Guidelines on Estimating the Size of

Populations Most at Risk to HIV

UNAI

DS/W

HO W

orki

ng G

roup

on

Glo

bal H

IV/A

IDS

and

STI S

urve

illan

ceFor further information, contact:World Health OrganizationDepartment of HIV/AIDS20, avenue Appia CH-1211 Geneva 27 Switzerland

E-mail: hiv-aids@who.inthttp://www.who.int/hiv/en

ISBN 978 92 4 1599580

WHO Library Cataloguing-in-Publication Data

Co-published by World Health Organization and UNAIDS. 1.HIV infections – epidemiology. 2.HIV seroprevalence. 3.Populations surveillance – methods. 4.Guidelines. I.UNAIDS/WHO Working Group on Global HIV/AIDS and STI Surveillance. II.World Health Organization. Dept of HIV/AIDS. III.UNAIDS.

ISBN 978 92 4 159958 0 (NLM classification: WC 503.41)

© World Health Organization 2010All rights reserved. Publications of the World Health Organization can be obtained from WHO Press, World Health Organization, 20 Avenue Appia, 1211 Geneva 27, Switzerland (tel.: +41 22 791 3264; fax: +41 22 791 4857; e-mail: bookorders@who.int). Requests for permission to reproduce or translate WHO publications – whether for sale or for noncommercial distribution – should be addressed to WHO Press, at the above address (fax: +41 22 791 4806; e-mail: permissions@who.int). The designations employed and the presentation of the material in this publication do not imply the expression of any opinion whatsoever on the part of the World Health Organization concerning the legal status of any country, territory, city or area or of its authorities, or concerning the delimitation of its frontiers or boundaries. Dotted lines on maps represent approximate border lines for which there may not yet be full agreement.

The mention of specific companies or of certain manufacturers’ products does not imply that they are endorsed or recommended by the World Health Organization in preference to others of a similar nature that are not mentioned. Errors and omissions excepted, the names of proprietary products are distinguished by initial capital letters.

All reasonable precautions have been taken by the World Health Organization to verify the information contained in this publication. However, the published material is being distributed without warranty of any kind, either expressed or implied. The responsibility for the interpretation and use of the material lies with the reader. In no event shall the World Health Organization be liable for damages arising from its use.

Des ign & layout : L’IV Com Sàrl, Le Mont-sur-Lausanne, Switzerland

Printed in

Guidelines on Estimating the Size of

Populations Most at Risk to HIVUN

AIDS

/WHO

Wor

king

Gro

up

on G

loba

l HIV

/AID

S an

d ST

I Sur

veill

ance

2

Table of contents

Acknowledgements 3

1Introduction 41.1 Purpose 41.2 Background 41.3 Terminology 41.4 SizeestimatesforunderstandingHIVsurveillance 51.5 Sizeestimatesforadvocacy 51.6 Sizeestimatesforprevention,careandtreatmentprogrammes 51.7 SizeestimatesforHIVprogrammeevaluation 61.8 Awordofcaution 61.9 Process:howtodevelopapopulationsizeestimate 7

2Preparetoconductpopulationsizeestimates,Steps1–4 82.1 Step1:Determinetheuseofthesizeestimate 82.2 Step2:Determinewhenthesizeestimatewillbeneeded 92.3 Step3:Definethepopulationandgeographicarea 9

2.3.1Definethepopulation 92.3.2Thinkaboutyourpopulation’sgeographicarea 102.3.3Improveyourdefinitionswithformativeresearchandmapping 10

2.4 Step4:Reviewavailabledata 112.4.1Whatdataareavailable? 112.4.2Whatprevioussizeestimatesareavailable? 12

3Chooseamethodthencollectdata,Steps5–6 133.1 Step5:Chooseamethodtocreateyourpopulationsizeestimate 13

3.1.1Accesstohiddenpopulations 133.1.2Accesstoclientsofinstitutions 133.1.3Overviewofthemethods 14

3.2 Methodsthatusedatacollectedfromthepopulationatrisk 143.2.1Censusandenumerationmethods 143.2.2Capture-recapturemethod 173.2.3Nominationmethods 223.2.4Multipliermethod 22

3.3 Methodbasedondatacollectedfromthegeneralpopulation 263.3.1Surveys 263.3.2Networkscale-up 27

3.4 Summaryofsizeestimationmethods 323.5 Guidanceonselectingmethodstoestimatethesizeofat-riskpopulations 333.6 Step6:Collectthedata 34

4Analyseanddisseminateresults,Steps7–10 354.1 Step7:Analysetheresults 35

4.1.1Samplingerrorandbiasinsizeestimates 354.1.2Howtouselocalstudyresultstoreachanationalestimate 364.1.3Usingmultiplesizeestimatestocreatethebestestimate 384.1.4Estimatingthepopulationsizeofmost-at-riskadolescents 39

4.2 Step8:Documenttheprocess 404.3 Step9:Disseminatetheresults 414.4 Step10:Usethesizeestimates 42

AppendixA.Usefuldatasourcesforsizeestimation 43

References 46

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

3

Acknowledgements

GlobalsurveillanceofHIVandsexuallytransmittedinfectionsisajointeffortoftheWorldHealthOrganization(WHO) and the Joint United Nations Programme on HIV/AIDS (UNAIDS). The UNAIDS/WHO WorkingGrouponGlobalHIV/AIDSandSTISurveillance,initiatedinNovember1996,isthemaincoordinationandimplementationmechanismforUNAIDSandWHOtocompilethebestinformationavailableandtoimprovethequalityofdataneededforinformeddecision-makingandplanningatthenational,regionalandgloballevels.

Theseguidelinesareanupdatetothepreviousdocument“EstimatingtheSizeofPopulationsatRisktoHIV:IssuesandMethods”whichwasproducedinJuly2003.

WHO and UNAIDS would like to thank Donna Stroup, Rob Lyerla, Mary Mahy and Carolyn Smith forpreparingthisdocument.TheGuidelinesweredevelopedinpartbasedonworkdonebyViriginaLooandTobiSaidelinpreparationforatrainingonsizeestimationfortheAsiaregioninJuly,2009.

SignificantcontributionswerealsoreceivedfromKeithSabin,LoriNewman,JesusM.GarciaCalleja,WHOheadquarters; Peter Ghys, Eleanor Gouws, Karen Stanecki, Rand Stoneburner, UNAIDS headquarters;EvelynKim,AbuAbdul-Quader,UnitedStatesCenters forDiseaseControlandPrevention;VirginiaLoo,TobiSaidel,PartnershipforEpidemicAnalysis;HeidiFrank,UniversityofCaliforniaSanFrancisco;membersof theUNAIDS/WHOWorkingGrouponHIVandSTISurveillanceandmembersof theSurveillanceandSurveysTechnicalWorkingGroup.

TheconceptsandmaterialpresentedinthispublicationwereinformedbysizeestimateworkshopsheldinBangkok,ThailandinJuly2009andBucharest,RomaniainOctober2009.

UNAIDS and WHO would like to acknowledge the contributions of the participants of these meetings,includingstaffofnationalAIDSprogrammesandexpertsinvariousfieldsrelatedtoHIVepidemiology.

4

1.1.PurposeMeasuring and understanding the impact and magnitude of the human immunodeficiency virus (HIV)epidemic presents many challenges. Yet without accurate measures and estimates of the impact andmagnitudeofHIV,itisimpossibleforcountriestocarryoutHIVprogrammeactivities,suchas:G Advocatingformost-at-riskpopulationsG PlanningandimplementingHIVprevention,careandtreatmentprogrammesG Evaluatingprogrammes.

EstablishingthesizeofpopulationsmostatrisktoHIVallowsepidemiologiststodevelopmodelswhichestimateandprojectHIVprevalence1orinformcountriesofthedistributionofHIVincidencewithintheircountry2.

UsethisguidelinetoconductpopulationsizeestimatestudiestomeasureandunderstandthepopulationsmostatrisktoHIVinyourcountry.Notethattheguidelinedoesnotcoverissuesaroundbehaviouralandbiologicalsurveillanceamongthesepopulations.RefertotheGuidelines on Surveillance on Most at Risk Populations and Second Generation Surveillanceinthissameseriesforadditionalinformation.

AParticipant’sManualandslidepresentationsfor traininghavebeendeveloped inconjunctionto theseGuidelinesandcanbefoundontheUNAIDSwebsite.AlternativematerialsforconductingtrainingonsizeestimatesinAsiaarealsoavailablefromPartnershipforEpidemicAnalysis.

1.2.BackgroundMost countries have developed surveillance systems for tracking HIV infection and the behaviours thatspreadHIV.However,countriesmaylackthecapacitytoestimatethesizeofpopulationswithbehavioursthatputthematincreasedriskforHIV.

Recognizingthis,aguidelineforEstimating the Size of Populations at Risk for HIVwasdevelopedin20033)

byFamilyHealthInternational,theImpactProject,theUnitedStatesAgencyforInternationalDevelopment,theJointUnitedNationsProgrammeonHIV/AIDS(UNAIDS),theWorldHealthOrganisationandtheUNDrugControlProgramme.

Thisdocumentupdatesthe2003publicationwithrecentlydevelopedmethodsandtechniques,includinghow to develop national estimates from local estimates. Country experience in using the methods ispresented.

1.3.TerminologyPopulations at increased risk, or most-at-risk, for HIV are often referred to as hidden or hard-to-reach.Thesepopulationsarecomposedof individualswhoengageinbehavioursthataresometimesillegalorstigmatizingsothesepopulationstendtoavoiddisclosure.

Populations most at risk to HIV are often reluctant to participate in activities or programmes that maypersonallyidentifythem,suchas:G HIVsurveillanceactivitiesG HIVprevention,careandtreatmentprogrammes.

1. Introduction

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

5

1.4 SizeestimatesforunderstandingHIVsurveillanceMost-at-riskpopulationsareofparticularimportanceforHIVsurveillance4, 5.KnowingthenumberofpeoplewithbehavioursthatputthematincreasedriskallowsepidemiologiststoestimatethefuturecourseoftheHIVepidemic.

TheUNAIDS/WHOWorkingGrouponGlobalHIV/AIDSandSTISurveillancehasidentifiedfourpopulationsofspecificimportanceforHIVsurveillance:G SexworkersG ClientsofsexworkersG PeoplewhoinjectdrugsG Menwhohavesexwithmen.

1.5.SizeestimatesforadvocacyToconvincepolicymakersandfundersoftheexistenceandmagnitudeofanypublichealthproblem,youneedtohaveagoodestimateofthesizeofthepopulationatrisk.Itiseasierforpotentialfunderstoneglecttheat-riskpopulationif:G dataareunavailableG thebasisoftheestimatesisnotclear.G inconsistenciesbetweenestimatesarenotexplained

Arguments to implementprevention,careand treatmentprogrammesaremorecompellingwhengoodestimates of sizes of at-risk populations are available. The estimates of size must be based on soundmethodsthatcanbereplicated.

Advocacyisimportantatdifferentlevels.Incountrieswithdecentralizedpublichealthsystemsanddecisionmaking, suchasChina, India, Indonesia,NigeriaandMexico, localgovernmentsmaysee limitations inusingnationaldatatoadvocateforandinfluencepublichealthactionintheirregion.Localinformationisneededtodevelopmoreappropriate interventionswhenepidemicsarediverseandvary fromregiontoregionwithinacountry.

1.6.Sizeestimatesforprevention,careandtreatmentprogrammesPlanningandimplementingprevention,careandtreatmenteffortsaremoredifficultwithpopulationsatincreasedrisktoHIVascomparedtointerventionsforthegeneralpublic.Governmentsmayfinditpoliticallychallengingtoinvestinservicesforpeoplewhoinjectdrugs,menwhohavesexwithmenandsexworkersand their clientsbecauseof thestigma toward thesegroups.Yetserving thesegroupshas thegreatestpotentialforcurbingtheepidemicinsomecountries.

Estimatesofpopulationsizeareneededtohelpwithdecisionsonhowresourcesshouldbeallocatedforbetterprogrammeplanningandmanagement.Forexample:G HIVprevalencedatamayshowthatinfectionamongmalesexworkersinacertainareais22percent,

whileitisonly11percentamongfemalesexworkers.Afirstinterpretationofthesedatamaysuggestthattwiceasmuchfundingshouldbetargetedtowardprogrammesformalesexworkersasforfemalesexworkers.

G However,ifyouthenlearnthatthisareahas5,000malesexworkersand50,000femalesexworkers,thenwecanestimatethattheareahas1,100malesexworkersand5,500femalesexworkersinfected.Ifmaleandfemalesexworkershave,onaverage,thesamenumberofclients,itwillbeappropriatetodedicatemorepreventionresourcestocommercialsexbetweenmenandwomenthantothatbetweenmenandmen.

6

PreventionofnewHIV infectionsrequiresprovidingservices tomost-at-riskpopulations. It isdifficult toplanforadequateservicesforaspecificpopulationifyouhavenoideahowmanypeopleareinvolved.Youmighthavethesequestions:G Howmanysexually-transmittedinfection(STI)screeningkitsareneededtoconductregularscreeningfor

allsexworkersinacommunity?G Howmanycleanneedlesareneededforaneedleexchangeprogrammeforthecountry’sdrugusers?G Howmanyoutreachworkersareneededifyourprogrammewishestocontact50percentofmenwho

havesexwithmenatleastonceamonth?

Countriesneedtodeterminewheretofocustheirfinancialresourcesandhowtocosttheirresponse.Todo this,anaccurateestimateof thesizeof thepopulation isneededtodetermine themagnitudeof theresponse6,7.

1.7.SizeestimatesforHIVprogrammeevaluationRecently,resourcesforHIVpreventionandriskreductionhaveshiftedawayfrompilotprogrammestowardlarger-scale prevention programmes. International donors expect measurable progress toward targets.Countriesable todocumentprogressaremore likely tobenefit fromfundingfrominternationaldonors.Documentingprogresswillinclude,inpart,accuratelyestimatingthesizeofclearlydefinedpopulationstoprovideameasureofHIVprevalence.

1.8.AwordofcautionHealthcare researchers assume that population size estimates will be used for a better public healthresponse.Rememberthatanumberofthesepopulationssharebehavioursthatareillegalorstigmatised.Sizeestimatesoftheat-riskpopulations(say,personswhoinjectdrugs)mayleadto:G unwantedorinaccuratereportinginthemediaorapunitiveresponsebylawenforcementG increasedstigmaanddiscrimination.

Takecarewiththedisseminationanduseoftheresultingsizeestimatesandthedatacollectedtocreatetheestimates.Takemeasurestoensuretheconstructiveuseofthedataifyouplantoestimatethesizeofapopulation.

RememberthatgoodpopulationsizeestimatesarenotsufficientformonitoringtheHIVepidemic.Thesedatamustbecombinedwithotherformsofsurveillancedatafrommostatriskpopulations.Similarly,sizeestimationshouldnotbeconsideredanintervention.Ifthereisnocommitmenttoprovideservices,donotwasteresourceswithrepeatedsizeestimationstudies.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

7

1.9.Process:howtodevelopapopulationsizeestimateFigure1.1belowshowsageneralprocessforestimatingthesizeofhiddenpopulations.Thetenstepsareshowninthreegeneralareas:prepare,chooseamethod/collectdatathenanalyse/disseminateresults.

Theremainderofthisdocumentcoversthetenstepsabove.Clearguidanceisprovidedonthedifferentmethodswiththeirstrengthsandweaknesses.G Prepare,Steps1–4,Section2G Chooseamethod/collectdata,Steps5–6,Section3G Analyse/disseminate,Steps7–10,Section4

Figure 1.1. Process for estimating the size of hidden populationsAdditional details provided later in this document

8

2.1.Step1:DeterminetheuseofthesizeestimateTheplanneduseoftheestimateinfluencesthemethodyouwillchooseinstep5.Forexample:G Ifyourpurposeistoestimatehowmanypeople injectdrugsinasinglecitysothatprogrammescan

providedrugtreatmentforall,onemethodwouldbeused.G IfyourpurposeistoprovideanestimateofnationalHIVinfectionorofthesizeofpopulationsat-risk,you

woulduseadifferentmethod.

