HOTSPOT ANALYSIS ON POVERTY, UNEMPLOYMENT, SECURITY IN JAVA, INDONESIA DIAN KUSUMANINGRUM

Transcription

1 HOTSPOT ANALYSIS ON POVERTY, UNEMPLOYMENT, AND FOOD SECURITY IN JAVA, INDONESIA DIAN KUSUMANINGRUM SEKOLAH PASCASARJANA INSTITUT PERTANIAN BOGOR BOGOR 20100

2 DECLARATION OF THESIS AND SOURCES OF INFORMATION I hereby declare that the thesis Hotspot Analysis on Poverty, Unemployment, and Food Security in Java, Indonesia is my own work with guidance of my supervisors and has not been proposed to any other Universities in any form. Sources of information derived or quoted from a published or not published work conducted by other authors are mentioned in the text and included in the Reference at the end of this thesis. Bogor, February 2010 Dian Kusumaningrum NRP G ii

3 ABSTRACT DIAN KUSUMANINGRUM. Hotspot Analysis on Poverty, Unemployment, and Food Security in Java, Indonesia. Supervised by ASEP SAEFUDDIN and MUHAMMAD NUR AIDI Eradicating extreme poverty and developing strategies for decent and productive work for youth is a very important issue in Indonesia. Hence, it would be important to conduct a research to evaluate the circumstances of these issues in Indonesia. Geoinformatics techniques can be explored further to obtain premises for decision making and finding methods for better strategic efforts. Satscan is a geoinformatics tool widely used in hotspots detection. In this research the hotspots obtained by Satscan was compared with the hotspots obtained from ULS and the official food scarcity and poverty map developed by the Food Security Agency. The hotspot related to poverty, food scarcity and poverty were mainly in Central Java, East Java, and Yogyakarta. Afterwards, main factors causing the hotspot were analyzed by Ordinal Logistic Regression Model. Factors related to the hotspot were school facilities, village trade, village industry, village services, slum areas, proportion of families without electricity, and credit facilities. Keywords: Food scarcity, poverty, unemployment, hotspot detection, ordinal logistic regression iii

4 SUMMARY DIAN KUSUMANINGRUM. Hotspot Analysis on Poverty, Unemployment, and Food Security in Java, Indonesia. Supervised by ASEP SAEFUDDIN and MUHAMMAD NUR AIDI Eradicating extreme poverty and developing strategies for decent and productive work for youth is a very important issue in Indonesia. Hence, it would be important to conduct a research to evaluate the circumstances of these issues in Indonesia. Geoinformatics techniques can be explored further to obtain premises for decision making and finding methods for better strategic efforts. Satscan is a geoinformatics tool widely used in hotspots detection. In this research the hotspots obtained by Satscan was compared with the hotspots obtained from Upper Level Set (ULS) Scan Statistic and the official food scarcity and poverty map developed by the Food Security Agency. Another problem faced by Satscan and ULS is the stability of the hotspot clusters obtained. Changing the maximum cluster size will lead to different hotspots. The default maximum-size setting of 50% seldom produces usable, informative results because the reported primary cluster often occupies a large proportion of the study area. Therefore, it is difficult to determine an optimal setting for scaling parameters. Thus a process for addressing the sensitivity issues of scan statistics method and enhancing the interpretation of scan statistics result was conducted (Chen et. all 2008). First, scan statistics methods were run multiple times, starting from a small maximum-size (5%) and systematically increased to the 50% default value. Second, the results were visualized in a map matrix for side-by-side comparison of different maximum-sizes. Third, the reliability of a region in a map was calculated and interpreted. Fourth, core clusters would be discriminated from heterogeneous clusters through interpretation of the reliability. Fifth, the interpretation of core clusters has been confirmed by comparing the results to other independent techniques and

5 consultation with domain experts. In this study for poverty and food scarcity, the results of Satscan and ULS were compared with the Food Security Map and Poverty Map accomplished by CBS and FSA. Based on the comparison of ULS and Satscan on the poverty and food scarcity case in 78 districts in Java, ULS showed more accurate and stable results. The stability can be seen from the average stability score of clusters. For poverty and food scarcity ULS had an average stability score above Satscan. Meanwhile accuracy can be seen from the precision of ULS and Satscan in detecting high percentages of poor and food scarcity cases. These areas are known as first priority areas in Food Security Agency and CBS. The percentages of accuracy of ULS in detecting high cases of poverty and food scarcity are higher than Satscan. Therefore ULS was used to detect hotspots of poverty, unemployment, and food scarcity in all areas of Java. The hotspot related to poverty, food scarcity and poverty were mainly in Central Java, East Java, and Yogyakarta. Areas such as Cilacap, Demak, Kab.Madiun, Kota Pekalongan, Kulon Progo, Pemalang, and Purworejo were identified as critical areas because these areas were poverty, unemployment, and food scarcity hotspots. Based on the stability analysis Cilacap was the core cluster of poverty, unemployment, and food scarcity. This indicated by using several maximum cluster sizes Artinya Cilacap was identified as a poverty, unemployment, and food scarcity hotspot. Hence, Cilacap should be given prioritization. Meanwhile Kota Batu, Salatiga City, and Serang City were not detected as critical areas. Main factors that caused the hotspot were analyzed by Ordinal Logistic Regression Model. Factors related to the hotspot were school facilities, village trade, village industry, village services, slum areas, proportion of families without electricity, and credit facilities. The increase of school facilities, stimulating

