MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 1 © 2024 International Bank for Reconstruction and Development/The World Bank 1818 H Street NW Washington DC 20433 Telephone: 202-473-1000 Internet: www.worldbank.org DISCLAIMER This work is a product of the staff of The World Bank. The findings, interpretations, and conclusions expressed in this work do not nec- essarily reflect the views of The World Bank, its Board of Executive Directors, or the governments they represent. The World Bank does not guarantee the accuracy, completeness, or currency of the data included in this work and does not assume responsibility for any errors, omissions, or discrepancies in the information, or liability with respect to the use of or failure to use the information, methods, processes, or conclusions set forth. 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. Nothing herein shall constitute or be construed or considered to be a limitation upon or waiver of the privileges and immunities of The World Bank, all of which are specifically reserved. ATTRIBUTION Please cite the work as follows: Majoka, Zaineb, Christina Wieser, Maria Qazi, David Guzman Fonseca, Thomas Pave Sohnesen, and Ibrahim Khan. 2024. Mind the Gap: Assessing Pakistan’s National Socio-Economic Registry (NSER). Washington, DC: World Bank RIGHTS AND PERMISSIONS The material in this work is subject to copyright. Because The World Bank encourages dissemination of its knowledge, this work may be reproduced, in whole or in part, for noncommercial purposes as long as full attribution to this work is given. Any queries on rights and licenses, including subsidiary rights, should be addressed to World Bank Publications, The World Bank Group, 1818 H Street NW, Washington, DC 20433, USA; fax: 202-522-2625; e-mail: pubrights@worldbank.org. Cover and Layout Design: Umaima Mughal Cover Illustration: © Freepik | (Freepik standard image license) 2 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 3 Table of Contents Acknowledgments ................................................................................................................................. 7 Abbreviations.......................................................................................................................................... 9 Executive Summary.................................................................................................................................. 10 1. Context............................................................................................................................................... 16 2. NSER Completeness ............................................................................................................................ 20 2.1 How Complete is NSER?........................................................................................................ 22 2.2 Are the Poor Included In NSER?.............................................................................................. 24 2.3 Why are People Not Included in NSER?................................................................................... 27 3. NSER Data Quality ............................................................................................................................... 33 3.1 Registration Modalities............................................................................................................. 34 3.2 Urban and Rural Classification................................................................................................. 35 3.3 Household Size...................................................................................................................... 38 4. NSER During Shock Response............................................................................................................. 40 5. Policy Recommendations ..................................................................................................................... 42 5.1 Keep Higher Coverage in the Bottom Four Quintiles.................................................................. 42 5.2 Overcome Existing Challenges Related to Access to NSER...................................................... 42 5.3 Improve Data Decay and Accuracy........................................................................................... 44 5.4 Continue Research on Effectiveness of Targeting Methodology & Data Collection Modalities...... 45 References.............................................................................................................................................. 47 Annex A - Sampling Methodology ............................................................................................................ 50 Annex B - Supplemental Figures & Tables ................................................................................................ 59 4 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Figures & Tables FIGURES ES1: Share of households matched in the NSER database......................................................................... 11 ES2: Share of households in the NSER by province.................................................................................... 11 ES3: Share of households aware of their NSER registration status ............................................................. 12 ES4: Share of households in the NSER by modality of registration............................................................... 12 1.1: Beneficiary incidence in the bottom quintile.......................................................................................... 18 2.1: Examples of Enumeration Areas (EAs) used for sampling...................................................................... 20 2.2: Share of households matched in the NSER database........................................................................... 22 2.3: Share of households matched in the NSER by province........................................................................ 23 2.4: Predicted probability of exclusion from the NSER by province............................................................... 23 2.5: Predicted probability of exclusion from the NSER by PMT deciles......................................................... 24 2.6: Share of households matched in the NSER by decile in each province.................................................. 25 2.7: Employment type by matching status.................................................................................................. 27 2.8: Reasons for not registering in the NSER.............................................................................................. 27 2.9: Share of households aware of the NSER registration status.................................................................. 28 2.10: Share of households in the NSER...................................................................................................... 29 2.11: Average distance from the NSER DRC.............................................................................................. 31 3.1: Distribution of self-reported urban status by Enumeration Area............................................................. 36 3.2: PMT distribution by different urbanization classification........................................................................ 37 4.1: Distribution of households by the latest date of information update in NSER......................................... 41 A1: An example of Enumeration Area (EA) used for sampling....................................................................... 50 TABLES 1.1: Comparison between NSER and the 2023 Census Population.............................................................. 19 2.1: NSER Survey Sample.......................................................................................................................... 21 2.2: Socioeconomic and demographic characteristics by matching status................................................... 26 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 5 3.1: Relationship between PMT scores by date of last update................................................................... 33 3.2: Relationship between PMT scores by modality of registration............................................................ 34 3.3: Share of urban and rural households................................................................................................. 37 3.4: Variation in household size by survey type......................................................................................... 38 A1: NSER Quality Review Survey Sample................................................................................................. 51 A2: Replacement of Tehsils and SSUs..................................................................................................... 52 A3: Refusal rates by PMT quintiles........................................................................................................... 52 A4: Reasons given for not being interviewed............................................................................................ 53 A5: Distribution of the weighted total survey population by province.......................................................... 54 A6: Distribution of NSER sample SSUs by province, urban and rural stratum, based on GHSL pre-dominant share, combining suburban with rural................................................................................................................. 55 A7: Distribution of NSER sample SSUs by province, urban and rural stratum, based on GHSL pre-dominant share, combining suburban with urban............................................................................................................... 56 A8: Distribution of the weighted total survey population from the NSER Survey data and the 2023 Pakistan Census total population, by province, urban and rural strata................................................................................. 57 A9: Weighted final NSER Survey total population and weighted number of households by province, urban and rural stratum..................................................................... 58 B1: PMT Indicators across matching categories....................................................................................... 59 B2: Marginal effects based on probit regression analysis.......................................................................... 60 B3: Comparison of household characteristics of matched and unmatched households with PMT below 32...................................................................................... 63 B4: CNIC unavailability by quintile........................................................................................................... 64 B5: Household size components of PMT and their weights....................................................................... 64 B6: Comparison of household characteristics of matched and unmatched households with PMT below 32...................................................................................... 65 6 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Acknowledgments This report was prepared by a World Bank team jointly led by Zaineb Majoka (Economist, Social Protection and Jobs) and Christina Wieser (Senior Economist, Poverty and Equity). The team also included David Guzman Fonseca (Consultant, Social Protection and Jobs), Thomas Pave Sohnesen (Consultant, Social Protection and Jobs), Maria Qazi (Economist, Poverty and Equity), and Ibrahim Khan (Economist, Poverty and Equity). The work was carried out under the guidance of Cem Mete (Practice Manager, Social Protection and Jobs) and Ximena Del Carpio (Practice Manager, Poverty and Equity). The team is indebted to Najy Benhassine (Country Director, Pakistan) and Nicole Klingen (Human Development Regional Director, South Asia) for their leadership and overall support of the report. The team is grateful to the following World Bank staff who provided critical inputs throughout the life cycle of this project: Phillippe Leite (Senior Economist, Social Protection and Jobs), Moritz Meyer (Senior Economist, Poverty and Equity), and Gul Najam Jamy (Consultant, Social Protection and Jobs). Dana Thomson worked on the sampling strategy and created maps using geospatial data, and Peter Fisker provided GIS support. David Megill worked on sampling weights to ensure that the final dataset was representative of Pakistan’s population according to the 2023 Pakistan Census. Yasmeen Baloch and Sharmeen Talpur provided excellent field supervision during the fieldwork, which included field monitoring and conducting back checks. Rida Ha- meed and Tehreem Fatima provided support with data cleaning and running high-frequency checks, which was instrumental in collecting high-quality data. Syed Farrukh Ansar and Kiran Shahzadi provided outstanding ad- ministrative support. The report is an output of a collaboration between the World Bank and the National Socio-Economic Registry (NSER) team at the Benazir Income Support Program (BISP). We offer our special thanks to Naveed Akbar (Di- rector General, NSER), who not only facilitated our access to the NSER administrative data and field offices but also provided invaluable guidance, support, and feedback throughout this study. BISP tehsil offices supported the data collection teams. Noman Ali and Maliha Arif (both BISP) worked closely with the World Bank team to get access to the required administrative data. The report benefited from the insights of peer reviewers: Mohammed Ihsan Ajwad (Senior Economist, HMNSP), Andrea Vermehren (Lead Social Protection Specialist, HAES1), and Nasir Iqbal (Associate Professor, PIDE). We also recognize Aldo Morri for his excellent editorial support and Umaima Mughal for the graphic design. Finally, MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 7 we extend our deepest gratitude to the survey respondents for their willingness to share their experiences, which have been instrumental in deepening our understanding. This work was generously supported by the Government of the United Kingdom of Great Britain and Northern Ireland, acting through the Foreign, Commonwealth and Development Office (FCDO) Trust Fund No. TF073431. The team also wishes to recognize the generous award of a grant from the World Bank’s Rapid Social Response Adaptive and Dynamic Social Protection (RSR-ADSP) Umbrella Trust Fund Program, which is supported by the Russian Federation, United Kingdom, Norway, Sweden, Australia, Denmark, the Bill and Melinda Gates Foun- dation, USAID, GHR Foundation and UBS Optimus Foundation. 8 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Abbreviations BISP Benazir Income Support Programme CCT Conditional Cash Transfer CNIC Computerized National Identity Card DoU Degrees of Urbanization DRC Dynamic Registration Center EA Enumeration Area FGD Focus Group Discussion GHSL Global Human Settlement Layer GoP Government of Pakistan HIES Household Integrated Economic Survey ICT Islamabad Capital Territory KP Khyber Pakhtunkhwa MNA Member of the National Assembly NADRA National Database and Registration Authority NSER National Socio-Economic Registry PMT Proxy Means Test PSU Primary Sampling Unit SSU Secondary Sampling Unit UCT Unconditional Cash Transfer WB World Bank MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 9 Executive Summary The Benazir Income Support Programme (BISP) was initiated in 2008 in Pakistan to aid the poor and vulnerable, particularly in response to the global financial crisis. It began with an unconditional cash transfer program, now known as Kafaalat, which has grown to support 9.3 million families by 2024. BISP also introduced conditional cash transfers to promote health and education. Over 15 years, BISP has enhanced its delivery systems, including identification, social registries, payment, and information systems, using technology and data to improve efficiency and effectiveness. The National Socio-Economic Registry (NSER), established by BISP, is a key component for deter- mining eligibility for Kafaalat through data collected on household characteristics and assets. Initial- ly, enrollment was based on political recommendations, but has shifted to data-driven methods like the Proxy Means Test (PMT), which proved more effective. Regular updates to the NSER are crucial for accuracy, with the latest update in 2021 registering 35 million households. In 2023, NSER moved to an on-demand registration system to maintain up-to-date records, with more than 40 million families registered by 2024. NSER’s accuracy is critical as it underpins the inclusion of households in over 30 social protection programs. The NSER’s completeness is essential to ensure broad coverage for social protection, while its accuracy of targeting ensures the inclusion of the most vulnerable households into social protection programs. The analysis in this report assesses the completeness and accu- racy of the NSER, focusing on factors that can lead to exclusion from the registry. It also sheds light on how data collection methods can affect PMT scores, which directly affects targeting accuracy. The report utilizes two data sources to evaluate the completeness and accuracy of Pakistan’s NSER: the NSER Quality Review Survey (QRS) and the NSER’s own administrative data. The QRS, conducted between July and October 2023, interviewed 59,456 households across four provinces and the Islamabad Capital Terri- tory (ICT), collecting detailed demographic and socioeconomic data to calculate PMT scores. The NSER admin- istrative data provided additional household information, including registration modality and date of last update. 