Wewilllearnmoreaboutthosemethodsinthenextunit.Fornow,rememberthatpopulationsizeestimatesaremeanttoquantifytheproblem.Thatis,youaretryingtoaccuratelycount/estimatepopulationsize,notprovideaccess to themforprogrammesandservices.Theestimatealonewillnotsolve theproblemorprovideaccesstopopulationsathighestrisk.

Doyouhavesufficientresourcestomountanewdatacollectioneffort?Determinethisupfront.Ifyoudonothavesufficientresourcesandcanjustuseexistingdata,adjustyourplanonhowtopresenttheestimate.

2. Prepare to conduct population size estimates, Steps 1–4

Thefirstfourstepsoftheprocessaregeneralpreparationsyouwillundertakewhenyouareplanningtodoapopulationsizeestimate.Itmaybeusefultodeveloptheprotocolforyoursizeestimatesexerciseasyoucompletethesesteps.

Section 4 of this guideline provides advice on documenting your population size estimation exercise,includinghowtoaddtoyouroriginalprotocolsothatotherstudyteamsoryourownteamcanreproduceyourresults.Reviewthatsectionbeforeyoubeginyourestimationexercisesothatyouhaveaclearpictureofwhereyouaregoing.

Prepare to conduct the study

1. Determinetheuseofthesizeestimate

2. Determinewhenthesizeestimatewillbeneeded

3. Definethepopulationandgeographicarea

4. Reviewexistingsizeestimates

Figure 2.1. Prepare to conduct your population size estimate

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

9

2.2.Step2:DeterminewhenthesizeestimatewillbeneededWhenwillasizeestimatebeneededinthecomingyears?Findthisouttodeterminewhentoconductyourestimation.Asdescribedearlier,therecanbemultiplepurposesorusesforapopulationsizeestimate.Itmaybeneededmorethanonce.Forexample:G Considerthetimingofthenationalstrategicplanningprocess.Apopulationsizeestimateprovidescritical

informationduringtheseplanningperiods.G Sizeestimatesarenecessaryfornationalcostingexercisesandjustifyingfundingrequests.G Carryoutasizeestimatetoinformprogrammeandsurveillanceefforts.G CoordinateyoursizeestimationplanswithplannedHIVmonitoringandevaluationactivities.G Determinehouseholdsurveyschedules.

Thetimingoftheseactivitieswillhelpyoudeterminewhentoconductapopulationsizeestimation.Trytodoanewestimationeverytwoyearsbecausepopulationsizechangesovertime.

2.3.Step3:DefinethepopulationandgeographicareaDefiningthepopulationandgeographicareaposesanumberofchallengeswhicharedescribedbelow.Itwillbeimportantthatallstakeholdersagreeontheseaspectsbeforemovingforwardwithyourestimation.

2.3.1. Define the populationOften, the most difficult problem in population size estimation is defining the population. For HIV, thegeneralconcerniswithpeoplewhosebehaviourputsthematincreasedriskofinfection.

Notallmembersofapopulationareatthesamelevelofrisk.Forexample,unprotectedanalsexbetweentwomeninamonogamousrelationshipislessriskythanamanhavingunprotectedanalsexwithamalesexworker.Transsexuals,gayorbisexualmenhavedifferentlevelsofrisk,yetallareinthecategoryofmenwhohavesexwithmen.

Inanotherexample,thepopulationofpeoplewhoinjectdrugsmayincludehealthcareworkerswhohaveaccesstosterileequipmentandastrongmotivationtoconcealtheirdruguse.ThispopulationhaslittleriskofHIVinfection.Youwouldprobablynotcapturethisgroupinyourpopulationsizeestimateofpersonswhoinjectdrugs.

Age is important when defining the population for a size estimate. Programme managers and policymakersoftenneedageinformationtodesigneffectiveprogrammes.Forexample,insomecountriesalargeproportionofsexworkersareunder18.Ifyourestimatedoesnotincludepersonsunder18,youwillhaveasignificantundercountofsexworkers.Ifpeopleunder18arespecificallynotincludedintheestimateforethicalreasons,statethisclearlyupfrontinyourplanandwhendisseminatingyourresults.

Thinkabouthowtoentirelycaptureyourpopulationofinterest.Forexample,malesmayacquiredrugsfortheirfemalepartners.Tofindthishiddenpopulationoffemaleswhoinjectdrugs,youwouldhavetoaskmaleinjectorswhethertheyprocuredrugsforafemalepartner.

Keep inmind that riskbehaviours changeover time.Thebehavioursmaybeaffectedby theeconomy,tourism, politics, disasters, seasonal migration and successful prevention efforts. This will affect theprevalenceofriskbehaviours.Sincebehavioursdefinethepopulation,thesizeofthepopulationwillchange.

Herearesomeexamplesofchanges:G Alocalelectionmayleadtoincreasedenforcementoflawsagainstsellingsex.Thesexworkersmaygo

toothercitiesorstopsellingsextemporarily.G An outreach programme targeting persons who inject drugs may change risk behaviours. This may

changethenumberofpersonsconsideredtobeinjectingdrugusers.

10

ThepopulationdefinitionshouldreflectthepopulationwhichisofinterestandshouldbedirectlyrelatedtothebehaviourthatresultsinthetransmissionofHIV.Forexample,G “menwhohavehadanalsexwithothermeninthepastsixmonths”wouldbeamoreaccuratedefinition

than“prisoners”G “sexworkerswhohavereceivedcashforvaginalsexinthepasttwoweeks”wouldbeamoreaccurate

definitionthan“womenfrequentingbars”G “men who have paid for sex in the past one year” would be a more accurate definition than “truck

drivers”

Thedefinitionshouldbeveryspecific(includingcriteriarelatedtofrequencyorhowrecentlytheyhavedonethebehaviour)andshouldberelevanttothepurposeofdoingthesizeestimate.

In your results, acknowledge populations you may not have captured either by the way the population was defined, or the limitations of the method used.

Itisoftennecessarytouseproxydefinitionsforatriskpopulationswhicharenotadistinctsocialgroup.A proxy definition uses a socio-demographic characteristic of a group, such as occupation, or placesassociatedwithriskbehaviourwhereriskgroupsareoftenfound(suchasmenatbeerhalls,malemigrantslivingindormitories,etc.).TheproxydefinitionisnotthecauseoftheincreasedrisktoHIV.Forexample,truckersareoftenusedasaproxydefinitionforclientsofsexworkers,becausesomestudiesshowahigherproportionoftruckersreportedbeingclientsofsexworkersthanmeninthegeneralpopulation.However,drivingatruckonitsownisnotariskforacquiringHIV.

Aproxydefinitionisalmostalwaysimperfect.Somepeoplewhomeettheproxydefinitionmaynotengageintheriskbehaviour,andviceversa,somepeoplewhohavetheriskbehaviourmaynotmeettheproxydefinition.

Theproxydefinitionisonlyuseful if thereisevidencethatahighproportionof individualsinthegrouppracticethehighriskbehaviourofinterest.Whenusingdatafromproxygroupstodescribetheepidemic,beclearwhyaproxygroupisadoptedanddocumentanylocaldatathatdemonstratetheproxygroupdoesdefineapopulationwithhigherriskbehaviours.

2.3.2. Think about your population’s geographic areaPopulation members at different locations may have different behaviours. These differences will beimportantforplanningthesizeestimationexercise.

Youmaywonderifthedifferencesamonglocationsmakeitimpossibletogeneraliseestimatesfromonecitytotherestofthecountry.Theansweris:youwillneedtoadjustyourestimatesbasedonthesedifferences.For example, you may need to stratify areas of high, medium and low prevalence of risk behaviour oraggregateestimatesfromkeyprovincestocreateanationalestimate.Laterinthisguideline,weshowhowtoaggregateestimates.

Whenwesayaggregate,wemeancombineseverallocalestimates.Thisisnotaseasyasitsounds.Localestimatestendtofocusonthetotalnumberofpeopleneedingservicesoveragivenplanningcycle,suchasayear.Butthetargetpopulationmaynotbeinthegeographicareaforthewholetime.Forexample:G Sexworkersmayonlysellsex inacity forsixmonthsbeforemoving toanothercitywhere theyare

considered“newstock”andcancommandhigherprices.Ifyouweretodoapopulationsizeestimationofthesesexworkers,yourannualtotalwouldbetwiceashighasthetotalatanyonetime.

G Sexworkerswhoworkinthecapitalonweekdaysmighttraveltoresortislandstoserveholidayclientsonweekends.Nationalestimatesbasedonasumofcapitalcity+resortislandsestimateswillcountthesamewomenmorethanonce,astheymovebackandforthbetweencapitalcityandresortislands.

The migration described above will likely result in inaccurate estimates. You will need to understandmigrationinthesexindustryandotherpredictorsofmobilitytomakegoodestimatesinsuchsituations.

2.3.3. Improve your definitions with formative research and mapping When you are preparing to do a size estimate, formative research should be used to improve yourunderstandingofthepopulationofinterest.Formativeresearchisresearchconductedduringtheplanningofyourstudytodeterminethebestwaystoreachthepopulation.Theresultsofthisresearchshouldhelp

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

11

youdecideonthepopulationdefinitionandthegeographicdefinition.

Formativeresearchofteninvolvesqualitativetechniquessuchasopenendedinterviews,observation,focusgroupdiscussions.Formativeresearchforsizeestimationmightinclude:G talkingtomembersofthepopulationatincreasedrisktoHIV,G talkingtopersonswhoprovideservicestothatpopulation,personswhoresideorworkinareaswhere

thepopulationcongregatesG observingthepopulationG readingexistingliteratureonthepopulation.

Byconductingformativeresearchitispossibletolearn:G whetherthepopulationisvisibleG whichsub-groupsofthepopulationarenotvisibleG wherethepopulationcongregatesG wherethepopulationreceivesservicesG whattimeofdaythepopulationisapproachablefordatacollectionG howthepopulationnetworksG whothegatekeepersaretothepopulationG howtheyreactandinteractwithpublicofficialssuchassurveyimplementersorpolice

Mostimportantly,thisinformationwillhelpyoudeterminepossibledatasourcesandsamplingmethods.

Formostsizeestimationexercises,geographic mappingwillalsobeuseful.Geographicmappingdescribestheuniverseofplaceswherethepopulationcongregates.Geographicmappingcanalsoprovidearoughestimateofthepopulationsizeandcharacteristicsofthelocationswherethepopulationcongregates.

Mappingisaprocessortoolandisnotasizeestimationmethodonitsown.Itisoftenusedwithcensusandenumerationbutisalsousefultohelpwiththeothermethodsdescribedinthisdocument.

Mappingisalsoessentialforplanningprogrammesandservicesforatriskpopulations.Interactionwithmembers of the population or persons familiar with the population will be useful when designing andimprovingHIVpreventionprogrammes.

Insummary,formativeresearchprovidesthesocialmappingforthesizeestimationexercise.Itwillhelpyoudefineanddescribethepopulationofinterest,itwillhelpyouunderstandthefactorswhichinfluencetheirbehavior,anddeterminethebestwaytoreachthepopulation.Whilegeographicmappingwillprovidethephysicaldescriptionandthecharacteristicsoftheareawhereyouwillbeworking.

2.4.Step4:ReviewavailabledataIn estimating the size of most-at-risk populations, different approaches or methods will be needed indifferentsituations.Youwillneedtoadaptyourplansandmethodstomakethemsuitableforyourcountryandpopulations.

2.4.1. What data are available?Many of the methods described later in this guideline rely on data taken from existing sources. Beforeundertakingapopulationsizeestimate,uncoverexistingdatafordifferentgeographicareas.

The2003versionofthisguidelineprovidedatooltohelpcountriesorganizedatasourcesaccordingtothepopulation.ThattoolisprovidedinthisdocumentasAppendixA.Thetoolisusefulforanyonethinkingaboutanestimationproject.

Findoutifthedataareaccessible.Donotassumethatonegovernmentagencyhasacomprehensiveideaaboutwhatdataarecollectedbyotheragencies.Forexample:G Aministryofpublichealthmaynothaveaccesstodatacollectedbylawenforcement.G Lawenforcementmayhesitatetosharenumbersbecausetheyfeelthatthepresenceofsexworkersor

drugusersreflectsafailureontheirpart.G Clinicsmaybereluctanttosharedataduetoconfidentialityconcerns.

12

Nowevaluatethedata.Aretheyappropriateforpopulationsizeestimation?G Dothedataallowidentificationofmembersoftheparticularpopulation?Forexample,prisonrecords

maynotallowidentificationofpeoplewhousedrugssincemanydrugusersinprisonsmaybechargedwithotheroffenses.Thisproducesunacceptablemissingdatathatwillbiaspopulationsizeestimates.

G Isthedataqualitygood?Forexample,workersinclinicsmaynotpursueriskfactorinformationforfearofalienatingpeopleintreatment.Thisproducesunacceptablemissingdatathatwillbiaspopulationsizeestimates.

G Dolegalorotherregulationspreventtheuseofexistingdatasourcesbypublichealthanalysts?

Ifexistingdataarejudgedtobeinadequateorinsufficient,considerplanningnewdatacollectionactivitiesinthecountry.Again,thinkaboutexistingactivities.Forexample,asurveycanberevisedwithadditionalquestionstoprovideinformationforpopulationsizeestimates.Youmightuse:G anationalcensusG anHIVsurveillanceactivityG anationalhealthstatussurvey.

2.4.2. What previous size estimates are available?Continue your preparations by studying existing population size estimates. When you review previousestimates,consider:G themethodusedG thedefinitionofthepopulationG theresultsG howtheestimatewasused.

Findoutwhatchallengeswereencounteredandovercomeinprevioussizeestimatesstudies.Trytothinkofwaystoavoidtheseissuesinfuturestudies.

Examinethesourceofexistingestimates.Evaluatepossibleconflictsofinterest.Itiscommonlyassumedthatcalculationsunderestimatethesizeofpopulations-at-riskbecauseofconcernsofstigma8.Butorganizationsmayhaveaninterest inhighestimatesforthepopulationstheyservebecauseahigherestimatemeansmorefundingforactivitiesandprogrammes.

Addressing ethical issuesThis is an essential part of your preparation if you plan to collect data on populations of personsatan increased riskofHIV. Inmanycases, theHIVsurveillance targetpopulationare, themselves,vulnerablepopulations.Collectionandstorageofdataonindividualsandtheirriskbehavioursmayplaceexcessriskofharmtothesepopulationsduetostigmatization,economiclossorlegalliability 9.Legalprotectionsofconfidentialitymaybechanging10.

G Give target populations special protection. Offer privacy during data collection. Ensure dataconfidentiality measures are in place. Respect for individual privacy creates a perception ofconfidentialitythatenhancesthecompletenessofreporting.

G Theethicalprincipalof“beneficence”,ordonoharm,extendstoprovidingbenefitstosurveillancesubjects.Thesebenefitsinclude,ataminimum,providing:• informationaboutHIVandAIDS• counsellingandtreatmenttotheextentpossiblewithlocalresourcesorparticipationinfuture

services.G Youngpeopleareparticularlyvulnerabletoexploitation,abuseandotherharmfuloutcomes.Ifyou

plantocollectdatafromadolescents,consultguidelinesforthis 11.G Includerepresentativesorlegalcounselfromthecommunityrepresentedbythetargetpopulation

whenyouareplanningasizeestimation.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

13

Table3.1showsthepossiblemethodsyoumightemploy.Eachmethodisthendescribedinthepagesthatfollowthetable.

