of merel& tabulating the results
from the screening inter"iew)
To study #$C !hara!teristi!s, ho)ever, a se!ond3phase survey )ould most li.ely e
needed, e!ause it !ould e too urdensome to have a detailed 4uestionnaire re #$C
appended to the +,S "or other household% survey that is used for the rider.
Thus, for surveying #$C !hara!teristi!s a se!ond survey "se!ond phase of sampling%
is done, in )hi!h follo)3up intervie)s in all
6=
the households identified as housing #$C
or other vulnerale !hildren are !ondu!ted for the fo!used study. The detailed
4uestionnaire on #$C is only administered at the se!ond phase sample.
6D
2roup 0uarters and institutions - pre"alence
?irst, )e dis!uss the sample0!ensus plan to estimate the prevalen!e of #$C living in
institutions. As mentioned already, it may not e ne!essary to use sampling at all. -n
many developing !ountries )here orphans are !ared for y a surviving parent or other
relatives, there are fe) formal institutions for housing them or other vulnerale !hildren.
-n su!h !ountries it is re!ommended to !ondu!t a !ensus of the !omparatively fe)
residential institutions that do e2ist. This s!enario is very li.ely to apply for any survey
!onfined to geographi!al su3areas of a !ountry su!h as parti!ular provin!es or !ities, if
not for a national survey.
To !ontrol !osts, a sample of institutions )ould, ho)ever, e ne!essary )hen there
are large numers * say, 6D0 or more 3 of institutions and other residential fa!ilities that
house #$C. A sample of aout =0 to D0 institutions is re!ommended. Note that =0 is the
6=
-t is of !ourse possile to !hoose a su3sample of the households, as des!ried in the illustration,
espe!ially if sustantially more #$C )ere identified than e2pe!ted in the first3phase, host survey.
6D
The detailed #$C module may, instead, e administered during the initial "and only% visit to the
household )henever an #$C is found. This )ould li.ely e the preferred approa!h if all su!h #$C )ere
to e intervie)ed )ithout any further su3sampling at a se!ond phase.
6B
re!ommended minimum, ut this numer )ould apply no matter )hether there are 6D0
institutions in the universe or 6000.
To estimate prevalen!e, a census of the #$C in the selected institutions )ould e
!ondu!ted, that is, a simple !ount of the residents under 6A )ith !olle!tion of only
minimal detailed data other than age. Note, for this reason there is only a single stage of
sample sele!tion * the institutions themselves and they )ould thus e sele!ted on the
asis of si/e strata rather than y proaility proportionate to si/e "pps%. More
!ompli!ated stratifi!ation !riteria su!h as !ost per !hild per year are not re!ommended for
the reason that information ne!essary to !reate su!h strata are not li.ely to e feasily
otainale.
To !reate si/e strata means it is ne!essary to otain figures on si/e "numer of #$C%
from institution offi!ials. #taining su!h figures is some)hat less prolemati! in
fa!ilities )here only #$C are housed, sin!e offi!ials may have !urrent !ounts or rosters
availale. -n other fa!ilities su!h as detention !enters and military arra!.s oth #$C
and adults )ould e residents. -n those !ases, administrative re!ords on age may not e
readily availale and it is ne!essary to )or. )ith institutional offi!ials to otain a !ount
of the #$C. Su!h a !ount, for use as the measure of si/e in ea!h institution, !an e
appro2imate.
3ppro4imate numers of #$C )ould then suffi!e for assigning an institution to its
stratum "and the design )ould not e ruined if a fe) institutions are assigned to the
)rong stratum%. The e2a!t definition of )hat !onstitutes small, medium or large )ould
no dout vary y !ountry, ut a rule3of3thum might e as follo)s5
C!"rt 3, Si*e Str"t" for Institutions# Ot!er Grou 6u"rters
Stratum Si/e "Numer of #$C% -llustrative Sample Sele!tion
6G

Small @nder 20 6 in 60 PS1s
Medium 203<< 6 in D PS1s
:arge 600 or More Sele!t all
-n !reating the strata the institutions should e arranged in a geographi! fashion
)ithin ea!h stratum y uran3rural and y provin!e prior to sample sele!tion, in order to
ensure geographi! spread of the ensuing sample. Sele!tion )ithin ea!h stratum )ould e
systemati! a!!ording the sele!tion rates illustrated in Chart 2.
The se!ond3stage !ensus !ounts )ould e inflated y the )eight asso!iated )ith the
sample institution, depending upon its stratum. -n the aove e2ample, the )eights )ould
e 60, D and 6, respe!tively, for small, medium and large institutions. All #$C in the
sample institutions from the small stratum )ould have a )eight of 60, for e2ample.
6G
Sampling fra!tions, in pra!ti!e, depend on the total numer of PS1s )anted for the sample and their
average si/es.
6A
-n !ountries )here oys and girls are housed in separate institutions, the sample plan
should !all for ea!h to e treated separately and independently. This is espe!ially
important )hen e4ually reliale data are )anted for oys and for girls, in )hi!h !ase the
sample si/e is the same for oth.
2roup 0uarters and institutions / OVC characteristics
Se!ond, )e !onsider sampling plans for estimating !hara!teristi!s. ;e!all from Chart
6, )e need a minimum of 600 !ases )hen this is a lin.ed survey )ith a household and
homeless !omponent1 =00 )hen a stand3alone survey. There are t)o !ases to !onsider
for sele!ting the appropriate sample of #$C. #ne is the situation )here the institutions
in the universe are so fe) that a !ensus of them is ta.en. -n that !ase, to illustrate,
suppose there are 6A institutions in )hi!h the !ensus )as !arried out, and a total of <00
#$C are housed in them. A !omplete roster of the #$C from ea!h institution must e
!ompiled. Then, the sample of, say, =00 #$C should e sele!ted systemati!ally from the
!omined list of all <00. The sampling interval is e4ual to <000=00 "in this illustration%,
or 2.2D1 a random starting numer is !hosen, using a tale of random numers et)een
0.06 and 2.2D. See the step3y3step pro!edures in Appendi2 C for an illustration of
systemati! sampling from a list.
The se!ond !ase o!!urs )hen the sample of #$C "again, say, =00% is ta.en from a
sample of institutions. Sele!tion in this !ase )ould e ased on a t)o3stage sample
design. The first stage should e a sample of institutions sele!ted systemati!ally using
proaility proportionate to the si/e of the institution. Again, as aove, this means )e
must otain figures on si/e "appro2imate numer of #$C% from institution offi!ials.
A fi4ed numer of #$C from ea!h sample institution )ould e sele!ted to parti!ipate
in the intervie) for the #$C survey. Again, the numer of institutions to sele!t should
e, at minimum, =0. Suppose in this !ase )e intend to sele!t =00 #$C1 )ith a minimum
of =0 institutions, the !luster si/e )ould e 60, that is, 60 #$C )ould e !hosen in ea!h
sample institution, using systemati! sele!tion from a !omplete roster of #$C in that
institution.
As mentioned previously, independent sampling of oys and girls in separate strata
)ill li.ely e ne!essary in !ountries )here they are housed in institutions separated y
gender.

7ro!edures for implementation of pps sampling of institutions and systemati!
sampling of residents are given in Appendi2 C in the step)ise methodology.
.omeless children
-n many !ountries, the numer of homeless !hildren may e !onsidered so fe) that it
)ould not e feasile to survey them e!ause of the te!hni!al !omple2ity of doing so.
