Queensland Centre for Population ResearchThe IMAGE Inventory:A User GuideMartin Bell, Aude Bernard, Elin Charles-Edwards, DorotaKupiszewska, Marek Kupiszewski, John Stillwell, Yu Zhu,Philipp Ueffing and Alison BoothFebruary 2015Working Paper No 2015/01School of Geography, Planning andEnvironmental

Contents12Introduction . 31.1The IMAGE Project . 31.2The IMAGE Inventory . 41.3Outline. 4Impediments to cross-national comparisons of migration . 52.1Dimensions of migration . 52.2Comparability of migration data . 92.3Type of migration data. 92.4Observation interval . 102.5Spatial framework . 113Review of internal migration data collection practice . 114Assembling the IMAGE Inventory . 135IMAGE Inventory: Files and Content. 1465.1Files . 145.2Content . 155.2.1Inventory Overview . 155.2.2Census Inventory. 165.2.3Register Inventory . 195.2.4Survey Inventory . 20Access to the IMAGE Inventory . 21References . 242

1 Introduction1.1 The IMAGE ProjectThe IMAGE project (Internal Migration Around the GlobE) is an international program ofresearch which aims to facilitate comparisons of internal migration between countries byimplementing a set of robust indicators that measure different dimensions of populationmobility. Comparative analysis calls for careful consideration of differences in the nature ofdata collected in countries around the world (Bell et al. 2002). These differences arecomplicated by the limited availability of migration data, as national migration statistics arerarely available in a readily accessible format (Bell et al. 2014). Moreover, even where dataare made available, formats vary widely from one country to another. Analysts also confronta number of challenges in deriving rigorous measures of migration.To address the above challenges, the IMAGE project has been developed around a numberof discrete modules, including a global survey of internal migration data, the assembly of acomprehensive repository, and the development of specialised software and analytical tools,as shown in Figure 1. The first step to the IMAGE project was a global inventory of migrationdata collection practice in the 193 UN member states (Bell et al. 2014). Building on thisinventory, a repository of internal migration data has been assembled, currently covering135 countries. In tandem with the IMAGE Repository, a suite of analytical software, theIMAGE Studio, has been developed to facilitate the spatial analysis and modelling of internalmigration by allowing the computation of a range of migration measures (Daras 2014;Stillwell et al. 2014).The IMAGE InventoryReview of internal migrationdata collection practice in the193 UN member statesThe IMAGE RepositoryGlobal collection ofpopulation and internalmigration data and GISboundariesThe IMAGE StudioAnalytical software tocompute internal migrationmeasures and address keymethodological issues3

Figure 1 The IMAGE Project Framework1.2 The IMAGE InventoryThe IMAGE Inventory is a database for developed for the use by the research communityinterested in the study of internal migration that can be downloaded from the IMAGEwebsite ( It is based on findings from the InternalMigration Around the GlobE project, which established an inventory of internal migrationdata collections for the 193 UN member States. The IMAGE Inventory covers three mainsources of internal migration data: population censuses, population registers andadministrative collections, and national sample surveys. It distinguishes data from the latesttwo UN census rounds: the 2000 round (1995-2004) and the 2010 round (2005-2014), butalso includes information on a number earlier censuses, particularly for countries whichhave not undertaken a recent census. The Inventory was last updated in February 2015.The content of the IMAGE Inventory has been carefully selected to provide information thatis thought to be of most value to migration scholars. It contains information on the types ofmigration data collected, the interval over which migration is measured and the spatialframeworks employed, but does not attempt to provide an exhaustive list of all data itemsin any individual data collection. It is important to recognise that data collection does notguarantee dissemination. For example, statistical agencies may code and release migrationdata at spatial scales other than those used for collection purposes. Moreover, relativelylittle data on internal migration is routinely made available in statistical publications orelectronic form, so potential users will often need to inquire further to establish theavailability of items particular data. The IMAGE Inventory essentially represents a first portof call. Nevertheless, in order to facilitate access data, the IMAGE project has alsoestablished a Repository of internal migration data, which currently holds internal migrationdata sets for 135 countries. Access to some data from the IMAGE Repository can beprovided under certain conditions. The Inventory includes links to the IMAGE Repository.1.3 OutlineThis document is a user guide to the IMAGE Inventory and is organised as follows. Section 2summarises the main impediments to comparing migration between nations. Section 3summarises current data collection practice in countries around the world. Section 4discusses the strategy employed to assemble the IMAGE Inventory, with regard to thecharacteristics of the migration data selected and the data sources used. Section 5introduces the content and structure of the Inventory and lists the different files availablefor each country and each data source. Finally, Section 6 explains how to access data fromthe IMAGE Inventory.4