3.1.Step5:ChooseamethodtocreateyourpopulationsizeestimateThesystematicapplicationofmethodswillencouragemoreusefulat-riskpopulationsizeestimatesforHIVworldwide.Bewaryofsimplebeforeandaftercomparisonsorreportingselectiveestimates.

3.1.1. Access to hidden populationsPopulationsatincreasedriskforHIVareoftenreferredtoashiddenorhard-to-reach.However,somehiddenpopulationsareeasiertoaccessthanothersasshownbelow.

Lesshiddensexworkers Morehiddensexworkers

Sex workers based in brothels: clientscometothebrothel.Yourstudywillbeconductedinthebrothel.

Call girls: theyhavenofixedlocationbutgowherevertheclientwantstomeet.Youwillnotbeabletogotoanyparticularlocationtofindthem.

Itisimportantthatyouunderstandandconsiderthelevelofaccesswhenyouarechoosingasizeestimationmethod.

3.1.2. Access to clients of institutionsInstitutionsthatcanprovideaccesstoindividualsortheirrecordsmayhaveclientsthatareatincreasedriskforHIV.Forexample:G drugtreatmentclinicsG emergencywards

3. Choose a method then collect data, Steps 5–6

Inthenexttwostepsoftheprocess,youwillchooseamethodologythatfitsyourtargetpopulationthencollectdata.

Figure 3.1. Choose a method then compile or collect data

Choose a method/collect data

5. Chooseamethodthatwillbeusedtodevelopyourpopulationsizeestimate

6. Compileexistingdataandcollectadditionaldataifneeded

14

G thejusticesystemG STIclinicsG publicinstitutionssuchasschools.

Accesstoindividualsthroughinstitutionsmaybeeasierthanlocatingthepopulationinthefield.Howeverrememberthatifyouaredrawingasamplefromaninstitutionthesamplemaynotberepresentative.Forexample:G Newerusersanduserswhoarenotdependentoncriminalactivitytosupporttheirdrugusewillnotbe

well-representedinjailsandthecriminaljusticesystem.G IDUintreatmentarenotcurrentlyinjecting,intheoryG Emergencywardswillover-representusersoftoxicsubstances.G STIserviceswillrepresentsexworkerswiththeriskiestbehaviours.

Yourknowledgeofinstitutionalpoliciesconcerningdataaccessandconfidentialitywillhelpyoutoeffectivelycollectdataandestimatethepopulationsize.

3.1.3. Overview of the methodsTable3.1introducestwocategoriesofmethods:G Methodsbasedondatacollectedfromanat-riskpopulationG Methodsbasedondatacollectedfromthegeneralpopulation.

Table 3.1. Categories of methods for estimating population size

Category1:Methodsbasedondatacollectedfromat-riskpopulation

Methodname Shortdescription

Census/enumeration, section 3.2.1 of this guideline

Census counts all members of the population.

Enumeration maps an area, counts a fraction of the population in selected areas, and inflates the value to create an estimate.

Capture-recapture, section 3.2.2 Calculates the total size of a population based on two independent captures of population members. The number of members captured in both samples is used to derive an estimate of the total number in the population.

Multiplier, section 3.2.4 Compares two independent sources of data for populations to estimate the total number in the population.

Category2:Methodsbasedondatacollectedfromthegeneralpopulation

Population survey, section 3.3.1 Includes questions on high risk behaviours in general population survey

Network scale-up, section 3.3.2 Includes questions on high risk behaviours of respondents’ acquaintances in general population survey

3.2.Methodsthatusedatacollectedfromthepopulationatrisk

3.2.1. Census and enumeration methodsThesimplestmethodsinthisguidelinearethecensusandenumerationmethods.

How these methods workCensusmethodstrytocounteveryindividualinanat-riskpopulation.Thisrequiresdevelopingacompletelistofplacesthatthepopulationmaycongregate.Forexample,youmightvisiteverybrothelinthecityorcountryandcollectinformationonthenumberofsexworkersbasedineachbrothel.Thiscounthastotakeplaceinaveryshortperiodoftime.Otherwise,sexworkersmovingbetweensitesmayleadtodoublecounting.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

15

Enumerationmethodsareverysimilar.Insteadofcountingeveryindividual:G Startwithasamplingframeorlist.Asamplingframeisthecompletelistofindividuals(orsites)from

whichasamplecanbechosen.G Chooseasampleofunits(suchasbrothelsorshootinggalleries)fromthatlistG Countonlytheindividualsinthosechosenunits.G Scaleupthenumbercountedaccordingtothesizeandstructureofthesampleframe.

Forexample,forasizeestimateofsexworkersinbrothelsinyourarea:G breakthepopulationintoquadrantsofacityordistrictswithinacountryG countthenumberofbrothelsineachareaG visitathirdofthebrothels(chosenrandomly)togetanaveragenumberofworkersperbrothelG multiplytheaveragenumberofworkersperbrothelbythetotalnumberofbrothelscounted.

Inthisexample,thesamplingframeisyourlistofallbrothelsorageographicbreakdownofthecountryorcommunity.

Bymappingbeforeyoubegin,youcanbesuretocoverallpossiblegeographicareas.

Anotherapproachwouldbetoseparate(stratify)thetotalpopulationofsexworkersbythevenueswheretheyselltheirservices(sostreet-based,brothel-based,callgirlsandsoforth).Whenyoumaporstratify,youensurethatyouhavecountedthefullpopulation.

Strengths and weaknessesCensus and enumeration methods are straightforward to calculate. They have the advantage of beingunderstoodbypolicymakerswhomaynotbeexperts inpublichealth statisticalor samplingmethods.Wherealistorsamplingframeexistsandwherethepopulationofinterestiswelldefinedandaccessible,thecensusmethodislesstimeandresource-consumingthanothermethods.

Thecensusmethoddoesnotperformaswellforhiddenpopulationsorforsituationswherethepopulationatriskisgeographicallydispersed.Inthesesituations,thecountcannotbecompletedinashortenoughtimetocompensateformigrationsoyoumaycountindividualstwoormoretimes.Yourestimatewillbetoohigh.Also,acensusisexpensivetoconduct.

Theenumerationmethodsharessomeofthestrengthsandweaknessesofthecensusmethod.However,since enumeration covers a fraction of the population, it usually requires fewer resources and is lessexpensivetoconduct.Withwell-trainedcommunityguidescoveringsmallareas,enumerationcanprovidebetteraccesstohiddenpopulations.Ifyouchoosetousetheenumerationmethod,assesswhetherdatacollectedfromtheregionsortypesofestablishmentsdifferinimportantways.

Ifthepopulationisveryhardtoreach,censusandenumerationmethodstendtounderestimatepopulationsizewhencomparedtoothermethods.However,ifthepopulationispoorlydefinedandpersonswhoarenottrulypartofthepopulationarecapturedinthecount,thepopulationwillbeoverestimated.Ifthecensusorenumerationisconductedoveraperiodoftimeindividualsmightbecountedtwiceleadingtoanoverestimate.

Detailed example of census and enumerationProgrammemanagersinacitysuspectthatthenumberoffemalesexworkershasrisenoverthepast15years.Theprogrammemanagersneedanaccurate,updatedestimate.TheyplantousetheestimatetoapplyforfundingfrominternationalAIDSorganizationsforoutreachandpreventionservicesforfemalesexworkers.Theyalsowanttoknowwhetherthereportedfemalesexworkerpopulationmightvarybytypeofestablishmentsothatfutureservicescanbetargetedcorrectly.

Twomethodswereusedtoderivetheestimate:Onewasacensusandtheotherwasacombinationofcensusandenumerationmethods.

Using the census method to estimate the number of female sex workers1. Fourtrainedstaffvisitedeveryentertainmentestablishmentwithinthestudyarea2. Thesestaffcountedtheindividualsexworkers.

16

3. Theyalsoaskedmanagershowmanysexworkerswereoutwithclientsornotworkingforhealthreasons.

Atotalof3,521sexworkerswereidentified.Theywerefoundasfollows:G 42%inkaraokecentresG 26%inhairsalonsG 7%street-basedG Theremainderwerefoundinmassagecentresandnightclubs.Thus, the census method suggested that the population size of sex workers who tended to gather in or near these kinds of establishments was 3,521.

Using the census and enumeration method to estimate the number of female sex workers1. Theinvestigatorsobtainedalistofallregisteredestablishmentsinthestudyarea.Therewere:

G27karaokecentresG31hairsalonsG16massagecentresG42nightclubs.

2. Field workers verified the list of establishments by visiting each of the establishments over athree-dayperiod.Theyfoundthatfiveofthekaraokecentreshadclosedandthreenewoneshadopened.

3. Fieldworkersvisitedarandomsampleof5karaokecentres,6hairsalons,4massagecentresand8nightclubsinoneday.Theycountedthenumberofsexworkerspresentandaskedaboutregularemployeeswhowereabsentforhealthorotherreasons.

4. Whileverifyingthelist,thefieldworkershadaskedtheownerstoreportonthenumberofsexworkers intheirestablishment.Thisprovidedacensusestimateofthenumberofsexworkersbasedonownersreports(similartothecensusinthefirstexample).

5. Thefieldworkersfoundthatinnocasedidthenumberofsexworkerstheycountedinthesamplevarybymorethan5%fromthetotalreportedbytheestablishmentowner.

6. Thefieldworkersrecordedthenumberofsexworkerswiththepossibleunderreportingmarginoferrorasfollows:GKaraokecentres:25centres,674workers,nodiscrepanciesGHairsalons:31hairsalons,723workers,discrepancies1-4%(730to752)GMassagecentres:16centres,512workers,discrepancies1-3%(517to527)GNightclubs:42clubs,1,227workers,discrepancies2-5%(1,251to1,288)

Thestudyteampreparedtwoestimatesfromtheenumerationdatacollectedandcomparedthetwoestimates.

Estimate and error calculation 1:Thisapproachmakestwoassumptions:G ThelocationssampledarearandomsamplefromallsuchlocationsG Thereisnovariabilityinthediscrepanciesorlikelihoodofreportingcorrectlybytypeofestablishment

(thatis,anyobservedvariationinunderreportingisduetochance).If we agree to these assumptions, then we need an overall estimate of underreporting for allestablishments.Ifwegobacktothediscrepanciesfortheindividualcentresvisited,wecanderiveanoverallestimateofunderreportingof3.2%(theaverageofthediscrepancies).Ourestimateisthesumofallnumbersreportedfromtheestablishments,alongwitharangeoferrorof3.2%:674+723+512+1,227=3,136witharangefrom3,036to3,236.

Estimate and error calculation 2: Thisapproachassumesthattheobservedvariationinthelikelihoodofreportingcorrectlyreflectstruedifferences.Inthiscase,weshouldapplythediscrepanciesforeachtypeofestablishmentseparatelythenaddtheestimates.G First,anupperlimitforallestablishmentscanbeobtainedbysummingtheupperestimates: 674+752+527+1,288=3,241G Similarly,alowerpointestimatecanbeobtainedfromaddingthelowerlimitfiguresfromalltypes

ofestablishments:674+730+517+1,251=3,172G Areasonablesingleestimatewouldbethemidpointofthetwo:(3,241+3172)/2=3,206

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

17

Itisrarethatoneestimateisseentobethecorrectestimate.Threedifferentestimateswerederivedinthisexample:theestimatefromthecensus,theestimateusingthefirsterroradjustmentandtheestimateusingtheseconderroradjustment.Eachprovidedadifferentvalue.Theinvestigatorsthenconsideredthestrengthsandweaknessesofeachmethodaswellastheassumptionstodeterminethe best estimate. Using an average of the methods, the investigators estimated that there wereapproximately3,350sexworkersinthecity.

3.2.2. Capture-recapture methodCapture-recapturetechniqueswerefirstusedin1662toestimatethepopulationofLondon.Itwasnotuntil150yearslaterthatLaPlacelaidoutthemathematicalformulationforcapture-recapture.Intheearly1900s,themethodwasadapted tostudywildlifepopulations 12, 13.Themethodhasbeenknownas theLincoln-Petersonestimator inwildlife,Chandra-Sekar-Demingmethod indemography,andsometimes thedual-systemestimator14.Othertermssometimesusedinclude“markandrecapture”and“captureandrelease”15.

How this method worksAsimpleexampleoftheprocedureisasfollows:G MapallthesiteswherethepopulationcanbefoundG Gotothesitesandtagallofthemembersofthepopulationatthesite(eithergivethemacardorsome

memorablegift).Keepacountofthepersonstagged.G Returntothesitesaweeklaterandretagallofthepersonsencountered.G Inthesecondvisit,count:

• memberswhowerecountedinthefirstsample• memberswhoarebeingcountedforthefirsttimeinthesecondsample.

Insituationswhereitisnotfeasibletovisitallofthesitesorallofthesitesarenotknownavariationofthismethodcanbeused.G Selectasampleofindividualsfromthepopulation.Ideallythesamplewillberandom,witheachmember

of thepopulationhavinganequalchanceofbeingselected.Thissamplecanbea listofsexworkersattendinganSTIclinic,orasurvey.

G Notepersonsselectedinsomefashion(forclinicattendeesyoumighthavetheirnamesorclinicpatientidentifiernumber,orsurveyrespondentsmighthavebeengivenacardorwillremembercompletingthesurvey).

G Collectasecondsampleatalatertime.Thesecondsampleshouldbeindependentofthefirstsample(eithersamplefromadifferentclinicorinstitutionorconductadifferentsurvey).

G Inthesecondsample,determine:• Howmanypeoplewerealsocountedinthefirstsample(sexworkerswhovisitedtheSTIclinicor

respondentsinterviewedinthefirstsurvey)• Howmanypeoplearebeingcountedforthefirsttimeinthesecondsample.

G Thenumberofindividualsobservedineachsampleandthenumberinbothsamplesisrecorded.

Thesenumbersareusedtoestimate population size.This isdonebymultiplyingthenumbercapturedinthefirstsampleby thenumbercaptured in the secondsampleanddividingbythenumber captured in bothsamples. (See below formathematicalexplanation.)

Figure 3.2. Illustrating the capture-recapture method

Notcaptured

Capturedinlist1only

Capturedinlist2only

Capturedinlist1andlist2

Notcaptured

List1

List2

18

Inat-riskpopulationestimates,twogeneralapproachestocapture-recapturehavebeenused:G Inthedirect contact approach,thefieldteamcontactsmembersofthetargetpopulation,forexamplethe

peopleinterviewedinthesurvey.Atalaterpoint,thefieldteamrevisitsthepopulationthroughadifferentindependentcontact(suchasasecondsurvey),andaskswhethertheyrespondedtothefirstsurvey.Ifso,theyarerecorded.

G Inthe no direct contact approach,theteamusesexistinglists(forexampleSTIclinicregistrationdataandabrothelregistry)todetermineifthesameindividualiscapturedonbothlists.

Therearesomeimportantassumptionsinthecapture-recapturemethod:

1.The population is closed. That is, the population available to be captured in the second sampling(recapture) includes exactly the same set of individuals as it did for the first. There is no in- or out-migration. This assumption is easily violated in studies of persons who inject drugs or sex workers,wherethereislargeturnover(peoplejoiningorleavingthepopulation)andoftenlotsofmovement.

Thechangeinpopulationbetweensample1andsample2canbecausedbyseveralthings,forexample:G peoplewhoinjectdrugswhoareincludedinthefirstsamplearemorelikelythanotherstoleavethe

populationbymovingaway,dyingorceasingtouse.G newdrugusersmightenterthepopulationG Peoplewhoinjectdrugswhoattendtreatmentprogrammesmaybemorelikelytoreducetheiruseof

drugsforaperiod.

