Report No: ACS9402 . EU Accession Countries Poverty Mapping of New Members in EU Completion Memo . June 11, 2014 . ECSP3 EUROPE AND CENTRAL ASIA . . Document of the World Bank . Standard Disclaimer: . This volume is a product of the staff of the International Bank for Reconstruction and Development/ The World Bank. The findings, interpretations, and conclusions expressed in this paper do not necessarily reflect the views of the Executive Directors of The World Bank or the governments they represent. The World Bank does not guarantee the accuracy of the data included in this work. The boundaries, colors, denominations, and other information shown on any map in this work do not imply any judgment on the part of The World Bank concerning the legal status of any territory or the endorsement or acceptance of such boundaries. . Copyright Statement: . The material in this publication is copyrighted. Copying and/or transmitting portions or all of this work without permission may be a violation of applicable law. The International Bank for Reconstruction and Development/ The World Bank encourages dissemination of its work and will normally grant permission to reproduce portions of the work promptly. For permission to photocopy or reprint any part of this work, please send a request with complete information to the Copyright Clearance Center, Inc., 222 Rosewood Drive, Danvers, MA 01923, USA, telephone 978-750-8400, fax 978-750-4470, http://www.copyright.com/. All other queries on rights and licenses, including subsidiary rights, should be addressed to the Office of the Publisher, The World Bank, 1818 H Street NW, Washington, DC 20433, USA, fax 202-522-2422, e-mail pubrights@worldbank.org. Poverty Mapping in the New Member States of the EU (P130960) Completion Memo June 11, 2014 Background According to the latest data from Eurostat, more than 124 million people in the European Union—almost 25 percent of EU citizens—are at risk of poverty or social exclusion1. One of the five headline targets of the Europe 2020 strategy is to reduce the number of people living at risk of poverty or social exclusion by 20 million by the year 2020. In its 2014–2020 multiannual financial framework the EU has budgeted one trillion euros to support growth and jobs and reduce poverty and social exclusion. Success depends on developing the right policies and programs and targeting them effectively. Just as rates of poverty and social exclusion vary widely across EU member states, there is also a high degree of variability in living standards within member states. It follows that promoting convergence of living standards across the EU requires detailed knowledge of the disparities in living standards within each member state, and especially in those member states with high levels of poverty and social exclusion. In 2011 the European Commission (EC) and the World Bank entered into an agreement to construct small area estimation poverty maps for the new EU member states. The objective is to help the EC and EU member states target funds and programs in the 2014–2020 budget cycle most efficiently to the areas in highest need. These high-resolution poverty maps combine information from recent national population censuses and EU Statistics on Income and Living Conditions (EU-SILC) household surveys2 to estimate the rates of monetary poverty for small geographic areas such as counties, districts, or municipalities. In previous years, the EC has had to rely on less detailed data and maps at the “NUTS 2” level (for example, the eight development regions in Romania) for program planning and the allocation of EU funds.3 The greater geographic disaggregation of the new poverty maps reveals which parts of these larger regions have particularly high rates of poverty and require greater attention for poverty reduction programs; and combined with data on population size, they also provide information on where most of the poor are located. The poverty maps not only help guide allocations of EU funds, but may also be used for decision making and policies at the national and sub-national levels in each of the EU member states. 1 "People at risk of poverty or social exclusion" - Statistics Explained (2014/2/1) 2 For most countries the 2011 population census was used along with EU-SILC from 2011 or 2012. See Table 1 for information about the specific years and levels of aggregation used in each country. 3 The NUTS (Nomenclature des Unités Territoriales Statistiques) classification is a hierarchical system of dividing up the economic territory of the European Union for the development of regional statistics, regional socioeconomic analysis, and the framing of EU regional policies. To date the NUTS 2 classification has been used for determining eligibility for aid from European Structural Funds. Below the NUTS 3 classification areas are defined according to Local Administrative Units (LAU). Most EU member states have LAU 1 and LAU 2 divisions, but some only have LAU 2. The first phase of the project, from February through November 2012, was a methodological pilot study that tested slightly different approaches to small area poverty estimation to arrive at a preferred method of estimating the EU poverty maps. The pilot study was carried out in Slovenia and Denmark, because in addition to having the EU-SILC and population census data, they also have available complete sets of household income data from tax registries. The various methodologies were validated by first estimating the poverty rates in small areas as if only the EU-SILC and census data were available, and then comparing it to the poverty rates calculated from the income tax data, which served as the benchmark reference. The pilot phase was guided by a Steering Committee, which was made up of the leading experts in small area estimation from European universities and national statistical institutes (NSIs); senior staff from Eurostat, DG Employment, and DG Regional Policy; and chaired by Directors of DG Employment and DG Regional Policy. The pilot study report is among the project deliverables, and is described briefly in section 2 below. The second phase of the