6 economical potential of a village in industry and services, decreased the possibility of an area to become critical areas. The government should give more attention to credit facilities, economical potential of a village in trade, villages without electricity, and small scale farm industry. It turned out that the increase of these factors increased the possibility of a municipality to become a critical area. From this study it was pointed out that credit facilities, farm Industry and trade in a village did not show indication that it could improve the welfare of people living in critical areas. Hence, these factors should be revitalized. Areas that had a high ratio of families living without electricity were also critical points in solving the problem of poverty, unemployment, and food scarcity. Therefore the government should have given more attention to people who lived in these areas. Keywords: Food scarcity, poverty, unemployment, hotspot detection, ordinal logistic regression

7 RINGKASAN DIAN KUSUMANINGRUM. Hotspot Analysis on Poverty, Unemployment, and Food Security in Java, Indonesia. Dibimbing oleh ASEP SAEFUDDIN dan MUHAMMAD NUR AIDI Pemberantasan kemiskinan dan kerawanan pangan serta mengembangkan strategi untuk mengatasi permasalahan pengangguran pada usaha produktif untuk kaum muda adalah masalah yang sangat penting di Indonesia. Oleh karena itu, sangatlah penting untuk melakukan riset untuk mengevaluasi kondisi permasalahan tersebut di Indonesia. Teknik Geoinformatika dapat digunakan untuk memperoleh daerah kritis (hotspot) yang sangat penting dalam pengambilan keputusan dan menentukan upaya strategis yang lebih baik. Satscan adalah alat geoinformatika yang banyak digunakan untuk mendeteksi hotspot. Namun dalam beberapa penelitian, telah ditemukan kekurangan metode Satscan ketika mendeteksi hotspot di suatu area yang bentuknya tidak teratur. Oleh karena itu metode Satscan akan dibandingkan dengan Upper Level Set (ULS) Scan Statistics dan Peta Kerawanan Pangan dan Peta Kemiskinan yang dikembangkan oleh Dewan Ketahanan Pangan. Permasalahan lain dalam Satscan dan ULS adalah kestabilan dari gerombol hotspot yang diperoleh dari kedua metode tersebut. Perubahan nilai ukuran gerombol maksimum (maximum cluster size) menyebabkan perubahan gerombol hotspot yang dieroleh. Ukuran gerombol maksimum 50% (default) menghasilkan gerombol hotspot yang kurang informatif karena gerombol hotspot utama yang diperoleh sering kali menempati sebagian besar daerah studi. Sehingga seringkali sangatlah sulit untuk menentukan ukuran gerombol maksimum yang optimal. Sebuah proses untuk menangani isu-isu sensitivitas metode scan statistik dan meningkatkan interpretasi hasil scan statistik telah diusulkan (Chen et. all 2008). Pertama, metode scan statistik harus disimulasikan berulang kali, dimulai dari ukuran maksimum yang kecil dan meningkat menjadi iv

8 50% (default). Kedua, hasil metode scan statistik harus digambarkan dalam peta matriks untuk perbandingan hasil hotspot dari ukuran gerombol maksimum yang berbeda. Ketiga, menghitung dan menafsirkan nilai kestabilan suatu daerah di peta. Keempat, kelompok inti harus dibedakan dari kelompok heterogen melalui interpretasi dari nilai kestabilan. Kelima, penafsiran kelompok-kelompok inti harus dibandingkan dengan hasil yang diperoleh dari teknik lainnya dan konsultasi dengan para pakar. Berdasrkan perbandingan ULS dan Satscan pada kasus kemiskinan dan kerawanan pangan pada 78 kabupaten di Jawa, ULS cenderung menunjukkan hasil yang lebih akurat dan stabil. Kestabilan dapat dilihat dari hasil perhitungan nilai rata-rata stabilitas gerombol. Untuk kasus kemiskinan dan kerawanan pangan ULS memiliki nilai rata-rata stabilitas gerombol yang cenderung lebih tinggi dari Satscan. Sedangkan akurasi dapat dilihat dari persentase ketepatan ULS dan Satscan dalam mendeteksi daerah-daerah yang memiliki persentase kemiskinan atau kerawanan pangan yang tinggi. Daerah-daerah tersebut oleh Badan Ketahanan Pagan maupun BPS dinyatakan sebagai daerah prioritas pertama. Persenatse akurasi ULS dalam mendeteksi kemiskinan dan kerawanan pangan lebih tinggi dari Satscan. Oleh karena itu ULS digunakan untuk menentukan hotspot kemiskinan, kerawanan pangan, dan pengangguran untuk seluruh kota dan kabupaten di Jawa. Mayoritas hotspot kemiskinan, kerawanan pangan dan pengangguran berada di Jawa Tengah, Jawa Timur, dan Yogyakarta. Daerah Cilacap, Demak, Kab.Madiun, Kota Pekalongan, Kulon Progo, Pemalang, dan Purworejo merupakan daerah kritis krena terdeteksi sebagai daerah hotspot kemiskinan, kerawanan pangan dan pengangguran. Berdasarkan analisis kestabilan, Cilacap merupakan daerah inti kemiskinan, pengangguran dan kerawanan pangan. Artinya dengan menggunakan berbagai ukuran maksimum gerombol yang berbeda daerah Cilacap terdeksi sebagai hotspot kemiskinan, pengangguan dan v