6<
'e suggest sample si/es of 6G00 homeless persons in order to lo!ate =00 street #$C, for
e2ample1 !learly, this does not ma.e sense if the total numer of su!h persons is only a
fe) hundred. -n other !ountries, ho)ever, )here there are large numers of street
!hildren in ig !ities, their e2!lusion )ould produ!e "6% signifi!ant under3estimation )ith
respe!t to #$C prevalen!e and "2% serious ias )ith respe!t to their !hara!teristi!s,
espe!ially for surveys fo!used on ,-$0A-+S.
;elated to the last point is the 4uestion of the appropriate survey !overage area for
homeless youth. As suggested, this su3population of youth is li.ely to e heavily
!on!entrated in large !ities. The !omparatively small numers of them to e found in
rural areas, to)ns or small !ities proaly does not (ustify !ondu!ting national or regional
surveys for homeless youth.
6B
?inally, lo!ating street youth for surveys )ill li.ely re4uire that some intervie)ing
ta.e pla!e in the middle of the night, )hi!h !an e intrusive and dangerous. -t is
re!ommended that pilot tests e !arried out on a small numer of !ases to as!ertain the
prolems asso!iated )ith su!h intervie)ing !onditions, as )ell as overall feasiility,
efore a full3s!ale survey is mounted.
'hen the de!ision is ta.en to survey homeless youth, a t)o3stage sampling plan
should e implemented. The first stage is a sele!tion of the time3lo!ation PS1s that )ere
des!ried in the se!tion on sampling frames. The se!ond stage of sampling !onsists of
posting intervie)ers at the site for the time interval designated and intervie)ing all "ta.e3
all approa!h% the youth that are present or arrive at the site during that period. A .ey
feature of the design is ensuring that sampling ta.es pla!e over a fi2ed time interval
)hi!h is the same for every sample PS1.
&e!ause of the unusual definition of the PS1s and the )ay they are !onstru!ted, it is
not pra!ti!al to sample a fi2ed numer of street #$C or other homeless at ea!h sele!ted
site. The numer of homeless is apt to vary !onsideraly from one day to the ne2t at a
given site. And even if an appro2imate measure of si/e !an e assigned to a site it is
li.ely to e unstale in !ertain )ays5 the ma(ority of homeless persons may not e )ithin
the survey age group, may not e #$C, et!. As a result, the ultimate numer of street
!hildren in the sample !annot e !ontrolled pre!isely, although that numer !an e
targeted and, hopefully, !losely appro2imated.
As re!ommended previously, the PS1s should e defined in terms of sites )here
street !hildren sleep, e2!luding other sites )here they may !ongregate for their )a.ing
a!tivities. As su!h, ea!h sleep lo!ation should e divided along the time dimension in =
to G hour segments for PS1 !onstru!tion. ?or e2ample, four PS1s might e formed as
follo)s5
@nder City &ridge * G a.m. to noon
6B
An e2!eption is !ountries )here homeless youth are .no)n to live in mining sites or 4uarries )hi!h are
typi!ally lo!ated in rural areas1 a !ase !an e made for !ondu!ting an #$C homeless !omponent in su!h
!ountries "or provin!es affe!ted%.
20
@nder City &ridge * noon to G p.m.
@nder City &ridge * G p.m. to midnight
@nder City &ridge * Midnight to G a.m.
Similar PS1s )ould e !onstru!ted for all other sleep lo!ations identified y .ey
informants. See ?orm S6 in Appendi2 C on step3y3step pro!edures for the 4uestions to
use in in4uiring of offi!ials and other .ey informants aout pla!es )here street !hildren
sleep.
-n all PS1s, so !onstru!ted, a measure of si/e must e determined. This is an
appro2imate !ount of the numer of homeless persons e2pe!ted to e at a parti!ular
lo!ation during the time interval spe!ified. -t is a tedious ut ne!essary step in order to e
ale to impose as mu!h !ontrol as possile over the ultimate sample si/e for the survey.
To estalish the measure of si/e re4uires pre3survey field )or. involving a visit to ea!h
site and !ounting, roughly, the numer of homeless people that !ome and go during the
spe!ified time interval. As implied, the !ount need not e e2a!t and it )ill e unstale
any)ay, ut still it is ne!essary to .no) )hether a PS1 should e assigned a measure of,
say, D or D0.
PS1s )ould e stratified a!!ording to si/e !ategories ased on their measures of si/e
* small, medium and large, similar to the institutional strata des!ried aove. Again, the
e2a!t si/e !ategories )ould e defined at the !ountry level, ut a rule3of3thum is given
in Chart 3.
C!"rt 1, Si*e Str"t" for Ti)e.Loc"tion PSUs 7!ere Ho)e%ess Persons S%ee
Stratum Si/e "Numer of ,omeless 7ersons% -llustrative Sample Sele!tion

Small PS1s @nder D 6 in 60 PS1s
Medium PS1s D32D 6 in D PS1s
:arge PS1s More than 2D 6 in 2 PS1s
#ther parameters of sampling to e !onsidered are sample si/e, numer of PS1s,
!luster si/e. ?or stand3alone surveys targeted solely on homeless youth, sample si/es
should e a minimum of 6G00 homeless persons, on the grounds that only aout =00 of
them )ould e #$C in the age group "Chart 6%. 'hen the survey is intended to estimate
homeless youth as part of a larger study to estimate the overall si/e of the #$C
population, the sample si/e ne!essary is =00 homeless persons "again, 600 e2pe!ted to e
#$C under 6A%.
Cluster si/e and numer of PS1s are intert)ined and involve the numer of homeless
youth e2pe!ted, on average, to sho) up at a given site during the time interval. The
!luster si/e has t)o dimensions to !onsider. #ne is the numer of homeless people at the
PS1 site and the other is the numer of those that are #$C. -n any !ase, a short set of
s!reening 4uestions is ne!essary )hen !ondu!ting the intervie) in order to eliminate any
homeless person )ho is out3of3s!ope for the survey. There is the issue of )hether to
s!reen for age y 4uestioning ea!h person in the sample PS1 or, instead, relying upon
26
oservation y the intervie)er. The first is more e2pensive and reliale, )hile the se!ond
is the reverse. A !ompromise pro!edure )ould e to forego 4uestioning adults )ho are
!learly older than 6B ut administer a simple 4uestion aout age for those )hose
appearan!e is more amiguous.
Administration of the s!reener should last only t)o or three minutes. The main
intervie) 4uestionnaire intended for the "#$C% target group should determine the !luster
si/e. And this is done in terms of ho) many #$C intervie)s !an e !ondu!ted at a given
site in the allotted time. That numer )ill vary of !ourse, dependent on su!h fa!tors as
intervie) length and numer of intervie)ers to e posted at the site. See illustration for
!al!ulating average PS1 si/e in the step3y3step pro!edures of Appendi2 C.
As already mentioned the intervie)ers should e posted for a fi2ed interval of time.
That interval must e the same as the interval used to define the 7S@s, )hi!h is
re!ommended at = to G hours. At ea!h sele!ted PS1 an intervie) is sought from all #$C
present or arriving during that time. -n !ases )here there are e2pe!ted to e a large
numer of #$C present, the intervie) team should e e2panded a!!ordingly. ?or
e2ample, if t)o intervie)ers !an intervie) 6D #$C in G hours and 30 #$C might e
e2pe!ted at a given site, the intervie) staff should e douled for that site.
-t is reasonale to 4uestion )hy not impose a ma2imum numer of intervie)s.