10 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) KEY FINDINGS 84 percent of the total population in The NSER is more likely to include poor Pakistan is covered in the NSER, a households with 90 percent of the higher coverage than in other countries poorest decile included in the NSER How Complete Is NSER? The NSER has a high level of coverage, with 84 percent of the population included in the NSER. The coverage of NSER is high compared to other countries, where the average coverage rate in countries with social registry is around 41 percent, while NSER covers 84 percent of Pakistan’s population (Figure ES1). However, as NSER is transitioning to an on-demand registration system, the overall coverage rates are likely to decrease as those who are better off, may not want to register. Are the Poor Included in NSER? The NSER is more likely to include poor households. NSER coverage is higher for poorer households in Pa- kistan, except in Balochistan where coverage is uniformly low across all levels of well-being due to governance and political challenges. Nationally, 90 percent of the poorest decile is included in the NSER but only 67 percent of the richest decile is included (Figure ES2). FIGURE ES1 FIGURE ES2 Share of households matched in the NSER database Share of households in the NSER by province Source: Authors’ calculations based on the NSER QRS and NSER adminis- trative data Note: Households were categorized as matched if a match for their CNIC was found in the NSER administrative data. Households were categorized as  eview Survey Source: Authors’ calculations based on NSER Quality R having an invalid CNIC after validating the shared CNIC numbers with NAD- RA. A household can have multiple CNICs and, therefore, belong to several groups. In this table, the categories supersede each other in the following order: matched, not matched, invalid CNIC, CNIC not shared. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 11 Despite the high overall coverage, over 2.2 million households from the bottom 40 percent of the pop- ulation are still not included in the NSER. This means they are excluded from social protection and disaster response programs. Specifically, 1.6 million households below the PMT score threshold of 32 for Kafaalat inclusion are missing from the NSER. It is also important to underscore that data on bottom 40 percent is not available for other countries, so it is hard to conclude where NSER stands in comparison. However, the overall coverage num- bers, as mentioned above, indicate that the NSER is performing in terms of coverage better than most countries. Why Are People Not Included in NSER? The report highlights two challenges for inclusion of the poor in the NSER: (i) Lack of awareness of the NSER, (ii) modality of registration and (iii) lack of identification documents. Awareness about the NSER registration process in Pakistan is low. Nearly half of the households not registered with NSER were unaware of how to do so, and only 27 percent knew how to contact BISP. The reg- istration process is seen as cumbersome, often requiring multiple visits to NSER Dynamic Registration Centers (DRCs), which are plagued by long wait times. Furthermore, many households that are registered with NSER are not aware of their registration status; only 42 percent of those in the NSER knew they were registered, with the figure dropping to 13 percent in Balochistan (Figure ES3). However, among Kafaalat beneficiaries, awareness of registration status is higher at 69 percent. Modality of registration matters for inclusion in the NSER and coverage rates are better in areas with dynamic registration, with areas using this modality showing a 97 percent inclusion rate (Figure ES4). Dynamic FIGURE ES3 FIGURE ES4 Share of households aware of their NSER registration Share of households in the NSER by modality status of registration Review Survey Source: Authors’ calculations based on NSER Quality   eview Survey Source: Authors’ calculations based on NSER Quality R 12 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) registration is more cost-effective than data collection through door-to-door surveys. Yet, dynamic registration transfers some costs to households, who incur travel expenses and potential loss of income by forgoing work to travel to DRC. The current setup of one DRC per Tehsil leads to capacity issues, resulting in overcrowding and long wait times. However, door-to-door registration is more effective in reaching remote households. Using a mixed approach that includes both dynamic registration coupled with door-to-door registration for poorer, remote, and marginalized communities could enhance coverage, especially for marginalized communities. One of the barriers to registration is lack of relevant documentation required to register in the NSER such as CNIC, B-form, and updated household information in the NADRA database. Among the bottom quintile, 10.7 percent of households did not have a CNIC (see Annex B, Table B4 for CNIC by quintile). While not having a CNIC can prevent a household from registering in the NSER, it can also lead to exclusion from other entitlements as access to public and private goods and services (such as voting, buying a sim card, opening a bank account, accessing other safety net programs, etc.), as they now increasingly require having a CNIC. What Is the Quality of NSER Data? Data quality and accuracy are crucial to the effectiveness of PMT-based targeting. The report explores three areas that affect data decay due to irregular updates and changes in variable definitions: (i) registration modality, (ii) the urban/rural classification, and (iii) household size. The registration modality significantly impacts the accuracy of data collected for the NSER. Data collected through desk-based kiosks or dynamic registration centers have a lower correlation with accuracy compared to data from door-to-door visits. This is partly because door-to-door methods allow interviewers to observe household characteristics directly while also giving the respondent privacy that often influences how people report their socio-economic status. The NSER uses self-reported data to classify households as urban or rural, an important variable in the PMT. The self-reported data is based on perception and may not align with the official categorization. Moreover, Pakistan has also experienced rapid urbanization and development in secondary cities. Differences in the cost of living between urban and rural areas are notable, with urban households generally having access to better services but also facing higher living costs. Data show large inconsistencies in self-reported urbanization data, with substantial variation within the geographic areas, which can lead to discrepancies in PMT scores and affect eligibility for social protection programs. The NSER QRS indicates a lower average household size than the NSER administrative data, which could affect the targeting accuracy of the PMT scores. The PMT formula incorporates household size in MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 13 different ways, which can either increase or decrease PMT scores. A smaller recorded household size can lower the PMT score, potentially allowing more households to qualify for assistance. Conversely, if a missing house- hold member is an adult with literacy, their inclusion could raise the PMT score, potentially leading to a net pos- itive effect. Variations in household size may stem from differences in definitions across databases. The NSER Administration data previously differentiated between “household” and “family,” but with dynamic registration, it now uses NADRA’s database, which only includes members related by blood or marriage, resulting in smaller household sizes. To ensure consistency across household size, as NSER has transitioned to the dynamic data registration process, it triangulates the data on household members with the NADRA database. The report highlights the importance of adaptive safety nets in Pakistan, particularly in the context of recent economic, health, and climatic shocks. These safety nets are designed to be scalable and flexible, providing swift and effective support to the most vulnerable populations during crises. The 2022 floods in Paki- stan demonstrated the effectiveness of the NSER as a shock-responsive safety net, with a higher percentage of flood-affected households being registered, more aware of their NSER status, and more likely to have received assistance, especially among the poorer quintiles. POLICY RECOMMENDATIONS This report presents a positive evaluation of NSER completeness and accuracy, and several important policy messages emerge. 1. KEEPING A HIGH COVERAGE OF THE BOTTOM FOUR QUINTILES In Pakistan, the national social registry, the NSER, covers 84 percent of the total population, a proportion much higher than in other countries. As the Poverty, Equity, and Resilience Assessment (World Bank, forthcoming) shows, a significant portion of Pakistani households are at risk of poverty, with one-third of non-poor house- holds vulnerable to falling into poverty as of 2019–20. Climate change, particularly through extreme weather events, exacerbates this vulnerability, impacting health, food security, and livelihoods, and disproportionately affecting marginalized groups. The NSER’s utility in disaster response, particularly for the poorest, is affirmed by its performance during the 2022 floods. Strengthening its shock-response capacity through better data in- tegration with disaster management and targeted support for displaced households is recommended. In light of the large vulnerabilities of households falling into poverty and at risk of climate vulnerabilities, it is crucial to maintain high coverage of NSER to use as the social registry in the aftermath of shocks. 2. DESPITE HIGH COVERAGE, SOME CHALLENGES REMAIN Despite the high overall coverage of 84 percent, 2.2 million households from the bottom 40 percent of the population are still not included in the NSER. This means they are excluded from social protection and disaster 14 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) response programs. The challenges of inclusion relate to lack of awareness of the NSER, remoteness of loca- tion, and modality of registration. To solve some of these challenges, several recommendations emerge: • Adapt a mixed approach to registration modality. BISP has planned to further decentralize the NSER registration process by establishing Dynamic Registration Centers at Union Council Level. While this will help address barriers related to mobility and proximity, it may not be as effective in Balochistan given large distances and difficult terrain. Therefore, especially in areas with poorer coverage, combining door- to-door survey with dynamic registration could improve NSER registration rates. • Implement communication campaigns to improve NSER awareness. Enhanced communication efforts can boost NSER awareness and registration, especially among the poorest and most marginal- ized. Learning from Rwanda’s Umurenge Program and Brazil’s Bolsa Familia, which use diverse media and transparency tools to engage communities, can guide effective campaign strategies. • Improve outreach to remote and marginalized communities. Targeted registration initiatives in Ba- lochistan and remote rural areas can increase NSER inclusion, particularly for marginalized groups such as transgender individuals and those from lower occupational castes. 3. IMPROVING THE DECAY OF DATA ACROSS QUINTILES AND CONTINUOUS MONITORING OF DATA ACCURACY ACROSS REGISTRATION METHODS CAN ENSURE DATA INTEGRITY • Improve NSER data accuracy with enumerator trainings and spot checks. Addressing interviewer biases and privacy issues through demographic recording, specialized training, and recognizing house- hold reporting variances can improve data accuracy. Standardizing questionnaires and documenting enumerator assignments are also recommended. Training DRC staff for diverse community service and conducting regular randomized spot checks on a small sample of registered households, as done in Brazil and other countries, can further improve data integrity. • Utilize urban/rural definition that better captures differences in cost of living and living condi- tions. To improve targeting in Pakistan’s NSER, it’s important to redefine urban and rural classifications to reflect true living conditions and service access, which will enhance the PMT formula’s accuracy. • Continue sharing data to create a powerful evidence base for decision-making. To enhance de- cision-making, maintain high data quality, and regularly share data across institutions for comprehensive insights and potential improvements in NSER updates and targeting. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 15 1. Context Pakistan launched its Benazir Income Support Identity Cards (CNICs) and linking them with the Na- Programme (BISP) in 2008 as an autonomous in- tional Socio-Economic Registry (NSER) and payment stitution to deliver programs targeting poor and systems. Factors such as complementary legal and vulnerable beneficiaries. The unconditional cash institutional arrangements, technology adoption, and transfer (UCT) program— also called BISP 1 at that the development of dynamic delivery systems have time, but now referred to as Kafaalat—was launched improved the effectiveness and efficiency of its pro- in response to the soaring food and fuel crisis during grams over the years (Guven et al., 2024). the global financial crisis of 2008. Since then, the number of beneficiaries has increased from 1.7 mil- BISP established the NSER as its social regis- lion families in 2008 to 9.3 million in 2024, covering try to support intake, registration, and determi- roughly 20 percent of Pakistanis. BISP launched two nation of eligibility for its UCT program, Kafaalat. complementary conditional cash transfer (CCT) pro- Needing a more objective and efficient method of tar- grams (Nashunuma and Waseela-e-Taleem ) promot- geting, BISP collaborated with the National Database ing human capital accumulation by encouraging BISP and Registration Authority (NADRA) to launch a na- Kafaalat families to use health and education services. tional door-to-door survey in 2010 to collect data on variables to calculate Proxy Means Test (PMT) score:2 Pakistan has significantly improved its social (i) household and individual characteristics; (ii) own- protection system over the past 15 years. In ad- ership of durable goods and housing characteristics; dition to expanding program coverage of BISP, the and (iii) ownership of productive assets, especially delivery systems now are more accessible and re- land holding, livestock, and farm equipment. The 25.5 sponsive to the needs of people. Delivery systems— million households registered in the database cov- foundational for providing social protection benefits ered about 87 percent of the population (Guven et and services—encompass identification systems, al., 2024). Each registered household was required to social registries, payment systems, and management have at least one member with a CNIC. Based on this information systems. BISP has increasingly used data data, the PMT score for each household was calculat- and technology to improve the performance of these ed to determine enrollment eligibility for Kafaalat. systems by leveraging the Computerized National 1 From here on in this report, BISP refers to the national safety net 2 PMT uses proxy variables to estimate household welfare, which is institution, which operates under the patronage of the Prime Min- then used to predict poverty. In Pakistan’s case selection of these ister office. Its main objective is to implement the Government of variables was based on an analysis using the 2007-2008 House- Pakistan’s (GoP) pro-poor social policies. hold Integrated Economic Survey (HIES) 16 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Using the NSER as its social registry, BISP sig- In 2023, NSER transitioned to an on-demand nificantly improved household inclusion and tar- registration system to promote more regular geting. The first beneficiaries in 2008 were enrolled in data updates. The dynamic registration system Kafaalat based on the recommendations of Members aims to completely update NSER entries every four of the National Assembly (MNAs). A World Bank (WB) 3 years, thereby cost-effectively reducing inclusion and assessment found that this type of targeting method exclusion errors. By 2024, BISP had set up 647 Dy- performed poorly compared to data-driven targeting namic Registration Centers (DRC)—at least one in methods such as PMT scores (Nikitin et al., n.d.), a each Tehsil—where households can register if they finding in line with similar assessments from other were missed during the 2016-2021 update, or up- countries (Alatas et al., 2012). date their information if: (i) their last update was more than three years ago; (ii) there has been a change Considering the dynamic nature of household in their household composition; or (iii) they reside in composition and circumstances, regularly up- shock-affected areas. Since its launch in 2023, 4.9 dating this information is crucial to reduce inclu- million families have registered or updated their data sion and exclusion errors. In 2016, BISP launched using DRCs. In total, NSER has more than 40 million a second door-to-door survey to update the NSER, families registered. complemented with desk-based registration in 2020, due to concerns related to the exclusion of poor and Over time, NSER has become the backbone of vulnerable households. Households were not allowed an increasing number of social protection pro- to register or update their information if they were grams in Pakistan, which underscores the im- missed during the first door-to-door survey round. portance of its completeness and accuracy. The 2016 survey used the latest technology to im- Having a complete and accurate NSER is important prove data quality, 4 but due to other logistical issues, for a number of reasons. First, over time, several it was not able to complete the door-to-door survey. provincial and federal programs have started using In 2020, BISP launched temporary desk-based regis- NSER data to identify and enroll beneficiaries, with tration kiosks where households could register in the now more than 30 programs use NSER for targeting. NSER if they had not been registered during the door- NSER completeness and quality directly affect the ef- to-door survey. The NSER update was finally con- fectiveness