9 kerawanan pangan oleh karean itu daerah Cilacap harus diutamakan. Sedangkan Kota Batu, Kota Salatiga, dan Kota Serang merupakan daerah yang tidak kritis. Factor-faktor yang menyebabkan hotspot kemiskinan, kerawanan pangan dan pengangguran telah dianalisis dengan menggunakan Model Regresi Logistik Ordinal. Faktor-faktor yang terkait dengan hotspot-hotspot tersebut adalah rasio fasilitas sekolah, rasio potensi desa perdagangan, rasio potensi desa industri, rasio potensi desa jasa, rasio fasilitas kredit, dan rasio industri kecil di bidang pertanian, dan proporsi keluarga yang tidak memiliki listrik. Peningkatan jumlah fasilitas sekolah dan peningkatan rasio potensi ekonomi dari suatu desa di bidang industri dan jasa dapat mengurangi kemungkinan suatu daerah untuk menjadi daerah kritis. Sedangkan peningkatan fasilitas kredit, potensi ekonomi dari sebuah desa di bidang perdagangan, desa-desa tanpa listrik PLN, dan industri pertanian meningkatkan kemungkinan daerah untuk menjadi daerah kritis. Fasilitas kredit, potensi ekonomi di bidang perdagangan dan industri pertanian di sebuah desa belum dapat meningkatkan kesejahteraan penduduk di daerah-daerah kritis secara signifikan. Oleh karena itu fasilitas kredit, perindustrian dan perdagangan pertanian harus direvitalisasi. Daerah yang memiliki rasio keluarga yang hidup tanpa listrik PLN tinggi juga merupakan daerah yang memerlukan perhatian khusus dalam upaya memecahkan masalah kemiskinan, pengangguran, dan kelangkaan pangan. Jika pemerintah ingin memecahkan masalah kemiskinan, pengangguran, dan kerawanan pangan maka factor-faktor tersebut perlu mendapatkan perhatian khusus. Kata Kunci: Kerawanan pangan, kemiskinan, pengangguran, pendeteksian hotspot, regresi logistik ordinal vi

10 Copy Right BAU, 2010 The copy right is protected by the Indonesian law It is prohibited to cite parts or all parts of this thesis without including or mentioning the source. Citation is only for educational purposes, research, writing papers, preparation of reports, preparation of criticism or review of a problem; and citation doesn t harm reasonable interest of Bogor Agricultural University (BAU). It is prohibited to announce or reproduce parts or or all parts of this thesis in any form without permission of BAU.

11 HOTSPOT ANALYSIS ON POVERTY, UNEMPLOYMENT, AND FOOD SECURITY IN JAVA, INDONESIA DIAN KUSUMANINGRUM Thesis as a requirement for Master Degree on Statistics GRADUATE SCHOOL BOGOR AGRICULTURAL UNIVERSITY 2010

12 Title : Hotspot Analysis on Poverty, Unemployment, and a Food Security in Java, Indonesia Name NRP Study Program : Dian Kusumaningrum : G : Statistics Approved by, Supervising Commission, Dr. Asep Saefuddin, M.Sc Chairman Dr. Ir. Muhammad Nur Aidi, M.Si Member Acknowledged by, Head of Statistics Program Dr.Ir. Aji Hamim Wigena, M.Sc Examination Date: 12 Februari 2010 Graduation Date:

13 ACKNOWLEDGEMENT Alhamdulillahi rabbil alamin, praise and grateful to Allah SWT for all the blessing that made this thesis entitled Hotspot Analysis on Poverty, Unemployment, and Food Security in Java, Indonesia completed. This thesis could not have been completed without the support of many people. Therefore, in this opportunity, the author would like to acknowledge the following people: 1. Dr. Ir. Asep Saefuddin, M.Sc. and Dr. Ir. Muhammad Nur Aidi, MSi as my supervisors, for the valuable advice, theoretical guidance, support, and opinion during this research. 2. Dewan Ketahanan Pangan, Dr. Hari Wijayanto and Swastika Andi for their valuable contribution in providing data for this research. 3. Prof. GP Patil and associates in Pennsylvania State University for their valuable contribution in this study 4. Higher Education Directorate for funding my research through Hibah Penelitian Tim Pascasarjana 5. PHK-A2, Higher Education Directorate, for their valuable support in funding my studies through PHK-A2 scholorship 6. All lecturers at Department of Statistics for sharing their knowledge and support and also the staff of Department of Statistics for helping me in my study. 7. My beloved family and husband for all their patience, prayer, supports, love, and advices. 8. All my friends in Statistics 2007 class, especially my sisters Halimatus Sadiah and Triyani thank you for being there to understand the essence of struggle. And many others whom I could not mention one by one in this opportunity thank you for everything. Hopefully, this thesis will be useful for the reseacheres and others who need the information in this thesis. viii

14 ABOUT THE AUTHOR The author was born on June 4 th 1981 as the first child of two children of Hermanto and Karyani. The author has graduated from SD Polisi IV Bogor in 1993, afterwards graduated from SMP Negeri IV Bogor in 1996, and graduated from SMU Negeri IV Semarang in After having the opportunity to have a brief education in Diponegoro University Semarang, the author enrolled in Statistics Department Bogor Agricultural University in 2000 through a National Selection Test to Enter State Universities (UMPTN) and took social economics as her minor. In 2007 she had the opportunity to continue her studies in Statistics at Post Graduate School of Bogor Agricultural University and married to Tosan Wiar Ramdhani in the same year. The author started to be involved with social studies related research in 2005 when she had the opportunity to become a research assistant in UNESCAP-CAPSA and since 2006 the author has been working as a Staff Lecturer at Department of Statistics Bogor Agricultural University. ix

15 TABLE OF CONTENTS LIST OF TABLES...xi LIST OF FIGURES...xi LIST OF APPENDENCIES...xii I. INTRODUCTION 1.1. Background Objectives... 3 II. LITERATURE REVIEW 2.1. Poverty Unemployment Food Scarcity Hotspot Hotspot Detection Method Scan Statistics (Satscan) Upper Level Satcan (ULS) Bernoulli Models Monte Carlo-Based Hypothesis Testing Hotspot Evaluation Joint Hotspots Ordinal Logistic Regression Testing the Model Significance Assumption of Logistic Regression Model Validation III. METHODOLOGY 3.1. Source of Data Method IV. RESULTS AND DISCUSSION 4.1. Data Description Satscan and ULS Evaluation Poverty Food Scarcity Joint Poverty, Food Scarcity, and Unemployment Hotspots Determining Factors Causing Poverty, Food Insecurity, and Unemployment V. CONCLUSION AND RECOMMENDATIONS 5.1. Conclusion Recommendation Page x