>stimation e!omes prolemati!, in terms of a!!ura!y, )hen there is a !ut3off. &y
!ontrast, an uniased estimate !an e made "for oth prevalen!e and !hara!teristi!s%
)hen all intervie)ers, in ea!h PS1, are posted at the site for a fi2ed interval of time and
intervie) all in3s!ope persons )ho sho) up. Su3sampling the #$C at a site is also
prolemati! )ith its impli!ations for listing them first and then applying systemati!
sele!tion pro!edures on the spot. Moreover, during a given interval, ne) arrivals )ould
e e2pe!ted throughout so that their total numer, )hi!h is needed to !al!ulate the
sampling interval, !annot e .no)n until the time interval e2pires * a parado2.
Some youth )ould e e2pe!ted to sleep at the same site or different sites )ithin the
duration of the survey. Juestions must e added to the survey instrument to identify su!h
!ases in order to 8do)n3)eight9 the results for a!!urate estimation.
To illustrate5 -t is 4uite li.ely the same site )ill fall into sample more than on!e,
sin!e the site )ill !omprise = PS1s ased on time of day if G3hour time intervals
are used or G if =3hour intervals are used. Any youth a!!ustomed to sleeping in
the same pla!e has a good !han!e of sho)ing up at the lo!ation at different time
intervals, thus giving him0her multiple !han!es of sele!tion. 'hen that o!!urs,
the youth should e intervie)ed only on!e. A some)hat more diffi!ult
dupli!ation prolem to sort out !on!erns those youth that sleep in different sites
from one day to the ne2t. To over!ome this prolem )e need to ta.e into a!!ount
the length of the survey period * one )ee., t)o, et!. :et us assume 2 )ee.s for
the e2ample. -t is ne!essary to as. ea!h respondent ho) many other pla!es he0she
usually sleeps during a 23)ee. period. ,e )ould also e as.ed to identify those
lo!ations. Those lo!ations )ould e mat!hed * a!. in the offi!e * to !ompare
22
against the full list of sites that ma.e up the PS1 frame. ?or ea!h mat!h, the
survey )eight for that respondent must e do)n3)eighted y the fa!tor, 60t,
)here t is the numer of mat!hing sites.
-f the survey of homeless youth is eing done in !omination )ith another that !overs
#$C in households or institutional fa!ilities, dupli!ation of !overage !an also o!!ur. -n
that !ase, it is ne!essary to as. ea!h youth intervie)ed at the sample site )hether he0she
has a usual pla!e of residen!e * again in order to do)n3)eight the results.
Docu)ent"tion "nd e$"%u"tion
The sampling te!hni!ian should ta.e ne!essary steps to !arefully do!ument not only
the sample plan for the #$C survey, )hether lin.ed or stand3alone surveys, ut also its
implementation. Sample designs often re4uire adaptation at some stage of the field )or.,
e!ause of unforeseen situations that arise in the !ondu!t of the survey. -t is important to
re!ord 3 step3y3step 3 all the pro!edures used in !arrying out the sample plan to ma.e
sure the implementation is faithful to the design. 'hen it is not, it is even more
important to do!ument all the departures from the design, even minor ones. This
information is ne!essary later at the analysis stage, in !ase any ad(ustments need to e
made1 ut it is also indispensale for planning future surveys.
To evaluate the results of the survey, sampling errors should e estimated. The
sampling error, or standard error, is the measure that allo)s the !onfiden!e interval to e
!onstru!ted around the estimate, so that users !an evaluate ho) reliale the data are.
Standard errors are estimated from the survey data themselves, )henever the design
adheres to proaility sampling methodology, y using fairly !omple2 mathemati!al
formulas and pro!edures. The pro!edures used must refle!t the a!tual design employed
in the survey. The latter is often a diffi!ult tas. to perform and is fre4uently done only
)ith the assistan!e of a sampling e2pert.
There are !omputer soft)are pa!.ages that may e used to !al!ulate the standard
errors, in lieu of designing a dedi!ated varian!e
6A
estimation program. -nformation aout
many of the pa!.ages !an e found on the -nternet in!luding 8'esvar9 from 'estat
Corporation and "'estat.!om% 8$7:K9 from the @.S. &ureau of the Census
"Census.gov%. Those pa!.ages, li.e most others, are suited espe!ially to household
surveys. They )ould have to e adapted, again )ith the help of an e2pert, for appli!ation
to institutional or homeless surveys1 other)ise, dedi!ated !omputer programs may e
ne!essary for those !omponents.
6A
The term 8varian!e9 refers to the s4uare of the standard error.
23
Aendi2 A, P"r")eters for Fi(ures of T"&%es + "nd 3 of Te2t
This appendi2 sho)s the estimation formulas used to derive the figures pertaining to
household survey sample si/es appearing in Tales 6 and 2 of the te2t, in!luding the
parameters used and assumptions made. ?or #$C in institutions or homeless the sample
si/e is not !al!ulated per se, ut rather is presented as the minimum ne!essary to produ!e
fairly reliale !ross3taulations for analysis in the !ase of measurement for !hara!teristi!s
or the minimum ne!essary to round out a household survey )hen total prevalen!e is the
o(e!tive.
Estimating the parameter, p
-n order to !al!ulate sample si/es, it is first ne!essary to have an appro2imate estimate
of the parameter, p, "defined elo)% that one is a!tually trying to estimate from the
survey. The estimation formula for p in Tale 6 is
p L "0.D=% pM "0.BD%, )here E6F
p is the parameter to e !al!ulated and is the estimated proportion of #$C 036B
living in households to total population1
pM is the estimated proportion of !hildren 036B )ho are orphans, ta.en from the
estimates for ea!h !ontinent provided in 8Children on the &rin.9
"@NA-+S0@N-C>?0@SA-+ 2002%1 those estimates are for orphans 036= years old
and it is assumed here that the estimated proportion for 036B is the same1
0.D= is the average estimated proportion of !hildren 036B in the total population,
)here the assumption is made that .03 is the proportion for ea!h single year of age
in developing !ountries1 thus 0.D= L 6A 2 .031 and
0.BD is the assumed proportion of #$C that are li"ing in households1 in this
!onte2t it is thought that most of the #$C in households are orphans and that the
remaining 2D per!ent )ill e oth orphans and other vulnerale !hildren living in
group 4uarters or homeless.
Estimating sample sizes
The estimation formula for the sample si/e, n
h
, in Tale 6 is
n
h
L 3.A= "p% "63p% "f% "6.6%0 G "e
2
%, )here E2F
n
h
is the parameter to e !al!ulated and is the sample si/e in terms of households1
3.A= is the s4uare of the z3statisti! "6.<G% )hi!h defines the <D per!ent level of
!onfiden!e1
2=
p is from e4uation E6F1
f is the sample design effe!t, assumed to e 6.D * a value )hi!h implies moderate
!lustering e!ause although #$C may !on!entrate in !ertain neighorhoods or
villages their overall proportion is !omparatively small1
6.6 is a multiplier to a!!ount for an e2pe!ted rate of non3response of 60 per!ent1
e is the margin of error to e attained, set at a moderate level of pre!ision, )hi!h
is 6D per!ent of p1 thus e L 0.6Dp1 and
G is the average household si/e in developing !ountries.
The estimation formula for the numer of households ne!essary to lo!ate varying
numers of #$C age 036B, found in Tale 2 is
5 L 60Gp, )here E3F
5 is the numer of households to e !al!ulated1
6 is the desired numer of #$C to e sampled and intervie)ed and it varies from
=00 to 60001 and
p is from e4uation E6F.