2. Identifyinginformationiscollectedinbothsamples.Individualscapturedinbothsamplescanbematched.

3.Captureinthesecondsampleisindependentofsampleinthefirst.Thatis,peopleinthefirstsamplearenotmoreorlesslikelytobeincludedinthesecondsamplethanpeoplewhowerenotincludedinthefirstsample.

Ifbeingincludedinthefirstsampleincreasesaperson’schanceofbeingincludedinthesecondsample,thetotalpopulationwillbeunderestimated.Forexample,ifthestudyteamreturnstothesamestreetcornerorbrotheltorecapturesexworkers,certainsexworkerswillprobablybefavouredintherecapturesample.Techniqueshavebeendevelopedtodetectdependenciesbetweensamples16.

4.Eachpersoninthepopulationshouldhaveanequalchanceofbeingincludedinthesample.Thiswouldsuggestthatthesampleisrandom.

5.Capture-recaptureestimatesbasedonsmallsamplesortoofewmatchedindividualscanbemisleading17.Make sure there are enough members in the samples to produce meaningful results.

Table 3.2. Avoid violating assumptions in your capture-recapture

Assumption…andeffectifitisviolated

Howtoplanyourstudytoavoidviolatingtheassumption

The population is closed (there is no in or out migration). If this assumption is violated:G A decrease in the population size for the second

sample will produce an overestimate of the population size.

G An increase in the population size for the second sample will produce an underestimate of the population size.

G Shorten the time between samples.G Avoid sampling on unusual days, such as festival

days. G Carefully define the boundaries of selected sites. G Enlist community support.G Ensure your team understands the points above.

Matching is reliable (you can identify persons captured in both samples).

G If you do not identify a match, your estimate will be too high

G If you identify a match incorrectly, your estimate will be too low

Collect sufficient identifying data for each member of your samples so that you can match who has been captured before.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

19

Every member has an equal chance of being caught in a sample.

G If some members of the population are less likely to be included in the sample, your results will likely be an undercount.

G Before your study begins, investigate how the local community defines the population.

G Carefully select study sites.

The two samples are independent (selection in the first capture is not related to selection in the second capture).

G If not, your results will be biased.

G Use separate teams to collect each sample. G Use different informants/local guides for each

sample.

The sample size for each capture is large enough to be meaningful.

G If not, your results will not be precise.

Increase the size of the target population for both lists (or survey).

Based on table created by Donna Stroup, Data for Solutions.

Using the capture-recapture method with a random sample from a population of interestThebasiccapture-recapturemethodologybeginswitharandomsamplefromthepopulationofinterest18.Iftheassumptionsstatedhold,theestimatedpopulationsizeis:

€

N =MCR

,

whereN =EstimateoftotalpopulationsizeM=Totalnumberofpeople“captured”and“marked”onthefirstvisitC =Totalnumberofpeople“captured”and“marked”onthesecondvisitR =Numberofpeoplecapturedonthefirstvisitthatwerethenrecapturedonthesecondvisit(i.e.,includedinbothsamples).

InFigure3.3below,a2x2table:G ThetoprowofthetableincludesallpeoplecapturedinthefirstsampleG ThefirstcolumnincludesallpeoplecapturedinthesecondsampleG Thetotalnumber,N,includesallthoseinbothsamplesaswellasthosemissedbybothsamples.

Figure 3.3. Capture-recapture data from a random sample of a population of interest

Were they capturedin first sample?

Were they captured in second sample?

Yes

No

R b

c x

Yes No

M=R+b

C=R+c N=R+b+c+x

Tocalculatethesamplesizerequiredforthismethodyoushouldconsiderthefollowing.Thecombinedtotalofsample1andsample2shouldbelargerthanthetotalnumberexpectedinthepopulation(M+C>N)andthenumbercapturedinbothsamplesislargerthan7(R>7)19.Naturallyyouwillnotknowthetotalnumberofpeopleinthepopulation(N)soyouwillneedtomakearoughguess.

20

Strengths and weaknessesAsimpletwo-samplecapture-recapturemethodisrelativelyeasytouse.Thismethoddoesnotrequiremuchdata.Italsodoesnotrequirestatisticalexpertise.

Butthemethodreliesonassumptionsthatarehardtomeet:G twosamplesmustbeindependentandnotcorrelatedG eachpopulationmemberhasanequal,orknown,chanceofselectionG eachmembermustbecorrectlyidentifiedas‘capture’or‘recapture’G nomajorin/outmigrationmayoccurG thesamplesizesofeachcapturemustbelargeenoughtobemeaningful.

Detailed exampleAsimpleexampleofcapture-recapturedata.

Astudyteamisusingthecapture-recapturemethodtodeterminethesizeofanat-riskpopulation.1.Thestudyteammappedtheareatheywishtostudy,2.Onememberoftheteamgoestotheareato“capture”individuals:

G 50individualsaremarkedonthefirstvisit,M3.Asecondteammembergoestotheareatorecaptureindividuals:

G 25ofthoseindividualscapturedonthefirstvisitarerecapturedonthesecondvisit,R

Thefieldworkerconcludesthattheprobabilityofcapturingapreviously-markedindividualonthesecondvisitis:R / M=25/50=0.50.

Thefieldworkerassumesontheseconddaythatallindividualsintheactualpopulation,N,havethesamecaptureprobabilityastherecapturedindividuals.Thefieldworkerthinksonthesecondvisit, “I know that today I recaptured50%of thepeople Imarkedduringmyfirst visit.Today Iprobablyalsocaptured50%oftheindividualsthatIdidnotmarkonmyfirstvisit.Infact,todayIprobablycaptured50%ofalltheindividualspresentinthestudysiteregardlessofwhetherornotthoseindividualsweremarkedonmyfirstvisit.”Thiscanbeexpressedas:

CN

RM

=

Using the capture-recapture method with programme dataToimplementcapture-recaptureusingprogrammedata, identifytheindividualscapturedintwodatasources.

Beclearandspecificwhenyoudecidehowtomatch.Agoodwaytoproceedistoadoptarigiddefinitionanddotheanalysis.Thenrelaxthematchingcriteria,repeattheanalysis,andcomparetheresults20.

Unlessyourdatasetsareenormous,thebestwaytomatchismanually.Spreadsheetsandelectronicdatabasescanhelpbysortingindifferentways.G Twolistssortedbysexanddateofbirthmayassistinmatchingindividuals.Birthdatesmaydiffer

slightlyoragecomputedfrombirthdatemaydifferfromreportedage.G Ifnamesareavailable,theymaynotbemuchhelpbecausesomenamescanbesosimilarorpeople

giveadifferentversionoftheirnameatdifferentvisits.

It isalsopossible tocalculateaconfidence interval togivea rangeoferror for theestimateof total

populationsize: 95 1 96% . ( )CI N Var N= ±

WhereVar(N)iscalculatedas: MC M R C RR

• − • −( ) ( )3

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

21

Detailed example of capture-recapture method with programme data 21

AcountryisexperiencingrapidlyexpandinginjectiondruguseassociatedwithHIV.Thecountryneedstoestimateofthenumberofinjectingdruguserstoevaluatethefeasibilityofinterventionprograms.Thehealthministryhasdataavailablefromtwoexistingdatasources:G Datasourceoneiscomprisedofrecordsfromasocialinsurancesystembasedonresidency.This

database includes information on out-patient and in-patient/hospital care and on reimbursedmedications.Thus,itcouldbeusedforpeoplereceivingdrugtreatmentorwhohavehadadrugoverdose.

G Datasourcetwoisapolicedatabasewithinformationoncriminaloffencesincludingthemeansofadministeringillicitdrugs(thatis,injectedornot).

G Bothdatasourcescontaininformationongender,day/month/yearofbirthandinitials.G Investigators decide to restrict analysis to persons 15-44 years old. Records outside this age

range,recordswithoutfullidentifyinginformationormultiplerecordswiththesameuniquesetofidentifiersaretobedeletedfromtheanalysis.

G Insurancerecordsidentify1299injectingdrugusers.G Policerecordsidentify5311.G 873personsareidentifiedinbothdatasources.

Fromthesedata,theteampreparesFigure3.4.

Figure 3.4. Capture-recapture using programme data from police and insurance record

Insurance

Police records

Yes

No

873 b

c x

Yes No

M=1299

C=5311 N=R+b+c+x

Thus,thetotalnumberofinjectiondrugusersinthepopulationisestimatedasN=5311x1299/873=7,903

Theteam’sstatisticiandeterminesthatthevariance,ameasureoftheuncertaintyoftheestimate,is:

MC M R C R

R• − • − • − • −( ) ( ) ( ) ( ) ,3 3

1299 53111299 873 5311 873873

19 604= =

The95%confidenceintervalforthisestimateis:N+1.96√19604=N+1.96*140orfrom7,629to8,177.

These are reasonable calculations if the assumptions hold: both samples are selected randomlyandthetwosourcesareindependent.Theteamassumesnorelationshipbetweenapersonhavingaccesstothesocialinsurancesystemandbeingincludedinthepolicedatabase.Iftheassumptionofindependenceisnotsatisfied,theestimatemaybebiased.Aviableinterpretationoftheseresultsistoreport,“weestimatethenumberofIDU,aged15–44,inthecountryfallsbetween7500and8200in2009”.Thisissufficientlyaccurateformostprogrammeplanning.

22

3.2.3. Nomination methodsThesemethodsaresometimesusedinconjunctionwithmultiplierorcapturerecapturemethodsbecauseno alternatives are available to survey hidden populations. However, nomination methods are notrecommendedforestimatingthesizeofahiddenpopulation.

How these methods workNominationmethodsstartwithalimitedbutvisibleandaccessiblepartofalargerpopulationsuchas:G drugusersintreatmentprogramsG menwhoattendopenlygayestablishments.

Thesepersonsarecontactedandaskedtoprovideeithercontactinformationorreferotherindividualswhosharetheirriskbehaviour.Ifthereferredindividualscomein,theyareaskedtoprovideothernamesorreferothermembersandsoon.Forthisreason,variationsofthismethodareoftencalledrespondent-driven,snowballorchainsamplingmethods.

Strengths and weaknessesNomination methodsgive thepromise ofprovidingaccess tohiddenpopulationsbecause they rely onmembers of the population to help find additional members 22. However, caution is needed with thesemethods.

Most hidden populations are usually composed of persons whose behaviour is the most illegal orstigmatized.Individualswhoparticipateinillegalandstigmatisedbehaviourwillnotwanttogivenamesandidentifyinginformationforfearofadverseconsequences.Ontheotherhand,thosepopulationstendtobehighlyconnected.Referralsmaybeduplicates 23.Thus, thecollectionof identifying information isessentialinordertoeliminateduplicatedpersons.

Nomination methods start with visible members of the group who may not be representative of thecompletepopulationatrisk24.Take,forexample,theproblemofdevelopinganHIVpreventionprogrammefor injectingdrugusers inacity. Ifwestartwith initial contactswithpersonswho injectdrugsandareenrolledinaprivatedrugtreatmentprogramme,theymightbemoreaffluentthanotherdrugusersinthepopulation.Thusnominationmethodsthatrelyonthistreatmentprogrammeastheonlypointofentrymaymissotherpartsofthedruguserpopulationentirely.

Theutilityofnominationmethodsdependsontheconnectivityofmembersofthepopulation.Anysamplefromanominationmethodwillover-representthosewithlargepersonalnetworksbecausetheywillhavealargernumberofpathsleadingtothem.Thus,moresociallyisolatedmembersofthepopulationwithlowlevelsofinteractionwithothernetworkswillbeneglectedinthistypeofsampling.

Toovercometheselectionbiasesdescribedabove,multiplesamplesandsophisticatedstatisticalmethodsareneededduringdataanalysis 25.

Nomination methods are useful for conducting formative research as part of pre-surveillance activities or in gaining access to a population in need of services. But for estimating population size, these methods are not recommended.

3.2.4. Multiplier methodThemultipliermethodishighlydependentonthequalityoftheexistingdata.Youwillneedtoreviewhowtheexistingdatawerecollectedbeforeyouusethedatatoproduceestimates.

How this method worksThemethodreliesontwosourcesofdata.G Thefirstsourceshouldbeacountorlistingfromprogrammedataincludingonlythepopulationwhose

sizeisbeingestimated(numberofsexworkerswhoattendedanSTIclinicinthelastmonth,numberofpersonswhoinjectdrugsvisitinganeedleexchangeprogramme)

G Thesecondsourceshouldbearepresentativesurveyofthepopulationswhosesizeisbeingestimated.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

23

Inthesurveyasktherespondentswhethertheyreceivedtheservice.Dividethenumberwhoreceivedtheservicebytheproportionreportingreceivingtheserviceinthesurveytoestimatethepopulationsize.Thiscanbeexpressedas:

S =# of key population in STI registry records

% population rreported being registered

Thebasicprincipleisthis.Forthenumberofpeoplebeingestimatedinthepopulation:

Those who appear at a

specific institution during a certain time period, such

as sex workers at STI clinics

=

the proportion of the population who attended

the institution x

the total size of the population

Forexample:G IfthenumberofsexworkerswhosoughtcareatanSTIclinicin2005isknowntobe1,000fromclinic

recordsG Ifapproximately10%ofsexworkersattendedanSTIclinicin2005(fromasurveyofsexworkers)G ThentheSTIcare-seekingfigurecanbemultipliedby10(ordividedby10%)togetanestimateofthesize

ofthesexworkerpopulation.

Similarly,ifyouhavealistofSTIclinicattendeesofwhomonlyaproportionaresexworkersyoucouldusethefollowing.

SPP

M= 1

2•

Where:S =estimatedtotalnumberofsexworkersP1 =proportionofsexworkersonalistofSTIclinicattendeesP2 =proportionofsexworkerswhoattendedtheSTIclinicamongacross-sectionalsurveyofsexworkersM =numberofindividualsontheSTIclinicattendeelist

This estimate is mathematically equivalent to a capture-recapture calculation but the interpretation issomewhatdifferent.

Table3.3providesasamplelistofdatasourcesforthemultipliermethodwhenthetargetgroupisinjectingdrugusers.

Table 3.3. Potential data sources for estimate of persons who inject drugs 26

DataSource Example

Drug treatment program Drug users attending treatment agencies or in residential care

Drug agencies Drug users contacted by outreach workers

Needle exchange programs Drug users registered at needle-exchange programs

Hospital/ER Drug users needing emergency treatment due to overdose

Laboratories Drug users tested for HIV, HCV, or Hep B virus

Police/prisons Drug users arrested for drug use or other crimes

Probation Drug users on probation

Social services Drug users assisted by social services

Mortality statistics Deaths due to opiate overdose

24

Another version of multiplier method involves the distribution of a unique object to members of thepopulation.Similartoacapture-recapturemethod,usethosereceivingtheuniqueobjectasthecountofthefirstpopulation.

Usetheuniqueobjectmultiplierwhenservicesdonotexistforthepopulationofinterest.Thestudyteamcontrols:G howmanyobjectsarehandedouttothepopulationG whattypeofobjectishandedoutG whoreceivestheobject(amemberofthedefinedpopulation)G whentheobjectishandedout(inthetimeperiodofthesurvey).

Inasurveyaskhowmanyrespondentsreceivedtheuniqueobject.Usetheformulaabovetoestimatethepopulationsize.

Strengths and weaknessesThe multiplier method is preferable to census and enumeration methods when the sampling frame isquestionableorwhenthepopulationisdifficulttoreach.Themultipliermethodisstraightforwardifdatasourcesareavailable.Thismethodisflexibleandusefulinmanycircumstances.

However,thetwosourcesofdata:G Mustbeindependent(everyonewithachancetobeonthelistshouldhaveachancetobeinthesurvey).