project, from October 2012 through June 2014, produced the small area poverty maps, which were developed in close cooperation with counterparts in individual EU member states. Constructing the poverty maps from microdata required access to data not available from Eurostat. This typically required conducting the analysis on-site at national statistical offices, with widely varying degrees of involvement by NSO analysts. It also meant that the preferred approach of using survey and census microdata was sometimes not possible, in which case the team resorted to using aggregate, or area-level, data to estimate poverty rates for small areas. After obtaining initial results, the draft poverty maps were shared and discussed with in-country experts, both as an informal validation of the results and to open discussions about the likely underlying causes for the observed spatial patterns of poverty and potential policy remedies. The team then revised the analysis, taking into account the comments received. Small area poverty maps have been produced for seven new EU member states: Estonia, Hungary, Latvia, Poland, Romania, Slovakia, and Slovenia. Under this project, maps were also to be produced for Bulgaria, the Czech Republic, and Lithuania. This has not been possible to date because of delays in census data processing by NSIs, and ongoing negotiations about the data confidentiality agreements required by the NSIs. The EC has expressed an interest in financing a follow-on activity, with one of the top agenda items being completing the set of poverty maps for these countries. Croatia, which was not included in the agreement with the EC, has expressed an interest in working with the Bank to produce a poverty map, and this is likely to occur in 2014 with funding from the EC and the Croatian government. This completion memo briefly summarizes the outputs from this project, describes some of the dissemination and outreach activities, and reviews some of the lessons learned. Summary of Outputs Methodological Pilot Study Report The methodological pilot study used validation methods to compare slightly different approaches to estimating poverty for small areas. The principal comparison was between Elbers-Lanjouw-Lanjouw (ELL) method that has been the main approach used by the World Bank and its partners for approximately the past 15 years, and an extension to ELL proposed by Molina and Rao (2010) commonly known as Empirical Best (EB). Under each of the ELL and EB umbrellas are a host of other assumptions and choices which are also testable, such as how to model the random error associated with location, or how to handle heteroskedastic errors when modeling household-level income. By testing selected variations of ELL and EB on data for which income data were available for the entire population it was possible to see which approaches came closer to the “true” poverty rates as measured directly by the income registry data. The main result of the pilot study was that all of the methods tested provide reasonable estimates of poverty rates at the municipality level, which are the first and second Local Administrative Units (LAU 1 and LAU 2) in Denmark and Slovenia, respectively. A key criterion proposed by the European Commission was “Of the X poorest areas identified by the income tax registry data, how many are correctly identified by each poverty mapping method?” All of the six estimation methods tested correctly identified 80 to 90 percent of the municipalities measured as poorest according to the income tax data. The rare opportunity to do a validation study also provided additional valuable insights that have helped improve our understanding of small area poverty estimation. In particular, the study generated new knowledge about choices concerning incorporation of survey sampling weights in regressions, accounting for heterskedasticity in household income models, and whether the distribution of model parameters should be estimated parametrically or via bootstrap methods. Results4: This output contributed to deepening knowledge, and to generating innovative approaches and solutions. It supported a critical examination of poverty mapping methodologies, including a serious self-examination of methods used within the World Bank Group and their applicability in a European setting, where sample designs are quite different from those found in low income countries. The exercise benefited not only from the knowledge of technical experts in academia and European NSIs, but also from users of this type of information in the European Commission. Beyond addressing the primary comparison of ELL and EB methods, the pilot study has also helped spur new research and developments in small area poverty estimation, including much greater dialogue among the various camps involved. Small Area Poverty Maps and Tables The main output requested by the client (the European Commission) was the poverty maps for new EU member states and the corresponding point estimates of the risk of poverty rates and the associated confidence intervals. These appear in the technical reports and country briefs discussed below. The following table summarizes the level of spatial disaggregation achieved for each of the poverty maps, using both the NUTS/LAU classification system and the common names for the respective regions in each country. The table also indicates the main data sources, and whether the poverty maps are based on microdata or aggregate data. 4 A summary of results, using the results framework from the project concept note, appears at the end of this memo. Table 1: Level of aggregation and data sources for poverty maps Country Level of poverty map Data sources and comments Estonia Municipalities, with some grouping Microdata using 2012 EU-SILC and 2011 of small municipalities (total of 132 Population Census areas, approximately corresponding to LAU 2) Hungary General Electoral Districts (total of Microdata using 2005 EU-SILC and 2005 174 areas, approximately “micro-census”. To be updated with 2011 corresponding to LAU 1) data in 2014. Latvia Statistical regions (total of 6 areas, Aggregate (area-level) data from 2012 