2 Impediments to cross-national comparisons of migrationThree main issues stand in the way of effective cross-national comparisons of internalmigration: the multifaceted nature of migration itself; the choice of migration indicators;and the widespread variation in the type of migration data that are collected. In anycomparative analysis, a crucial first decision concerns the particular aspect of migration tobe explored. Four discrete dimensions of migration can be recognised: intensity, impact,distance and connectivity (Bell et al. 2002), each of which call for somewhat different formsof data and methods of analysis. The IMAGE Repository has been assembled to facilitatecomparisons on all these aspects of migration, so an understanding of the four dimensionsof migration, the associated data and the migration indicators is essential to appreciate theway in which the Repository has been designed.2.1 Dimensions of migrationFour broad dimensions of internal migration can be recognised, each of which providesinsights into a different aspect of migration at a particular spatial scale. These are:(1) migration intensity, which indicates the overall level or incidence of migrationwithin a country;(2) migration impact, which captures the extent of population redistribution throughmigration;(3) migration distance, which indicates the distance decay associated with populationmovement; and(4) migration connectivity, which reveals the way migration serves to link cities andregions.Each of these dimensions can be captured using a number of statistical indicators and eachof these indicators requires somewhat different forms of data. Table 1 lists in summary formthe suite of 15 migration measures proposed by Bell et al. (2002) and sets out the datarequired for their computation. Table 1 also indicates which of these measures aregenerated in the IMAGE Studio.For the purposes of discussion, it is useful to identify three broad forms in which migrationdata are commonly available:(1) Origin-destination matricesAlso described as flow matrices, these contain region-to-region migration flows in which,by convention, rows represent origins and columns represent destinations. The diagonalelement of flow matrices commonly indicates the number of people who changedresidence but remained in the same region at the start and end of the observation,interval but it may also contain non-movers, or a subset of people who moved betweensmaller zones within the region.(2) Marginal totals5

In some cases, full origin-destination matrices are not available, but data may beprovided on the aggregate number of arrivals and departures for each region. Theseeffectively represent the marginal (row and column) totals of a full origin-destinationflow matrix and are also referred to as zonal inflows and outflows. While theseaggregates do not provide information on flows between specific origin-destinationpairs, they can be used to generate aggregate system-wide measures of migrationimpact, and are also commonly available disaggregated by age and sex. It is important toknow whether or not the marginal totals include or exclude flows taking place withinregions.(3) National migration countsCount data comprise a single figure which indicates the total number of movers ormoves between regions within a country, irrespective of origin and destination. Theymay also include information on changes of address that occurred within regions. Theydo not provide any information on the origin or destination of flows, but may bedisaggregated by age and sex.Allied to these three forms of migration data are two other types of information that arerequired to compute some of the 15 indicators associated with the four dimensions ofmigration listed above. These are:(1) digital boundaries matching the regions against which migration is recorded, which areused to compute migration distances, calibrate spatial interaction models and drive thespatial aggregation facility in the IMAGE Studio, and(2) populations at risk, which are used to compute migration rates and probabilities.Table 1 shows how these data combine to generate the various migration metrics. Withrespect to intensity measures, for example, the Crude Migration Intensity, which is obtainedby dividing the number of migrants by the population at risk, can be computed withmigration data of any format, because it simply requires the aggregate number of migrants.This may be directly available in the form of a national migration count, but it can also bederived by summation from an origin-destination matrix or from marginal totals. Otherintensity measures require migration data disaggregated by age, which may be available inflow matrices but are more commonly held only in the form of nationwide migration countsor marginal totals. Migration impact measures relate to individual regions so theircomputation requires data on inter-regional flows, which can be obtained either fromorigin-destination matrices or marginal totals. Distance measures are generated as Euclidiandistances between the region centroids by the IMAGE Studio, but it is also possible to inputa matrix of distance values assembled independently. Connectivity measures, on the otherhand, require complete origin-destination flow matrices, and cannot be computed frommarginal totals or from count data.6