When the assumptions do not apply
?or !ountries in )hi!h one or more of the assumptions dis!ussed aove do not hold,
simple sustitutions may easily e made in the formulas to arrive at more a!!urate figures
on sample si/e. ?or e2ample, the average household si/e may e larger or smaller than
G.01 non3response may e e2pe!ted around D per!ent instead of 601 and the value of pM for
a parti!ular !ountry !an e more pre!isely !omputed than y using the !ontinental
average )hi!h is )hat these formulas aove assume.
>a!h !ountry )ould e e2pe!ted to have e2a!t figures from its !ensus on the
per!entage of the population that is under 6A, so that 0.D= may e sustituted for. Some
!ountries may have more pre!ise information on the proportion of #$C that reside in
households, in )hi!h !ase the value of 0.BD in formula E6F )ould !hange. -t is
re!ommended ho)ever that no !hange e made for the z3statisti! value of 6.<G, )hi!h is
the international standard. The design effe!t, f, should also e left at 6.D as the default
value. -t is also re!ommended that e e defined as 0.6Dp e2!ept in !ases )here udgets
)ould permit a more stringent margin of error, in )hi!h !ase it may e redu!ed to 0.62p
or .6p. Su!h redu!tions in the margin of error )ill yield dramati!ally higher sample si/es
ho)ever.
2D
Aendi2 B, C"%cu%"tin( Sur$e' 7ei(!ts
Analysis of the survey data !ondu!ted a!!ording to the pro!edures des!ried in this
do!ument )ill generally re4uire )eighting * inflation of the ra) data y the inverse of
the proailities of sele!tion for the sample !ases. 'eighting is ne!essary e!ause, )ith
fe) e2!eptions, the sample plans dis!ussed are not self3)eighting.
Household Component
@se of the rider approa!h, in )hi!h #$C survey instruments are atta!hed to an
e2isting host survey, means that the )eights for the #$C identified and surveyed are the
same as those for the host survey. ,en!e, it is only ne!essary to otain the )eights
asso!iated )ith the host survey and apply them. Those )eights )ill e ased on the
design of the host survey, )hi!h )ill no dout differ in every !ountry that applies the
#$C sampling guidelines. ?re4uently, ho)ever, the host survey design may e a self3
)eighting one in )hi!h ea!h sample household and0or person has the same )eight. >ven
in that instan!e, ho)ever, it is ne!essary to use that )eight )hen !omining the #$C
results )ith surveys from institutions and homeless !hildren.
There are t)o spe!ial situations that need attention for )eighting in the household
!omponent * situation in )hi!h the host survey )eights !annot e used as is. ?irst, the
guidelines mention that su3sampling of #$C identified from the host survey may e
used )henever the numer is mu!h larger than e2pe!ted. -n that !ase the host survey
)eight is multiplied y k, )here k is the inverse of the sampling fra!tion used to su3
sample for the #$C intervie)s. ?or e2ample, if only 2 in 3 of the host survey
households !ontaining #$C are intervie)ed for the #$C survey, then the appropriate
)eight )ould e w 2 6.D, )here w is the host survey )eight and 6.D is the inverse of the
su3sample fra!tion, 203. Note that w may e variale, depending on )hether the host
survey is self3)eighting1 if it is not self3)eighting ut varies say, y strata or uran3rural,
then there )ould e a different value of w for ea!h su!h stratum.
The se!ond situation !on!erns the !ase of village sampling )here heads of the
villages sele!ted in the "usually% first stage of the host survey are as.ed aout #$C living
in !ommunal arrangements. #$C identified in this !ase )ould re!eive the )eight
asso!iated )ith sele!tion of the village. Again, it is ne!essary to otain this )eight from
the host survey, ut in many household surveys it is e4ual to5
w
i
7 89#a m
i
', )here
w
i
is the )eight of the i
th
sample village determined from the host survey,
8 is the total numer of households in the sampling frame for the host
survey and is e4ual to the sum of the m
i
values,
a is the numer of first3stage units "villages% sele!ted in the host survey,
2G
m
i
is the measure of si/e "numer of households% in the i
th
sample village.
-f the #$C identified from the !ommunal sour!e are su3sampled efore
intervie)ing, the host survey )eight, w
i
, is multiplied y k "des!ried aove%.
Institutional Component
'hen a !ensus of institutions is ta.en to otain the !ount of #$C for prevalen!e
estimation, no sampling is involved nor is )eighting.
-f the institutions are systemati!ally sampled for #$C prevalen!e estimates, ea!h
#$C in a given sample institution re!eives a )eight of *
s
, )hi!h is the sample interval for
the s
th
institutional stratum "small, medium or large%. ?or e2ample, sele!tion of
institutions in the small stratum at 6 in 6D )ould result in a )eight of 6D for ea!h #$C in
that institution "re!all that for prevalen!e, all the #$C in the sample institutions )ould e
in!luded in the survey intervie)%.
'hen males and females are sele!ted either in separate strata or as separate domains,
the )eight must e !al!ulated separately for ea!h.
?or #$C !hara!teristi!s, t)o !ases o!!ur. ?irst, )hen institutions are fe) and a
!ensus of them is ta.en, the sample )eight is !al!ulated on the fra!tion of #$C sele!ted
"re!all this is a !onstant fra!tion over all institutions !omined%. The )eight is simply the
sampling interval, *1 there )ill e t)o su!h intervals, ho)ever, )hen oys and girls are
sampled separately.
-n the se!ond !ase, involving a large numer of institutions, a t)o3stage sample is
done in )hi!h institutions are first sele!ted systemati!ally )ith pps and a fi2ed numer of
#$C is sampled systemati!ally )ithin ea!h sample institution. The )eight is then e4ual
to5
w
:
7 ,9#b n
:
'#n;
:
9c', )here
w
:
is the )eight of the (
th
sample #$C,
, is the total numer of #$C in the sampling frame and is e4ual to the
sum of the n
i
values, that is, sum of the institution measures of si/e,
b is the numer of institutions sele!ted in the survey,
n
:
is the estimated measure of si/e in the (
th
sample institution,
n;
:
is the a!tual !ount of #$C in the (
th
sample institution,
c is the fi2ed numer of #$C sele!ted in ea!h sample institution.
2B
Again, the aove )ould e repeated for oys and girls separately )hen they are
sampled from t)o independent sets of institutions. ?or a given stratum "that is, male
institutions or female institutions% the )eight is appro2imately the same for all #$C in
the sample, to the e2tent that the estimated measure of si/e is the same as the !ount of
#$C in a given institution. ,en!e, the sample is appro2imately ut not pre!isely self3
)eighting. #!!asionally, the total numer of #$C residing in a given institution may e
less than the value, c, in )hi!h !ase all the #$C at that institution are intervie)ed.
Cal!ulating the )eight for that institution then re4uires sustitution of the a!tual value of
c rather than the pre3determined, fi2ed value * another reason )hy the sample )ill not e
pre!isely self3)eighting.
Homeless Component
The asi! )eight for the homeless !omponent is very simple e!ause of the )ay the
time3lo!ation PS1s are !onstru!ted and sampled "even though that pro!ess is laorious%.
There are three )eights, one for ea!h of the three strata "small, medium, large%. >a!h is
given y *
s
, or simply the sampling interval for ea!h stratum. >a!h #$C in a given PS1
re!eives the same )eight for that PS1.
Some sample #$C )ill need to have their )eights modified y a fa!tor to !orre!t for
multiple sleeping sites they )ill have used during the survey period. -n that !ase, the
final )eight is e4ual to *
s
multiplied y60t, )here t is the numer of sites he0she may have
used.