corresponding to NUTS 3) EU-SILC and 2011 population census. Plan to update with microdata in 2014. Poland Statistical sub-regions (total of 66 Aggregate (area-level) data from 2011 areas, corresponding to NUTS 3) EU-SILC and 2011 population census. Romania Counties and Bucharest (total of 42 Microdata using data from 2011 EU-SILC and areas, corresponding to NUTS 3) 2011 population census. Slovakia Districts (total of 79 areas, Aggregate (area-level) data from 2011 corresponding to LAU 1) EU-SILC and 2011 population census. Slovenia Municipalities, with some grouping Microdata using data from 2010 EU-SILC and of small municipalities (total of 195 2011 population census. areas, approximately corresponding to LAU 2) Results: This activity helped to deepen knowledge and inform development financing. It has been widely assumed that there is a considerable amount of spatial variation in poverty within the highly aggregated NUTS 2 regions that are used for EU regional policies and programs. However, concrete and sufficiently precise information about the nature of these regional disparities has been lacking. By uncovering the heterogeneity behind aggregate averages, these results permit more accurate identification of the areas in greatest need of development assistance, whether from EU funds or other sources. They also complement other information on the correlates of poverty that will help inform the design of regional policies and programs. Technical Reports The team prepared short technical reports for the poverty maps in each country. These are oriented toward a statistically and econometrically inclined audience. They were specifically requested by Eurostat as a deliverable to help them evaluate and document the technical soundness of the analytical work, in the same vein as the “Quality Reports” that Eurostat and NSIs produce to accompany their survey data results. For each country, the technical report describes (a) the spatial divisions in the country and the target level for the poverty map, (b) the modeling approach, (c) the specific data sources used, and (d) the small area estimation results, including regression results, point estimates and confidence intervals (or standard errors) for poverty rates, and regression diagnostics. It should be noted that the Slovenia report is still in draft form, and will be revised in June to (a) take into account new analysis that was carried out in May 2014, and (b) revise the organization of the paper to conform more closely with that of the reports for other countries. Results: This output has helped to deepen knowledge and generate innovate approaches and solutions. The technical reports document the methods used and results obtained, and demonstrate the knowledge gains from using the small area estimation approach. The technical reports also demonstrate the results that are achievable in less-than-ideal circumstances, such as when there is no access to census microdata. Country Poverty Mapping Briefs To complement the technical reports, the team has decided to prepare a set of two-page summaries of the poverty map results, one for each country. These were not originally specified as deliverables for the project, but are potentially valuable vehicles for communicating the project’s findings to general audiences that are interested in poverty and policy, but not the technical details of the estimation process. The briefs provide information about the motivations for producing the poverty maps, summarize some of the main results in the country, and briefly outline how the poverty maps may help inform policies and programs. Draft briefs for Hungary, Romania, and Slovakia are included in the output package for the decision meeting. The team would especially appreciate guidance from the meeting on the content and presentation of the briefs, after which briefs will be developed for countries that have poverty maps. Results: This output is expected to inform policies and strategies, and also to deepen knowledge. So far only one of the poverty mapping briefs has been released: it was distributed in connection with a dissemination event in Romania, and has also been shared with senior staff at the European Commission. In both cases it was well-received, and this form of output is expected to grab the interests of diverse stakeholders in the EC, national and sub-national governments, and within the World Bank Group, concisely conveying information about the distribution of poverty that is relevant to policies and strategies across different sectors. Innovation and Knowledge Impacts Since the late 1990s the World Bank Group has played a pioneering role in the small area estimation of poverty using microdata, led by cutting-edge research in DECRG and widespread empirical application of the methodology in more than 60 countries. The methodology was also adopted by numerous researchers and analysts outside the World Bank Group. This project presented new challenges and opportunities to break new ground and improve the way that poverty mapping is practiced in the World Bank and elsewhere. The Empirical Best (EB) method had been proposed in the statistical literature as an enhancement to the Elbers-Lanjouw-Lanjouw (ELL) approach used by the World Bank. The advantage of EB appeared to be of marginal consequence in low income countries (because sample surveys usually cover only a small proportion of the small areas for which poverty is estimated), but potentially significant in the context of the survey sampling schemes used in European countries. This project provided an opportunity to rigorously test empirical applications of ELL and EB and their variants, using unique data sets in Denmark and Slovenia to validate and compare the different approaches. The validation study, aided by the guidance of the experts on the Steering Committee, was able to quantify the impact of adding EB to the ELL framework. The systematic validation testing also furthered the understanding of other elements of the poverty mapping methodology. It is not an overstatement to say that this component of the project has helped change the way the World Bank approaches