Table 1 shows that all measures of impact, distance and connectivity can be generated inthe IMAGE Studio, with the exception of the Migration-Weighted Gini which is excludedbecause of the high computational load leading to long processing times. As for measuresof intensity, the IMAGE Studio generates only the Crude Migration Intensity. Othermeasures of intensity require migration data disaggregated by age and are not computed inthe IMAGE Studio.The IMAGE Studio incorporates a spatial aggregation routine which was designed to assist ingenerating estimates of migration intensity that are comparable between countries, and toexplore the scale and pattern effects of the Modifiable Areal Unit Problem (MAUP) (Stillwellet al. 2014). Provided the requisite types and forms of information are available, the Studiocan be employed to generate migration metrics for any size of migration matrix based on aset of Basic Spatial Units (BSUs).7

NTable 1 Data needed for computation of migration measures in the IMAGE StudioMIGRATION MEASUREPopulation atriskOrigin-destinationmatrix Measuresgeneratedin theIMAGE StudioCrude Migration IntensityCMI Standardised Migration IntensitySMI Gross Migraproduction RateGMR Migration ExpectancyME Intensity at Migration PeakIMP Age at Migration PeakAMP Migration Effectiveness IndexMEIAggregate Net Migration RateANMRMedian Distance MovedMDMean Distance MovedMDMDistance Decay ParameterbIndex of Migration ConnectivityIMC Index of Migration InequalityIMI Migration-Weighted GiniMWF Coefficient of VariationACV Theil IndexTHEIL DATA REQUIREMENTData format1National count dataZonal in and outflowsTotalby age digitalboundaries digitalboundaries digitalboundaries sex-specific measures of intensity can be computed when national count data are disaggregated by sex.8

2.2 Comparability of migration dataComparability between countries is complicated by the fact that, irrespective of dataformat, migration can be measured in various ways using different instruments ranging frompopulation registers and administrative records, to censuses and surveys. As a result,significant variation exists between countries with regard to the type of migration datacollected, the time interval over which migration is measured, and the spatial frameworksused.2.3 Type of migration dataMigration can be measured as an event or transition, or by reference to duration ofresidence. Event data are usually associated with population registers and are the mostcommon form of internal migration data available in many European countries. Transitiondata measure migration by comparing place of residence at two points in time and are thetype of data most commonly derived from censuses and surveys (Bell et al. 2014). Becauseof the way they are measured, events and transitions count different things; populationregisters count migrations while censuses count migrants. The difference is importantbecause transition data fail to capture return and onwards moves that occur within theobservation interval, and therefore undercount the number of migration events. There arealso differences in the treatment of migration among those who are born or die in theinterval, as well as in the inclusion or exclusion of immigrants (Bell and Rees 2006), andthese may vary further between individual countries. The impact of these differences onoverall migration intensities is small over relatively short intervals (Long and Boertlein 1990)but increases as the observation interval lengthens, and care is also needed to eliminate orcontrol for variations in population coverage (Boden et al. 1991).Censuses around the world also commonly collect data on duration of residence in thecurrent dwelling or locality. By filtering duration data for fixed durations of residence, it ispossible to derive a surrogate estimate of the number of moves that have occurred within agiven interval, comparable to the migration count data commonly derived from events ortransitions. Duration of residence data are also commonly collected in association withinformation on previous place of residence, to generate origin-destination matrices thateffectively capture each respondent’s last move. In this instance, duration of residence canbe used as a filter to generate a migration flow matrix, which is broadly comparable to theconventional migration transition. However, lack of precision in the measurement ofduration and ambiguity in the locality to which it applies necessitate caution in the use ofthese data to make cross-national comparisons (Bell et al. 2014).Table 2 summarises the principal differences between events, transitions, duration and lastmove data. It shows that duration and last move data essentially represent a hybrid in termsof population and migration coverage. They fail to capture multiple migration events thatoccur within the observation interval but capture return moves missed by transition data.9