of these programs. Second, given Paki- cluded in 2021 with 35 million households registered, stan’s challenging economic situation and limited fis- covering around 87 percent of the population (Guven cal space (World Bank, 2024a), Pakistan’s is shifting et al., 2024). policies to move from general subsidies to targeted transfers that will rely on NSER data targeting. Third, Pakistan has undergone multiple crises over the last 3   Each MNA shared a list of 8,000 potential beneficiaries from their respective constituencies. five years, including the COVID-19 pandemic, the 2022 floods, and record-breaking inflation. Poverty   For example, computer-assisted personal interviewing (CAPI) 4 was used, which significantly reduced the data processing time is projected to have substantially increased (World MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 17 Bank, 2024b), and, with it, the need for expanding FIGURE 1.1 the social protection system. Beneficiary incidence in the bottom quintile The accuracy of social protection program tar- geting relies on the completeness and quality of NSER data. Based on the data from the Household Integrated Economic Survey (HIES) 2018-19, 5 we conclude that NSER accurately targets beneficiaries for BISP’s Kafaalat, where 40 percent of total social assistance beneficiaries belong to the poorest 25 per- cent of Pakistanis, a share higher than majority of oth- er countries’ social protection programs (Figure 1.1). Source: ASPIRE data on Beneficiary Incidence in 1st quintile – All Social The completeness of the NSER is vital to ensure many Assistance. Latest year available people are registered for potential inclusion in current Note: Beneficiary incidence is defined as the percentage of program bene- and future social protection programs. The accuracy ficiaries in a quintile relative to the total number of beneficiaries in the pop- ulation. of the NSER is important to ensure the poorest, most vulnerable households can be targeted. the lowest share of its population registered in the NSER, only second to ICT. Coverage rates are highest The latest NSER data on registered people sug- in Khyber Pakhtunkhwa (KP) at 94 percent. While we gests a significant gap in the total population would not expect a social registry to cover a popula- based on the 2023 Census population count. tion fully—typically, individuals at the upper end of the Overall, the NSER includes 203.5 million people, or welfare distribution are not covered—low coverage 41.4 million families, covering 82 percent of the cen- rates could potentially undermine the targeting effec- sus population. Table 1.1 compares the number of tiveness of the NSER, especially if it excludes relative- families, households,6 and individuals in NSER with ly poorer households. the 2023 census in each province. Balochistan has This report assesses the completeness and accura- 5   In the absence of a recent household survey data to determine cy of the NSER, focusing on factors that can lead the welfare of households in Pakistan, it is hard to conduct external to exclusion from the registry. It also sheds light on validity checks on the accuracy of NSER. how data collection methods can affect PMT scores, 6   The NSER differentiates between families and households. While which directly affects targeting accuracy. The rest of household constitutes people who share a single housing unit and share the same kitchen regardless of their relationship to one an- the report is organized as follows: Section 2 gaug- other, a family is defined in relation to an ever-married female within es the completeness of NSER based on welfare and a household whereby there can be more than 1 family in a house- hold. The average number of families in a household in NSER data household characteristics. Section 3 looks at the ac- is 1.3. However, as NSER is transitioning to dynamic registration, curacy of NSER data by assessing registration modal- it has updated its definition of family to include family members as they are registered in NADRA (more details on this in Section 3.3). ity, the urban/rural classification of households, and 18 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) household size. Section 4 assesses the effectiveness of the NSER to serve as a robust registry during the 2022 floods. Section 5 concludes with policy recom- mendations on how to improve NSER. TABLE 1.1 Comparison between NSER and the 2023 Census Population NSER NSER 2023 Census Pakistan Coverage Families Individuals Households Individuals (Ratio of individuals) Balochistan 2,095,661 9,431,298 2,318,519 14,894,402 63% ICT 258,991 1,044,650 411,518 2,363,863 44% KP 7,000,107 38,357,049 5,883,007 40,856,097 94% Punjab 20,992,189 102,559,911 19,855,902 127,688,922 80% Sindh 9,931,693 45,948,757 9,871,620 55,696,147 82% Total 40,278,641 197,341,665 38,340,566 241,499,431 82% Average household/family size 4.9 6.3 (using “Total” row) Source: NSER Administrative data retrieved in January 2024 and 2023 Census of Pakistan. Note: NSER records data at the family level, whereas Census Pakistan records data at the household level. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 19 2. NSER Completeness This report combines two critical sources of mographics and socio-economic indicators was then data to gauge the NSER’s completeness and ac- used to calculate the PMT scores. curacy: the World Bank’s NSER Quality Review Sur- vey (henceforth, “the survey data”) and administrative The NSER QRS relied on an innovative sampling data from the NSER itself (henceforth, “the NSER ad- approach, combining geospatial and gridded ministrative data”). population estimates. We first selected our areas for inclusion in the survey by selecting Tehsils within each The NSER Quality Review Survey (QRS) collect- province in proportion to the total number of house- ed data in face-to-face questions between July holds in a Tehsil from 2017 Census data. 7 We then and October 2023, whereby all households with- systematically selected clusters—small geographic in a small geographic area were interviewed. It areas—within each Tehsil. The cluster size was based surveyed 59,456 households covering four provinc- on population estimates from a population grid data- es (Sindh, Punjab, Khyber Pakhtunkhwa or KP, and base, with roughly 1,000 individuals (or 250 house- Balochistan) and Islamabad Capital Territory (ICT). holds) living within a cluster. Figure 2.1 illustrates ex- The questionnaire collected information on house- amples of clusters in rural (left panel) and urban areas hold roster, demographics, asset profile, and access (right panel). After listing every structure within each to safety nets and other services. It also collected cluster to characterize structures into dwellings and CNIC numbers wherever available and other contact information such as address, GPS location, and cell phone numbers. The information on household de- 7   At the time of sampling design, 2023 Census data was not avail- able. FIGURE 2.1 Examples of Enumeration Areas (EAs) used for sampling 20 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) non-dwellings, every household within each dwelling through a door-to-door survey, teacher-based mod- was interviewed. This resulted in a census-type ex- el,9 desk-based registration drive, or through DRCs); ercise in 339 clusters. (Table 2.1). 8 Annex A contains (ii) date when the household registered or last updated more details on the sampling methodology. its data in the registry; and (iii) select household-lev- el variables. At the time of retrieving the NSER ad- TABLE 2.1 ministrative data, the oldest records were from 2016, NSER Survey Sample whereas the latest records were from 2023. Number Number Number of Province of of interviewed Combining survey and administrative data sample sample sample tehsils clusters households proved a powerful tool to meet the objectives of this study. On average, there were 1.6 CNICs shared Balochistan 6 21 2,911 for each surveyed household. However, about 7 per- ICT 2 3 459 cent of households did not share any CNIC. CNICs KP 16 45 8,044 for which no match was found in the NSER were then shared with the National Database and Registration Punjab 49 175 30,564 Authority (NADRA) to confirm if they were valid CNICs. Sindh 28 95 17,478 In total, 4 percent of the households had shared inval- id CNIC numbers. The NSER administrative data for Total 101 339 59,456 the matched CNICs was merged with the survey data to construct four categories of matches for analysis: The NSER administrative data from the NSER database contains information on household de- 1. Matched: Households in the QRS data with mographic and socioeconomic indicators. The CNICs matched in the NSER database. NSER database has at least one CNIC available for 2. Not matched: Households in the QRS data with each household. The CNIC information collected in CNICs that were NOT matched in the NSER data- the NSER QRS was used to retrieve household infor- base. mation from the NSER database. For each matched 3. I nvalid CNIC: Households in the QRS data that CNIC, BISP shared information on: (i) the modality of shared a CNIC number, but it was found to be registration (that is, whether the household registered invalid after cross-checking NADRA. 4. CNIC not shared: Households in the QRS data did not share a CNIC. 8   While the selection of clusters was random, several clusters had to be replaced due to security concerns or inability to receive per- mits from local authorities. This sampling approach is still novel but 9   During the 2016-2020 door-to-door survey, BISP hired survey is more commonly used as in most cases, geospatial data provides firms to collect data. But in areas with security concerns, private a better estimate of population distribution than the census data, research firms were not allowed to conduct door-to-door survey. In especially if the most recent census survey is a couple of years old. these areas, BISP relied on public school teachers to collect data, It also allows a segmentation of areas when digital shapefiles may who were trained by BISP and followed the same door-to-door not be available. survey protocols. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 21 This analysis only used the following NSER ad- FIGURE 2.2 ministrative data for CNICs whose match was Share of households matched in the NSER database found in it. The NSER administrative data included information on matching status, date of last update or registration, modality of registration, location (urban or rural), whether the household is a Kafaalat benefi- ciary, and whether the household received assistance after the floods in 2022. The PMT scores are based on the survey data unless indicated otherwise. 2.1 How Complete Is NSER? Source: Authors’ calculations based on the NSER QRS and NSER admin- istrative data Combining survey and administrative data shows Note: Households were categorized as matched if a match for their CNIC that the NSER has a very high coverage rate. The was found in the NSER administrative data. Households were categorized as having an invalid CNIC after validating the shared CNIC numbers with comparison found about 84 percent of household NADRA. A household can have multiple CNICs and, therefore, belong to several groups. In this table, the categories supersede each other in the respondents matched in the NSER (Figure 2.2). The following order: Matched, not matched, invalid CNIC, CNIC not shared. unavailability of CNICs was the most common rea- son for not finding a match. Coverage estimates from cent of the total population, the coverage of NSER this study are consistent with the administrative data is significantly higher than registries in other coun- in Table 2.1 where the NSER covers around 81 per- tries. According to the Global Database on Social cent of the population. While 84 percent of the total Registries, which includes 62 countries, the aver- respondents were found in the NSER, it was 90 per- age coverage is 41 percent. Only 10 countries have cent among the households who shared their CNIC registries that cover a higher share of the popula- information. Around 8.2 million households (16.3 per- tion. Some of such examples include Rwanda (97 cent) were not found in the NSER because of vari- percent), Lesotho (86 percent), Costa Rica (98 per- ous reasons: CNIC was not matched in the NSER (6 cent), Argentina (93 percent), Chile (87 percent), percent), the CNIC number shared was invalid (3.8 Azerbaijan (87 percent), and Lao PDR (86 percent). percent), and no CNIC number was shared (6.5 per- cent). Among unmatched, not sharing the CNIC in- We use a probit regression analysis10 to com- formation was the most common reason. However, plement the findings from descriptive statistics. it is unclear if these households did not have a CNIC or were unwilling to share the CNIC number due to privacy concerns.   The probit model uses exclusion from the NSER as the depen- 10 dent variable while controlling for characteristics of the household head, location, household size, caste, wealth, most prevalent data In comparison to other countries, the cover- collection modality, and distance from the closest NSER DRC. Re- gressions are run at national as well as provincial level and by mo- age of NSER is among the highest. At 84 per- dality of registration. 22 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) This helps identify how various demographic and so- programs as well as any support following shocks cioeconomic characteristics relate to the probability (such as COVID-19 or floods). This can have a com- of exclusion from the NSER. The findings provide an pounding harmful effect on household well-being. ‘ opportunity to identify factors that can lead to sys- temic exclusion. Regression analysis confirms that household location significantly affects the probability of There is a significant variation in NSER matching NSER exclusion. Households in rural areas have a across provinces, with especially low coverage slightly higher probability of exclusion as compared to in Balochistan. The share of households matched their urban counterparts: 13.6 percent vs. 10.5 per- in the NSER varies between 90 percent in KP and 71 cent respectively. While rural households experience percent in Balochistan (Figure 2.3). However, if only a higher probability of exclusion, low coverage is par- considering those households who shared a CNIC, ticularly a concern in Balochistan: Figure 2.4 shows the likelihood of finding a match in the NSER increas- that households in Balochistan have a 26 percent es significantly. In this context, a low NSER cover- probability of exclusion from the NSER whereas it is age rate is particularly concerning for Balochistan only 11 percent for households in KP (see Table B2 in because it also has the highest poverty rates and Annex B for detailed regression results). To address exclusion from the NSER leaves a household unable these gaps, BISP implements a new initiative by de- to access an increasing number of social protection ploying 25 mobile vans to support the dynamic NSER update, 19 of which are in Balochistan, 5 in Sindh, FIGURE 2.3 and 1 in ICT. Since households without a CNIC can- Share of households matched in the NSER by province not register in the NSER, these mobile vans also offer CNIC services. Currently, NSER DRCs are at the Teh- sil level. To complement this, these vans are stationed FIGURE 2.4 Predicted probability of exclusion from the NSER by province Source: Authors’ calculations based on the NSER QRS and NSER adminis- trative data Note: Households were categorized as matched if a match for their CNIC was found in the NSER administrative data. Households were categorized as having an invalid CNIC after validating the shared CNIC numbers with NAD- RA. A household can have multiple CNICs and, therefore, belong to several groups. In this table, the categories supersede each other in the following order: Matched, not matched, invalid CNIC, CNIC not shared.  eview Survey Source: Authors’ calculations based on NSER Quality R MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 23 at the Union Council (UC) level and sometimes at the 2.2 Are the Poor Included in NSER? village level. This is especially important as coverage rates are likely to decrease with a complete transition Poor households are more likely to be included to the dynamic registration approach. 11 in the NSER, a positive finding from a social pro- tection perspective. Based on the regression anal- The strategic integration of mobile registration ysis, households from the poorest decile are only 6 alongside dynamic registration is expected to percent likely to be missing from the NSER. In con- address some of the limitations inherent in the trast, households in the richest decile are 31 percent desk-based registration modality currently in likely to be excluded from the NSER (Figure 2.5). place at the DRCs. While dynamic registration aims to streamline beneficiary data updates, its reliance on FIGURE 2.5 demand-based registration at specific locations pos- Predicted probability of exclusion from the NSER by PMT es accessibility challenges, especially to households deciles in remote or underserved areas. The introduction of mobile registration in areas of Balochistan and Sindh is expected to increase coverage, as was experienced in Brazil, where Bolsa Familia reduced exclusion errors among homeless populations, riverine communities, and ethnic minorities through an active search strate- gy implemented between 2011 and 2014 (Leite et al., 2017). The number of registered families increased from 200,000 in 2011 to 1.9 million in 2015 (Social Protection, 2016 and World Bank, 2019). However, in Pakistan, this initiative is temporary and, while it may address mobility related constraints, other challenges may still prevent households from registering. These  eview Survey Source: Authors’ calculations based on NSER Quality R include lack of trust in state institutions; restrictive gender norms that prevent women from obtaining a Poorer households are more likely to be matched CNIC and registering in the NSER; and other indirect in the NSER, a trend that holds in most provinc- costs incurred by the household. es. Figure 2.6 shows that NSER coverage is signifi- cantly higher among poorer households, especially in KP and Sindh. However, Balochistan has no signifi- 11   This approach is similar to Brazil’s Bolsa Familia—where social workers were sent to the country’s most remote areas between cant variation across deciles: coverage levels are low 2011-2014 to reduce exclusion errors among the homeless, riv- irrespective of household well-being, except for those erine population, and ethnic minorities (Leite et al., 2017) —which significantly increased the registration of marginalized groups in the top decile. This points to issues related to weak (World Bank 2019). 