16 REFERENCES LIST OF TABLES 1. Multicriteria Hotspot Category List of Explanatory Variables per District and Its Sector The Percentage of Poor in Scan Window A and B Performance of Satscan and ULS Food Insecurity Hotspots Joint Hotspots in Java Ordinal Logistic Regression Table LIST OF FIGURES 1. Scan statistics zonation for circle (left) and space-time cylinders (right) Connectivity of tessellated regions A confidence set of hotspots on the ULS tree A logic process for analytics of scan statistics result Geoinformatics Research on Poverty, Unemployment, and Food Security Road Map Main Source of Fix Income of Poor Households Share of Food Income Poverty s ULS and Satscan Performance with a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5% Circle Scanning Window and a Comparison of ULS and Satscan Hotspots Food Scarcity s ULS and Satscan Performance with a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5% Multicriteria Hotspots Distribution in Java xi

17 LIST OF APPENDENCIES 1. Municipalities with the Highest Poverty Rate in Java Municipalities with the Highest Food Insecurity Household Rate in Java Municipalities with High Unemployment Rate in Java Side-by-side Comparison Map of different Poverty Hotspots Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Satscan and ULS Poverty Hotspots Comparison Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Comparison of ULS and Satscan Poverty Hotspots with Critical Poverty Areas based on the Food Security Map (Food Security Agency) Side-by-side comparison Map of different Food Scarcity Hotspots Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Satscan and ULS Food Scarcity Hotspots Comparison Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Indicator used for Food Insecurity Atlas Poverty Map of Poverty, Food Insecurity, and Unemployment in Java (2005) Using ULS with a Maximum Cluster Size of 50% The Stability of Poverty, Food Insecurity, and Unemployment Hotspots in Java (2005) Using ULS with a Maximum Cluster Size of 50%, 40%, 30%, 20%, 10% and 5% Multicritria Poverty, Food Insecurity, and Unemployment in Java (2005) Using ULS with a Maximum Cluster Size of 50% Multicriteria Poverty, Food Security, and Unemployment Map (2005) Distribution of Multicriteria Hotspots Based on Province Variables used in the Ordinal Logistic Model Correlation of Independent Variables used in the Ordinal Logistic Model Ordinal Logistic Regression.. 69 xii

18 1 I. INTRODUCTION 1.1. Background Based on The United Nations (2000), The Millennium Development Goals (MDGs) comprise eight goals that emerged from the 2000 Millennium Summit of world leaders in New York. The MDGs provide a set of time-bound and measurable targets for combating poverty, hunger, illiteracy, disease, discrimination against women, and environmental degradation. Eradicating extreme poverty and hunger is the main goal of MDGs. Meanwhile, in cooperation with developing countries, developing and implementing strategies for decent and productive work for youth is incorporated in the 8 th goal to develop a global partnership for development. Hence, it would be important to conduct a research to evaluate the circumstances of these goals in Indonesia. The focus of this research would be Java Island where it represents 60% of Indonesia s total population. Nevertheless, Based on Central Bureau of Statistics (CBS) 2005, at least 50% of the poor are found in Java Island. The Semeru Research Institute (2002) pointed out that poverty is a significant and complex problem caused by the combination of cultural, social, political, and economic factors. Hence, poverty reduction strategies and programs require an integrated approach and must be implemented in stages that are both well-planned and sustainable. In Indonesia, poverty, unemployment, and food scarcity are the three main problems that the government faces. Strategic efforts to reduce poverty, unemployment, and food scarcity in many regions have been done, though the result was not quite satisfying. Observing poverty, unemployment, and food scarcity must be done comprehensively and holistically. Poverty, unemployment, and food scarcity data are gathered periodically at various levels and regions to observe the effectiveness of poverty, unemployment, and food scarcity reduction programs. Maps of poverty and unemployment in Indonesia are also available and have been studied from various aspects. There are many government institutions and research institutes that conduct studies on unemployment, poverty, and food scarcity mapping, however studies that emphasize on the spatial and geoinformatic techniques are uncommon.

19 2 Nevertheless, these techniques can be explored further to obtain premises for decision making and finding methods for quicker and better strategic effort. Using geographical maps of poverty, unemployment, and food scarcity based on the location (spatial) attribute provides further information. Poverty, unemployment, and food scarcity are the three variables that are spatially related within themselves as well as with other variables. Spatial relationship is considered an important aspect in solving problems related to poverty, unemployment, and food security in Indonesia. Information technology has enabled us to study the condition of poverty and unemployment and combine it with geographical information. Statistical techniques can be adopted to produce additional information by making use of the spatial concepts among variables. Therefore, practically, there is no technological resistance for us to obtain more and better knowledge of geographical information and also data which have already existed. Data is transformed into information, and then becomes knowledge. Hence, when the data has geographical information, it has developed into geoinformatics. The main things that are visible by the use of the geoinformatics techniques are: 1. Identify unemployment, poverty, and food scarcity hotspot by using scan statistics method. The focus is on hotspots where a cluster was characterized by unemployment, poverty, and food scarcity and which are considered as an outbreak compared to other regions. Regions with high proportions of unemployment, poverty, and food scarcity were not hotspots if the surrounding areas of these regions also had high proportions of unemployment, poverty, and food scarcity. The output of statistical test is applicable to determine whether a cluster region is a hotspot. Information about the existence of hotspots will be vital in the effort of determining the regional priority of unemployment, poverty, and food scarcity reduction. 2. Identify joint hotspots. After knowing the poverty, unemployment, and food scarcity hotspot, this research took a step forward to analyze the relationship between poverty, unemployment, and food scarcity