poverty mapping. These innovations are already beginning to spread outside the World Bank, through our engagement with the Steering Committee, counterparts in NSIs, and enhancements to the PovMap software produced and supported by the World Bank. Results: This activity has developed and fostered innovation in the area of small area poverty estimation. The innovation is embedded in the small area poverty estimates based on microdata that are produced in the outputs of this project. The innovation and knowledge spillovers are also evident in the skills transfer that occurred with partners in NSIs and the updating of PovMap software. It has also gained prominence in academic research around the world by catalyzing greater communication and collaboration across different methodological “camps,” including featured sessions at prominent academic conferences. Updated PovMap software In 2004 the World Bank research department developed special-purpose software, known as PovMap, for the estimation of poverty in small areas. The main advantage of using PovMap instead of standard statistical software is its speed and efficiency in performing hundreds of simulations on large census data sets containing millions of observations and its handling of complex error structures. Its graphical user interface also simplifies operations for the user, while retaining the required rigor and complexity of the computations as background operations. PovMap is freely available from the DEC web site. From the outset, this project supported new development and enhancements to PovMap software. Initially this was to facilitate addressing the methodological questions posed in the pilot study, and it continued throughout the project as computational methods were refined and enhancements were added. Results: This activity has generated innovative approaches and solutions. It has brought new features to the PovMap software package, a public good that is used by analysts both inside and outside the WBG, incorporating the latest methods (including some still in the experimental stage), adding a great deal of new flexibility, and producing useful new regression diagnostic output. Dissemination and Outreach There have been several forms of dissemination and outreach, with more planned in 2014. These have taken place in individual new EU member states and at the European Commission. The research-focused aspects of the project have been presented at numerous academic conferences. The following is a brief summary of the dissemination and outreach efforts that took place under this project:  Inception presentations and meetings in individual countries: During 2012, meetings were held in each of the ten countries with key partners, such as NSIs and government ministries (most often the ministry of labor and ministry of regional development, and sometimes the ministry of finance). These presentations explained the objectives and structure of the project, and served as a springboard for discussions about how the results could be used in that country, as well as the practical considerations for carrying out the work.  Preliminary results seminars in individual countries: As poverty mapping analysis took place in the various countries (October 2012 to May 2014), preliminary results were presented in seminars attended by NSI staff, and occasionally staff from relevant ministries. These served as an informal validation of the results, and also helped to discuss and understand the possible reasons for the observed patterns of poverty across the country.  Results seminars in individual countries: These seminars were follow-ups to the preliminary seminars, and were convened after additional analysis was done, taking into account the comments received. One of the particularly noteworthy examples was the seminar in Bucharest, held in March 2014, which was held jointly with the local EU delegation and was well-covered by the media and attended by approximately 60 officials representing government ministries, international organizations, and bilateral development partners.  Workshop with EC country desk officers: In April 2013 a full-day workshop was held in Brussels with approximately 50 country desk officers from DG Employment and DG Regional Policy. The workshop presented the poverty mapping results to available date, as well as related geo-referenced information on poverty correlates such as employment, business activity, education, and demographics. The focus was on interpreting the results and how to use the results in the EC’s programmatic work. A similar workshop was held again in March 2014, led by senior staff from the EC.  High-level presentation at the EC: In October 2013 the project results were presented at a high-level seminar that was co-chaired by the Directors General of DG Employment and DG Regional Policy. Approximately 60 EC staff attended the event, which generated lively debate and follow-up discussions.  Web content: The full-scale web outreach has not begun, but two examples may serve as useful guides for future work. One is a blog post by the ECCU5 Country Director on “What can the EU learn from poverty maps?” (http://blogs.worldbank.org/futuredevelopment/what-can-eu-learn-poverty-maps), which frames the results in the context of the EU’s 2014-2020 financial framework. The other example is from Statistics Estonia, which was the first of the countries to feature the poverty mapping results on their web page, including incorporating the small area poverty estimates in their official online database (http://www.stat.ee/74507). Future dissemination and outreach will build on these efforts, and benefit from guidance by EXT staff at headquarters and in the country offices. Further outreach is planned in individual countries (including events outside the capital cities), at the European Commission, and on the web. It is also planned to feature the poverty maps in an upcoming edition of the EU11 Regular Economic Report, mostly likely this winter’s issue. Lessons Learned As this is a large project involving close collaboration among the World Bank, the European Commission, and authorities in individual