24 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) governance or political instability, which affect every- not included in the NSER database. The total number one irrespective of welfare. Balochistan is mired with of households missing from the NSER is 7.9 million. unique governance challenges, such as weak institu- tional capacity, especially in terms of service delivery, Households registered in NSER are poorer ac- lack of trust in institutions, and recurring political con- cording to numerous characteristics. Households flict and instability (Kakar, 2023; Ahmed and Qayyum, matched between NSER QRS data and the NSER ad- 2023; and Bashir et al., 2023). From a social protec- ministrative data are poorer according to their PMT tion perspective, low coverage in the upper part of the score and other socioeconomic indicators. The av- distribution is less of a concern. erage PMT score of matched households is 34.9, significantly lower than that of households not in the FIGURE 2.6 NSER. A breakdown of socioeconomic indicators Share of households matched in the NSER by decile in (that feed into the PMT score) confirms that matched each province households have worse socioeconomic outcomes than those not matched, including slightly larger household sizes and fewer rooms per person in their dwellings (Table 2.2). Households matched in the NSER are more like- ly to live in dwellings of worse quality with lim- ited access to water, sanitation, and hygiene (WASH). If comparing only those households with a CNIC, Table 2.2 shows that households matched in the NSER are more likely to live with kaccha walls12 (39.7 percent) than unmatched households (30.8 per- Review Survey Source: Authors’ calculations based on NSER Quality  cent) or roofs made of wood or bamboo (36 and 27.7 percent). Only 21.4 percent of the households in the While overall NSER coverage is high, more than NSER have access to toilets with flush, compared to 2.2 million households are missing from the bot- 34.1 percent of those who were not matched. This, tom of the distribution. Roughly 2 million house- combined with the fact that they are also less likely holds are not included in the NSER from the bottom 40 to have access to drinking water from a motorized percent of the population. This is concerning as NSER pump, indicates poorer access to WASH. exclusion excludes such households from social pro- tection or disaster response programs. When looking at households with a PMT score below 32—the score determining the threshold for inclusion into Kafaa-   Walls are referred to as kaccha if they are made of raw bricks, 12 lat —we find that 1.6 million of these households are mud, wood, or bamboo. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 25 TABLE 2.2 Socioeconomic and demographic characteristics by matching status CNIC not   Matched Invalid CNIC Not matched shared PMT 34.95 39.62 39.42 40.83 Rural 63.40% 76.20% 68.00% 64.20% Urban 36.60% 23.80% 32.00% 35.80% Household size 5.05 4.8 4.67 4.68 Dependency ratio 0.4 0.37 0.36 0.35 With children <5 47.00% 40.60% 43.60% 43.40% With a person with disability 14.90% 13.20% 12.40% 12.00% Number of household members per room 3.7 3.2 3.1 2.7 PROVINCE Balochistan 4.30% 4.90% 6.10% 14.20% ICT 0.90% 1.60% 1.20% 2.10% KP 19.00% 7.40% 12.20% 12.90% Punjab 52.40% 67.30% 59.30% 47.80% Sindh 23.40% 18.80% 21.20% 23.00% HOUSEHOLD HEAD Male 3.10% 3.00% 4.20% 4.00% Female 21.40% 11.30% 13.20% 15.50% Transgender 13.50% 21.30% 19.50% 20.10% Illiterate 13.00% 6.90% 9.10% 11.00% Employed 23.30% 43.00% 36.80% 43.40% Moved from another district 36.30% 22.50% 27.70% 24.80% Affected by floods in July 2022 59.60% 76.50% 68.50% 70.60% Has access to the internet 39.70% 22.90% 30.80% 29.00% Has no electricity connection 21.40% 37.80% 34.10% 37.60% Roof material (concrete or cement) 41.00% 26.70% 32.50% 24.90% Roof material (wood or bamboo) 22.30% 30.90% 25.00% 21.00% Wall material (Pakka) 14.50% 10.50% 15.70% 21.20% Wall material (Kaccha) 21.10% 39.90% 30.40% 34.50% Household has improved toilet (with a flush) 3.90% 6.60% 5.00% 5.30% Drinking water source (Hand pump) 41.00% 26.70% 32.50% 24.90% Drinking water source (Motorized) 22.30% 30.90% 25.00% 21.00% Drinking water source (Open/closed well, pond, canal) 14.50% 10.50% 15.70% 21.20% Owns a fridge 21.10% 39.90% 30.40% 34.50% Agriculture land owned (>5 acres) 3.90% 6.60% 5.00% 5.30% Source: Authors’ calculations based on the NSER Quality Review Survey Note: The table presents a subset of PMT indicators. For other PMT indicators, please see Annex B. Dependency ratio is defined as the ratio of the number of household members less than 15 and above 65 to the total household size. Wall material is referred to as Pakka if it is made of burnt bricks or stones and kaccha if it is made of raw bricks, mud, wood, or bamboo. 26 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) While employment rates of household heads found in the NSER (see Table B3 in Annex B for de- are similar across matching categories, NS- tailed comparison). However, a few significant differ- ER-matched households are more likely to en- ences exist. For example, 90.4 percent of matched gage in seasonal employment. Seasonal em- households report not having access to toilets, in ployment is often an unstable source of income contrast to 82.9 percent of unmatched households. and makes households more vulnerable to income Matched households are also likely to live in dwellings shocks. Around 55 percent of households matched in of worse quality: 50.2 percent reported roof material the NSER engage in seasonal employment, compared to be wood or bamboo as compared to 45.6 percent to 47 percent of those who were not matched in the of unmatched households. NSER (Figure 2.7). This underscores the findings that households registered in the NSER are poorer. (For 2.3 Why Are People Not Included in NSER? other PMT indicators across matching categories see Table B1 in Annex B). NSER AWARENESS A comparison between matched and unmatched Awareness about the NSER registration process poor households (those below PMT score 32) remains low, with almost half of the unmatched shows little difference in characteristics. Most households “not aware.” Among the households BISP safety net programs targeted poor and vul- not in the NSER, 47 percent reported they did not nerable households, defined based on a PMT score know how to register (Figure 2.8). Survey respon- threshold of 32. For households below this thresh- dents were also asked whether they knew how to old—those considered poor—on average, matched contact BISP, with only 27 percent responding “yes”. households are slightly worse off than those not Based on focus group discussions conducted in KP FIGURE 2.7 FIGURE 2.8 Employment type by matching status Reasons for not registering in the NSER Source: Authors’ calculations based on NSER Quality  Review Survey  eview Survey Source: Authors’ calculations based on NSER Quality R MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 27 with those not NSER-registered, participants said of non-beneficiaries. This difference is not surprising, they found the registration process to be tedious, of- as Kafaalat benefits are tangible and create a direct ten requiring multiple visits to the NSER DRCs, where link between their action (registration) and its out- wait times are often very long and there is no guar- come (benefit). These findings underscore the need antee of being interviewed on the same day. Some for stronger outreach and information dissemination participants were unaware of the DRCs and assumed to ensure that all registered households are adequate- that they had to register only at NADRA. Many partic- ly informed about their status. ipants did not know the type of documents needed to register (Majoka and Armaghan, 2024). This problem Despite BISP’s better communication with its indicates inadequate communication. program beneficiaries, much of the information about the program is still being spread through Many NSER-registered households did not know informal channels like word of mouth. Focus they were registered, indicating weak communi- group discussion (FGD) participants in all provinc- cation and a lack of public understanding. Over- es reported that they primarily learned about BISP all, only 42 percent of those found in the NSER knew through conversations with others, often receiving in- they were registered, with awareness particularly low complete or inaccurate information. Additionally, par- in Balochistan, where just 13 percent of households ticipants shared that after door-to-door surveys, they were aware of their registration status (Figure 2.9). In frequently revisited registration centers due to confu- contrast, awareness was significantly higher among sion about next steps or outcomes. Many continued Kafaalat beneficiaries, with 68.6 percent knowing visiting even after registering, unsure of their eligibility they were registered, compared to only 32.5 percent or registration status. In some cases, misinformation from neighbors and shopkeepers further complicated FIGURE 2.9 the registration process, leaving households uncertain Share of households aware of the NSER registration about their standing within the program (Majoka and status Armaghan, 2024). Lack of information can also lead to mistrust in institutions. While the use of NSER for targeting is designed to be objective and impartial, when house- holds are unable to access information that can help them understand these processes, it can lead to a lack of trust and a perception of corruption and unfairness. For example, focus group discussion participants per- ceived the targeting process to favor those with polit- ical connections. These perceptions and experiences Review Survey Source: Authors’ calculations based on NSER Quality  28 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) undermine trust in the fairness and impartiality of the MODALITY OF REGISTRATION registration process (Majoka and Armaghan, 2024). Coverage rates are better in areas with dynamic The NSER process evaluation corroborates registration, a finding confirmed by the regres- these findings. The evaluation found that a lack of sion analysis. Almost all households (96.5 percent) effective awareness and communication channels in areas where dynamic registration was the most and absent radio broadcasts and electronic media common modality were found in the NSER (Figure hinder stakeholder engagement. It also found that 2.10, Panel A). Regression confirms that the prob- feedback mechanisms, such as “hotlines” and sug- ability of exclusion is only 3 percent in areas where gestion boxes, did not exist, limiting communication DRC is the most common modality, compared to 20 about concerns and clarifications (Khan, 2023). The percent in areas where door-to-door surveys are most process evaluation also highlights the importance of common (Figure 2.10, Panel B) improving the coherence and structure of information, particularly to include those with limited literacy. Re- FIGURE 2.10 latedly, the process evaluation found inadequate at- Share of households in the NSER tention to cultural sensitivities. Panel A: By modality of registration As NSER transitions to dynamic registration, it will have to make more concerted efforts to en- sure households know how and where to regis- ter. The communication campaign for Cadastro Uni- co, Brazil’s social registry, has significantly evolved over time, relying on multiple channels to ensure that Panel B: Predicted probability of exclusion from NSER the target population is not only aware of the program but also knows how to contact the implementation agency. It also deployed mobile teams and boats to reach out to poor and marginalized groups proac- tively. As a result, the number of registered families belonging to such groups increased from 200,000 in 2011 to 1.9 million in 2015 (Social Protection, 2016 and World Bank, 2019).  eview Survey Source: Authors’ calculations based on NSER Quality R MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 29 However, as NSER transitions to a desk-based However, it is worth noting that with dynamic registra- modality of data collection, it is likely that cover- tion, households trying to register have to incur trav- age rates will reduce. Currently, NSER covers 4 out el and other direct and indirect costs, and they often of 5 Pakistanis. However, in other countries, where have to forego work to visit DRCs. FGDs participants households can only register through desk-based in Majoka and Armaghan (2024) highlighted the sig- modalities, coverage rates are much lower. For ex- nificant burdens of transportation costs and the need ample, the social registry covers only 40 percent of to sacrifice work and domestic responsibilities to visit the population in Brazil and only 50 percent in Türkiye a NSER DRC. 13 In this context, the cost of data col- (Leite et al., 2017). NSER aims to finish updating all lection is now partially transferred to the households. its data through desk-based modality by September 2024 and it is expected that coverage rates will de- With only one DRC in each Tehsil, these centers crease. However, from a social protection perspec- have limited capacity to serve a large number tive, it is important to ensure that coverage rates of households. Currently, there are only 650 DRCs, remain high among the poor and vulnerable. NSER which limits the number of households that can regis- is already planning to increase its local presence by ter in a day. According to the latest estimate, around setting up registration centers at the Union Council 79,000 households register or update their data ev- (UC) level, which will make it more convenient for ery day through BISP’s DRCs (around 105 surveys households to register in the NSER. However, poorer per day per DRC). This often leads to overcrowding households face higher barriers to accessing public and long wait times. FGDs participants also shared services not only due to mobility constraints but also that they often visit these centers multiple times and due to the opportunity cost of visiting government of- spend the entire day there due to crowding and long fices, as it often requires foregoing work. To ensure lines. This makes the registration process tedious and such households don’t fall through the cracks, NSER time-consuming (Majoka and Armaghan, 2024). How- may have to combine desk-based registration with ever, BISP plans to roll out at least one DRC in each door-to-door data collection Union Council, which will significantly increase its ca- pacity to register or update the data of households. Compared to door-to-door surveys, dynamic reg- istration allows more regular updates and intake Yet, door-to-door registration is slightly better and has decreased the cost of data collection at capturing households farther from urban cen- for BISP. Data collection through DRCs costs, on ters. Public service delivery is often harder in remote average, US$0.53 per survey, compared to US$2.47 or difficult-to-reach areas. The NSER DRCs are mostly for each door-to-door survey in 2011. Either mode of data collection is significantly less costly than in other 13  The only data available on cost of travel, food, and foregone countries. For example, in Niger, the cost per house- work is from Shabbar’s research (forthcoming) who calculates the hold to collect data for PMT through door-to-door average cost incurred by BISP beneficiaries when they travel to col- lect their benefit. It can be as high as 15 percent of the total benefit surveys was US$6.8 (Premand and Schnitzer, 2018). value. 