20 3 hotspots. One of the questions raised was If an area was a poverty hotspot, would it also be a unemployment and food scarcity hotspot? Or was there any relation between poverty, unemployment, and food scarcity hotspots. 3. Identify variables that can spatially be an influence on the unemployment, poverty, and food scarcity hotspots. The ordinal logistic regression model was used to obtain variables that had significant effect on the hotspots of poverty, unemployment, and food scarcity. Information on these variables will be useful to determinate the type of poverty and unemployment reduction effort that is in line with the goals. 4. Identify the effectiveness of unemployment, food scarcity, and poverty reduction efforts. The efforts for Poverty, unemployment, and food scarcity reduction program and activity in selected regions will have an impact, especially for the nearby regions. But how far (in space context) has the effort had affected the society must be examined further on. Information on the level of extent can be used to determine the amount of locations where these programs can be implemented. This is an advantage because it reduces the risk of overlapping programs at a particular region and avoid certain regions from being neglected The most recent method used to identify a candidate hotspot is Satscan. However, in several researches it was found that Satscan had several limitations, such as the circles that have been used for the scanning window caused low power for detection of arbitrarily shaped cluster. Hence, Upper Level Set (ULS) Scan Statistics has been used as a comparison to detect hotspots Objectives Poverty, unemployment, and food scarcity have been a major problem in many developing countries, such as Indonesia. Finding the right methods to solve these problems would be promising. This research incorporated spatial attribute in

21 4 order to produce better outcomes. Through geoinformatics techniques the issues below were inquired: 1. Compared ULS and Satscan methods in detecting poverty and food scarcity hotspots in order to decide which method was more sufficient. 2. Identified the hotspots of unemployment, poverty, and food scarcity using the selected method based on the first objective. 3. Developed a method to identify joint hotspots 4. Identified variables that have a significant influence on the unemployment, poverty, and food scarcity joint hotspots

22 5 II. LITERATURE REVIEW 2.1. Poverty In general poverty is the state of being without the necessities of daily living, often associated with need, hardship and lack of resources across a wide range of circumstances. A person living in the condition of poverty is said to be poor or impoverished (The Free Dictionary). There are also several definitions on Poverty made by the NGOs and GoI Institutions. According to the Central Bureau of Statistics (CBS) a household is categorized as poor if they have an income per capita below the poverty line. The poverty line is a measure of the amount of money a government or a society believes is necessary for a person to live at a minimum level of subsistence or standard of living. Meanwhile, The World Bank defines extreme poverty as a person living on less than US$ 1 per day, and moderate poverty as a person living less than US$ 2 a day. Pre-prosperous family (Keluarga Pra Sejahtera) is a family that is unable to fulfil a minimum of basic human needs including spiritual needs, food, clothes, home, education and health. Prosperous family 1 (Keluarga Sejahtera I) is a family that is already able to fulfil its basic human needs, but unable to fulfil higher human needs (BKKBN 2004) Unemployment Unemployment occur when a person is available to work and seeking work but currently without work. The prevalence of unemployment is usually measured using the proportion of unemployed people, which is defined as the percentage of those in the labour force who are unemployed. The proportion of unemployed people is also used in economic studies and economic indices (Wikipedia 2009). In seven years between 1994 and 2001, CBS changed its definition of open unemployment twice. First, in 1994, CBS removed the qualifying time period of actively seeking work. Prior to 1994, a person was considered to be actively looking for work if s/he had actually looked for work during the week preceding the survey. Starting from 1994, a person is considered to be actively looking for work if s/he had looked for work, regardless of when the last time s/he actually actively looked for work, as long as s/he is still waiting for the result of the job

23 6 search. In 2001, CBS altered the definition of unemployment to include three more groups of the unemployed on top of the traditionally measured-unemployed, which is defined as part of the labour force who are not working and actively looking for work. The three additional groups of unemployed people are: (i) those who are not working and are not actively looking for work because they do not believe work is available (discouraged workers), (ii) those who already have jobs but have not started working, and (iii) those who are preparing a business. Prior to 2001, these groups of people would be considered out of the labour force, hence not included in the calculation of proportion of unemployed people. The first group makes up the majority of the additional persons considered unemployed under this new definition of open unemployment (Semeru 2005) Food Scarcity Hunger is widely known as an extreme manifestation of a prolonged hunger or sudden occurrence of food insecurity. Food insecurity exists when people are undernourished as a result of unavailability of food, lack of social or economic access to adequate food, and inadequate food consumption and adsorption. Food insecurity can be a short term or long term phenomena. Macro level food security, in terms of self-sufficiency at a national or regional level does not guarantee a household or individual level of food security. Hence food security is a multidimensional issue and needs further in depth studies on host parameters (DKP 2005) Ariani (2006) referred that World Bank since 1986 categorized food insecurity into chronic food insecurity and transitory/occasional food insecurity. Chronic food insecurity is a condition of frequent food scarcity over a certain period of time. On a household level, chronic food insecurity indicates that the share of food owned is slight lower than needed. Meanwhile on an individual level, chronic food insecurity is a situation where food consumption is lower than the biological needs. Transitory/occasional food insecurity is a cyclical food insecurity that occurs when there is a shock or an unexpected event. Transitory/occasional food insecurity is also categorized into two specific