Similarly,everyoneon the list shouldbeamemberof thepopulationyouare trying toestimateandthispopulationshouldalsobecapturedrandomlyinthesurvey.Inreality,mostsurveyswillhavesomeamountofselectionbiasandthesurveyteamneedstodecidewhetherthatbiasisindependentofthelikelihoodofbeingincludedonthelist.• The first data source (multiplier) need not be random but should be specific to the group being

estimated.Thatis,iftheteamisusingSTIclinicdatatoestimatesizeofthesexworkerpopulation,theymustexcludenonsex-workersfromthelist.Thisisindirectcontrasttocapture-recaptureestimates.

• Thesecondsource(thesurvey)shouldberandomandencompassthegroupinthemultiplierbutitcanincludeothersaswell.Thatis,itcanincludebothbrothelandstreet-basedsexworkersevenifthemultiplierincludesonlybrothel-basedsexworkers.

G Mustdefinethepopulationinthesameway(thetwopopulationsforthedatasourcesareequivalent)G Musthavealignedtimeperiods,agerangesandgeographicareas

Keepinmind,datacollectedfromexistingsourcesmaybeinaccurate.

Detailed examples

Detailed example 1: Sex workers in China ConsideranexamplewithsexworkersinChina27.Twosurveyswerecarriedoutinoneprovince:G Inthefirstsurvey,92femaleSTIclinicclients(atoneof16registeredSTIclinics)wereenrolledafter

informedconsent.ThetotalnumberoffemaleSTIclinicattendees(842women)over6monthswasdetermined frommedical records.Of the92 femaleSTI clinicattendees interviewed,45 (48.9%)wereclassifiedassexworkers. It isassumed that thissampleof92 is representativeof the842womenattendingtheclinic.

G A second survey was conducted among the community female sex worker population throughanonymousinterviewsattheirplaceofwork.TheywereaskedwhethertheyhadvisitedanyofalistofSTIclinicsinthepastthreemonths.Fromthesurveyofsexworkersinthecommunity,16.2%(47/327)hadvisitedtheSTIclinicfromthefirstsurvey.Thus,theestimatedsizeofthefemalesexworkerpopulationis:

S = =48 916 2

842 2 500. %. %

,•

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

25

Detailed example 2: Using programme based and unique object multipliers in India 28

Programmemanagersusedtwomultipliers toestimatethesizeof thesexworkerpopulations in6statesinIndia.Theyconductedaseriesofintegratedbiologicalandbehaviouralsurveysamongsexworkerstousewiththemultipliers.

Datasource1:ProgrammebasedmultipliersanduniqueobjectmultipliersG Multiplier1camefromservicestatisticsrecordedbyorganizationsworkingwithfemalesexworkersG Multiplier2camefromauniqueobjectdistributedtofemalesexworkers

Datasource2:SurveyG The integratedbiologicalandbehaviouralsurveysweresampledusingeitherrespondent-driven

samplingor time-locationsampling.Thesesamplingmethodsapproximateprobabilitysamplingmethodstoobtainarandomsample.

Thequestionsusedinthesurveyweredesignedtobecompatiblewiththedataroutinelycollectedandavailablefromlocalserviceproviders.Indicatorsincluded:G proportionreportingbeingregisteredwiththeserviceproviderG proportionreportingcontactbyapeerinthepastmonthG proportionreportingreceivingaprojecthealthcardinthepastyearG proportionvisitingtheserviceproviderinthepastyearorpastthreemonths.

Theseindicatorsformultiplierspresentedsomechallenges.Forexamplewhenusingregistration with the service provider,itwasfoundthat:G SomeserviceproviderrequiredregistrationwhileothersdidnotG Someserviceprovidergaveoutregistrationcards,othersdidnotG Sometimescommunitymembersknewtheywereregisteredwithaserviceprovider,othertimes

theydidnot.(Thismayhavebeenduetothevariabilityinthedegreeofbrandingoftheserviceproviderrecognizablebythecommunity.)

Inusingvisit to clinic in past three months,duplicationwasaproblem:G Only service providers that did individual tracking could provide information on the number of

individualswhohadvisitedtheclinicduringagiventimeframeG Othershadinformationonnumberofvisits,butnotonnumberofindividuals.

Due toanticipatedchallengeswith thesemethods,anadditionalmultiplierwasused thatcouldbecontrolledbythesurveyteam.Thiswasknownastheuniqueobjectmultiplier.Theteamdistributedanobjectinadvanceofthesurvey.

Inseveralofthedistricts,theobjectdistributedwasakeychain,designedtobeuniquelymemorable.Thekeychainwasdistributedbeforethesurveytopersonswithintheboundsofthesurveycoverageareawhomatchedthedefinitionofthepopulationwhosesizewasbeingestimated.Respondentswereaskedinthesurveyiftheyhadreceivedthekeychain.

In the majority of cases, both the programme based multipliers and the unique object multipliercombinedwiththesurveyyieldedlowersizeestimatesthanexistingdatafromprogrammemappingdata.Reasonsforthisdiscrepancycouldhaveincludedproblemswiththedifferentdatasources.Potentialproblemswithdatasource1:G IneligiblepeoplewereincludedintheprogrammecountsG UniqueobjectweredistributedtoineligiblepeoplePotentialproblemswithdatasource2:G Selectionbiasinthesurveyleadingtonon-independencebetweendatasources.Thiscouldhappen

ifthoseincontactwiththeserviceprovideraremorelikelytobeincludedinthesurveythanthosenotintheprogramme.

G Thesurveyquestionswerenotspecificoradequatelymatchedtotheprogramme-basedmultipliers

26

G People reported having received a unique object or being in a programme when they had notreceivedtheobjectorwerenotintheprogramme

G Thesurveysamplewasnottrulyrandom.

Although thekeychainwasnot randomlydistributed, thisdidnotviolate theassumptions for themethod.

Themainsafeguardagainstthesebiasesistomakesuretheprobabilitysurveyisascloseaspossibletobeingrandom.Theuniqueobjectmethodhasthepotentialadvantageofbeingeasiertocontrolbythesurveyteamsothatsomebiasesareavoided.

3.3.MethodbasedondatacollectedfromthegeneralpopulationPopulationsmostatrisktoHIVareoftenhiddenbecausetheyarestigmatizedorengagedinillegalactivities.Membersofthesepopulationsareespeciallyhesitanttoidentifytheirbehavioursforfearofpunishment.Thus conducting surveys of such individuals can be challenging as they often do not want to identifythemselvestoastudyteam.3.3.1. SurveysIn most circumstances, surveys directly asking about extremely high risk behaviours are also notrecommended for estimating the size of most-at-risk populations. As described above populations atincreasedrisk toHIVare likely toavoidansweringsuchquestions truthfully. Inadditionbehaviours thatputpeopleatincreasedrisktoHIVareoftensorarethataverylargesamplesizewouldberequiredfromasurveytoestablishtheprevalenceofsuchbehaviourswithinapopulation.Thedescriptionofthemethodisincludedintheguidelinesbecauseitisacommonmethodforcollectingpublichealthdataandcaninraresituationsbeusedtocollectbehaviourswithlowstigma.

How this method worksSurveysofthegeneralpopulationareverycommoninmostcountries.Theyaremostoftenadministeredtoresidentsofahouseholddrawnfromasampleframethatisrepresentativeatanationalorregionallevel:G Inindustrializedcountries,telephonesurveysarepossible 29.G Indevelopingcountries,dataaregenerallycollectedbysurveyteamsvisitinghouseholdsdoingface-to-

faceinterviews30.G Youthinschoolcanbereachedthroughschool-basedsurveys.Becarefultoconsidertherepresentativeness

ofindividualsattendingschoolversustheremainingpopulation31ofthesameagethatisnotinschool.To estimate the size of the hidden population, respondents in a general household survey are asked iftheyinjectdrugs,sellsex,purchasesex,or,ifmale,havesexwithothermen.Thesearenotalwayseasyquestions to insert in a survey given the stigma, discrimination and illegality of these behaviours. Thewordingandlocationofthesequestionsinthesurveyareimportantaspectstoconsider.

Strengths and weaknessesSurveysaregenerallyeasytoimplement.Surveysarelongstandingmethodsinthestatisticalliterature 32,soresultswillberelativelyeasytoanalyseanddefendandarepoliticallyinfluential.Ingeneral,itisfairlyeasytofindasampleframeforageneralpopulationsurvey.

Surveysare lessusefulwhenbehaviour is rarebecause itmaynotbe reflected in thesampleselected.Thoseatriskmaynotbefoundinhouseholds,schoolsorotherinstitutions.Inaddition,ifbehaviourhasbeenstigmatizedwithinasociety,respondentswillbelesstruthfulwiththeinterviewer,especially if theinterviewisnotconductedinaconfidentialsetting 33.

Thus,inmostsettings,itisdifficulttocreateapopulationsizeestimatebasedondirectquestioningabouthighlystigmatizedbehaviourswithinahousehold-basedsurvey.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

27

3.3.2. Network scale-upThe network scale-up method is currently being considered for estimating the size of hard-to-reachpopulations. (Adjustments for theknownbiases in thismethodarestillbeingdevelopedsothemethodshouldberegardedas“under-development”.)Themethodusesinformationcollectedingeneralpopulationhousehold surveys to estimate the size of hidden populations. However, instead of asking about therespondent’sownbehaviour,thismethodasksaboutthebehaviouroftherespondent’sacquaintances.

Basedontheaveragenumberofindividualsthatrespondentsknowinhiddenpopulationsandtheaveragepersonalnetworksize,theproportionofpeopleinthemostatriskpopulationisestimated.

How this method worksTherearethreestepstothenetworkscale-upmethod:1. Estimatetheaveragepersonalnetworksizeofthegeneralpopulation.2. Askthegeneralpopulationhowmanyindividualstheyknowineachofthehiddenpopulationsofinterest.3. Calculatetheestimatedpopulationsizeandadjustforknownbiases.

Forexample:G Ifarespondentknows300peopleandtwoofthosepeopleinjectdrugs,youcanestimatethat2/300thof

thegeneralpopulationareinjectingdrugusers.G Whenyoucombinethatestimatewiththetotalpopulationsizeofthecountry,say300million,youcan

estimatethattwomillionpeopleinjectdrugsinthecountry.G Yourestimatecanbeimprovedbyaveragingovermanyrespondentswithdifferentnetworksizesand

numberofpersonstheyknowwhoinjectdrugs.

Step 1 – Determine personal network sizeInstep1,weestimatepersonalnetworksize.Thatis,howmanypeopledoestherespondentknow?Inmostcultures,theideaof“knowing”someoneorcountingwhoweknowisnotspecific34:G Doeswhoyouknowapplytocurrentacquaintancesoreveryoneyouhaveknowninyourlifetime?G Howwellmustyouknowsomeonetocounttheminyournetwork?G Ifyounamesomeoneasyouracquaintance,musttheyalsonameyouasoneoftheirs?

Howyoudefineapersonalnetworkwillhavetobedefinedforyourestimatethenusedconsistentlyovertime.Previousnetworkscale-upstudieshaveusedthisdefinition:

“Apersonwhoshouldbecountedinyourpersonalnetworkissomeonewhoknowsyouandyouknowthem,bysightandname.Youcancontactthemortheycancontactyou.Youhavetalkedtothemwithinthelasttwoyears.ThepersonlivesinX”.(Xbeingthespecificareaofreference.)

Twomethodshavebeenexploredforestimatingpersonalnetworksize:thesummation method (sectionabelow)andtheknown population method(sectionbbelow).Thesummationmethodshouldbeusedinsituationswherestatisticalresourcesorrecordsareunreliableorunavailable.

A. The Summation Method of Estimating Personal Network Size Inthismethod,youwillaskrespondentsforadirectestimateoftheirpersonalnetworksize.Tobreakthisdownintoamanageabletask,therespondentisaskedtocounthowmanyacquaintanceshehasineachofasetofmutually-exclusive,butexhaustive,categories.Bysummingupthenumberofacquaintancesineachcategorywehaveadirectestimateofthenumberofpeopletherespondentknows.

Asaruleofthumb,peopleareabletocountupto20individualswithoutwritingoutalist.Ifacategoryislikelytoroutinelycontainmorethan20people,sub-dividethecategory.ApartiallistofpossiblecategoriesisprovidedinTable3.4below.Again,trytoensurethatthecategoriesoverlapaslittleaspossible.

28

Table 3.4. Possible categories of acquaintances for the summation method

G ImmediatefamilyG Otherbirthfamily/familyofspouse/partnerG Co-workersG OtherpeopleatworkG BestfriendsG Peopleknownthroughhobbies/recreationG Peopleknownthrough...(religiousorganizations,neighbourhoods,school)G PeopleknownthroughothersG ChildhoodacquaintancesorfriendsG Peoplewhoprovideaservice

Oneoptiontogetaccurateanswersonapersonalnetworksizeistoprovideavisualprompt.Therespondentswillknowinadvancewhatcategorieswillbecoveredandshouldbeabletoavoidcountinganacquaintanceinmultiplecategories.

Clearly,thechoiceofcategoriesforasummationmethodofestimatingaveragenetworksizeisculturallydependent.Developthislistatthecountrylevelandtestittoensurethatthelistavoidsoverlapasmuchaspossibleandisexhaustive.

B. The Known Population Method of Estimating Personal Network Size Another alternative for estimating personal network size is the known population method. You will askrespondentsaboutthenumberofpeopletheyknowinspecificpopulationsforwhichatruevalueisknown.

Forexampleifcensusdatashowthereare3,200adultsnamedMichaelinthecountrywith300,000people.ThemeannumberofacquaintancesnamedMichael(calculatedfromtherespondents)is5.57.Theestimatedpersonalnetworksizecanbecalculatedas:

5 573200

300 000 522. ,• ≈

Continuingwiththisexample,imagineasurveydoneinthegeneralpopulationof300,000individuals.Inaddition to thesetofquestions toestimate respondents’personalnetworksize (suchas thenumberofacquaintancesnamedMichael),respondentswerealsoaskedhowmanypeopletheyknewwhowereborninadifferentcountry.Toestimateaveragenetworksize,thesamecalculationthatwasdoneforthenameMichaelisdoneforeachoftheknownpopulations.

Tipsonselectingknownpopulations:G Ideally,20to30knownpopulationswillbeusedtocreateareliableestimateofpersonalnetworksize.G Tomakeafairlyaccurateestimate,theknownpopulationsshouldbeonaverage0.1%to4%ofthetotal

population.G Researchhasshownthat thismethodworksbest if theknownpopulationsaresimilar to thegeneral

population(sameagegroup,samesex).

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

29

SubpopulationsusedinahypotheticalstudyaregiveninTable3.5below.

Table 3.5. Known population method: examples of sub-populations used to estimate average network size

Sub-populationSizeofsub-populationinthecountry

Meannumberknowntorespondents

Had a child in last 12 months 4,000 3.20

Has diabetes 6,500 2.43

Opened business in last 12 months 630 0.83

Moved house in last 12 months 8,200 1.68

Voted for X in last election 20,000 3.33

Was born in a different country 22,000 4.76

Is widowed & < 65 years of age 3,300 2.83

Is named Michael 3,200 5.57

Thismethodallowsnotonlyforquantificationofuncertaintybutalsoisarealitycheckoftheestimatesofhiddenpopulationsize.WecantesttheaccuracyoftheknownpopulationmethodbyestimatingthenumberofadultsnamedMichaelfromourknownpopulationsandcomparingthatvaluetothevalueprovidedbythecensus.Naturallyiftheaccuracyoftheknownpopulationsisnotgooditwillbiastheresults.

AnestimateofaveragesocialnetworksizeintheUnitedStateswasfoundtobe29035, 36,butlittleworkhasbeendoneininternationalcontexts.Thevalueof290wasfoundusingboththesummationmethodandtheknownpopulationsmethod.

Step 2 – Ask the general population how many people they know in the hidden populationThenetworkscale-upmethodrequiresaskingpeopleinthegeneralpopulationhowmanypeopletheyknowinthepopulationsmostatrisktoHIV37.