30 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) set up in urban areas. For this analysis, their location the effectiveness of door-to-door surveys in register- is used as a reference point to ascertain the remote- ing the poorest and remote households, opting for a ness of a household’s location. It is surprising that mixed approach where door-to-door surveys comple- households farther from the NSER centers are more ment dynamic registration in remote and poorer re- likely to be in the NSER. However, door-to-door sur- gions could increase registration for the most margin- veys perform slightly better in registering households alized communities. farther away as compared to desk-based registration, especially in rural areas. This finding is confirmed by An additional barrier to registration that NSER the regression analysis (see Annex B, Table B2) that faces is the lack of relevant documentation re- shows that urban households are less likely to be ex- quired to register in the NSER such as CNIC, cluded in case of desk-based registration. The aver- B-form, and updated household information in age distance of matched households from their clos- the NADRA database. Among the bottom quintile, est NSER DRC is 21.6 km for door-to-door surveys 10.7 percent of households did not have a CNIC (see and 20 km for dynamic registration (Figure 2.11). Table B4 in Annex B for CNIC by quintile). While not having a CNIC can prevent a household from regis- Using mixed approaches to registering house- tering in the NSER, it can also lead to exclusion from holds in NSER can be most effective. Considering other entitlements as access to public and private goods and services (such as voting, buying a sim card, opening a bank account, accessing other safety FIGURE 2.11 net programs, etc.), as they now increasingly require Average distance from the NSER DRC having a CNIC. Larger households are more likely to be exclud- ed from the NSER. An increase in the household size by one member increases the probability of ex- clusion from the NSER by 1.9 percent. This is not surprising considering that larger families have higher mobility constraints due to care responsibilities and transportation access. The regression analysis also shows that some marginalized communities are also likely to be missing from the NSER. For example, households Source: Authors’ calculations based on the NSER Quality Review Survey with transgender heads are 5.3 percentage points Note: In this graph, desk-based registration methods refer to registration by either dynamic registration centers or temporary desks set up in 2020. Door- more likely to be missing compared to households to-door survey includes the teacher model. with male heads. This effect gets stronger for door- MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 31 to-door surveys. We also observe a similar trend among households belonging to marginalized occu- pational castes that are more likely to be excluded from the NSER when the modality of registration is door-to-door. One explanation could be that these communities often live in clusters and are often more disconnected from services. If this is because of the enumerator bias, dynamic registration centers offer an opportunity to improve their access, especially by better training the DRC staff. A surprising finding from the regression analysis shows that generally, gender, age, marital sta- tus, or literacy does not impact the probability of exclusion from the NSER. The regression analysis shows that factors such as the gender of the house- hold head, age, marital status, and literacy rate, have no statistically significant impact on the likelihood of exclusion from the NSER. It may indicate that oth- er factors or unmeasured variables may play a more significant role but such an inquiry is outside of the scope of this analysis. 32 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 3. NSER Data Quality Multiple data quality-related factors can affect changes in variable definitions. targeting performance. The accuracy and effec- tiveness of PMT-based targeting relies heavily on The correlation between PMT scores in the NSER data quality. If data are not accurate, due to being administrative database and the NSER Quality outdated, for example, the performance of the PMT Review Survey data is remarkably low, pointing model reduces. Aiken et al. (2023) identify two phe- to data “decay”. The NSER administrative dataset nomena that reduce the effectiveness of a PMT mod- included PMT scores for each matched household. el in predicting poverty. First is data “decay”, which The NSER QRS collected data on variables to cal- refers to outdated data due to changing household culate the PMT scores. A comparison between the socio-economic conditions. Second is model decay, two scores—the PMT scores from the NSER data which occurs if consumption data used to create and the PMT scores calculated from the NSER QRS the PMT model loses its effectiveness in predicting data—shows a weak (albeit positive) correlation. Most poverty. They estimate that the predictive power of a households in the NSER updated their data more PMT model reduces by 1.5 to 1.9 percentage points than three years ago. Yet, variable inputs into PMT with each passing year. In Pakistan, the PMT formula score calculation can change, especially those re- is based on the HIES 2013-14. 14 Given the multi- lated to household composition, such as changes in ple crises that have affected households in Pakistan, household members. However, it is particularly puz- especially since 2019, the model may be outdated. zling that the correlation coefficients are smaller for However, in the absence of a recent nationally rep- resentative household survey, we cannot gauge the TABLE 3.1 PMT model’s predictive power. We explore factors Relationship between PMT scores by date of last update that affect data decay due to irregular updates and Year of NSER registration Correlation PMT in NSER and in Survey 2017 0.4825*** 2018 0.4172*** 14  The PMT formula was first developed using the HIES 2007/08. Due to the dynamic nature of poverty and its correlates, in 2015, 2019 0.5477*** the formula was revised based on the HIES 2013/14, primarily to 2020 0.4636*** improve coverage of the poor. The proxy indicators and their rela- 2021 0.4614*** tive weights were reviewed based on the correlation between indi- 2022 0.3050*** cators and per adult equivalent household consumption to improve coverage of BISP among the poor in urban areas. The PMT score 2023 0.3692*** was then recalibrated to correct for discrepancies in estimating Source: Authors’ calculations based on NSER Quality Review Survey and poverty rates at provincial levels. The PMT formula was again val- NSER administrative data idated using HIES 2018-19, but there were no statistically signifi- cant differences, therefore no revisions were made to the relative Note: Asterisks denote the following levels of significance: * p<0.10, **p<0.05, weights. *** p<0.01 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 33 data collected more recently in 2022 and 2023 (Table ators interview male respondents. DRCs have male 3.1). This was also the time when BISP transitioned to and female enumerators, but respondents are not dynamic registration. given a choice to pick. Moreover, the DRCs are often over-crowded and do not offer much privacy. These 3.1 Registration Modalities factors can influence respondents as multiple studies suggest that respondents alter responses based on Registration modality can significantly affect re- the gender of the interviewer, with women potential- sponses and data accuracy. Table 3.2 shows that ly underreporting poverty levels to male interviewers the strength of correlation is lower for data collected due to cultural norms (Benstead, 2014; Blaydes and through desk-based kiosks or dynamic registration Gillum, 2013). centers. It underscores the importance of data col- lection methods. For example, during door-to-door Other factors related to data collection methods visits, interviewers can visually observe many of the can also induce respondent bias and decrease household characteristics and follow-up responses data accuracy. Masselus and Fiala (2024) highlight accordingly. In desk-based registration, this is not that female respondents may report lower household feasible. For example, in Pakistan, gender segrega- income when surveyed alone compared to interviews tion is generally observed in public spaces. The gen- with their spouses. Other enumerator characteristics, der of the enumerator can induce a response bias. such as urban or rural background, can also influ- In door-to-door registration, female enumerators typi- ence responses, exacerbating disparities in poverty cally interview female respondents and male enumer- measures (Huddy et al., 1997; Kane and Macaulay, 1993). The FGD respondents reported concerns re- garding men asking women questions related to TABLE 3.2 household composition, which highlights the need for Relationship between PMT scores by modality of registra- tion gender-sensitive data collection methods and com- prehensive training for staff at DRCs to enhance their Correlation PMT in NSER Registration Modalities ability to manage interviews sensitively and effectively, and in Survey Desk Data 0.3851*** especially in crowded and less private environments (Majoka and Armaghan, 2024). Furthermore, as sug- Dynamic Registry 0.3681*** gested by Martinelli and Parker (2009), inherent bi- Door to door 0.4847*** ases reveal that individuals may refrain from disclos- Teacher Model Data 0.4883*** ing their true poverty status when they perceive their neighbors to be less impoverished, highlighting the Source: Authors’ calculations based on NSER Quality Review Survey and complex interplay of social dynamics in reporting pov- NSER administrative data erty levels. Note: Asterisks denote the following levels of significance: * p<0.10, **p<0.05, *** p<0.01 However, if these differences are due to misre- 34 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) porting, this can be addressed through random data sources such as spatial poverty and accessibility spot checks and other similar measures. There is maps to validate larger trends observed in the NSER. a possibility of misreporting of assets since respon- dents are aware that their responses have a direct 3.2 Urban and Rural Classification bearing on their selection for benefits. Like in other countries, Pakistan can also institute cross-checks Classification of urban or rural household status to validate the self-reported data. Brazil has routine for the PMT model is challenging. In an environ- monitoring activities such as spot checks, random ment of rapid peri-urban growth and development of home visits, and audits to ensure program integrity secondary cities, it is important that a social regis- and identify discrepancies. During face-to-face ses- try adequately captures location-based differences sions for self-declaration of socioeconomic condi- in population density, which affects living standards, tions, individuals are also informed about verification prices, and access to services (Grover et al., 2022). measures, including the consequences of false dec- NSER administrative data relies on self-reported data larations such as benefits suspension and penalties to classify households into urban and rural catego- affecting all family members (Leite et al., 2022). Sim- ries.15 In contrast, HIES data—on which the PMT ilarly, in Türkiye and Ukraine, regular in-person evalu- model is based—uses census block ID to classify ations are conducted by local inspection officials who households into urban and rural. Both methods of verify household data (Barca and Hebber 2020). classification are problematic. First, the existing in- formation on urbanization from self-reported and ad- Pakistan can also explore data verification by tri- ministrative sources exhibit significant gaps and vari- angulating NSER data with other administrative ation. Second, the administrative definition of urban/ data sources such as vehicle registration, tax re- rural is outdated.16 Between 1975 and 2020, Paki- cords, etc. High-resolution poverty maps combined stan experienced one of the fastest urbanization rates with administrative records have been utilized for veri- fication and outreach to specific populations in Brazil, 15   At the time of registration, households are asked whether they and allowing for interoperability of data stores in Chile live in urban or rural areas. NSER enumerators are supposed to and Türkiye has led to efficiency gains (Leite et al., cross-check this information by asking whether the household re- ceives any municipal services. A household will be classified as 2020). While Pakistan has CNIC-linked tax, immov- urban only if it receives municipal services. However, this question able asset and mobile connection records, there are is not part of the official NSER questionnaire and hence cannot be validated. Moreover, urban areas in South Asia have a large and large gaps in documentation, especially at the lower growing number of informal settlements that also do not receive end of the welfare spectrum, as well as a restricted municipal services. level of interoperability of these data sources. Invest- 16   Based on the European Union Global Human Settlement Layer ments to rectify these data gaps and strengthen ad- (GHSL) data, which uses degrees of urbanization instead of a bina- ry urban-rural category, only 18.8 percent of surveyed households ministrative linkages to the social registry can create are identified as rural whereas a majority of the households fall into opportunities significantly improved verification sys- the “suburban” category. In contrast, 2023 Census data classifies 61.2 percent of households as rural. This indicates that the admin- tems. Related to this is enhancing the use of external istrative classification in Pakistan is outdated. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 35 in South Asia, creating rapidly developing centers of living in urban areas are likely to have access to bet- economic activity (World Bank, 2024c). However, the ter housing, more consistent access to electricity, and official urban-versus-rural definition uses an adminis- improved water than their rural counterparts, but this trative classification that ignores intermediate spaces comes at a cost, increasing the cost of living. between dispersed rural areas and dense towns. Self-reported information on urbanization is Differences in the urban-rural cost of living are highly inconsistent and variable, even for house- significant, particularly when measured using holds within the same enumeration area (EA), as Fig- CPI inflation. There is a significant dispersion in in- ure 3.1 highlights, indicating the absence of clear flation rates faced by households, including urban household consensus on whether they are classified and rural areas (Baez et al., 2021; Kishwar et al., as urban or rural.17 For nearly three-quarters of all 2024). These differences stem from several factors, EAs, the share of households reporting themselves as including differing household consumption patterns “urban” in the NSER can vary anywhere between 20 in urban and rural areas and access to higher-qual- ity services such as housing, water, and electricity in urban areas. Rural households, on average, allocate   “Mixed” Households have multiple responses in the NSER, with 17 different members reporting different urban/rural status. This could a larger share of consumption to food. In contrast, be because of (i) difference in self-perception; (ii) migration patterns urban households tend to spend more on housing where a household member has moved to a different location but has not updated their NSER data; or (iii) change in household com- and related services. Households of a similar profile position due to marriage and patrilocal household setup. FIGURE 3.1 Distribution of self-reported urban status by Enumeration Area Source: Authors’ calculations based on the NSER Quality Review Survey and NSER Administrative Data Note: The urban/rural category was available only for the households matched in the NSER. 36 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) and 80 percent. These inconsistencies in the data can suming all other inputs remain unchanged. Only 1.2 lead to significant variations and discrepancies even percent of households move across the PMT thresh- within the same enumeration area, creating distor- old of 32 in either direction. This suggests that the tions in the PMT scores, directly affecting households’ PMT score is very stable to even large reconstitutions eligibility for social protection programs. Consequent- in households’ urban status. ly, misreporting urban or rural status can result in the misallocation of resources, where ineligible house- The above analysis highlights the need to un- holds might receive aid while those genuinely in need derstand further the role and measurement of could be excluded. urbanization in the PMT formula. Households in urban areas are qualitatively different from their rural While aggregating household urban-rural status counterparts in terms of their livelihoods, incomes, at the EA level results in significant re-classifica- prices, and access to health and education services tion, the effect on the distribution of PMT scores and markets. All these factors affect the welfare of and targeting is very small. To address high vari- households, which underscores the importance of ation across EAs, households are assigned urban or understanding the interplay between household wel- rural locations based on the majority response within fare and its circumstances. their respective EA. Table 3.3 shows that about 1 in 4 households would be reclassified from rural to urban FIGURE 3.2 (or vice versa) based on the majority status. However, PMT distribution by different urbanization classification this reclassification does not significantly change the overall distribution of the PMT Score (Figure 3.2), as- TABLE 3.3 Share of urban and rural households Share of Households Majority Status in an EA  Urban  Rural  Urban  66.4%  12.5%  Self-Reporting in NSER Administrative Data  Rural  11.4%  65.5%  Source: Authors’ calculations based on NSER Quality Review Survey Source: Authors’ calculations based on the NSER Quality Review Survey Note: The percentages exclude the sample of households for which the clas- sification is missing, are “mixed” or were not matched in the NSER Admin- istrative Data MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 37 TABLE 3.4 Variation in household size by survey type NSER Quality Average Household Size Census 2023 HIES 2018 NSER Administrative Data Review Survey Balochistan 6.4 8.1 6.1 5.6 ICT 5.7 4.8 5.3 KP 6.9 7.5 4.8 6.8 Punjab 6.4 5.8 5.0 5.3 Sindh 5.6 6.2 5.0 5.4 Pakistan 6.3 6.2 5.0 5.6 Source: Authors’ calculations based on NSER Quality Review Survey, NSER Administrative Data, HIES 2018-19, and Pakistan Census Data 2023. 