Combining results for estimation
'eighting )ill almost surely e re4uired )hen the three !omponents are !omined in
a lin.ed survey for estimates of total #$C or their !hara!teristi!s. This is e!ause ea!h
!omponent entails an independent sampling frame and design, and the proailities of
sele!tion )ill ne!essarily differ among the three !omponents. The )eights are !al!ulated
separately, in a!!ordan!e )ith the des!ription aove, for ea!h !omponent and simply
applied to the data files to produ!e the !omined estimates.
2A
Aendi2 C, Ste8ise Procedures for S")%e Se%ection in OVC Sur$e's
This appendi2 provides step3y3step pro!edures for implementing the sampling
methodologies des!ried and proposed in the te2t. :ess attention is devoted to household
sampling than to the other !omponents e!ause it is re!ommended that the household
!omponent of #$C surveys e done through the method of appending a rider
4uestionnaire module to an e2isting household survey, the design of )hi!h is more or less
pre3determined.
The reader is referred to the main te2t to find the !onte2t and the !onditions under
)hi!h the various methodologies given here, in!luding the appropriate use of lin.ed
surveys of the three !omponents versus stand3alone surveys, should e applied.
Household Component
&e!ause the rider approa!h is the re!ommended strategy to e used for the household
!omponent, the steps involve sampling3related a!tivities in the !onte2t of the host survey.
ST>7 6 * ;evie) sample design of host survey for adheren!e to measurement o(e!tives
of #$C survey.
ST>7 2 * >2amine design to ensure that proaility methods are used at all stages of
sele!tion.
ST>7 3 * Che!. that sample si/e is large enough to otain re4uired numer of
households "see Tales 6 and 2%1 note, espe!ially, the larger sample si/es that are needed
for oy3girl estimates as separate, e4ually reliale domains.
ST>7 = * >nsure that !luster si/es are toleraly small su!h that sample design effe!t for
#$C estimates is aout 2.0 or less.
ST>7 D * >2amine design for !ompleteness of !overage and e prepared to des!rie in
survey do!umentation any important groups that may not e !overed y survey "for
e2ample, nomadi! households, oat people, areas e2!luded for se!urity reasons, et!.%.
ST>7 G * >2amine design to determine if any su3populations, su!h as uran areas, are
over3sampled and e prepared to ad(ust sampling a!!ordingly for #$C rider.
ST>7 B * 'or. )ith host survey team to ensure that !hild3headed households are not
e2!luded from !overage.
ST>7 A * 'or. )ith host survey team to identify all first stage sampling units "host
survey PS1s 3 villages, uran neighorhoods, et!.%5
"a% Administer spe!ial 4uestionnaire to head of ea!h sample village or
neighorhood to find out )hether orphans or other !hildren outside family
2<
!are may reside in !ommunal living arrangements rather than in parti!ular
households.
"% -dentify all su!h !hildren under 6A years old and tally them ")ith PS1
)eight% for prevalen!e estimation1 intervie) them "or su3sample% for
detailed !hara!teristi!s.
ST>7 < * Administer s!reener 4uestionnaire as rider to host survey, 8-s there any !hild
under 6A living in this household )hose mother is deadN9 E;epeat for 8father9F. ,ote
that the screener will identif& orphans but not other "ulnerable children who ma& be
li"ing in households)
ST>7 60 * ?or all households identified as having at least one orphan, re!ord on the
s!reener 4uestionnaire appropriate identifi!ation fa!ts "PS1 numer, provin!e, distri!t
and other administrative !odes, address0lo!ation of household, name of household head%,
in order to e ale to find the household again for the #$C follo)3up survey.
ST>7 66 * Tally the s!reener 4uestionnaires )ith their appropriate )eighting fa!tors
from the host survey "see Appendi2 &% to estimate prevalen!e or si/e of #$C population.
ST>7 62 * ?or studies of #$C !hara!teristi!s, !ondu!t follo)3up intervie)s using #$C
module5
"a% 'ith all households )here orphans are found through rider 4uestionnaire
"% #r, systemati! su3sample of those households if their numer is
sustantially larger than e2pe!ted.
ST>7 63 * -f su3sampling is used in step 62, re!ord su3sampling fra!tion so that proper
)eighting !an e implemented at data analysis phase of operations.
ST>7 6= * Maintain detailed do!umentation of the sampling operations and pro!edures
that are used, noting espe!ially any implementation features that depart from the design.
Institutional Component
ST>7 6 * +evelop sampling frame y !onta!ting government offi!ials, NG#s, religious
leaders and others )ho are .no)ledgeale aout institutions and other group living
4uarters )here #$C reside.
ST>7 2 * Ma.e appropriate !onta!ts at national level su!h as ministries of health, so!ial
)elfare or others that deal )ith youth1 national head4uarters of NG#s, national
head4uarters of religious groups.
ST>7 3 * Ma.e similar !onta!ts also at provin!e level and at distri!t level.
ST>7 = * Through !onta!ts aove, !ompile !omprehensive list of all institutions and
group living 4uarters that house #$C.
30
?ollo)ing is a suggested 4uestionnaire, ?orm GJ6, to use )hen !onta!ting
government offi!ials and other e2perts for this purpose.
I%%ustr"ti$e For) +, I%%ustr"ti$e D"t" Co%%ection For) to Co)i%e Grou 6u"rters
?orm GJ6 * Compilation
of :ist of -nstitutions and
other Group :iving
Juarters Ono)n to ,ouse
#rphans, other $ulnerale
Children
E*ntroduce &ourself and
e4plain the purpose of the
sur"e&F.
3dminister this form to officials of the ministries of health- social ser"ices
and others working with &outh< to religious leaders and to ,2Os that work
with &outh) Complete a separate form for each contact) Continue on
additional forms as necessar&)
This form re0uires obtaining an estimate of the number of OVC under the
age of #======' in each residential facilit&) The estimate ma& be an
appro4imation if e4act figures are not readil& a"ailable from the
respondent)
'e must !ompile a list of all the orphanages, institutions and other group living arrangements )here
orphans and other !hildren outside traditional family !are are .no)n to reside so that )e may !ondu!t the
survey.
6. Are there any
government3run orphanages
in the "!ountry, !ity,
provin!e, distri!t%N
:ist them )ith name, address0lo!ation.
6. PPPPPPPPPPPPPPPPP
2. PPPPPPPPPPPPPPPPP
3. PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
2. Are there (uvenile !enters
for youth in troule )ith the
la)N
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
3. Are there institutions for
adults, su!h as (ails, that
may house youth in!luding
#$C as )ellN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
=. Are there informal group
homes that provide foster
!are for #$C that you .no)
aoutN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
D. Are there any orphanages
that are run y religious
organi/ationsN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
G. Are there military
arra!.s or !amps )here
youth in!luding orphans
may liveN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
B. Are there any other types
of institutions for housing
young people in!luding
#$C that you .no) aoutN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
A. E3sk especiall& in cities-
rural communitiesF Are
there 8drop3in9 !enters
)here orphaned !hildren
my liveN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
Numer of #$C
6.PPPP
2.PPPP
3.PPPP
?orm GJ6 is illustrative and )ould e adapted as ne!essary to fit a parti!ular
!ountryIs !onditions. The age group of #$C is not spe!ified in the form, sin!e it may
36
vary from !ountry3to3!ountry from under 6A to under 6D. &ut in adapting ?orm GJ6 for
your !ountry the defined age group should e spe!ified in the form so that respondents
!an ans)er relevantly.
EOVC Pre$"%enceF
ST>7 D * -f !ompilation of !omplete list results in less than 6D0 institutions prepare to
!ondu!t !omplete !ensus of #$C to estimate si/e, or prevalen!e, of #$C population.