European countries it should not be surprising that many of the lessons concern relationships and organization. Although all partners share considerable common ground on the objective of improving the well-being of poor people, substantial institutional barriers may make working together difficult. Each organization has its own sets of rules and practices that may be internally consistent but also incompatible with those of its external partners. In this project this occurred from setting up the financing arrangements, carried through to resolving adequate confidentiality safeguards to work with sensitive data, and continues with dissemination of the results in some countries. Through good-faith negotiation it has usually been possible to accommodate the concerns of all parties. Whenever possible, this project tried to build on existing precedent and to establish agreements that were sufficiently general to be effective for future projects, without reinventing the wheel. Nevertheless, the investment of time and energy required on these matters should not be underestimated. Related to the institutional complexity, the degree to which working arrangements, analytical approaches, and skills transfer had to be customized in each of the countries is noteworthy. Despite common EU objectives (e.g., income convergence) and Eurostat technical standards (e.g., the EU-SILC survey), one size does not fit all across the new member states. The engagement with each NSI was unique, with varying levels of staff expertise and time available to work on the project, data readiness and access, and even NSI management confidence in the whole notion of estimating poverty risk in small areas by “indirect” methods. Getting traction on poverty issues can be difficult in the upper-middle and high income countries of the new EU member states, especially among national authorities. It may be even more difficult when the proximate initiative appears to come from Brussels and Washington rather than domestically. That said, this project has shown that it is possible to attract the necessary attention when the information is rich and presented in an accessible and appealing manner. In this instance, regional development—a very active policy area at the EU, national, and subnational levels—proved to be a useful entry point in several countries. It may be more common knowledge than a “lesson learned,” but the hands-on and intensive engagement of the European Commission bears noting. The Commission came to the WBG because of its extensive experience in poverty mapping and because the EC lacked the in-house resources to produce the poverty maps itself. Our counterparts in the Commission at all levels have been discerning consumers of the team’s work, contributing to improvements in the team’s work throughout the process, while remaining cognizant of the constraints imposed by the institutional barriers noted earlier. The exchange of ideas and information has been extremely constructive, and we expect it will continue to be so as the poverty maps are increasingly used in the EC’s operations. Next Steps The team and the European Commission have keen mutual interest in building on the investments made via this project. This includes completing some outputs that were envisioned under this project, but that were delayed because of exogenous factors. Some of this work can be undertaken within the existing trust fund, which closes on July 31, 2014; other work would be supported by a new trust fund (that is, a new child trust fund under the Europe 2020 programmatic trust fund) that is currently being discussed by the Bank and the Commission. The following activities are planned for completion under the current trust fund.  The current draft technical report for Slovenia will be finalized, as will the poverty mapping briefs for all countries with completed poverty maps.  Continued outreach and dissemination, as outlined in section 3. The following activities have been identified in discussions with the EC as the top priorities under a new trust fund.  Developing a small area poverty map for Croatia. The team is working with the CMU to investigate the possibility of jump-starting this activity in summer 2014 through a Miscellaneous Reimbursable Agreement with the Government of Croatia, with subsequent support through the EC trust fund.  The team will work with the NSI in Bulgaria to complete poverty maps at the municipality level (LAU 1), and work with the Hungary Central Statistical Office to update their poverty map using the 2011 population census and the 2012 EU-SILC survey.  Completing poverty maps in the Czech Republic and Lithuania, where work up to this point has been stalled by the inability to come to resolve outstanding issues with data confidentiality agreements.  Achieving further disaggregation of the existing poverty maps in selected countries, particularly Latvia.  Using the findings from the poverty mapping analysis to develop a system for monitoring changes in the risk of poverty in small areas over time. Table 2: Results matrix (from results framework in project concept note) Intermediate Indicators Activities Outcomes Development  Mobilization of non-Bank  Small area poverty maps and financing informed resources informed tables  Government expenditure informed Policy/strategy  Government policy/strategy  Small area poverty maps and informed informed tables  Public debate  Country poverty mapping stimulated/initiated briefs Knowledge  Facilitated exchange of best  Methodological pilot study deepened practice with clients  Small area poverty maps and  Facilitated exchange of best tables practice with partners  Technical reports  Country poverty mapping briefs Innovative  New innovative approach  Methodological pilot study approaches and fostered  Technical reports solutions generated  New innovative approach  Changes in World Bank developed analytical approach to poverty mapping  Updated PovMap software Attachments (also filed in WBDocs)  Methodological pilot study report  Technical reports: Estonia, Hungary, Latvia, Poland, Romania, Slovakia, Slovenia  Poverty map briefs: Hungary, Romania, Slovakia