3.3 Household Size and in some cases, these variables have a positive effect on PMT calculation, whereas in other cas- The average household size recorded in the es, this effect is negative (see Table B5 in Annex B NSER is lower relative to both administrative da- for a list of variables in the PMT formula that uses tabases, as recorded through the 2017 and 2023 household size). census, as well as the HIES 2018-19 (Table 3.4), which can have implications on targeting. Nota- Such recording differences can have significant ble here is that (except for Balochistan), the NSER implications on PMT scores, but due to the bidi- Quality Review Survey records the lowest average rectional nature effect on household size and its household size compared to the NSER administrative components, it is less clear whether a missed data for the same set of households. These differ- member(s) is likely to push a household below ences persist even when the average household size or above the PMT threshold. The direct mechanical in each quintile is compared across data sources.18 effect of recording an increase in household size for The size and composition of households are import- the median household is a decrease in PMT score ant welfare determinants, with clear implications on of 2.9 in rural areas and 3.3 in urban areas, which both the ability to generate income and the division is large enough to push vulnerable households into of accumulated endowments. The PMT formula uses qualifying for BISP Kafaalat . In theory, other indirect variables related to household size in multiple ways, channels could offset this negative change; for ex- ample, if the missing member was a literate adult, the associated marginal increase would be approximately   The average household size in the bottom quintile in 7.22 in 18 5.4, resulting in a net positive effect on PMT. HIES, 7.18 in NSER QRS, and 6.41 in the NSER Administrative data. 38 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) The variation in household size could be due dated civil registration data. While it may serve as an to differences in definitions across databases. incentive for households to keep civil registration data Census and household surveys such as HIES de- updated, it will still misrepresent how people actually fine a “household” as a group of people who share a live with each other. Overall, these results emphasize common kitchen and consume their meals together. the need to improve the measurement and updating This is also the same definition that the NSER Quality of household composition numbers in registries. Review Survey used. However, the NSER Administra- tion data records this information differently. Before transitioning to the dynamic registration, it differen- tiated between “household” and “family” where the “household” definition was the same as in the other surveys, but a “family” referred to a married female member and those household members who were re- lated to her by marriage or by blood. Based on this definition, a household could have multiple families. However, dynamic registration, now retrieved infor- mation on household members from NADRA’s data- base. It is no surprise that the average household size with dynamic registration is even smaller since only members related by blood or marriage, and are also registered with NADRA, will now be considered part of the “household”.19 It is important to keep the definition in the NSER Administrative Data consistent with the HIES 2018-19. The aim of using NADRA’s data on family members is to reduce the incidence of misreporting, but it has its caveats. One example is that Pakistani household composition is not strictly based on re- lationships by marriage or by blood, and individuals may reside with distant relatives or family friends. Another example is that NADRA data assumes that people registering in the NSER will always have up-   This presupposes 100 percent coverage of CNICs and up to 19 date data on marriage, birth, and death. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 39 4. NSER During Shock Response Safety nets become increasingly important as aware of their status in the NSER, and are more likely shock-response mechanisms during heightened to have received assistance after the floods. Among economic climate shocks. Pakistan has faced a flood-affected households, this trend of inclusion was myriad of economic, health, and climatic shocks in stronger for poorer quintiles of households than richer the past five years. Shock-responsive safety nets can households. Within the set of households that re- support the most vulnerable populations in the after- ported exposure to floods, the poorest quintile has math of shocks. They are social protection programs the highest share that self-report information to NSER designed to support vulnerable populations in the and receive benefits from its schemes. This trend de- event of a shock quickly and seamlessly. These safe- creases when moving up the PMT score distribution, ty nets are “adaptive”, meaning they can be scaled- suggesting that flood registration targeting was pro- up or redirected to meet increased demand or target gressive in its reach, enrollment, and disbursement to those most affected by the shock. The goal is to mit- the shock-vulnerable poor. igate shock damage on the poorest and most vulner- able segments of the population by providing timely Controlling for demographics and other factors, and effective support. Developing a shock-responsive flood-affected households are more likely to be safety net requires a strong social registry and good covered in the NSER, suggesting that the 2022 enrollment, delivery, and payment systems, often le- Flood Social Protection Response was success- veraging technology. ful. In 2022, the Government of Pakistan (GOP) ear- marked PKR 35 billion to aid flood response, in par- The 2022 floods provide a unique opportunity to ticular, to enroll in NSER and target and disburse cash better evaluate NSER’s effectiveness during a transfers to approximately 2.7 million households in devastating shock.20 About 20 percent of house- flood-affected areas (SDPI 2022). Results from probit holds report being directly harmed by floods during regressions on self-reported flood exposure find an 2022. These households are more likely to be regis- 8.5 percentage point higher probability of registration tered in the NSER (89 versus 82 percent), are more in the NSER for those affected by floods (see Table B6 in Annex B for regression results). This could be due to the targeted registration drives that BISP im- 20   The exposure to floods is measured through self-reporting in the survey, specifically through the question “Was your household plemented to ensure that flood-affected households or community directly affected by the floods during the last mon- could register and/or update their information in the soon”. While this question allows households to respond based on their experience of any flooding in that time period, the enu- NSER in the aftermath of floods. However, our anal- merators were specifically instructed to ensure that respondents understood this question to refer to the 2022 floods. ysis also shows that within flood-affected districts, 40 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) households, in general, were more likely to be in the to floods. While it shows that NSER’s targeted reg- NSER, suggesting a positive externality of these reg- istration drives were successful in ensuring inclusion istration drives. of households, it also it also shows that NSER lacks up to date ex ante data, which can help in identifying A slightly higher share of affected households and targeting shock vulnerable households before the had their NSER data updated during or after the shock hits. This can ensure that by the time shock flood period relative to those not exposed, indi- hits, climate vulnerable households have already built cating a need for regular ex-ante data updates. resilience to cope with shocks. Figure 4.1 shows the distribution of households by the latest date of information update in the registry. Data was updated for 23 percent of flood-affected during and after the flood period (June to November 2022), compared to 19 percent for those not exposed FIGURE 4.1 Distribution of households by the latest date of information update in NSER MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 41 5. Policy Recommendations This study presents a positive evaluation of NSER more likely to (i) know about the NSER, (ii) be in- completeness and accuracy and several important cluded in the NSER, (iii) have their information up- policy priorities emerge. dated during or after the floods, and (iv) receive cash transfers. Among flood-affected house- holds, these trends are stronger for those in low- 5.1. Keep Higher Coverage in the er quintiles, meaning that it is working to reach Bottom Four Quintiles the poorest households. This proves the value of continuing to invest in NSER as the main nation- In Pakistan, the national social registry, the NSER, wide social registry for shock-response use. To covers 84 percent of the total population, a propor- this end, improving linkages between NSER and tion much higher than in other countries. As the Pov- disaster management authorities by consolidat- erty, Equity, and Resilience Assessment (World Bank, ing the registry to a digitized hazard database forthcoming) shows, a significant portion of Pakistani with regularized mechanisms for data sharing. households are at risk of poverty, with one-third of These efforts must be accompanied by a rec- non-poor households vulnerable to falling into poverty ognition of the differentiated needs of shock-in- as of 2019–20. Climate change, particularly through duced internally displaced households as an extreme weather events, exacerbates this vulnerabil- especially vulnerable group and ensuring their ity, impacting health, food security, and livelihoods, inclusion in the NSER through targeted outreach. and disproportionately affecting marginalized groups. 5.2. Overcome Existing Challenges Continue strengthening NSER use as the social Related to Access to NSER registry in the aftermath of shocks Despite the high overall coverage of 84 percent, over 2.2 million households from the bottom 40 percent of • NSER, a good instrument for shock protec- the population are still not included in the NSER. This tion, should be further strengthened to focus means they are excluded from social protection and on shock-response. This study shows that the disaster response programs. The challenges of inclu- NSER was a good instrument for assisting house- sion relate to a lack of awareness of the NSER and holds during the aftermath of the 2022 floods. the modality of registration. To solve some of these Households harmed by the floods in 2022 were challenges, several recommendations emerge: 42 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) sion. Communications campaigns targeting different stakeholder groups, complement- Adopt mixed approaches ing broad-based communication, can raise to registration modality awareness about and registration in the NSER. • Mixed approaches to registering house- • Other countries provide excellent examples holds in NSER can be most effective. Con- of strong communication campaigns. Rwan- sidering the effectiveness of door-to-door sur- da’s Umurenge Program employs various mass veys in registering households in relatively remote communication strategies to inform and engage locations, opting for a mixed approach where all communities, including nonrecipient house- door-to-door surveys complement dynamic reg- holds, about the program’s objectives, policies, istration in remote and poorer regions could in- and procedures (Nsour et al., 2020). These in- crease registration in the poorest communities. clude radio and television shows featuring gov- ernment representatives and beneficiaries, work- • Dynamic registration is the modality BISP shops for local officials, distribution of program most recently implemented, but data accu- brochures in local languages, and periodic arti- racy may suffer. The correlation between PMT cles placed in national newspapers and electron- scores based on the NSER QRS and NSER ad- ic newsletters. Brazil’s Bolsa Familia program has ministrative data is weak, especially in later years evolved its communication strategies over time, when BISP switched to a dynamic, desk-based utilizing various channels—such as websites, so- registration model. Continuing to monitor data cial workers, letters, social media, mobile appli- accuracy for different registration modalities can cations, and call centers—along with transparen- ensure that data accuracy does not suffer. cy measures like a Federal Transparency Website and a weekly newsletter, to enhance social con- trol, participation, and information access (Social Improve awareness of NSER by implementing Protection, 2016). communication campaigns Improve outreach for registration to remote, • Stronger communication campaigns can marginalized communities improve awareness of NSER and its regis- tration process. The study points towards low awareness of NSER and its registration pro- • Targeted registration drives in Balochistan cess. Strengthening communication and out- and remote rural areas all over Pakistan can reach, particularly targeting the poorest, most improve NSER registration at the bottom of marginalized communities, can improve inclu- the distribution. This study highlights significant MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 43 gaps in NSER registration in Balochistan. Specific A mbler et al. (2021) suggest acknowledging 2. mobilization efforts to register households in the spousal disagreement in reporting asset owner- province and increase access to CNICs can im- ship and decision-making, emphasizing the need prove NSER coverage. This study also showed for comprehensive data collection methods that that certain marginalized communities, such as capture diverse perspectives within households. transgender individuals, are more likely to be ex- cluded. Prioritizing registration campaigns in re- 3. Kilic and Pave Sohnesen (2019) highlight the im- mote rural areas and marginalized communities portance of thinking through the questionnaire (transgender individuals or people from lower oc- design in influencing reporting behaviors and ad- cupational castes, for example) through mobile vocating for standardized approaches to survey registration drives can improve NSER inclusion. administration to minimize reporting differences. 5.3. Improve Data Decay and Accuracy 4. DiMaio and Fiala (2020) underscore the necessity of cautious interpretation of collected data, especial- Improving the decay of data across quintiles and ly in sensitive contexts, and recommend transpar- continuous monitoring of data accuracy across ent documentation of enumerator assignment pro- registration methods can ensure data integrity. cesses to assess and mitigate enumeration effects. To achieve this, several recommendations emerge: • Training DRC staff to serve people from dif- ferent backgrounds, castes, and communi- Improve data accuracy ties can improve data accuracy. The biggest concern is the extent to which NSER excludes • Evaluating and improving factors that affect some of the most marginalized communities. The data accuracy can improve targeting perfor- prejudices and opinions of inadequately trained mance. This study highlighted interviewers’ gen- staff can create a stigma with the registration pro- der and privacy concerns in DRCs as the main cess leading to low uptake among the most vul- aspects affecting data accuracy. Several recom- nerable (Hossain 2011). With a stronger focus on mendations emerge from the literature to address dynamic desk-based registration, sensitizing DRC these concerns and improve data accuracy: staff to the challenges these communities face and training them to serve members from marginalized 1. Flores-Macias and Lawson (2008) and Fisher et communities can improve registration results. al. (2010) propose that researchers record inter- viewers’ demographic characteristics and imple- • Incorporating randomized spot checks to ment context-sensitive training to mitigate gender validate household information will improve and interview biases. data accuracy through desk-based approach- es. As in other countries, such as Brazil, Türkiye, 44 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) and Ukraine, Pakistan can also institute random spot checks to validate self-reported data. As Continue sharing data to create a powerful mentioned earlier, Brazil conducts routine mon- evidence base for decision-making itoring activities, such as spot checks, random home visits, and audits to ensure program in- • Improve data quality and continue sharing tegrity and identify discrepancies. Furthermore, data to create a strong evidence base to im- during face-to-face sessions for self-declaration prove decision-making. Linking survey and ad- of socioeconomic conditions, individuals are in- ministrative data provides richer insights on com- formed about verification measures, including the pleteness and accuracy than either data source consequences of false declarations, such as ben- alone. Based on this experience, continuing to efits suspension and penalties (Leite et al., 2022). share data across institutions regularly is key to gaining a better understanding of possible im- provements in data collection and improvements Utilize urban/rural definition that captures differences in living conditions in updating the NSER. Based on preliminary find- ings on the NSER administrative data, an end- to-end review of how administrative data is pro- • In rapid urbanization, capturing differenc- cessed and coded and how the PMT formula is es in cost of living and living standards applied could improve targeting performance. across rural and urban spaces is critical to improve targeting. The NSER classification of households’ urban or rural status (and the 5.4. Continue Research on Effectiveness of PMT) is based on administrative considerations Targeting Methodology & Data Collection that does not truly reflect the differences in liv- Modalities ing conditions across space in Pakistan. In future discussions on updating the PMT formula, espe- This study also uncovers several important top- cially in a context where households self-report ics for additional research related to data qual- urban or rural status at the DRCs, applying an ity and targeting. Future work should focus on: urban/rural definition that more realistically re- flects actual differences in living conditions and • How data collection modality affects data access to services can improve targeting. These quality. As NSER transitioned to desk-based could include proxy measures for differences in data collection, the quality of data has suffered. location characteristics that affect welfare, such This requires further inquiry into the impact of as prices, access to services, or distance mea- changing data collection modality on data qual- sures to schools, markets, and health centers. ity, with a focus on identifying variables that are MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 45 more likely to be misreported and assessing the • Understanding climate vulnerabilities and impact of misreporting on targeting decisions. finding ways to incorporate it into the PMT formula. Since Pakistan is increasing- • How transition to dynamic registration may ly vulnerable to climate shocks, the NSER can affect coverage. As observed in other countries be leveraged to design programs to help cli- with desk-based data collection methods, the mate-vulnerable households prepare, cope coverage rates are significantly lower. However, with, and mitigate future shock damage. for social protection purposes, it is important not to lose coverage of poor and vulnerable house- holds. As NSER completes its update through desk-based methods, an assessment of change in coverage will help identify coverage gaps. • How data accuracy affects PMT mod- el validity for targeting. The PMT formula is based on 2013/14 survey data. Considering the multiple crises that have affected the coun- try, proxies for household welfare and their rel- ative weights may have changed. A better un- derstanding of how updating the PMT formula could improve targeting represents a crucial next step in improving targeting performance. • Understanding and identifying inconsis- tencies in definitions across NSER Admin- istrative Data and survey data. This analysis focuses only on a few indicators from the Admin- istrative data (primarily household size and urban/ rural classification), but there are more than 20 indicators that feed input into the PMT score. An in-depth, end-to-end review of the data collection and PMT calculation process will help further un- derstand the validity of PMT data and the PMT model. 