Despite these measurement differences national count data can be generated from event,transition, duration or last move data. Origin-destination matrices, zonal inflows andoutflows can be generated from all data types, except when duration data are collected inisolation from information on previous place of residence.Table 2 Population coverage of migration flow matrices by data typeElementEventsMigrationAll moves undertakenwithin the intervalPopulationIndividuals in thepopulation during theobservation intervalImmigrantsEmigrantsBorn inintervalDie duringintervalIncludedIncludedTransitionsChanges of residencebetween the start and endof the intervalIndividuals alive andresident in the country atboth the beginning and endof the interval.ExcludedExcludedLast dedExcludedMost recent moves within thedefined duration of residenceIndividuals resident in thecountry at the end of theinterval.IncludedExcludedSource: Bell et al. (under review)2.4 Observation intervalMigration can be measured over a range of observation intervals, which may be of a fixed(defined) or variable length and the IMAGE Repository incorporates flow matrices pertainingto these intervals.Countries which measure migration over a fixed interval commonly use one or five years asthe interval length, but other intervals ranging from two to ten years are also employed (Bellet al., 2014). Migration data measured over intervals of different length are not readilycomparable due to the effects of chronic and repeat movement. The consequence is thatfive-year migration data are not equivalent to five times the one-year migration data, withempirical evidence showing that the ratio between one and five-year transition rates variesover time and space. While approximate conversions have been proposed, there is nostraightforward analytical solution to harmonise these data (Kitsul and Philipov 1981;Rogers et al. 2003; Rogerson 1990). Data on migration events, derived from populationregisters and administrative collections commonly refer to a single-year interval.Some countries measure migration by comparing place of residence with place of birth,which delivers a measure of lifetime migration, and this in fact is the most commonmeasure of internal migration collected by censuses worldwide (Bell et al. 2014). Lifetimemigration data provide useful insights into the cumulative impact of migration over apopulation’s collective lifetime. However, because individuals have been exposed tomigration for varying periods, differences in age structure prejudice comparability between10

countries. Moreover, lifetime migration data offer limited insights into contemporarymigration processes.A third approach to the collection of migration data is based on asking each person’sprevious place of residence (PPR), irrespective of when the migration look place. As notedabove, these data can be filtered by duration of residence, if collected, to generate asurrogate estimate that approximates a fixed interval transition.2.5 Spatial frameworkCountries also vary widely in regard to the number of spatial units into which they aredivided and which are used to record migration. Some countries record all changes ofaddress, including those that take place within the same region, but for most it is the changeof address that crosses regional boundaries that is recorded in the migration flow matrixdescribed earlier. The level at spatial scale at which data are collected, as apparent on theCensus form or other record, does not necessarily correspond to the level at which the dataare subsequently coded or made available. At the same time, migration flow matrices forthe same country, source and time interval may be made available at multiple levels ofaggregation.3 Review of internal migration data collection practicePrevious sections have highlighted the diversity of ways in which migration is collected andmeasured. The IMAGE Inventory covers three main sources of internal migration data:population censuses, population registers and administrative collections, and nationalsample surveys. Five main strategies were used to identify the forms of data on internalmigration collected by individual countries around the world: (1) mining of statisticalorganisation websites; (2) review of prior inventories and papers; (3) questionnaire surveyof statistics agencies; (4) analysis of country census and survey forms; and (5) advice froman international network of scholars. The review process was conducted in 2013 for the 193UN member states, and resulted in two main outputs. First, a synthesis and assessment ofglobal data collections was produced (Bell et al. 2014), which provides summary statistics onthe number of countries collecting migration data, the sources used, and the types ofmigration data collected. Second, a detailed inventory was assembled, which providesinformation on the types of migration data collected for each 193 UN member state. Beforeintroducing the IMAGE Inventory, the remainder of this section provides an overview of thefindings published in Bell et al. (2014).The IMAGE survey shows that most countries rely on population censuses to measureinternal migration (158 countries). As shown in Table 3, nationally representative surveysare also widely used (110 countries), while population registers and administrative data11