24 7 categories, which are caused by repeated or cyclical factors and temporary or unpredicted factors Hotspot Hotspot is defined as something unusual, anomaly, aberration, outbreak, elevated cluster, or critical area (Patil and Taillie 2004). Meanwhile according to Harran e.t. all (2006) hotspots are locations or regions that have consistently high levels of occurrences (such as the total amount of poor, unemployed, or people that suffer from food scarcity) and may have characteristics unlike those of surrounding areas. Hotspot clusters were generated by setting the relative risk in some areas to be larger than one and (Song and Kulldorff 2003). Furthermore a poverty hotspot represents an area characterized by certain local characteristics which could also expand and affect other neighbouring areas (Betti et. all 2006) Hotspot Detection Method Hotspot detection method contains three components, which include a) identifying hotspot candidate, b) evaluating the statistical significant hotspot, and c) estimate the covariance related with hotspot. In Indonesia nowadays, the most recent method used to identify a candidate hotspot is spatial scan statistics. In Bungsu (2006) it is stated that spatial scan statistics suffers from several limitations, such as the circles that have been used for the scanning window caused low power for detection of arbitrarily shaped cluster. Hence Upper Level Satcan (ULS) will be used as a comparison to detect arbitrarily shaped hotspots. Likelihood ratio, relative risk, and hypothesis testing based on montecarlo simulation are techniques used to evaluate a candidate hotspot Scan Statistics (Satscan) Scan statistic is a statistical method used to detect clusters in a cluster process. Spatial scan statistic is used to determine whether a spatial cluster process contains a localized cluster of points somewhere in a region of interest. The spatial scan statistic deals with the following situation. A region R of euclidian

25 8 space is subdivided into cells defined (denote by A). Data are available in the form of a count on each cell A. In addition, A size value P A is associated with each cell. The cell sizes P A are assumed to be known and fixed, while the cell counts N A are independent random variables. The spatial scan statistic seeks to identify clusters of cells that have an elevated response compared with the rest of the region. Elevated response means large values for the rates, r A = N A / P A, instead of for the raw counts N A. Cell counts are thus adjusted for cell sizes before comparing cell responses. Kulldorf (1997) presented the following algorithm for a circular window of fixed diameter d on a homogeneous Poisson/Bernoulli (assuming homogeneous variance) process: 1. Pick a grid point. Calculate the distance to the different population points and sort those in increasing order. Memorize the sorted population points in an array 2. Repeat step 1 for each grid point 3. Pick a grid point 4. Create a circle cantered at the grid point and continuously increase the radius. For each population entering the circle, update the number of cases n and measure the population N W inside the circular area W 5. Repeat step 3 and 4 for each grid point. Report the largest likelihood based on all (n, N W ) pairs as the scan statistics, where the likelihood is calculated according to equation 6. Repeat steps 3 to 5 for each monte carlo replication The relative risk is a non-negative number, representing how much more common a case is in the location and time period compared to the baseline. Setting a value of one is equivalent of not doing any adjustments and a value of less than one to adjust for lower risk A value of greater than one is used to adjust for an increased risk. A cluster with a relative risk (RR) value greater than one is defined as a candidate of hotspot. A relative risk of zero is used to adjust for missing data for that particular time and location (Kulldorff 2006). The relative risk is calculated by (Kulldorff 2006) nz RR = where nz is the number of E(c) observed cases, and E (c) is the expected number of cases in a location which is

26 9 C calculated by E ( c) = p where p is the number of population in the cluster P of interest, while C and P are the total number of cases and total number of population. Available scan statistic software is known to have several limitations. First, circles have been used for the scanning window, resulting in low power for detection of irregularly shaped clusters (Figure 1). Second, the response variable has been defined on the cells of a tessellated geographic region, preventing application to responses defined on a network (stream network, water distribution system, highway system, etc.). Third, response distributions have been taken as discrete (specifically, binomial or Poisson). Finally, the traditional scan statistic gives only a cluster estimate for the hotspot but does not attempt to assess estimation uncertainty (Patil 2006) Time Cholera outbreak along a river flood-plain Small circles miss much of the outbreak Large circles include many unwanted cells Space Outbreak expanding in time Small cylinders miss much of the outbreak Large cylinders include many unwanted cells Figure 1 Scan statistic zonation for circles (left) and space-time cylinders (right) 2.7. Upper Level Set (ULS) Scan Statistics Patil (2006) acknowledged a new version of the spatial scan statistic designed for detection of hotspots of arbitrary shapes and for data defined either on a tessellation or a network. This version looks for hotspots from among all connected components of upper level sets of the response rate and is therefore called the upper level set (ULS) scan statistic. The method is adaptive with respect to hotspot shape since candidate hotspots have their shapes determined by the data rather than by some a priori prescription like circles or ellipses. This data dependence is taken into account in the Monte Carlo simulations used to

27 10 determine null distributions for hypothesis testing. The performance of the ULS scanning tool was compared with spatial scan statistic. The key element here is enumeration of a searchable list of candidate zones Z. Figure 2 Connectivity for tessellated regions. The collection of shaded cells on the left is connected and, therefore, constitutes a zone. The collection on the right is not connected. A zone is, first of all, a collection of vertices from the abstract graph. Secondly, those vertices should be connected (Fig. 2) because a geographically scattered collection of vertices would not be a reasonable candidate for a hotspot. Even with this connectedness limitation, the number of candidate zones is too large for a maximum likelihood search in all but the smallest of graphs. We reduce the list of zones to searchable size in the following way. The response rate at vertex a is G = Y / A. These rates determine a function a a a G a defined over the vertices in the graph. This function has only finitely many values (called levels) and each level g determines an upper level set defined by Ug a: G g. Upper level sets do not have to be connected but each upper level set can be decomposed into the disjoint union of connected components. The list of candidate zones Z for the ULS scan statistic consists of all connected components of all upper level sets. This list of candidate zones is denoted by Ω ULS. The zones in ΩULS are certainly plausible as potential hotspots since they are portions of upper level sets. Their number is small enough for practical maximum likelihood search, the size of a U g ΩULS does not exceed the number of vertices in the abstract graph (e.g., the number of cells in the tessellation). Finally, ΩULS becomes a tree under set inclusion, thus facilitating computer representation. This tree is called the ULS-tree (Figure 3); its nodes are the zones Z Ω ULS and are therefore collections of vertices from the abstract graph. Leaf