Respondentsareaskedhowmanypeopletheyknowwhoinjectdrugs,sellsex,purchasesex,or,aremenwhohavesexwithmen.Thesequestionsarenotalwayseasytoinsertinasurveygiventhecommonstigmaandillegalityofthesebehaviours.Thewordingandlocationofthesequestionsinthesurveyareimportantaspectstoconsider.G Interviewersshouldbetrainedonhowtoaskthesequestions.G Theconfidentialityoftheinterviewmustbeassured.G Thewordingofthesequestionsiscritical.Ifthewordingislefttotheinterviewertodeterminetherewill

bevariationsintheresponsesduetotheinterpretationofthepeoplewhoareclassifiedinthehiddenpopulations.

Step 3 – Calculate the estimated population size and adjust for known biasesInstep3,calculatethesizeestimatebydividingtheaveragenumberofpeopleintheknownpopulationsbytheaveragenetworksizeandmultiplyingbythetotaladultpopulation.

Thereareseveralproblemsassociatedwiththeapproachesdescribedforestimatingpersonalnetworksizeandtheresultingsizeestimates:G Thesizeofanetworkmayvaryamongindividuals.G Arespondentmaybeunawarethatsomeoneinhis/hernetworkisamemberofthepopulationofinterest

(knownas“transmissionerror”)38.G Thepositionofarespondentmaycausehim/hertoknowfewermembersofthepopulationthanwould

beexpected(barriereffects).Forexample,peopleinruralareasmaybelesslikelytoknowsomeonewhoinjectsdrugs39.

G Somepopulationsmightnotadmitknowingindividualswiththehiddenbehaviours.

30

Currenteffortsarefocusedonhowtoadjustfortheseerrors.Forexamplestudiesareunderwaytomeasuretransmissionerrorbasedonsurveysofmostatriskpopulations.Inadditionstatisticiansarelookingintothesamplesizerequiredtoconductanetworkscaleupsurveyandtheestimationofvariance.

Figure3.5belowdisplaystheconceptofthenetworkscale-upmethod.

G ThewholeboxisthetotalpopulationT.G cisoneindividual’sacquaintances(orpersonalnetworksize)G marepersonswhoinjectdrugsamongthoseacquaintancesG Eisthesizeofthehiddenpopulation(thevalueofinterest)G Nisthetotalnumberofpeopleinthesurvey

c=respondent’snetwork

Respondent

m=hiddenpopulationknowntorespondent

E=hiddenpopulation

Figure 3.5. How the network scale-up method works

We can then estimate E using the below formula where the subscripts are the survey respondents,1throughN.

Strengths and weaknessesThetechniquemayhavesignificantadvantagesoverexistingmethods:G Itdoesnotrequiremembersofhiddenpopulationstoidentifythemselvestoasurveyteam.G Thequestionscanbeincorporatedintoexistinghouseholdsurveyssoestimatescanbegeneratedatthe

levelofthosesurveys,typicallynationalorprovincial.G Themethodcancreatesizeestimationsformultiplehiddenpopulationsinonesurvey.

Atthetimeofpublicationofthisguideline,therearestillanumberofadjustmentsrequiredforestimatesproducedfromnetworkscale-up:G Adjustmentstoaccountforthebarriereffect–somesubgroupsmaynotassociatewithmembersofthe

generalpopulation.G Adjustmentstoaccountforthetransmissioneffect–Arespondentmaybeunawaresomeoneinhis/her

networkengagesinthebehaviourofinterest.

Additional Information on the network scale-up method is available at http://nersp.osg.ufl.edu/~ufruss/scale-up.htm

v

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

31

Detailed example of network scale up ArecentpilotsurveyofnetworkscaleupwasconductedinonecityinAsia.(Thedatapresentedherearefictitioussincetheresultshavenotbeenpublished.)

Step 1: Estimate the average personal network size:1.Thestudyteamdecidedtoestimatethepersonalnetworksizeusingboththesummationmethod

andtheknownpopulationmethod.2.Forthesummationmethod,formativeresearchwasconductedtodeterminethemostappropriate,

mutuallyexclusive,yetexhaustivelistofpossibleacquaintancecategories.Seventeencategorieswereestablished.

3.Questionsaboutthe17categoriesofacquaintanceswereaddedtoageneralpopulationsurvey.Forexample:G Howmanyadultsdoyouknowwhoarepartofyourimmediatefamily?G Howmanyadultsdoyouknowthroughyourwork?

4.Respondentsweretoldnottocountthesamepersoninmorethanonecategory.InadditiontheyweretoldonlytocountpeoplewholiveinthespecifiedCity.

5.The study team summed the categories for each respondent and an average network size wasestimatedforthecity.G Average personal network size using summation (mean value of all respondents): 131

6. In addition, 40 known populations were initially identified. These populations were narroweddownto20populationsbasedontheavailabilityofrecentstatisticsforthecityandwhetherthepopulationswerebetweenapproximately0.2%and4%oftheadultpopulation.

7.Questionswereaddedtothesurveyabouthowmanyacquaintancestherespondenthadineachofthe20knownpopulations.Forexample:G HowmanyadultsdoyouknowwhoarenamedMichael?G Howmanyadultsdoyouknowwhoarepracticingdoctors?Respondentsweretoldtheycouldcountpeoplemorethanonce.Theywereremindedtoonlycountpeoplewholivedinthespecifiedcity.

8.The study team compared the estimated number of people in the known populations with thestatisticsavailablefromthatcityforthesameyear.Thestudyteamwasabletoestimatetheaveragepersonalnetworkusingtheknownpopulationstatisticsandtherespondents’answers.G Averagenumberofpracticingdoctor’sknowntotherespondents(m):4.2G Percentofpracticingdoctor’softheadultpopulation(E/T):3%G Known populations average personal network size (c): c = T/E x m = 1/0.03 x 4.2 = 140

Step 2: Collect information on the most at risk populationsInadditiontothequestionstoestimatepersonalnetworksize,4questionsnecessarytocapturethenumberofacquaintancesinthemostatriskpopulationswereaddedtothesurvey.Thesequestionswerecarefullywordedbasedonadvicefromdifferentstakeholders.Forexample:

G How many adults do you know who inject drugs? Average value (0.397)G How many women do you know who sell sex? Average value (1.82)

Step 3: Calculate the size estimate and adjust for known biasesUsing the known population estimate of the personal network size, the study team calculated theproportionoftheaveragepersonalnetworkthatwasmadeupofadultswhoinjectdrugsorsoldsex.

Theadultpopulationinthecityisapproximately600,000.Thestudyteamestimated:G thenumberofpersonswhoinjectdrugsinthecity:0.397/140x600,000=~1,700.G Thenumberofwomenwhosellsexinthecity:1.82/140x600,000=~7,800

An additional set of questions was added to the survey to adjust for the reduced likelihood thatsomeonewillshareaverystigmatizedbehaviourwiththeiracquaintances.Afteraskingtherespondentabouteachoftheknownpopulationstheywereaskedabouttheir“respect”foreachoftheknownpopulations.Forexample:

G Onascaleof1to5howmuchdoyourespectdoctors?G Onascaleof1to5howmuchdoyourespectmennamedMichael?

Theaveragevalueofthisresponseprovidedalevelofrespectforthedifferentpopulations.Forexampledoctorsweregivenaveryhighlevelofrespectwhilepersonswhoinjectdrugsweregivenaverylowlevel.Thestudyteamdevelopedanadjustmentfactorfortheresultsbasedontheprestigemeasure.The size estimate was reduced proportionally to correct for the lower likelihood of a respondentadmittingthatheorsheknewapersonwhoinjectsdrugs.

32

3.4.Summaryofsizeestimationmethods

Table3.6providesanoverviewofthedifferentmethodsandtheiradvantagesanddisadvantages.Usethistableasaquickreferencewhenchoosingasizeestimationmethod.

Category1:Methodsbasedondatacollectedinanat-riskpopulation

Methodnameanddescription Advantages Disadvantages

Census method counts all members of the population.

Enumeration develop a sampling frame then counts all members of the population at a sample of places listed in the sampling frame.

Census method is easy to explain as it simply attempts to count all members of the population

Enumeration method maps then covers just a fraction of the population.

GMost-at-risk populations are often hidden. Both methods will miss members of the population not visible to the public.

GCommunity guides are necessary to improve access.

GCensus is time-consuming and expensive to conduct

GEnumeration method requires a reliable sample frame of venues.

GOverestimate if population is mobile and double counted

GUnderestimate if populations are well hidden.

Capture-Recapture methods calculates the total size of a population based on two independent captures (samples) of population members: GCapture 1: ‘tag’ and count

number tagged. GCapture 2: ‘tag’: keep track of

who was ‘retagged’ and who is ‘first time tagged’.

GA simple two-sample capture-recapture method is relatively easy to use.

GDoes not require much data.

GDoes not require statistical expertise.

Relies on assumptions that are hard to meet in normal field conditions: GTwo samples are independent and not

correlated. GEach population member has an equal chance

of selection. GEach member is correctly identified as

‘capture’ or ‘recapture’. GNo major in/out migration is occurring.GSample size is large enough to be meaningful.

Multiplier methods compare two independent sources of data for most-at-risk populations GSource 1: count/listing of

persons who accessed a service

GSource 2: proportion of population who accessed service from representative survey of population of interest

GStraightforward if data sources are available.

GFlexible method, useful in many circumstances.

GThe two data sources must be independent. GThe data sources must define the population in

the same way. GTime periods, age ranges and geographic areas

from the two data sources are not always aligned.

GData collected from existing sources may be inaccurate.

Table 3.6. Summary of methods for estimating population size

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

33

Category2:Methodsbasedondatacollectedfromthegeneralpopulation

Methodnameanddescription Advantages Disadvantages

Population survey methods with general population behaviour questions. GDirectly ask respondents

whether they have specific behaviours that put them at increased risk to HIV (such as selling sex, injecting non-medical drugs, men having sex with other men)

GSurveys are common and familiar.

GStraightforward to analyse and easy to explain to data users.

G Difficult to use when the behaviours are rare or stigmatized.

G Only reaches people residing in households, schools or other institutions used to create the sampling frame.

G Respondents are unlikely to admit to high risk or stigmatized behaviours if interview is not confidential or if interviewer is not skilled at establishing trust and rapport.

Network scale-up methods are based on the idea that people’s social networks reflect the general population. GAsk a random sample in the

general population to estimate number of people they know, and how many of those people have the behaviour of interest.

GCan generate estimates from general population rather than hard-to-reach populations.

GIndividuals are often more likely to report on the behaviour of others instead of their own behaviour

GA single survey can be used to create size estimates for multiple hidden populations.

GAverage personal network size difficult to estimate.

GSubgroups may not associate with members of the general population.

GRespondent may be unaware someone in his/her network engages in behaviour of interest.

GRespondents may be hesitant to admit to knowing individuals with the specified behaviour.

Source: Adapted from “Monitoring and Evaluation Guidelines for HIV Prevention for Men Who Have Sex with Men”. MERG Technical Working Group on Most at Risk Populations. Draft December 2009.

3.5.Guidanceonselectingmethodstoestimatethesizeofat-riskpopulationsEstimationresultsaresubject topoliticalaswellasscientificuse.Thediscussions in thisdocumentareintendedonlyasgeneralguidance.Surveyteamsinvariousculturalsituationswillneedtomakeadaptationsfortheirlocale.

Estimating the size of sex work client populationsIncountrieswherebuyingorsellingsexisfairlyprevalentandnothighlystigmatized,ahouseholdsurveyisanappropriatemethodofestimation.

G Forexample,insomecountriesinAsia,5%to25%ofadultmenreporthavingpaidforsexinthepastyear. The proportion of respondents reporting buying sex can be applied to census denominators invariousstrata,suchasage.

G Ifpopulationsurveysalreadyexist,addingafewquestionscanbecosteffective.G For countries uncomfortable asking sensitive questions about paying for sex on surveys, multiplier

methodscanbeusedwithbehaviouraldataonnumberofpartnersfromsurveillanceofsexworkers.

Estimating the size of sex worker populationsG Censusmethodshavebeenshowntobeusefulforbrothel-basedsexworkers.G Useenumerationforsituationswherethereare largenumbersofvenuesandthesexworkersdonot

movequicklybetweenlocations.G Usecapture-recapturetoestimatethesizeofstreet-basedsexworkerpopulationswhenitisnotpossible

tocreatealistofvenuesorconductacensus.G Multipliermethodswillbeusefulforlocalestimateshoweveritmightbechallengingtofindlistsfrom

administrativesourcestoprovideamultiplierforanationalestimate.

34

Estimating the population size of persons who inject drugsG Generalpopulationsurveysaregenerallynotusefulforestimatingthesizeofinjectingdrugpopulations

becausethesamplesizeisnotlargeenoughtocapturedrugusers.Expectunderreportingwiththisgroupduetotheirfearoflegalrepercussions.

G Inmostcountries,therearemoreexistingdatasources(suchastreatmentdata,arrests,andregistries)ondrugusethanforotherriskbehaviours.Howevercareshouldbetakentomakesurethesesourcesarecompleteandaccurate.

G Capture-recapture methods should be considered in settings where programme data sources are ofreasonablygoodqualityandwhereinjectiondruguseisnotpunishablebyimprisonmentordeath.Anyprocedurerelyingonnamesorotheridentifyinginformationinacontextwheresuchriskbehaviourcouldresultinseverepunishmentislikelytoproduceinaccurateestimates.

G Multipliermethodscanbeusefulwheretreatmentservicerecordsareofgoodquality.Inthiscase,sincemultipliersvarybyplace,thenationalestimateshouldaggregateasmanylocalareaestimatesasareavailable.

Estimating the population size of men who have sex with menThepopulationofmenwhohavesexwithmenisoftenwellhiddenandoftennotcapturedinroutinedatacollection.

G Ifmenareopenabouthavingsexwithothermen,acensusconductedatgatheringlocationscouldbeuseful.

G Ifnoprogrammedatasourcesareavailable,acosteffectiveoptionistoincludesame-sexbehaviouronexistinggeneralpopulationsurveys.Howevertheseestimatesarelikelytobeunderreported,especiallyinsettingswheresuchbehavioursarehighlystigmatized.

3.6.Step6:CollectthedataNowthatyouhavedecidedonwhichmethodyouwilluseyouwilleithercompiletherequiredsourcestocreatetheestimateoryouwillneedtocollectdata.Thespecificinstructionsforcollectingdataareincludedinthedescriptionsofthemethodsintheprevioussection.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

35

4.1.Step7:AnalysetheresultsOncetheexerciseiscompletedandanestimatedpopulationsizeisavailabletheresultsmustbeinterpreted.Thisincludesexplainingtheconfidenceintheestimate,reconcilingmultipleestimates,extrapolatingtheresultstonationalestimates,andcreatinganysub-populationestimatesthatareneeded.

4.1.1. Sampling error and bias in size estimatesAnalyseandreportonthesamplingerrorofthesizeestimate.Eveninaperfectsurvey,asampleselectedrandomlyfromapopulationwillalmostneverbeexactlythesameastheentirepopulation.Thisistheresultofsamplingerror.Moststatisticalmethodsallowforestimationofsamplingerror(e.g.,thevariance,oraconfidenceinterval).

Youhaveprobablyheardorreadstatementssuchasthefollowing:G Thesurveywasbasedon570interviewsconductedbetweenMarch20and31.Thesamplingerrorwas

plusorminus4.5percentagepoints.G 48%feltthattheremaybetoomanysexworkersinthecountry....Thepollof996adultswasconducted

1-3Mayandhasamarginofsamplingerrorofplusorminus3percentagepoints.