"a% $isit every institution.
"% #tain only asi! information from ea!h young person, su!h as name, age, gender
and )hether one or oth parents are still alive.
ST>7 G * -f !ompilation of !omplete list results in 6D0 or more institutions, prepare to
sele!t sample of =03D0 institutions. This step would be repeated separatel& for bo&s and
girls in countries where the& are institutionalized separatel&)
"a% Assign institutions to small, medium or large stratum on asis of their
estimated si/e in terms of numer or #$C "see Chart 2%.
"% 7rior to sample sele!tion arrange institutions in geographi! fashion )ithin ea!h
stratum y uran3rural and y provin!e.
"!% ?igure sample sele!tion interval, * * per Chart 2 * for ea!h stratum to yield =0
"or D0% institutions.
"d% ?or ea!h stratum, sele!t a random start using a random numer tale and sele!t
6 in every * of the institutions. ?or e2ample, if there are 6A0 institutions in the
small stratum and you de!ide to sele!t 6A of them, then * L 60 and the random
start )ould e any randomly !hosen numer et)een 6 and 60.
"e% $isit every sampled institution ut !ondu!t census of #$C in those
institutions.
"f% #tain only asi! information from ea!h young person, su!h as name, age,
gender and )hether one or oth parents are still alive.
"g% 7repare prevalen!e estimates using, as )eights for ea!h stratum, the sampling
interval, *1 apply appropriate )eight to every #$C in a given institution
a!!ording to the )eight for that institution. ?or e2ample, if institution A is
sampled from the small stratum at the rate of 6 in 60, all #$C in that
institution are in!luded in the !ensus and ea!h one re!eives a )eight of 60 to
produ!e the prevalen!e estimates.
EOVC c!"r"cteristicsF
ST>7 B * +e!ide on sample si/e * minimum of 600 #$C for lin.ed survey and =00 for
stand3alone #$C survey "see Chart 6%.
ST>7 A * +oule the sample si/e if estimates are )anted for male3female separately.
ST>7 < * 'hen the total numer of institutions is so fe) "under 6D0% that all of them are
in!luded, prepare to sele!t a systemati! sample of #$C.
"a% $isit every institution.
"% #tain a roster of all the #$C at ea!h institution from the offi!ial in !harge, )ith
only asi! information for ea!h young person, su!h as name, age, gender.
32
"!% Arrange the rosters "male3female separately )hen re4uired% in a !ontinuous
stream "as though from a single sour!e%.
"d% Cal!ulate the sampling interval, *, e4ual to ,9n, )here , is the total numer of
#$C in all the institutions !omined and n is the sample si/e "for e2ample, 600 or
=00%. Cal!ulate separate intervals if male3female are to e separate domains. ?ind
a random start in a tale of random numers et)een 6 and *. An illustration
follo)s of systemati! sampling "the illustration is for only 20 male #$C in a
population of B2, to enale it all to fit on t)o pages%.
C!"rt 9, I%%ustr"tion of S'ste)"tic Se%ection of 34 M"%e OVC
N")e of F"ci%it' M"%e OVC No, Se%ection
R"ndo)
St"rt : 4,9
N")e of F"ci%it' M"%e OVC No, Se%ection
-nstitution 6 Name 06 Name 3B
Name 02 Name 3A
Name 03 Name 3<
Name 0= =.0 Name =0 =0
Name 0D -nstitution G Name =6
-nstitution 2 Name 0G Name =2
Name 0B Name =3
Name 0A B.G Name == =3.G
Name 0< Name =D
Name 60 Name =G
Name 66 66.2 Name =B =B.2
Name 62 -nstitution B Name =A
-nstitution 3 Name 63 Name =<
Name 6= Name D0
Name 6D 6=.A Name D6 D0.A
Name 6G Name D2
Name 6B Name D3
Name 6A 6A.= Name D= D=.=
Name 6< Name DD
Name 20 Name DG
Name 26 -nstitution A Name DB
Name 22 22 Name DA DA
-nstitution = Name 23 Name D<
Name 2= Name G0
Name 2D Name G6
Name 2G 2D.G Name G2 G6.G
Name 2B Name G3
Name 2A Name G=
Name 2< 2<.2 -nstitution < Name GD GD.2
Name 30 Name GG
Name 36 Name GB
Name 32 Name GA
-nstitution D Name 33 32.A Name G< GA.A
Name 3= Name B0
Name 3D Name B6
33
Name 3G 3G.= Name B2 B2.=
"e% >a!h #$C sele!ted is then eligile for the detailed intervie).
ST>7 60 * 'hen the numer of institutions is large enough to re4uire sampling, that is,
6D0 or more, prepare to sele!t a t)o3stage sample.
"a% 7rior to sample sele!tion arrange all institutions in geographi! order y
provin!e and )ithin provin!e y uran3rural.
"% Cumulate the measures of si/e !onse!utively and sele!t a systemati!, pps
"proaility proportionate to si/e% sample of =0 to D0 institutions. An
illustration is given in Chart D.
3=
C!"rt ;, I%%ustr"tion of S'ste)"tic pps Se%ection of 9; Institutions# Ot!er Grou
6u"rters
Administrative Area, Name of ?a!ility Measure of Si/e Cumulative Sample Sele!tion
"no. of #$C%
7rovin!e 06 @ran ?a!ility 06 22 22
?a!ility 02 B 2<
?a!ility 03 < 3A 36.6
?a!ility 0= 23 G6
?a!ility 0D 62 B3
?a!ility 0G G B<
?a!ility 0B 60 A< AB.2
?a!ility 0B 6G 60D
;ural ?a!ility 06 20 62D
?a!ility 02 20 6=D 6=3.3
?a!ility 03 A 6D3
?a!ility 0= D 6DA
7rovin!e 02 @ran ?a!ility 06 2D 6A3
?a!ility 02 6B 200 6<<.=
?a!ility 03 60 260
?a!ility 0= < 26<
?a!ility 0D 62 236
?a!ility 0G A 23<
?a!ility 0B < 2=A
?a!ility 0A G 2D=
?a!ility 0< 22 2BG 2DD.D
;ural ?a!ility 06 6= 2<0
?a!ility 02 20 360
?a!ility 03 26 336 366.G
7rovin!e 03 @ran ?a!ility 06 6G 3=B
?a!ility 02 2B 3B= 3GB.B
?a!ility 03 26 3<D
?a!ility 0= 23 =6A
?a!ility 0D B =2D =23.A
?a!ility 0G B =32
?a!ility 0B 66 ==3
?a!ility 0A 62 =DD
Q Q Q Q
Q Q Q Q
Q Q Q Q
7rovin!e 20 @ran ?a!ility 06 6B 2==B 2==3.=
?a!ility 02 6A 2=GD
?a!ility 03 62 2=BB
?a!ility 0= 60 2=AB
;ural ?a!ility 06 A 2=<D
?a!ility 02 < 2D0= 2=<<.D
?a!ility 03 62 2D6G
?a!ility 0= A 2D2=
SAM7:> 7A;AM>T>;S5 Sele!tion -nterval L 2D2=0=D, #; DG.61 ;andom Start L 36.6
Note that the same fa!ility may e sele!ted more than on!e if its measure of si/e e2!eeds the sampling
interval, *. -f that should happen, the numer of #$C to sele!t in su!h a fa!ility is doule for t)o
8hits,9 triple for three 8hits9 and so forth.