28 11 nodes are (typically) singleton vertices at which the response rate is a local maximum; the root node consists of all vertices in the abstract graph. Finding the connected components for an upper level set is essentially the issue of determining the transitive closure of the adjacency relation defined by the edges of the graph. Several generic algorithms are available in the computer science. Tessellated Region R MLE Junction Node Alternative Hotspot Delineation Alternative Hotspot Locus Figure 3 A confidence set of hotspots on the ULS tree. The different connected components correspond to different hotspot loci while the nodes within a connected component correspond to different delineations of that hotspot Bernoulli Models According to Kulldorff (1997), let X denote a spatial cluster process where X A is the random number of clusters in the set A G. As the window, moves over the study area it defines a collection Ω of zones Interchangeably, Z will be used to denote both a subset of G and a set of parameters defining the zones. For the Bernoulli model we consider only measures N such that N A is an integer for all subsets A G. Each unit of measures corresponds to an entity or individual that could be either one of two states. Individuals in one of these states are considered as clusters and the location of those individuals constitute the cluster process. In the model there is exactly one zone such that each individual within that zone has probability π 1 of being a cluster, while the probability for individuals outside the cluster is π 2. The probability for any one individual is independent from all the others. The null hypothesis is H 0 : π 1 = π 2

29 12 and the alternative hypothesis is H 1 : π 1 >π 2, Z Ω. Under H 0, X A ~ Bin(N A, π 1 ) A. Under H 1, X A ~ Bin(N A, π 1 ) A Z, and X A ~ Bin(N A, π 2 ) A Z c. Let n z denote the observed number of cases in zone Z and n G is the total number of cases. N Z is the total population in zone Z while N G is the total population. Hence the likelihood function for the Bernoulli model is expressed as, π, π π 1 π π 1 π To detect the zone that is most likely to be a cluster, we find the zone that maximizes the likelihood function. We do this in two steps. First we maximize the likelihood function conditioned on Z., π, π 1 1 when ; Next we find the solution : ` `. We are also interested in making statistical inference. Let π π,π,π statistic (λ) can be written as and the likelihood ratio test,,,, In order to find the value of the statistic test, we need a way to calculate the likelihood ratio as it is maximized over the collection of cluster in the alternative hypothesis. This might seem like a daunting task since the number of cluster could easily be infinite. Two properties allows us to reduce it to a finite problem. The number of observed clusters is always finite and for a fixed number of clusters the likelihood decreases as the measure of the moving window increases Monte Carlo-Based Hypothesis Testing Simulation is analytical method meant to imitate a real-life system, especially when other analyzes are too mathematically complex or too difficult to

30 13 reproduce (adithmc ). Monte Carlo simulation can be defined as a method to generate random sample data based on some known distribution for numerical experiments (Teknomo 2008). Once the value of the test statistic has been calculated, it easy to do the inference. We can t expect to find the distribution of the test statistic in closed analytical form. Instead we rely on Monte Carlo-Based hypothesis testing to test the hypothesis. With a Monte Carlo test, the significance of an observed test statistic calculated from a set data is assessed by comparing it with a distribution obtained by generating alternative sets of data from some assumed model. If the assumed model implies that all data orderings are equally likely then this amounts to a randomization distribution. In Kulldorff (1997) it was known that Monte Carlo-based hypothesis testing was proposed by Dwass (1957), who pointed out that probability of falsely rejecting the null hypothesis is exactly according to the significance level, in spite of the simulation involved. Mantel (1967) proposed its use in terms of spatial clusters processes, while Turnbull et. all (1990) was the first to use in the context of a multidimensional scan statistic. Monte Carlo hypothesis testing for a scan statistic is a four-step procedure: 1. Calculate the value of the test statistic for the real data. 2. Create a large number of random data sets generated under the null hypothesis. 3. Calculate the value of the test statistic for each of the random replications. 4. Sort the values of the test statistic from the real and random data sets, and note the rank of the one calculated from the real data sets. If it is ranked in the highest α percent, then reject the null hypothesis at α percent significance level. For example, when we condition on the total number of clusters n R. with 9999 such replications, the test is significant at the 5 percent level of α if the value of the test statistic for the real data sets is among the 500 highest values of the test statistic coming from the replications. The p-value is obtained through Monte Carlo hypothesis testing (Kulldorf 1997) by comparing the rank of the maximum likelihood from the real data sets

31 14 with the maximum likelihood from the random data sets. If this rank is R, then p- value = R / (1+ number of simulation) Hotspot Evaluation According to Chen et. all (2008) Satscan has made the spatial scan statistic widely accessible, substantially impacting numerous domains in which spatial clusters are of interest. However, two issues make using the method and interpreting its results nontrivial: (1) Satscan lacks cartographic support for understanding the clusters in geographic context and (2) results from the method are sensitive to the selection of scaling parameters, but the system provides no direct support for making these choices. Upper Level Set (ULS) Scan Statistics also suffer from these problems and each issue is discussed below. First, scan statistics methods does not provide cartographic support to view the identified clusters, nor a visual interface to explore cluster characteristics. Geographic information about the identified clusters (e.g., the centre location, the cluster radius, data entities included in each cluster) is available only as text. In order to visualize the geographic location and size of the clusters, a user must process the textual output and export it to GIS software (e.g. ArcGIS or Mapinfo). This is a time-consuming process and inhibits interactive exploration of multiple parameter configurations for interpretation of the results. Because of this limitation, researchers may choose default parameters or make some other arbitrary choices that do not reflect characteristics of the geographic phenomena. Second, it is difficult to determine an optimal setting for scaling parameters. Confusing and even misleading results are possible if the parameter choices are made arbitrarily. The focus of the research presented here is on the aforementioned maximum-size parameter. The default maximum-size setting of 50% seldom produces usable, informative results because the reported primary cluster often occupies a large proportion of the study area. The task of determining the most appropriate maximum-size is challenging. Thus a process for addressing the sensitivity issues of scan statistics method and enhancing the interpretation of scan statistics result was proposed and presented in Figure 4 (Chen et. all 2008).