feature in 50 countries and are the dominant collection practice in Europe. Most countries(109) draw migration statistics from multiple sources.Table 3 Countries collecting internal migration data since 1995 by region and sourceRegisterSurveyMultiplesourcesTotal countriescollecting dataTotalcountriesAfrica430Asia3715Europe3132Latin America320Northern America22Oceania131Total15850Source: IMAGE Inventory; see Bell et al. 4333214193RegionCensusAmong countries using census-based statistics, lifetime migration based on region of birth isthe most common migration measure worldwide (122 countries), but many countries alsomeasure migration by reference to last place of residence. A total of 52 countries measuremigration over a five-year interval, whereas 29 countries use a one-year interval. A total of71 countries collect data on duration of residence at their census, often in association withplace of last residence.Table 4 Countries collecting internal migration data in the 2000 UN Census round byregionType of DataObservation PeriodOtherOneFive Europe1341225Latin America217229Northern America1202Oceania28210Total295232122Source: IMAGE Inventory; see Bell et al. (2014)RegionLast move131810120255Duration a32353129213142The review of two large-scale survey programs in developing countries the USAID’sDemographic and Health Surveys (DHS) and the World Bank’s Living StandardsMeasurement Study (LSMS) showed that duration of current residence is also commonlyasked, together with place of previous residence. Duration of residence is available for atotal of 60 countries in the DHS and 14 countries in the LSMS. Among developed countries,the review of the European Union’s Labour Force Surveys and the American Community12

Survey showed that most countries utilising surveys measure migration by reference toplace of residence one year previously.Table 5 Countries collecting internal migration data by survey(s) by regionRegionDemographicand HealthSurveyAfrica38Asia18Europe3Latin America and Caribbean10Northern America0Oceania1Total70Source: IMAGE Inventory; see Bell et al. rnationalSurveyAll surveys0826020363824321222110Population registers and administrative records capture migration as an event (Rees et al.2000), although it is feasible to generate transition data from comparison of registers at twopoints in time. The IMAGE inventory identified 50 nations producing migration statisticsusing administrative records or population registers, 32 of them being in Europe.4 Assembling the IMAGE InventoryTo facilitate access to the findings of this review, a detailed inventory was assembled toprovide information on the types of migration data collected for each of 193 UN memberstates. It is important to note that not all data collected are subsequently released and thatthe IMAGE Inventory provides information on what is theoretically available. However, theInventory also contains information on register-based migration statistics known to beavailable.The IMAGE Inventory contains information on the types of migration data collected, theinterval over which migration is measured and the spatial frameworks employed, but doesnot attempt to provide an exhaustive list of all data items in any individual data collection.This limitation is particularly true in the case of surveys, since many provide informationsuch as detailed migration histories which cannot be adequately summarised here. Whilesome information items, such as the types of migration data collected, the interval overwhich migration is measured, and the spatial frameworks are common to censuses,registers and surveys, others are specific to particular data sources. For example,information on migrant characteristics, including age and sex, are reported only for registerswhereas reasons for moving are reported only for surveys.13

The IMAGE Inventory endeavours to provide a comprehensive global collection of internalmigration practices. While it provides complete coverage of censuses, population registersand administrative collections, a full inventory of migration surveys is out of reach. Theinventory focuses mainly on surveys conducted since 1995 that potentially facilitate crossnational comparison, such as the USAID's Demographic and Health Survey, the World Bank'sLiving Standards Measurement Study and the European Union Labour Force Surveys.5 IMAGE Inventory: Files and Content5.1 FilesThe three main sources of migration data - population censuses, population registers andadministrative collections, and national sample surveys – vary markedly in the way theymeasure migration. Because differences in the way migration is measured ultimately shapesthe utility of the data (Bell et al. 2014), the IMAGE Inventory has been structured aroundthree discrete components, one per data source. A fourth component provides a summaryof the findings across the three data sources and is intended to serve as a readily accessibleguide to the types of data available in each country. Reflecting this structure, the IMAGEInventory takes the form of four Excel spreadsheets in which rows present unique records,each pertaining to a particular country for a particular year, and columns sets out theparticular feature of the data collected for each country (e.g. observation interval). Thesefour Excel components of the Inventory are as follows:(1) Inventory OverviewThe overview file indicates whether a census, register or national survey is conducted ineach country and whether these instruments are used to collected any form ofmigration data. This file enables analysts to identify whether migration data arecollected in their country of interest, and from which type of data collection technique.Note that this file does not contain any information on the type of migration datacollected.(2) Census CatalogueThis file provides information on migration data collected at the census in each country.It contains fields which indicate, inter alia, the observation interval and spatial scale atwhich the data were collected, as well as general information about the census,including the da

of discrete modules, including a global survey of internal migration data, the assembly of a comprehensive repository, and the development of specialised software and analytical tools, as shown in Figure 1. The first step to the IMAGE project was a global inventory of migration data collec