3D
"!% -n ea!h sample fa!ility prepare to sele!t a fi2ed numer of #$C * say D males and
D females * ut the numer should e e4ual to n divided y the numer of institutions
sele!ted "for e2ample, if n is =00 and =0 fa!ilities are !hosen, then the fi2ed sample
si/e in ea!h fa!ility is 60%. 'hen there are fe)er #$C than the re4uired fi2ed
numer in a given fa!ility, intervie) all of them.
"d% #tain a roster of the #$C in ea!h sample institution and sele!t a systemati!
sample from ea!h one in the same )ay as des!ried aove in Steps < "a% through "e%.
ST>7 66 * Maintain detailed do!umentation of the sampling operations and pro!edures
that are used, noting espe!ially any implementation features that depart from the design.
Homeless Youth Component
ST>7 6 * +e!ide on important survey, as opposed to sample, parameters5 pre!ise
operational definition of street !hild or homeless !hild1 !overage only in large !ities
)here street !hildren !on!entrate or else)here.

ST>7 2 * 7repare to !onstru!t sampling frame of time3lo!ation PS1s.
ST>7 3 * +evelop sampling frame y !onta!ting government offi!ials, NG#s, religious
leaders and other .ey informants )ho are .no)ledgeale aout lo!ations )here homeless
youth are .no)n to sleep.
ST>7 = * Ma.e appropriate !onta!ts at national level su!h as ministries of health, so!ial
)elfare or others that deal )ith youth1 national head4uarters of NG#s, national
head4uarters of religious groups.
ST>7 D * Ma.e similar !onta!ts also at provin!e level and at distri!t level "or !ity
government level if survey is restri!ted to !ities%.
ST>7 G * Through !onta!ts aove, !ompile !omprehensive list of all lo!ations )here
#$C sleep.

?ollo)ing is a suggested 4uestionnaire, ?orm S6, to use )hen !onta!ting government
offi!ials and other e2perts for this purpose. The form is illustrative and )ould e adapted
as ne!essary to fit a parti!ular !ountryIs !onditions.
3G
I%%ustr"ti$e For) 3, I%%ustr"ti$e D"t" Co%%ection For) to Co)i%e S%ee Sites
?orm S6 * Compilation of :ist of Sleep
Sites for ,omeless Routh
>*ntroduce &ourself and e4plain the
purpose of the sur"e&?)
3dminister this form to officials of the ministries of health-
social ser"ices and others that work with &outh< to religious
leaders and to ,2Os that work with &outh) Complete a
separate form for each contact) Continue on additional forms
as necessar&)
'e must !ompile a list of all the pla!es )here street !hildren and other homeless youth are .no)n to sleep,
so that )e may find them to !ondu!t the survey.
6. Are there any shelters
6<
for homeless
persons, in!luding young people, in the
"!ountry, !ity, provin!e%N
:ist them )ith name, address0lo!ation.
=. PPPPPPPPPPPPPPPPP
D. PPPPPPPPPPPPPPPPP
G. PPPPPPPPPPPPPPPPP
2. Are there aandoned uildings used y
youth for sleepingN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
3. 'hat aout other en!losed stru!tures
su!h as shopping malls, transportation
terminalsN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
=. Any other en!losed stru!tures that you
.no) aoutN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
D. 'hat aout outsideN @nder ridgesN :ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
G. @nderground, in!luding se)age
tunnelsN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
B. S4uatter areasN :ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
A. #n the street, espe!ially out3of3the3
)ay spotsN
:ist them )ith name, address0lo!ation.
6.PPPPPPPPPPPPPPPPP
2.PPPPPPPPPPPPPPPPP
3.PPPPPPPPPPPPPPPPP
6<
,omeless shelters are in!luded as 8sleep sites9 for street !hildren on the grounds of their transient nature,
as opposed to eing !onsidered group 4uarters or institutions.
3B
ST>7 B * #n asis of responses to ?orm S6, !reate list of time3lo!ation PS1s, using G3
hour segments for time dimension.
20
An e2ample of four of the PS1s might e as
follo)s5
@nder City &ridge * G a.m. to noon
@nder City &ridge * noon to G p.m.
@nder City &ridge * G p.m. to midnight
@nder City &ridge * midnight to G a.m.
ST>7 A * 7repare to !ondu!t field )or. to otain measure of si/e for ea!h PS1 in the
entire universe of PS1s. ?igure )or.load in terms of numer of PS1s in universe and
field staff availale. Note that ea!h PS1 may re4uire G person3hours of )or. sin!e staff
)ill have to remain at site for G hours to otain !ount.
-llustration5 Suppose there are 300 PS1s and you intend to use 6D field personnel.
Then, 20 PS1s "300 S 6D% )ould e assigned per person1 the total time re4uired,
per )or.er, is 620 hours "20 2 G%, or the e4uivalent of 3 person3days for ea!h.
Altogether, it )ould e 6D 2 3, or =D person3days of )or. "in this illustration% to
!ompile the measures of si/e. Alternatively, the numer of field )or.ers re4uired
!an e !al!ulated instead through a !orresponding pro!ess if the numer of
person3days is fi2ed.

ST>7 < * $isit ea!h PS1
(
to otain appro2imate !ount of homeless persons present or
arriving during time interval1 do not attempt to s!reen for #$C status or age group "this
is done later%.
ST>7 60 * >stalish 3 strata of PS1s a!!ording to si/e !ategories ased on their
measures of si/e * small, medium and large1 see Chart 3 for rule3of3thum, though e2a!t
si/e !ategories )ould e de!ided differently for ea!h !ountry.
ST>7 66 * +e!ide upon target sample si/e. See Chart 65 minimum of =00 #$C for
!hara!teristi!s in a stand3alone survey, 600 for prevalen!e in a lin.ed survey.
ST>7 62 * Multiply sample si/e y = to determine numer of homeless persons to e
s!reened "only aout 2D per!ent of )hi!h are e2pe!ted to e #$C%. ,ote that in the
particular application for a countr& the ratio of homeless to OVC needs to be more
carefull& determined- rather than :ust assuming it is @) *n that case- the multiplier would
be different of course)
ST>7 63 * Cal!ulate average si/e, T, of PS1, in terms of numer of homeless persons.
This e4uals the total estimated measure of si/e over all PS1s divided y the numer of
PS1s. ?or e2ample, if total numer of homeless persons is A000 and there are 320 PS1s
in universe, then T is 2D "A000 S 320%.
20
Alternatively, G PS1s of =3hour time intervals may e !onstru!ted.
26
Note that this implies visiting the same location = times, in order to estalish the measure of si/e for ea!h
time interval.
3A
ST>7 6= * Cal!ulate numer of PS1s to sele!t to rea!h sample si/e )anted "Step 66%.
?or e2ample, if =00 #$C is the sample si/e, then G= PS1s )ould e needed, !al!ulated
as =00 S "0.2D 2 T%.
ST>7 6D 3 Cal!ulate the average PS1 si/e in ea!h stratum "same as Step 63 ut
separately y stratum%.
ST>7 6G * +istriute the numer of PS1s to sele!t from ea!h stratum a!!ording to the
average si/e, from Step 6D, and Chart 3.
-llustration5 suppose the values of T for ea!h stratum are =, 6G and =0,
respe!tively, for small, medium and large. ;e!all this is the average numer of
homeless persons, and the average numer of #$C is e2pe!ted to e aout 2D
per!ent of those numers, or, respe!tively, 6, = and 60. Suppose further that there
are 6G0 PS1s in stratum 6 "small%, 600 in stratum 2 "medium% and G0 in stratum 3
"large%. -f the sample si/e is =00, a plausile sample s!heme )ould e to sele!t 6
in 66 of the stratum 6 PS1s "yielding 6D or 6=%, 6 in = of the stratum 2 PS1s
"yielding 20% and one3half the stratum 3 PS1s "yielding 30%. The total numer of
PS1s )ould e GD "or G=% and the e4pected numer of #$C )ould e e4ual to
E"6D 2 6% U "20 2 =% U "30 2 60%F L 3<D.