46 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) References Ahmed, M., & Qayyum, A. (2023). Decentralisation’s Bashir, S., Khan, J., Danish, M., & Bashir, W. (2023). Effects on Health: Theory and Evidence from Baloch- Governance and development challenges in Baloch- istan, Pakistan. The Pakistan Development Review, istan: A comparative study with other provinces and 62(3), 331–355. way forward. International Journal of Contemporary Issues in Social Sciences, 2(4), 620–649. Aiken, E., Ohlenburg, T., & Blumenstock, J. (2023). Moving targets: When does a poverty prediction mod- Benstead, L. J. (2014). Effects of interviewer–respon- el need to be updated?. In Proceedings of the 6th dent gender interaction on attitudes toward women ACM SIGCAS/SIGCHI Conference on Computing and and politics: Findings from Morocco. International Sustainable Societies (pp. 117-117). Journal of Public Opinion Research, 26(3), 369-383. Ambler, K., Doss, C., Kieran, C., & Passarelli, S. Blaydes, L., & Gillum, R. M. (2013). Religiosity-of-In- (2021). He says, she says: Spousal disagreement in terviewer Effects: Assessing the Impact of Veiled survey measures of bargaining power. Economic De- Enumerators on Survey Response in Egypt. Pol- velopment and Cultural Change, 69(2), 765-788. itics and Religion, 6(3), 459–482. doi:10.1017/ S1755048312000557 Alatas, V., Banerjee, A., Hanna, R., Olken, B. A., & Tobias, J. (2012). Targeting the poor: Evidence from a Brown, C., Ravallion, M., & Van de Walle, D. (2018). field experiment in Indonesia. The American Econom- A poor means test? Econometric targeting in Africa. ic Review, 102(4):12061240. Journal of Development Economics, 134, 109-124. Baez, J. E., Inan, O. K. & Nebiler, M. (2021). Getting Fisher, M., Reimer, J. J., & Carr, E. R. (2010). Who Real: The Uneven Burden of Inflation across House- Should be Interviewed in Surveys of Household In- holds in Turkey. Policy Research Working Paper 9869. come? In World Development (Vol. 38, Issue 7, pp. World Bank Group. 966–973). Elsevier BV. Barca, V., Cord, L., Jones, N., & Wodon, Q. (2010). Flores-Macias, F., & Lawson, C. (2008). Effects of in- Evaluating the Impact of Service Delivery on the Lives terviewer gender on survey responses: Findings from of Beneficiaries: The Case of the Program of Targeted a household survey in Mexico. International journal of Social Assistance in Moldova. The World Bank. public opinion research, 20(1), 100-110. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 47 Grosh, M., Wai-Poi, M., & Tesliuc, E. (2022). Revisit- ly, 57(1), 1-28. ing Targeting in Social Assistance: A New Look at Old Dilemmas. Human Development Perspectives, Wash- Karachiwalla, N. (2019). A teacher unlike me: Social ington, DC: World Bank. distance, learning, and intergenerational mobility in developing countries. Economic Development and Grover, A., et al. (2022). Place, Productivity, and Pros- Cultural Change, 67 (2), 225-271 perity – Revisiting Spatially Targeted Policies for Re- gional Development. World Bank Group. Kilic, T., & Sohnesen, T. P. (2019). Same question but different answer: experimental evidence on question- Guven, M. U., Majoka, Z., & Jamy, G.M. (2024). The naire design’s impact on poverty measured by prox- Evolution of Benazir Income Support Programme’s ies. Review of Income and Wealth, 65(1), 144-165. Delivery Systems: Leveraging Digital Technology for Adaptive Social Protection in Pakistan (English). Kishwar, S., Meyer, M., & Nasir, M. (2024). Losing to Washington, D.C.: World Bank Group. inflation: Unmasking the unequal price effect of infla- tion across households. World Bank. Hossain, N. (2011). Extreme poor Adivasis and the problem of accessing social safety nets, Shiree Work- Leite, P., George, T., Sun, C., Jones, T., & Lindert, ing Paper No. 4. Shiree, Dhaka, Bangladesh. K. (2017). Social Registries for Social Assistance and Beyond. World Bank, Washington, DC. Huddy, L., Khatib, N., & Capelos, T. (1997). The role of interviewer gender in survey interviews: Does gen- Majoka, Z. & Armaghan, F. (2024). Beneficiary Jour- der of interviewer affect respondent’s answers? Public ney Mapping of the National Socio-Economic Regis- Opinion Quarterly, 61(2), 161-188. try Update. [Unpublished Manuscript]. Jacoby, H. G., & Mansuri, G. (2011). Crossing bound- Martinelli, C., & Parker, S. W. (2009). Deception and aries: gender, caste and schooling in rural Pakistan. Misreporting in a Social Program. Journal of the Euro- World Bank Policy Research Working Paper, (5710). pean Economic Association, 7(4), 886–908. Kakar, R. (2023). School Education in Post-18th Masselus, L., & Fiala, N. (2024). Whom to ask? Test- Amendment Balochistan: A Political Economy Per- ing respondent effects in household surveys. In Jour- spective. The Pakistan Development Review, 62, nal of Development Economics (Vol. 168, p. 103265). 467–492. Elsevier BV. Kane, E. W., & Macaulay, L. A. (1993). Interviewer Nikitin, D., Hou, X., & Mete, C. BISP 2009 Beneficiary gender and gender attitudes. Public Opinion Quarter- Survey Report. Online. 48 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Nsour, S., Anandan, V., Lindert, K., & Karippacheril, T. G. (2020). Outreach. In Sourcebook on the Foun- dations of Social Protection Delivery Systems (pp. 61–79). The World Bank. SDPI. (2022). Floods in Pakistan: Rethinking the hu- manitarian role. Online. Shabbar, S. (forthcoming). Help fee not so helpful: impact of Net Cash Transfers on labor supply of Paki- stani Households. Socialprotection.org. (2016, July 12). The Bolsa Fa- milia Program: Access to and dissemination of infor- mation [Webinar]. World Bank. (2019). Implementation Complementa- tion and Results Report - Second Bolsa Familia Proj- ect. Social Protection and Jobs Global Practice - Lat- in America and the Caribbean Region. World Bank. (2024a). Pakistan Development Update April 2024: Fiscal Impact of the Federal State Owned Enterprises. World Bank Publications. World Bank. (2024b). Macro Poverty Outlook Coun- try-by-country Analysis and Projections for the De- veloping World. Spring Meetings 2024. World Bank Publications. World Bank Publications. Available at: World Bank (2024c): Urbanization in Pakistan. World Bank Group, Washington DC. MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 49 Annex A - Sampling Methodology This study presents a positive evaluation of NSER as rivers and mountains. This ensures that the same completeness and accuracy and several important survey team can also interview households within the policy priorities emerge. EAs. Each EA was visually inspected for challenges for field teams. In some cases, manual adjustments 1. Creating a Sample Frame were made (for example, if an area included two sides of a river without a visible bridge to cross in between). As the 2023 census sample frame was not available This was based on visual inspection of the EAs by at the time of sample design, we used an alterna- sampling experts and the survey company, which has tive sample frame based on gridded population esti- permanent field workers throughout the country. Fig- mates.21 Gridded population estimates are estimates ure A1 illustrates the EA information used for review of the population on the ground for all locations based and implementation in the field. For each selected on modeling using satellite images combined with EA, all households within the EA were interviewed, census population data. The first step was calculating irrespective of the actual number of households. This the spatial boundaries for the primary sample frame meant that in some EAs, enumerators had to inter- at the Tehsil and District level, matching datasets of view more, and in others, less than 250 households. NSER/Census Tehsil names to Tehsil administrative Integrating this detailed mapping technology into names in shapefile maintained by the World Food Pro- survey and data collection methodologies rep- gramme (WFP) and the Office for the Coordination of Humanitarian Affairs (OCHA). In the second step, the FIGURE A1 team generated enumeration areas (EA) with an esti- An example of Enumeration Area (EA) used for sampling mated population of 1,000 people each (roughly 250 households) using a plugin for QGIS. The plug-in generates new EAs with a set target popu- lation (in this case, 1,000 people), taking into account existing boundaries such as administrative (tehsils , districts, and provinces) and physical boundaries such 21   For this work the distribution of population was taken from the Global Human Settlement Population layer (GHSP) from the Euro- pean Commission Joint Research Center (EC-JRC). The data has a resolution of ~100x100m and is estimated for the year 2025 50 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) resents a pivotal advance in the efficacy and preci- sampling frame stratified by Pakistan’s four provinces sion of enumerator operations. By creating com- (Punjab, Sindh, KP, and Balochistan) and the Islam- prehensive geo maps available in both physical abad Capital Territory (ICT). In the first stage, prima- and digital formats within the Computer-Assisted ry sampling units (PSUs) were chosen, focusing on Personal Interviews (CAPI) tools survey teams uti- tehsils within each province. Systematic selection lize, survey administrators can strategically mitigate with probability proportional to size (PPS) determined enumerator bias and optimize resource allocation. the sample tehsils within each province, where the Survey administrators can adeptly allocate resourc- size metric was based on household counts from the es and streamline enumerator routes to ensure ex- 2017 Pakistan Census. Subsequently, the sampling haustive EA coverage. Enumerators equipped with frame for the second stage was constructed using a these sophisticated maps can navigate designated population grid database specific to the sample te- areas more efficiently. Moreover, precisely delineating hsils. Secondary sampling units (SSUs) clusters were survey areas and interviewing all households within a identified within individual grid squares of each tehsil. pre-specified area amplifies operational efficiency and Systematic selection with PPS was again employed reduces sampling bias. to choose sample clusters within each tehsil, with size determined by estimated population figures from the 2. Sample Selection population grid database. In each selected cluster, a list of structures was created, with all listed residen- The study employed a stratified, two-stage sample tial structures (households) earmarked for interview. design to select households for analysis, with the TABLE A1 NSER Quality Review Survey Sample Number of Total households, Total population, Number of Number of Province interviewed sample 2017 Census 2017 Census sample tehsils sample clusters households 14,894,402 2,318,519 6 21 17,478 Balochistan 2,363,863 411,518 2 3 8,044 ICT 40,856,097 5,883,007 16 45 30,564 KP 127,688,922 19,855,902 49 175 2,911 Punjab 55,696,147 9,871,620 28 95 17,500 Sindh 241,499,431 38,340,566 101 339 Total MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 51 3. Survey Replacements and Refusals variables are available from satellite and GIS data and, therefore, available for all locations. Both vari- During survey implementation, several tehsils and nu- ables capture aspects of the localities that proxy for merous EAs were dropped from the sample. Some the degree of urbanization. were dropped due to security concerns; others were dropped because local authorities would not allow Across EAs, 19 percent of identified inhabited dwell- survey teams to implement the survey. To check for ings refused to participate in the survey. Pakistan sample selection bias of replaced areas, we com- has a history of high refusal rates. It is important to pared characteristics of the locations dropped with note that such refusals do not lead to survey bias if those that replaced them. Table A2 shows the statis- those refusing are similar to their immediate neigh- tics for both the tehsils and EAs that were dropped bors, as the sample weight corrects for this (through and replaced. The table shows no systematic differ- the Mhij and mhij terms in the formula below). ences between dropped and replaced locations for the distance to the nearest main road or the popu- Household surveys in all countries have difficulty lation density. This indicates that replacements have capturing the wealthiest households. Based on PMT not introduced bias into the sample. The selected scores, this survey has a higher refusal rate in EAs with higher consumption, though the differences TABLE A2 are not large (Table A3). Table A4 provides reasons Replacement of tehsils and SSUs households indicated for refusing interviews. Average Average TABLE A3 Number distance population of tehsil/ Refusal rates by PMT quintiles to nearest density SSUs main road (population in tehsils (meters) per ha) PMT consumption quintiles Mean EA refusal rate 5 2,837 56 Dropped tehsils 1 14% 5 2,860 53 New tehsils 2 15% 40 2,816 256 Dropped SSUs 3 16% 35 2,036 249 New SSUs 4 17% Notes. There are more dropped SSUs than replacements as some SSUs were replaced more than ones. Ei a SSU was dropped, and the replacement was also dropped. The difference in average distance to road and population density between dropped and replaced SSUs and Tehsils are not significant. 5 20% Distance to nearest main road was measured in meters from the boundary of the SSU, and main roads were defined as motorways, primary roads, or trunk roads in OpenStreetMap The table is based on household level data. 52 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) TABLE A4 Mh = total number of households in stratum Reasons given for not being interviewed (province) h, from the 2017 Pakistan Census No. of % of nhi = number of sample clusters (SSUs) selected Reason households responses in the i-th sample PSU of stratum h Declined 6,984 56.93 Phij = ea_pop_ghs = estimated population in the j-th No one at home 3,275 26.7 sample cluster (SSU) in the i-th sample PSU of Language barrier 53 0.43 stratum h, based on the population grid database Household head not at 963 7.85 home ∑ Phij = cumulated measure of size, or sum of the Respondent is busy and 238 1.94 cannot be interviewed population estimates P hij across all clusters in the grid sampling frame for the i-th sample PSU of Other 443 3.61 stratum h WEIGHTING METHODOLOGY Mhij = total number of dwellings (occupied house- holds) listed in the j-th sample cluster in the i-th Based on the sample design, overall sampling sample PSU of stratum h probabilities by sample cluster can be expressed as follows: m hij = number of households with completed inter- views in the j-th sample cluster in the i-th sample PSU of stratum h The basic household weight is simply the inverse where: of this probability, expressed as follows: p hij = overall probability of selection for the house- holds interviewed in the j-th sample cluster (SSU) in the i-th sample PSU ( tehsil or unit) in stratum where: (province) h Whij = overall basic weight for the sample households interviewed in the j-th sample cluster n h = number of sample PSUs (tehsils) selected in (SSU) in the i-th sample PSU (tehsil or unit) in province h stratum (province) h Mhi = total number of households in the i-th sample Although all households listed in each sample cluster PSU of stratum h, from the 2017 Pakistan Census were eligible to be interviewed, some of the house- MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 53 holds refused or did not have respondents available. The weight adjustment factor based on the 2023 Therefore, the last component of this weight rep- Pakistan Census population by province can be ex- resents an adjustment factor for nonresponse. pressed as follows: a. Adjustment of Weights Based on 2023 Pakistan Census Population Counts by where: Province A h = adjustment factor for the weights of the After calculating the basic design weights for the sam- sample households in province h ple households in the final survey dataset—based on the probabilities of selection and the nonresponse ad- PCh2023h = total population for province h from the justment factor described—the weighted total popu- 2023 Pakistan Census lation by province was tabulated from the survey data and