Thesedescriptionsarereportingconfidenceintervals.Thetechnicaldefinitionofa(95%)confidenceintervalis this: if you repeat the same data collection procedure many times, with the same methodology andsamesamplesize,approximately95%oftheintervalsthatyoucomputewillcontainthetruevalueforthepopulation.Theconfidenceintervalgivesussomeideaoftherangeoferrorthatmaybeexpectedforanestimate.

4. Analyse and disseminate results, Steps 7–10

Inthefinalphaseoftheprocess,youwillanalyseanddisseminateyourresults.Estimatesofthesizeofmost-at-riskpopulationsarelikelytobeuncertain.AswediscussedinSection3,eachmethodhasitsownspecificstrengthsandweaknesses.Youwillneedtokeeptheuncertaintyandbiasesinmindwhenitistimetoanalyseresults.Thissectiondescribessomeoftheissuestoconsiderandprovidesrealworldexamples.

Figure 4.1. Analyse and disseminate results

Analyse/disseminate results

7. Analyseandinterprettheresults

8. Documenttheprocessusedtoarriveatthesizeestimates

9. Disseminatestudyresultsappropriately

10.Usethesizeestimates

36

Weuseconfidenceintervalstocomparetheresultsofdifferentestimationactivities.Forexample:G Anestimateofthefemalesexworkerpopulationinaregion5yearsagowas5,767±215.G Anewestimateforthatsameregiontoday,usingthesameestimationmethodwas6,102±178.G Canwesaythatthesizeofthesexworkerpopulationhasincreased?G Clearly,thenewestimateof6,102isgreaterthanthepreviousoneof5,767.G However,ifweconsidertheconfidenceintervals,theestimatefiveyearsago(5,552to5,982)seemsto

overlaptheupdatedestimate(5,924to6,280).G Thuswewouldsaythattheestimatesarereallynodifferent,andtheapparentdifferenceinestimatescan

beexplainedbysamplingvariability.

Bias results when the data were collected incorrectly or the sampled population does not adequatelyrepresent the population of interest. Bias can result from several sources; the two most important forpopulationsizeestimationare:G Measurementbias–measurementsaretaken(questionsareasked)incorrectly.G Samplingbias–dataarecollectedfromanon-representativesample

Biascanbepresentinsurveysandotherdatasetsevenifsamplingandanalysisaredonecorrectly.Neitheralargesamplesizenorstatisticalmethodscancorrectforbias.Inmostcases,biascannotbequantitativelymeasureorcalculated.

Sincewecannotcontrolforbiasbylargersamplesorstatisticalmethods(thereisnocure),itisimportantto prevent it. This prevention is most effective if done prior to data collection by ensuring that surveyquestions are valid and reliable, using correct measurement techniques, and carrying out the samplingcorrectlyandrandomly.Thusitisrecommendedthatyouhaveexpertsreviewyoursurveyplans,provideongoingtrainingtofieldworkers,ensurefieldsupervision,andperforminterimqualitychecksondata.

Implicationsofnotknowingwhetheryouhavebiasinyoursurveyincludemakinginappropriatedecisionsaboutprogrammesbasedoninvalidresults.Youmight:G failtoprovideneededservices,G wasteresourcesonprovidingunneededservices,orG losecredibilitybyprovidinginvalidestimates.

Attheveryleast,youhavewastedtheresourcesandtimebecausetheresultsdonotreflectthetruesituationinthepopulation.

4.1.2. How to use local study results to reach a national estimate Mostofthemethodsdiscussedinthisguidelinearemoreeasilyappliedatthelocallevelthanatthenationallevel:G Programmedataareusuallyapplicabletoageographicregionthatissmallerthanacountry.G Itiseasiertodeveloplistsofvenueswherethepopulationsofinterestcongregateinacityordistrict.

Itwouldbemuchhardertodevelopanationallistofallvenuesthatsexworkersorpersonswhoinjectdrugsfrequent.

G Programmedatacollectionmaybemoreconsistentatalocallevel.Ifseveralregionaltreatmentclinicsprovidedata forpersonswho injectdrugs, thenmethodsand thequalityof recordkeepingmayvarywidelyfromregiontoregion.

Can population size estimates from local studies be used to extrapolate to a national population sizeestimate? In this case, extrapolate means using the data from some locations or areas to estimate forotherareas.Youwillalsoseethetermsynthetic estimationusedforthis.Thepurposeofextrapolationistogeneralizefromaseriesoflocalstudies.

Forexample,sayyouaredevelopinganestimateofthenationalpopulationsizeofsexworkers.Tobevalid,thelargerregionsshouldhavethesamedatasourcesasthelocalareas.Toextrapolate,youwould:G Map the largerarea, suchas country,breaking itdown into smallerareasandcombiningareaswith

similarcharacteristicsG Collectdatainsomeofthesmallareas(datamayalreadyexistorcollectnewdata)G Useanextrapolationprocedurewithyourdatatocreateanationalestimate.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

37

Extrapolationmethodsmaybesimpleorcomplex.Simpleextrapolationmightapplythesamepercentagetoallareas.Considertheexampleprovidedbelow.

Detailed example of a simple extrapolation from local estimates to a national estimate

Imagineyouliveinacountrythatwantstogetanationalestimateonpersonswhoinjectdrugssothatpreventionandtreatmentservicescanbeimproved.

A local study in Region R produced an estimate of 37,000 persons who inject drugs. You want toextrapolatethatfindingtotheentirecountry40.Youmaythink,“One-third(1/3)ofthepopulationofthecountryresidesinRegionR.SoIbelievethatthereare111,000(3x37,000)personswhoinjectdrugsinthecountry.”Butthereisaproblemwiththisapproach.Itassumesnoregionaldifferencesindruguse.IfRegionRisamajorurbanarea,portcityorbordercrossing,thenitmaynotberepresentativeofmoreruralregionsinyourcountry.

Nowsupposeyouobtainnationaldataondrugtreatmentandfataldrugoverdoses.ThesedatashowconsistentlythatabouthalfthepersonsintreatmentandhalfthedrugdeathsarefromregionR.Nowyouchangeyourassumption.Thenationalnumberofpersonswhoinjectdrugsisnotproportionaltopopulation,buttonumbersofdrugdeaths.Usingthisapproach,wewouldestimatethat2x37,000or74,000isthesizeofthepopulationofpeoplewhoinjectdrugsinyourcountry.

Morecomplexextrapolation tries toaccount forother factors: socio-economic factors,geographicarea,differentsub-populations.Askyourselfthesequestions:

G Is there geographic variability? If so, applying one percentage to a whole population may not beappropriate.

G Howlocalarethedata?Iftheareaistoolarge,thelocalestimatemaynotbetrulylocalandthusmaybeinaccurate.

G Inthelocalsurveys,dotheat-riskpopulationdefinitionsmatch?G Dothelocaldataapplytothespecificat-riskpopulationyouaredevelopinganestimatefor?Forexample,

doessexworkinurbanareasrefertoalltypesofsexwork:brothel-based,street-basedandentertainment-establishmentbased?Ordothelocaldatajustconsiderasubsetofthesegroups?

G Whattypesofmenwhohavesexwithmenareincludedinthelocaldatayouhaveobtained:higher-riskMSM,venue-basedMSM,anymanwhohassexwithaman?

Detailed example of a complex extrapolation of sex workers in IndonesiaOf Indonesia’s 440 districts, only some had data estimating the size of the sex worker population.However,anationalsurveyofvillageleaderswasconductedinvillagesineachofthe440districts.Inthissurvey,thefollowingquestionwasasked:“Are there sex work spots in your village?”

G InvestigatorscalculatedthepercentofdistrictvillageswhoseleaderssaidYEStothisquestion.G All440districtswererankedbythepercentageofvillagesinthatdistrictwithsexworkspots.This

rankingwasdistributedintoquintiles(thatis,ranked1,2,3,4or5).Inotherwords,districtswiththehighestproportionofvillageswithsexworkspotswereassignedtothehighestquintile,5,anddistrictswiththelowestproportionofvillageswithsexworkspotswereinthelowestquintile,1.

G Foreachdistrictwithsizeestimationdata,investigatorsusedthesedatatocalculatetheaveragepercentageoftheadultfemalepopulationthatarefemalesexworkers.

G Thesedatawereaggregatedtocomeupwithanaveragesizeofsexworkerpopulationforeachofthefivequintiles,rangingfrom0.05percentoftheadultfemalepopulationinQuintile1to0.73percentinQuintile5(seeFigure4.2).

G Theseaverageswerethenappliedtodistrictswithoutdatainthematchingquintilegroupasshownbelow.Table4.1showshowthiscalculationwasdoneforfourdistricts,basedontheknownsizeofthedistrict’sadultfemalepopulationanditsrankingbyquintile.

38

DistrictsWITHdirectsizedata*

DistrictsWITHNOdirectsizedata

Quintile1

Quintile2

Quintile3

Quintile4

Quintile5

* AveragesizeofFSWpopulationasapercentageoftheadultfemalepopulation

0.05%

0.73%

0.35%

0.10%

0.07%

Figure 4.2. Indonesia’s extrapolation for female sex workers

District

Adultfemalepop.size

(a)Quintile

(b)%fromquintile

(c)

Estimated#ofsexworkersindistrict

(a)X(c)

District A 250,456 2 0.07 175

District B 1,329,875 5 0.73 9708

District C 546,982 2 0.07 383

District D 356,968 3 0.10 357

Table 4.1. Estimated population size by district

4.1.3. Using multiple size estimates to create the best estimateThereisnoreasonforyourcountryorlocalareatolimititselftoasinglemethodforestimatingthesizeofapopulationmost-at-risktoHIV.Findasmanydatasourcesasyoucantoimproveyourestimate.G Usingestimatesfrommultiplemethodsallowsforchecksandbalances.Ifresultsarevastlydifferentwe

cangobackandconsidertheassumptionsandthemethodtofindoutwhichonewasincorrect.G Estimatesfrommultiplesourceswhicharesimilarwillimprovethecredibilityofthefinalestimate.

Beforeevaluatingdifferentestimatesitisimportanttomakesuretheestimatesarecomparable.Thismightbedocumentedbycreatingamatrixthatexplicitlydescribesthedifferentestimatesthatwillbecompared.G DescribethedefinitionofthepopulationforeachestimateG DescribethegeographicregioncoveredbytheestimateG DescribethemethodandthepossibleviolationsoftheassumptionsforthatmethodG Basedontheviolationsoftheassumptionsdocumentwhethertheestimateislikelytobeanoverestimate

oranunderestimateG Finallyincludetheestimatecreatedbyeachmethod

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

39

Suchamatrixwillprovideaclearandtransparentdescriptionoftheevidenceavailablefordeterminingthefinalestimate.

Detailed example of using multiple resultsAnunnamedcountrywasrecentlyestimatingthenumberofpersonswhoinjectdrugsinCityY.Theyhadanumberofdifferentdatasourcesfromwhichtocalculatesizeestimates.G Arecentgeneralpopulationsurveyhadaskedquestionsfornetworkscaleup.G Programme data provided information on the number of persons who were registered in their

“narcology”registry.G Programmedatawereavailableonthenumberofpeopleenrolledinaharmreductionproject.G 328personswhoinjectdrugswerehandedakeychain(oruniqueobject)byoutreachworkers.G A survey using respondent driven sampling provided information on the proportion of persons

injectingdrugswhowereenrolled in theharmreductionprogramme,whowereregisteredwithnarcology,andwhoreceivedtheharmreductionkeychain.

Thedifferentresultsforpersonswhoinjectdrugsshowedarangeofestimates.

MethodEstimatesofpersonswhoinjectdrugsinCityY

Network scale up (adjusted for stigma 7,896

Multiplier method (Needle exchange) 7,774

Multiplier method (unique object) 8,548–42,620

Multiplier method (Narcology) 3,483

Programme records (Narcology) 2,220

Theprogrammemanagerscomparedtheresultsanddetermined:G Resultswerefairlyconsistentbetweentheprogrammedatamultiplierandnetworkscale-up:7,774

and7,896G (Only3respondentsinthesurveyreportedreceivingkeychains.Thesmallnumberofkeychains

returnedinthesurveyresultedinaverylargeconfidenceintervalfortheuniqueobjectestimate:8,548–42,620,howeverthelowerboundisclosetotheotherestimates

G Thenarcologydata,evenwhencombinedwithamultiplier,givesaverylownumber:3,483.ThenarcologyregistrationislikelytobeincorrectforCityYsincetheregistryclassifiespeoplebywheretheyenrolledandnotwheretheyarecurrentlyliving.

Basedon theunderstandingof thedifferentbiasesandstrengthsandweaknessesof themethodstheprogrammemanagerswereabletodetermineabestestimate:theyestimatedthatthenumberofpersonswhoinjectdrugsinCityYwasapproximately8,000.

4.1.4. Estimating the population size of most-at-risk adolescents ProgrammesaimedatpreventingHIVinfectionareoftentargetedtospecificagegroups.Thisisespeciallytrueamongpopulationssuchassexworkersorinjectingdrugusers:G Programmesforrecentinitiatorsofthebehaviour(oryoungerpeople)focusonchanginglifestyles(such

asinterventionstodevelopalternativeworkskillsordrugtreatmentprogrammes).G Programmesforpersonswhohavehadthebehaviourforanumberofyears,andwhoaregenerallyolder,

tendtofocusonchangingspecificbehaviours(suchasincreasingcondomuseorusingsafeneedles).

Whenpossible,collectlocalsizeestimatesbydifferentagegroups.Forexample,inalocalstudy,considerthenumberofpeopleinthepopulationunderage25oroverage25.Thismaybechallengingifyouareusingthemethodsdescribedinthisguidelinebecauseageinformationisnotcommonlycollectedinthesemethods.

40

Anotheroptionistousedataavailablefromarecentsurveyofthesamepopulationtoidentifytheproportionofthepopulationindifferentagegroups.Forexampleifasurveywasconductedusingrespondent-drivensamplingofinjectingdrugusersinacity,and17percentofthosewhorespondedtothesurveywerelessthan25.Youcouldapplythatproportiontoyoursizeestimatesandestimatethenumberofyoungpeoplewhoinjectdrugs.(However,iftherearelargebiasestothesamplingstructuretheproportionsindifferentagegroupsmightbebiased.)

Beforeyoubeginestimationcalculations,besuretolearntheparameterssetforinclusioninthesurvey.Inmanysurveys,adolescentsundertheageof18arepurposefullyexcludedbecauseofethicalconcerns.

Youmayfindnosurveyexiststhatprovidesunbiasedestimatesoftheproportioninthepopulationunderage25.Ifso,consideraddingacomponenttotheproposedmethodsthatallowsyoutomakeanestimateofat-riskadolescents.G Ifenumerationorcensusmethodsareused,theadditionofasimpleclassificationbyagecouldbeadded

totheenumerationtool.G Ifauniqueidentifierisused,amarkontheobjectorcolouroftheobjectcouldbeusedtoidentifyifthe

recipientislessthan25yearsold.G Ifyouareusing themultipliermethod (recall that thismethodreliesonprogrammerecords),service

providersmightnotalwayshavetheageoftheattendees.G Ifnetworkscale-upisbeingused,youwouldhavetoaskanadditionalquestionaftereachpopulationof

interest.Forexample,“Amongthesexworkersthatyouknow,howmanyareunderage25?”Theresultsofthesequestionswillintroduceadditionalbiasessothismethodshouldonlybeusedasalastresort.

4.2.Step8:DocumenttheprocessThe most important step in size estimation comes after you create the size estimate. Consider how toimprovethelongtermuseoftheestimate.G Carefully document all details of how the method was carried out if the estimation method is to be

replicatedtoproducecomparableresults.G Useclearandappropriatelanguage.

Sizeestimatesaremoreusefulwhentheyareupdatedovertimeandcanbestudiedforchanges.Increasesordecreasesinthepopulationsizeofmost-at-riskpopulationsaremoreusefuliftheycanbeassociatedwithinterventions.