ST>7 6B * Sele!t the PS1s using systemati! sele!tion1 in ea!h stratum !hoose a random
start et)een 0.6 and the sampling interval and su!!essively add the sample interval to
designate the sele!ted PS1s. The methodology for systemati! sele!tion is similar to that
illustrated in Chart = of Step <"d% of the -nstitutional Component se!tion.
ST>7 6A * 7repare to !ondu!t the intervie)s in the sele!ted PS1s.
ST>7 6< * Set the time interval at G hours "or )hatever interval that )as used to estalish
the PS1s%
22
that intervie)ers are to remain at ea!h sele!ted site for intervie)ing. &e sure
that it is the same for every PS1.
ST>7 20 * 7ost an intervie)er "or team of intervie)ers% in ea!h sample PS1 site.
;ememer that the PS1s span all hours of the day and one3fourth of them are from
midnight to G a.m., so some intervie)ers must )or. in the middle of the night. The
intervie) team must remain at the site for the entire G3hour interval in order to ensure
a!!urate estimation later during the analysis stage.
Step 26 * S!reen the o!!upants of the sleep site for #$C status, espe!ially age,
eliminating those )ho are 6A and older. This may re4uire a s!reening 4uestionnaire for
persons )ho !annot e eliminated as adults !learly from oservation. 7erform this
a!tivity not only for o!!upants already present ut others )ho sho) up during the time
interval.
22
-f time3lo!ation PS1s, for e2ample, are estalished in =3hour segments, then the interval of time that
intervie)ers must e stationed for the survey at the sample sites is also = hours.
3<
ST>7 22 * Administer detailed #$C 4uestionnaire to all #$C identified in the PS1.
Note that the te2t re!ommends this 4uestionnaire in!lude 4uestions to as!ertain )hether
the respondent slept in other lo!ations during the survey period * say, 8last t)o )ee.s9 *
and, if so, their lo!ations, in order to use in )eighting at the analysis stage. See
illustration in the te2t, repeated elo).
To illustrate5 -t is 4uite li.ely the same site )ill fall into sample more than on!e,
sin!e the site )ill !omprise = PS1s ased on time of day. Any youth a!!ustomed
to sleeping in the same pla!e has a good !han!e of sho)ing up at the lo!ation at
different time intervals, thus giving him0her multiple !han!es of sele!tion. 'hen
that o!!urs, the youth should e intervie)ed only on!e. A some)hat more
diffi!ult dupli!ation prolem to sort out !on!erns those youth that sleep in
different sites from one day to the ne2t. To over!ome this prolem )e need to
ta.e into a!!ount the length of the survey period * one )ee., t)o, et!. :et us
assume 2 )ee.s for the e2ample. -t is ne!essary to as. ea!h respondent ho)
many other pla!es he0she usually sleeps during a 23)ee. period. She )ould also
e as.ed to identify those lo!ations.
ST>7 23 * E'hen survey is lin.ed onlyF. As. respondent if there is a usual pla!e of
residen!e )here he0she normally sleeps, information that is re4uired in order to un3
dupli!ate )ith the household or institutional frame.
ST>7 2= * -n offi!e operation, mat!h sites mentioned y respondents from Steps 22323
against master list of sites that ma.e up the PS1 frame. ?or ea!h mat!h, the survey
)eight for that respondent is do)n3)eighted y the fa!tor, 60t, )here t is the numer of
mat!hing sites.
ST>7 2D * Maintain detailed do!umentation of the sampling operations and pro!edures
that are used, noting espe!ially any implementation features that depart from the design.
=0
Aendi2 D, References "nd Ot!er Re"din(
&i!ero, G., S. ;utstein and O. Cohnson "2003%, 8+imensions of the >merging #rphans
Crisis in Su3Saharan Afri!a,9 Social Science and 8edicine, DG"G%5 623D362=B.
;otterdam5 >lsevier.
Centers for +isease Control "2002%, .andbook for Conducting Aouth Bisk Ceha"ior
Sur"e&s) Atlanta5 C+C5
Chandrase.ar, C. and +eming, '. >d)ards "6<=<%, 8#n a Method of >stimating &irth
and +eath ;ates and the >2tent of ;egistration,9 Dournal of the 3merican Statistical
3ssociation- 6<=<5 "==%, 606366D. EThis arti!le is sour!e of !apture * re!apture theory as
applied to human populationsF.
?amily ,ealth -nternational "2000%, Ceha"ioral Sur"eillance Sur"e&s #CSS', Guidelines
for ;epeated &ehavioral Surveys in 7opulations at ;is. of ,-$, ;esear!h Triangle 7ar.5
?,-.
Green)ell, O. ?ern "2002%, 8A +emographi! 7rofile of ;)andaIs Children in ;esidential
Care.9 ;)anda5 @N-C>?.
Oalton, Graham "6<<3%, Sampling Bare and Elusi"e Populations) Ne) Ror.5 @N
Statisti!al #ffi!e.
Oish, :eslie "6<GD%, Sur"e& Sampling. Ne) Ror.5 Cohn 'iley.
Ma!ro -nternational -n!. "6<<G%, Eemographic and .ealth Sur"e&s #E.S' Sampling
8anual, +,S3--- &asi! +o!umentation G. Calverton, M+5 Ma!ro -nternational -n!.
Turner, Anthony G. "2000%, Chapter =5 8+esigning and Sele!ting the Sample,9 Chapter G5
8Condu!ting the ?ield)or.9 and Appendi2 B5 8Sampling +etails,9 End-Eecade 8ultiple
*ndicator Sur"e& 8anual, 8onitoring Progress Toward the 2oals of the %%F World
Summit for Children. Ne) Ror.5 @N-C>?.
Turner, A., ;. Magnani and M. Shuai "6<<G% 8A Not Juite as Jui!. ut Mu!h Cleaner
Alternative to the >2panded 7rogramme on -mmuni/ation ">7-% Cluster Survey +esign,V
*nternational Dournal of Epidemiolog&, 2D"6%5 6<A3203. :ondon5 -C>.
Turner, Anthony "6<<A%, 8Some Notes on $illage Sampling,9 Proceedings of the Doint
*3SS9*3OS Conference on Statistics for Economic and Social Ee"elopment.
A4uas!alientes, Me2i!o.
@NA-+S0@N-C>?0@SA-+ "2002%, Children on the Crink (FF(, Coint ;eport on #$C
>stimates and 7rogram Strategies. Geneva, Ne) Ror., 'ashington5 @NA-+S.

=6
@NA-+S "2000%, ,ational 3*ES Programmes / 3 2uide to 8onitoring and E"aluation)
Geneva, Ne) Ror., 'ashington5 ',#, @N-C>?, @SA-+.
@NA-+S0@N-C>? "2003%, ;eport on 8Te!hni!al Consultation on -ndi!ators
+evelopment for Children #rphaned and Made $ulnerale y ,-$0A-+S,9 23= April
2003 in Gaorone, &ots)ana. Ne) Ror.5 @N-C>?.
'orld &an. "2000%, Child ,eeds 3ssessment Tool Git, Tas. ?or!e for Child Survival and
+evelopment. 'ashington5 'orld &an..
=2

Das könnte Ihnen auch gefallen