compared to corresponding 2023 Census totals. Whij = basic design weight for the sample Given the significant differences between weighted households in the j-th sample cluster in the i-th estimates of the total population for some provinces sample PSU (tehsil) in province h, as defined above and the corresponding Census figures, it was decided to adjust weighting based on the population counts Phijk = number of persons in the k-th sample from the 2023 Pakistan Census for each province. household in the j-th sample cluster of the i-th sample PSU in province h TABLE A5 Distribution of the weighted total survey population by province Province Weighted total population 2023 Census population Weight adjustment factor Balochistan 10,266,138 14,894,402 1.450828 ICT 3,124,986 2,363,863 0.75644 KP 34,660,046 40,856,097 1.178766 Punjab 144,769,364 127,688,922 0.882016 Sindh 74,792,908 55,696,147 0.744672 Total 267,613,442 241,499,431 54 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) The denominator of the adjustment factor Ah is the tum within each province. weighted total population in province h from the survey data using the basic design weights (adjusted The SSUs were defined based on the population grid for nonresponse). database, and it is not possible to link them directly to the EAs in the 2023 Census data to determine the Applying the adjustment factors to the weights within corresponding urban or rural classification. Howev- each province rendered the final weighted survey es- er, two sources of information are available for SSU timates of the total population by province consistent urban and rural classification. The first source is the with the corresponding 2023 Census figures. Global Human Settlement Layer (GHSL) estimate of the proportion share of population in each SSU by b. Urban and Rural Classification of Sample urban, suburban, and rural categories, based on the SSUs GIS information in the population grid database. The other source is based on survey enumerators’ obser- Given the differential characteristics of the Pakistan vations regarding the urban or rural status of the sam- population in urban and rural areas, it would be ideal ple households, for which there is an average of 20 to post-stratify the sample of each province by urban percent missing information. and rural areas. This would make it possible to fur- ther calibrate sample weights for the sample SSUs The more reliable GHSL share information was used by urban and rural areas in each province, improving to classify each SSU as urban or rural based on the survey estimation accuracy by reflecting the relative predominant share. We compared alternative classi- proportion of the population by urban and rural stra- fications treating the suburban share as either rural TABLE A6 Distribution of NSER sample SSUs by province, urban and rural stratum, based on GHSL pre-dominant share, combining suburban with rural Classification of sample SSUs % urban population, Province 2023 Census Urban Rural Total % urban SSUs Balochistan 14 7 21 66.70% 31.0% ICT 3 0 3 100.00% 46.9% KP 31 14 45 68.90% 15.0% Punjab 91 84 175 52.00% 40.7% Sindh 50 45 95 52.60% 53.7% Total 189 150 339 55.80% 38.8% MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 55 TABLE A7 Distribution of NSER sample SSUs by province, urban and rural stratum, based on GHSL pre-dominant share, combining suburban with urban Classification of sample SSUs % urban population, Province 2023 Census Urban Rural Total % urban SSUs Balochistan 17 4 21 81.0% 31.0% ICT 3 0 3 100.0% 46.9% KP 42 3 45 93.3% 15.0% Punjab 165 10 175 94.3% 40.7% Sindh 89 6 95 93.7% 53.7% Total 316 23 339 93.2% 38.8% or urban to examine corresponding distribution of the used to adjust the weights by urban and rural stratum sample SSUs within each province. We then com- within each province. pared the alternative distributions to the distribution of the population by urban and rural stratum in the c. Final Adjustment of NSER Sample Weights 2023 Pakistan Census results. Table A6 presents the distribution of sample SSUs by urban and rural strata The NSER sample weights within each province were within each province based on the classification com- already adjusted based on the 2023 Census popula- bining suburban with rural. Table A7 shows the cor- tion at the province level. A similar calibration proce- responding distribution of sample SSUs combining dure was used to further adjust the weights for the suburban with the urban stratum within each prov- urban and rural strata of each province based on the ince. In the case of ICT, the three sample SSUs are corresponding 2023 Census population. The final classified as urban under both classification methods. weight adjustment factor based on the 2023 Pakistan Table A7 shows that combining suburban with the ur- Census population by province urban and rural strata ban stratum results in 93.4 percent of sample SSUs can be expressed as follows: classified as urban, compared to an urban population of 38.8 percent in the 2023 Pakistan Census. Com- bining the suburban category with the rural stratum in where: Table A6 reduces the urban stratum to 55.8 percent of the sample SSUs, still higher than the Census ur- • A ph = adjustment factor for the weights of the ban share, but closer to the Census population dis- sample households in stratum h (urban, rural) tribution for most provinces. Therefore, the classifica- of province p tion of urban and rural SSUs shown in Table A6 was • P Ch2023ph = total population for stratum h of 56 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) province p from the 2023 Pakistan Census at the province level. Since ICT only has three sample urban SSUs, the weight is adjusted at the province • W’phij = previous adjusted weight for the sample level. Given that the weights were already adjusted at households in the j-th sample cluster in the i-th the province level, the final weight adjustment factor sample PSU (tehsil) in stratum h of province p for ICT is 1. • Pphijk = number of persons in the k-th sample Applying the adjustment factors to the weights within household in the j-th sample cluster of the i-th each province, urban and rural stratum, brought the sample PSU in stratum h of province p final weighted survey estimates of total population by province, urban and rural stratum in line to be con- The denominator of the adjustment factor Aph is the sistent with the corresponding 2023 Census figures. weighted total population in stratum h (urban, rural) of province p from the survey data using the previously d. Adjusted Weighted Estimates of Total adjusted weights. Number of Households and Average House- hold Size The final adjusted weights were calculated by multi- plying the weight adjustment factor for each stratum Despite this adjustment based on the 2023 Pakistan within each province by the previous weights adjusted Census counts of the total population by province, TABLE A8 Distribution of the weighted total survey population from the NSER Survey data and the 2023 Pakistan Census total popu- lation, by province, urban and rural strata 2023 Pakistan Weight adjustment Province and Stratum Weighted population Census population factor Balochistan Rural 3,331,530 10,282,574 3.086 Balochistan Urban 11,562,873 4,611,828 0.399 ICT 2,363,863 2,363,863 1.000 KP Rural 15,563,451 34,724,801 2.231 KP Urban 25,292,646 6,131,296 0.242 Punjab Rural 70,373,617 75,715,270 1.076 Punjab Urban 57,315,304 51,973,652 0.907 Sindh Rural 18,518,169 25,771,071 1.392 Sindh Urban 37,177,978 29,925,076 0.805 Total 241,499,431 241,499,431 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 57 urban and rural stratum, the weighted total number pends on interpretation of the enumerators. of households based on these weights was not con- sistent with the corresponding 2023 Census results. Table A8 shows that the weighted total population This is because the average number of persons per is consistent with the corresponding total popula- household from the survey data differs from the cor- tion in the 2023 Census by stratum. Table A9 shows responding 2023 Census average household size. It the weighted total population and the weighted to- is possible to adjust NSER Survey weights based on tal number of households by province, urban and the total number of households by province, urban rural stratum, using the final adjusted weights for and rural stratum from the 2023 Census, but it is the NSER Survey. Additionally, it shows the NSER more effective to adjust the weights based on total weighted average number of persons per household population, given that the Census focus on counting by stratum. persons while the number of households partly de- TABLE A9 Weighted final NSER Survey total population and weighted number of households by province, rural and urban stratum Final NSER weighted Final NSER weighted Average Province and Stratum population households household size Balochistan Rural 10,282,574 1,687,862 6.09 Balochistan Urban 4,611,828 761,138 6.06 ICT 2,363,863 493,771 4.79 KP Rural 34,724,800 7,313,090 4.75 KP Urban 6,131,296 1,293,437 4.74 Punjab Rural 75,715,271 15,214,592 4.98 Punjab Urban 51,973,652 10,496,558 4.95 Sindh Rural 25,771,072 5,058,882 5.09 Sindh Urban 29,925,076 6,109,047 4.9 Total 241,499,432 48,428,377 4.99 58 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Annex B - Supplemental Figures & Tables TABLE B1 PMT indicators across matching categories Matched Invalid CNIC Not matched CNIC not shared Rural 63% 76% 68% 64% Urban 37% 24% 32% 36% Household size 5.05 4.8 4.67 4.68 Number of adults in household (over 65) 0.17 0.2 0.16 0.12 Number of males ages 35 -65 0.63 0.68 0.61 0.59 Dependency ratio between 0.25 and 0.6 42.80% 40.70% 43.40% 44.00% Dependency ratio above 0.6 31.30% 27.50% 25.20% 22.80% Average number of adults (15-64) in a household 0.02 0.03 0.03 0.03 employed in the public sector Average number of adults (15-64) in a household that 0.47 0.59 0.57 0.59 can read with understanding in any language Owns an air cooler 5.60% 9.00% 9.40% 14.20% Owns a geyser or air conditioner 3.30% 7.50% 9.00% 11.20% Owns a telephone 2.90% 6.20% 5.90% 7.30% Owns cooking stove or cooking range 34.10% 46.50% 44.80% 49.40% Owns a heater 3.70% 5.20% 6.40% 15.20% Owns a freezer 4.10% 7.40% 8.70% 9.30% Owns a TV or VCR 27.50% 45.20% 37.90% 40.50% Owns a washing machine 30.20% 44.30% 39.50% 45.40% Owns a fan 80.10% 84.60% 82.10% 76.50% Owns a car 1.00% 3.40% 3.50% 5.60% Owns a motorcycle 28.10% 44.80% 39.90% 42.10% Owns cattle 6.30% 7.50% 6.00% 3.70% Owns buffalos 5.70% 7.10% 7.20% 4.80% Agriculture land owned (0-5 Acres) 3.70% 3.60% 3.60% 2.90% Source: Authors’ calculations based on NSER Quality Review Survey MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 59 TABLE B2 Marginal effects based on probit regression analysis Dependent Variable: Not Matched in NSER 1 2 3 4 5 6 7 8 National National National National Modality Modality Balochistan KP Punjab Sindh Door Desk HH Head is Male Ref. Ref. Ref. Ref. Ref. Ref. Ref. Ref. -0.005 -0.004 -0.005 -0.005 -0.066 -0.013 -0.012 0.031* HH Head is Female [0.010] [0.005] [0.005] [0.019] [0.040] [0.018] [0.009] [0.018] 0.054* 0.047 0.115*** - - 0.069 0.094* -0.008 HH Head is Transgender [0.031] [0.032] [0.040] - - [0.074] [0.052] [0.089] 0 0.004 0.004 0.007 -0.019 -0.008 0.003 0.018 Employment of HH [0.004] [0.006] [0.006] [0.006] [0.025] [0.013] [0.007] [0.015] 0.004 0.002 0 0.001 0.014 -0.033** 0.006 0.013 HH Head is Literate [0.006] [0.008] [0.011] [0.014] [0.024] [0.015] [0.006] [0.011] 0.005*** 0 0.004** 0.006* -0.003 -0.003 0.005*** 0.009*** Household Size - New [0.001] [0.002] [0.002] [0.004] [0.005] [0.004] [0.002] [0.003] Urban -0.033** -0.070*** -0.022 -0.086*** -0.168*** -0.063*** -0.045*** -0.002 [0.013] [0.007] [0.042] [0.010] [0.029] [0.014] [0.009] [0.015] -0.002 -0.007 0.011 -0.054*** - 0.048 -0.004 -0.038 Low Occupational Caste [0.004] [0.005] [0.008] [0.011] - [0.034] [0.010] [0.069] 0.01 0.001 -0.003 -0.005 - - -0.007 0.033 Low Marginalized Caste [0.018] [0.002] [0.006] [0.005] - - [0.024] [0.105] 60 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) Balochistan 0.140*** - 0.225*** 0.108*** - - - - [0.010] - [0.014] [0.009] - - - - ICT 0.088*** - 0.121*** 0.018 - - - - [0.008] - [0.019] [0.016] - - - - KP -0.015** - -0.035** -0.033*** - - - - [0.007] - [0.014] [0.006] - - - - Sindh Ref. Ref. Ref. Ref. - - - - Punjab 0.002 - 0.038*** -0.041*** - - - - [0.008] - [0.010] [0.002] - - - - Desk Data - - - - - - - - - - - - - - - - Dynamic -0.065*** -0.037*** - - -0.071*** -0.080*** -0.072*** 0.016*** Registry Data [0.010] [0.006] - - [0.015] [0.023] [0.007] [0.005] Door-to-Door 0.103*** 0.145*** - - 0.044*** 0.070** 0.138*** 0.159*** [0.030] [0.029] - - [0.014] [0.028] [0.008] [0.024] Teacher Model 0.060* 0.068*** - - - - 0.115*** 0.109*** Data [0.034] [0.015] - - - - [0.016] [0.009] 0 0 0 -0.003** -0.000** -0.002 -0.001*** 0.000** Distance in kms to NSER [0.000] [0.000] [0.000] [0.001] [0.000] [0.001] [0.000] [0.000] PMT Decile: 1 -0.069*** -0.036** -0.092*** -0.043* -0.130*** 0.04 -0.067*** -0.036 (Poorest) [0.021] [0.016] [0.030] [0.025] [0.026] [0.053] [0.015] [0.028] PMT Decile: 2 -0.068*** -0.045*** -0.081*** -0.050*** -0.108*** -0.014 -0.060*** -0.053* [0.016] [0.011] [0.024] [0.013] [0.022] [0.052] [0.013] [0.031] PMT Decile: 3 -0.042** -0.030** -0.046** -0.030** -0.067*** -0.002 -0.024* -0.057** [0.017] [0.013] [0.022] [0.012] [0.023] [0.054] [0.013] [0.026] MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 61 PMT Decile: 4 -0.011 -0.004 -0.017 -0.004 0.018 0.002 0.003 -0.059*** [0.015] [0.012] [0.021] [0.009] [0.055] [0.023] [0.013] [0.022] PMT Decile: 6 0.021*** 0.013 0.033*** -0.002 -0.026 0.016 0.035*** 0.011 [0.008] [0.009] [0.011] [0.008] [0.057] [0.022] [0.012] [0.022] PMT Decile: 7 0.038*** 0.027*** 0.048*** 0.018 -0.014 0.016 0.044*** 0.051** [0.006] [0.004] [0.005] [0.019] [0.059] [0.022] [0.012] [0.024] PMT Decile: 8 0.081*** 0.058*** 0.101*** 0.040* 0.021 0.079*** 0.086*** 0.068*** [0.005] [0.009] [0.019] [0.023] [0.052] [0.023] [0.012] [0.026] PMT Decile: 9 0.081*** 0.050*** 0.090*** 0.058*** 0.045 0.056** 0.096*** 0.060** [0.007] [0.015] [0.020] [0.021] [0.061] [0.022] [0.012] [0.028] PMT Decile: 10 0.148*** 0.107*** 0.157*** 0.118*** 0.186*** 0.094*** 0.156*** 0.142*** (Richest) [0.004] [0.010] [0.020] [0.023] [0.053] [0.024] [0.012] [0.027] District Fixed No Yes No No No No No No Effects N 50241 50241 32146 18175 2548 6332 26010 14884 Predicted 0.154 0.154 0.201 0.081 0.239 0.091 0.172 0.135 Probability Pseudo R2 0.102 0.146 0.054 0.219 0.094 0.132 0.107 0.1 * p<.10, ** p<.05, *** p<.01 Note: Marginalized castes are identified through their historical socio-occupational roles following  Karachiwalla (2019),  Jacoby and Mansuri (2011). Muslim Sheikhs are coded as a distinct group due to their historically marginalized status even among low caste groups. 62 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) TABLE B3 Comparison of household characteristics of matched and unmatched households with PMT below 32 Matched Unmatched Female-headed 13.5% 14.3% Household head employed 65.8% 64.8% Household head illiterate 66.1% 65.8% Average household size 6.5 6.8 Average number of rooms 1.42 1.57 Persons per room 5.27 5.08 Average distance from NSER center (km) 16.5% 15.3% Moved from another district 63.3% 62.8% At least one member with a disability 23.1 20.4 Households with child(ren) <5 2.9% 4.0% Owns a cellphone 95.0% 90.2% Owns a telephone 0.7% 1.1% Household has internet access 6.7% 5.6% Household with no electricity 19.2% 19.3% WATER SOURCE Hand pump 57.9% 42.8% Open or closed well 12.6% 17.2% No toilet with flush 90.4% 82.9% HOUSE STRUCTURE Wall material Pakka 44.0% 47.1% Kaccha 55.2% 52.5% ROOF MATERIAL Concrete or cement 9.9% 16.5% Wood or bamboo 50.2% 45.6% Girder/T-Iron 39.7% 37.7% Source: Authors’ calculations based on NSER Quality Review Survey MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 63 TABLE B4 CNIC unavailability by quintile Quintiles   1 2 3 4 5 CNIC Unavailable 10.7% 13.0% 16.9% 21.8% 37.6% TABLE B5 Household size components of PMT and their weights HH Size Component of PMT Score Weight Direct Channels Household size -0.1166 Household size squared 0.0030 Number of household members per room -0.0332 Number of household members per room in urban areas -0.0272 Indirect Channels Number of elderly adults in household (65+) -0.0159 Number of males ages 35-65 0.0225 Dependency ratio between 0.25 and 0.6 inclusive -0.0253 Dependency ratio above 0.6 -0.0348 Average number of adults (15-64 yrs) in a household employed 0.0530 in the public sector Average number of adults (15-64 yrs) in a household that can read 0.1727 with understanding in any language Source: Authors’ calculations based on NSER Quality Review Survey 64 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) TABLE B6 Comparison of household characteristics of matched and unmatched households with PMT below 32 Dependent variable: Not matched in the NSER 1 2 3 4 Modality: Modality: National National Door-to-Door Desk-based District FE -0.086 -0.073 -0.066 -0.025 Exposed to floods in last monsoon [0.023] [0.026] [0.017] [0.018] District Fixed E No No No Yes N 50241 32146 18175 50241 Predicted probability 0.154 0.201 0.081 0.153 Pseudo R2 0.109 0.057 0.24 0.147 Source: Authors’ calculations based on NSER Quality Review Survey Note: The table reports marginal effects from probit regressions on self-reported flood exposure. Each regression specification includes controls for household demographics, province, distance to NSER center and PMT deciles. Column (1) reports the marginal effects for the full sample, with additional controls for modality of registration. Columns (2) and (3) restrict the sample to the Door-to-Door (including Teacher Module) and Desk registration assumed NSER registration modalities. Column (4) reports marginal effects for the full sample, including district fixed effects. Standard errors are clustered at the province-level in paren- theses: * p<.10, ** p<.05, *** p<.01 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER) 65 66 MIND THE GAP: ASSESSING PAKISTAN’S NATIONAL SOCIO-ECONOMIC REGISTRY (NSER)