Keep in mind that the methods described in this guideline produce estimates with a large degree ofuncertainty.Thiscouldmakeitdifficulttomeasuresignificantchangesinthepopulationsize.Alsoconsiderchanges in the larger population (such as increasing or decreasing total population size or increasingnumbersofyoungpeople)whenyouarelookingforlongtermtrends.

Statingthesteps,assumptions,techniquesandcalculationstakentocreatethepopulationsizeestimatewillallowfutureresearchers(oreventhesameteam)toreplicatetheprocessinthefuture.Havingcomparablemeasuresofsizeestimatesshouldbeaveryhighpriorityforpolicyandprogrammemanagerswhoneedtomeasuretheimpactoftheirprogrammesandpolicies.

Startyourdocumentationwiththeprotocolasitwasinitiallydeveloped.Theprotocolmustinclude:G anexplanationofwhythepopulationwaschosenG thedefinitionusedforthepopulationG thegeographicareaoftheestimateG themethodchosenfortheestimateG theassumptionsrequiredforthemethodG anyviolationsofthoseassumptions.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

41

Clearlydescribethedatasourcesyouusedforthesizeestimate.Ifdecisionsweremadeaboutwhetherornottouseadatasource,includeadecisiontreediagramtodescribehowthosedecisionsweremade.

Inaddition,amendtheprotocoltoincludeexplanationsofanychallengesthatcameupduringtheestimationexerciseandhowtheywerehandled.Forexample,youmayhavehadchallengeswithsamplingorsurveyimplementation thatcouldhavebiased theresults.Datahandlinganddataprocessing issuesmayhaveresultedinchangestotheanalysisplan.

It is importanttodocumentwhichpartsoftheat-riskpopulationmaybemissedentirely.Forexampleiffemaledruguserswhoreceivethedrugsfromtheirpartnerswerenotcapturedinasurveyofindividualsatadrugtreatmentclinic,thedocumentationshouldhighlightthatsuchindividualswerenotcountedintheexercise.Suchissuesshouldbecarefullynotedinthedisseminationmaterialsaswell.

Finallytheresearchersshouldincludeasectioninthedocumentationonwhethertheybelievetheestimateisanunderoranoverestimateandshouldincludethereasonsforthatconclusion.

4.3.Step9:DisseminatetheresultsSize estimates of populations at increased risk to HIV can be politically sensitive. Also, the media maymisinterprettheresults.Whenyouaredisseminatingasizeestimate,studythewordingandmechanismscarefullybeforeyoureleaseresults.

Thereareseveralwaystoreleasethisinformation:G apressreleaseG atechnicalreportG abriefingtopolicymakersG abriefingwithmembersofthepopulationoranon-governmentorganisationthatprovidesservicesor

representsthatpopulation.

Itmaybeusefultodevelopatablethatlistsstakeholderswhoneedaccesstotheresultswiththemethodyoupropose for sharing the results, the timingof the releaseandanyotheruseful information.Timingthedisseminationactivitiesbasedon thepriority levelof thestakeholderallows thestudy teamtimetoincorporatefeedbackfromearliersessionsintoafinalreport.Also,creatingatablewillhelpyoutodeterminewhatproductsareneededfromtheestimationexercise.Table4.2providesanexample.

Table 4.2. Example of dissemination plan matrix

Stakeholder Methodofsharing Timingofdissemination

Civil society organisations, most-at-risk population Briefing First

Policy and programme managers Briefing, executive summary Second

Development partners Technical report Third

Media Press release Fourth

Sometimestheresultsof theestimationmaythreatenacommunity.Forexample, ifasmallcommunity(generalpopulation)learnsthatthereareover2,000personswhoinjectdrugsintheircity,theremightbeabacklashagainstpersonswhoinjectdrugs.

Incorporating messages about how to prevent or treat drug addiction in your final report might avoidsuchsituations.Programmemanagerswhoworkwithmost-at-riskpopulationsshouldbeinvolvedinthedevelopmentofthereportandotherproductsyouplantodisseminate.

42

4.4.Step10:UsethesizeestimatesUsingtheestimatesappropriatelyforprogrammeandplanningactivities iscritical.Thejobofthestudyteamwhocreatedthesizeestimatesisstillnotcomplete,evenafterithasbeendisseminated.

Sizeestimatesshouldbeusedfordesigninganddevelopinginterventionprogrammestohelpindividualsavoidriskybehaviours:G CondomdistributionG CleanneedleexchangeG DrugtreatmentprogrammesG Microfinanceeconomicopportunitiestofacilitatepreventing/leavingsexwork

Theestimatednumberofpeopleinthepopulationswillhelpdeterminethemagnitudeandtheresourcesneededforsuchinterventions.

Sizeestimatesshouldalsobeincludedintheprocessofknowingyourepidemic.Thesizeestimatescanbeusedincreatingnationalprevalenceestimatesinlowandconcentratedepidemiccountries.Inaddition,thesizeestimatesareneededforcreatingmodelsonwherefutureinfectionswilltakeplace.

Teamsdevelopingthenationalstrategicplanandcostingthoseplanswillneedtheestimatestodeterminetheresourcesrequiredformostatriskpopulations.

Finally the size estimates are often used as denominators for reporting on international monitoringindicators(suchastheUNGeneralAssemblySpecialSessiononHIVindicators).Whenapplyingforgrantsfrominternationalorganisations,countriesarerequestedtoincludeinformationonsizeestimatesfortheirmost-at-riskpopulations.

Theimplementersofthesizeestimateexerciseshouldideallyensurethatthesizeestimatesarebeingused,andbeingusedcorrectly,inthesedifferentapplications.

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

43

Appendix A. Useful data sources for size estimation

Table 1. Drug-related data

Data source Does it exist?For what cities/areas? Data quality?

If it exists, is it available for regular estimation work?

If not, how feasible to set up data collection?

Can you bring examples to the meeting?

Behavioural Sentinel Surveys (BSS)

Other focused surveys

Household surveys

HIV surveillance

Register of addicts

Treatment centre data

Police data, by reason for arrest

Court data, by reason for trial

Prison data, by reason for conviction

Hospital data on drug-related death

Mortality data on drug-related death

Health service data on positive drug tests

Employment agency data on positive drug tests

Data from needle exchange programmes

Data from outreach programmes

Research studies

Anything alse?

44

Table 2. Data related to sex work

Data source Does it exist?For what cities/areas? Data quality?

If it exists, is it available for regular estimation work?

If not, how feasible to set up data collection?

Can you bring examples to the meeting?

BSS among sex workers

BSS among clients

Other focused surveys among sex workers

Mapping of brothels/red light districts

Registry of brothels

Police data on brothels

Registry of “short stay” hotels

Registry of karaoke bards or “entertainment places”

Other focused surveys among clients

Household surveys of consumption of commercial sex

HIV surveillance

Register of sex workers

STI clinic data

Police data, by reason for arrest

Court data, by reason for trial

Prison data, by reason for conviction

Condom sales or distribution data

Data from outreach programmes

Research studies

Anything else?

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

45

Table 3. Data related to MSM/transvestites

Data source Does it exist?For what cities/areas? Data quality?

If it exists, is it available for regular estimation work?

If not, how feasible to set up data collection?

Can you bring examples to the meeting?

BSS among MSM/transvt

BSS among clients

Other focused surveys among MSM/transvt

Mapping of cruising areas

Registry of bars or “entertainment places”

Household surveys with questions about same-sex experience

HIV surveillance

MSM clinic data

STI clinic data

Data from outreach programmes

Research studies

Anything else?

46

1 Vandepitte,J.,etal.,Estimates of the number of female sex workers in different regions of the world.SexuallyTransmittedInfections2006,82(Supplement3):iii18-iii25.

2 Gouws,E.,etal.,Short term estimates of adult HIV incidence by mode of transmission: Kenya and Thailand as examples. SexuallyTransmittedInfections2006,82(Supplement3):iii51-iii55.

3 FHI, Impact,USAID,UNAIDS,WHO,andUNDCP.EstimatingtheSizeofPopulationsatRisk forHIV:IssuesandMethods.2003:03-36E.Availableathttp://data.unaids.org/pub/Manual/2003/20030701_gs_estpopulationsize_en.pdf.Accessed22February2010.

4 Lyerla R, Gouws E, Garcia-Calleja JM. The quality of sero-surveillance in low- and middle-incomecountries: status and trends through 2007. Sexually Transmitted Infections 2008, 84(Supplement1):i85-i91.

5 BarcalK,SchumacherJE,DumchevK,MorozLV.AsituationalpictureofHIV/AIDSandinjectiondruguseinVinnitsya,Ukraine.HarmReductionJournal2005,2:16doi:10.1186/1477-7517-2-16(accessed22February2010).

6 Heimer,R.andE.White,EstimationofthenumberofinjectiondrugusersinSt.Petersburg,Russia.DrugandAlcoholDependence,2010.InPress,CorrectedProof.

7 MastroTD,etal.EstimatingthenumberofHIV-infected injectiondrugusers inBangkok:acapture-recapturemethod.AmJPublHealth1994;1094-99.

8 BrownP.RateofHIVtransmissionamongAfricansinUK“underestimated”.BMJ2000;320(7237):735.

9 FairchildAL,GableL,GostinLO,BayerR,SweeneyP,JanssenRS.PublicGoods,PrivateData:HIVandtheHistory,Ethics,andUsesofIdentifiablePublicHealthInformation.PublicHealthReports2007;122:7-15.

10 Baker MG, Fidler DP. Global public health surveillance under new international health regulations.EmergInfectDis2006;12:1058-65.

11 Schenk K, Williamson J. 2005. Ethical Approaches to Gathering Information from Children andAdolescentsinInternationalSettings:GuidelinesandResources.Washington,DC:PopulationCouncil,2005.

12 Pollock KH. Modeling capture, recapture and removal statistics for estimation of demographicparametersforfishandwildlifepopulations.JAmerStatAssn1991;86:225-38.

13 Botero CA, Mudge AE, Koltz AM, Hochachkat WM, Veherncamp SL. How reliable are methods forestimatingrepertoiresize.Ethology2008;114:1227-38.

14 WittesJT,ColtonTandSidelVW.Capture-recapturemodelsforassessingthecompletenessofcaseascertainmentusingmultipleinformationsources.JChronicDis1974;27:25-36.

15 HookEB,RegalRR.Capture-recapturemethodsinepidemiology:methodsandlimitations.EpidemiolRev1995;17:243-64.(CorrigendainAmJEpidemiol1998;148:1219).

References

Estim

atin

g th

e si

ze o

f pop

ulat

ions

mos

t at r

isk

to H

IV

47

16 TillingK.Capture-recapturemethods-usefulormisleading?IntJEpidemiol2001;30:12-14.

17 Blanxart MF, Olmos JG, Serrano MLH, Cebollero MP. Validation of the capture-recapture method.Psicothema2000;12(Suppl2):231-5.

18 Millar T, Comingo-Salvany A, Eastwood C, Hay G. Glossary of terms relating to capture-recapturemethods.JEpidemiolandCommHealth2008;62(8):677-81.

19 Robson, et al. Sample Size in Petersen Mark-Recapture Experiments. Transactions of the AmericanFisheriesSociety,1964.93(3):p.215-226.

20 MastroTD,etal.EstimatingthenumberofHIV-infected injectiondrugusers inBangkok:acapture-recapturemethod.AmJPublHealth1994;1094-99.

21 UuskűlaA,RajaleidK,TaluA,AbelK,RüütelK,HayG.Estimatinginjectiondruguseprevalenceusingstatewiseadministrativedatasources:Estonia2004.AdditionResearchandTheory:2007;16:411-24.

22 HeckathornDD.Respondent-drivensamplingII:derivingvalidpopulationestimatesfromchain-referralsamplesofhiddenpopulations.SocialProblems2002;49:11-34.

23 OveF,SnijdersT.Estimatingthesizeofhiddenpopulationsusingsnowballsampling.JOfficialStat1994;10:53-67.

24 EricksonBH.Someproblemsofinferencefromchaindata.SociologicalMethodology1979:10;276-302.

25 HeckathornDD.Respondent-drivensamplingII.Derivingvalidpopulationestimatesfromchain-referralsamplesofhiddenpopulations.SocialProblems2002;49(1):11-34.

26 Hickmanetal.Estimatingdrugprevalence: reviewofmethodswithspecial referencetodevelopingcountries,BulletinonNarcotics,UNOfficeonDrugsandCrime,2004.

27 ZhangD,WangL,LvF,etal.AdvantagesandchallengesofusingcensusandmultipliermethodstoestimatethenumberoffemalesexworkersinaChinesecity.AIDSCare2007;19:17-9.

28 SaidelT.Estimatingpopulationsizesintheabsenceofa“goldstandard”:multipliermethodinIndia.PresentedatUNAIDS/WHOReferenceGroupMeeting,Amsterdam,December2008.

29 LinkM,BattagliaM,FrankelM,OsbornL,MokdadA.Address-basedversusrandom-digitdialsampling;comparisonofdataqualityfromBRFSSmailandtelephonesurveys.In:Proceedingsofthe2005FederalCommitteeonStatisticalMethodologyResearchConference[CD-ROM];Arlington,Virginia.Availableathttp://www.fcsm.gov/events/papers05.html.Accessed24February2009.

30 Measure DHS+ (2000) Demographic and Health Surveys. www.macroint.com/dhs 2000 MacroInternationalCalverton,MD.Accessed24February2009.

31 CDC. Methodology of the Youth Risk Behavior Surveillance System Morbidity & Mortality Weekly Report2004;53(RR-12):1–13.

32 KalsbeekWD.TheUseofSurveys inPublicHealthSurveillance:MonitoringHigh-RiskPopulations.Chapter inMonitoring theHealthofPopulations (BrookmeyerR,StroupDFeds).NewYork:OxfordUniversityPress,2004.

48

33 NathMB.AgenderedresponsetoHIV/AIDSinSouthAsiaandthePacific:insightsfromthepandemicinAfrica.Gender&Development.2006;14:11-22.

34 McCarty,C.,1997,Subgroupsinpersonalnetworks:Theinsideview.AmericanJournalofCommunityPsychology1997.

35 KochenM,ed.TheSmallWorld.NorwoodNJ:Ablex,1989.

36 KillworthPD,McCartyC,JohnsenEC,BernardHR,ShelleyGA.Investigatingthevariationinpersonalnetworksizeunderunknownerrorconditions.SociologicalMethodsandResearch2006;35:84-112.

37 BernardHR,JohnsenEC,RobinsonS.Esimatingthesizeofanaveragepersonalnetworkandofaneventsubpopulation:someempiricalresults.SocialScienceResearch1991;20:109-21.

38 Laumann EO, Gagnon JH, Michaels S., Michael RT, Schumm LP. Monitoring AIDS and other rarepopulationevents:Anetworkapproach,JournalofHealthandSocialBehavior1993;34:7-22.

39 KillworthPD,McCartyC,JohnsenEC,BernardHR,ShelleyGA.Investigatingthevariationinpersonalnetworksizeunderunknownerrorconditions.SociologicalMethodsandResearch2006;35:84-112.

40 Lynskey M, Hall W. Jurisdictional trends in opiod deaths 1988-1995. Australian and New ZealandJournalofPublicHealth.1998;23:519-24.

Guidelines on Estimating the Size of

Populations Most at Risk to HIV

UNAI

DS/W

HO W

orki

ng G

roup

on

Glo

bal H

IV/A

IDS

and

STI S

urve

illan

ce

For further information, contact:World Health OrganizationDepartment of HIV/AIDS20, avenue Appia CH-1211 Geneva 27 Switzerland

E-mail: hiv-aids@who.inthttp://www.who.int/hiv/en

ISBN 978 92 4 1599580