32 15 Run Multiple Maps Visualize and Compare Clusters on Map Calculate the Reliability of the Maps Identify Core Clusters Compare the results Figure 4 A logic process for analytics of scan statistics result First, scan statistics methods should be run multiple times, starting from a small maximum-size and increasing to the 50% default value. Second, the scan statistics methods results should be visualized in a map matrix for side-by-side comparison of different maximum-sizes. Third, calculate and interpret the reliability of a region in a map. Fourth, core clusters should be discriminated from heterogeneous clusters through interpretation of the reliability. Fifth, the interpretation of core clusters should be confirmed by comparing the results to other independent techniques and consultation with domain experts. When analyzing county-level data, scan statistics methods reports many statistically significant clusters that contain a relatively high proportion of lowrisk counties, which is describe as heterogeneous clusters. However, we noticed that there are often smaller, homogeneous subsets within heterogeneous clusters that exhibit values high enough to reject the null hypothesis on their own strength that is described as core clusters. To assist discrimination of stable, core clusters from heterogeneous and/or unstable ones, Jin Chen et. all (2008) has developed an advanced reliability visualization method. This method visualizes and calculates the reliability that a county is reported within a cluster when scan statistics methods is run multiple times with a systematically varying maximum-size

33 16 parameter. Reliability is estimated by the equation /, where R i is the reliability value for location i, S is the total number of scans, and C i is the number of scans for which that location i is within a significant cluster. The reliability measure has a value range from '0' to '1,' where '0' means that the location is not found in a significant cluster in any of the scans and '1' means that the location is within a significant cluster in all scans. The reliability score measures the stability of clusters reported by multiple scans. Reliability is distinct from the concept of validity, which is a measure of the probability that the cluster represents a true high-risk region. Therefore, the goal of reliability visualization is to identify stable core clusters rather than to evaluate the validity of the core clusters. Since we are applying it to the results of an analysis that measures validity, the end result is to identify the locations that are reliable within a significant high risk cluster Joint Hotspots Most of the time poverty, unemployment, and food security hotspot detection have been done by considering house hold income per capita, but we must keep in mind that these are multidimensional problem. Poverty, unemployment, and food security can be seen from monetary and non monetary dimension. The monetary dimension is based on income and consumption indicators. Meanwhile the non monitory indicators include health, education, social culture, land fertility, etc. Hence, poverty, unemployment, and food security are complex problems that require further research through joint hotspot detection. Hotspot detection using single criteria (household income per capita) based on CBS methods are suspected to be underestimated. Other criteria are needed and this is where joint hotspot detection techniques will be used for altering better results. In this research we will try to find the relationship between poverty, unemployment, and food scarcity hotspots. One of the questions being raised will be If an area is a poverty hotspot, would it also be a unemployment and food scarcity hotspot? Or is there any relation at all between poverty, unemployment, and food scarcity hotspots.

34 17 Table 1 Joint Hotspot Category Area Poor Food Scarcity Unemployment Response Variable Category Score A Yes (3) Yes (2) Yes(1) 6 B Yes (3) Yes (2) No (0) 5 C Yes (3) No (0) Yes (1) 4 D No (0) Yes (2) Yes (1) 3 E Yes (3) No (0) No (0) 3 F No (0) Yes(2) No (0) 2 G No (0) No (0) Yes (1) 1 H No (0) No (0) No (0) 0 To determine the importance between poverty, unemployment, and food security we will use the MDGs criteria: poverty reduction as target number one will be given a score of three, reducing the proportion of people who suffer from hunger as the second target will be given a score of two, and in cooperation with developing countries, develop and implement strategies for decent and productive work for youth as the 16 th target will be given a score of one (UN 2000). Based on the addition of these scores we will develop a final category of multiciriteria hotspots shown in Table 1. These categories will be used as a response variable for ordinal logistic regression model. These methods will be used to identify the factors that are of significant influence towards poverty, unemployment, and food scarcity hotspots Ordinal Logistic Regression Logistic regression extends categorical data analysis to data sets with binary response and one or more continuous factors (Freeman, 1987). Ordinal logistic regression perform logistic regression on an ordinal response variable. One way to use category ordering forms logit of cumulative probabilities for ordinal response Y with c categories, x are explanatory variables. The cumulative probability for each category can be formulated as: P( Y j x) = F ( x) = ( x) +... ( ) where j π +π 1 j x

35 18 π j(x) is the response probability of the j th category of explanatory variable x. Cumulative logits for each category j are defined as F j ( x) L j ( x) = ln ; where j = 1,2,..., c 1 1 Fj ( x) A model that simultaneously uses all cumulative logits can be written as Each cumulative logit has its own intercept. The are increasing in j, since P( Y j x) increases in j when x is fixed, and the logit is an increasing function of this probability (Agresti 2002). and are the maximum likelihood estimators for each and. These estimators represent the change in logits cumulative for each j category, if the other explanatory variables do not influence ˆ ( x). The interpretation of the is the change in logit cumulative for each j category, in other hand, odds ratio will change equal to exp for each change of explanatory variables x (Agresti 2002). The estimate value for P( Y j x) can be derived with inverse transformation of logit cumulative function, the result will be shown below. 1 where 1, 2,, 1, or 1 1 exp ; so that 1 1 exp L j Testing the Model Significance Likelihood ratio test of the overall model is used to assess parameter β i with hypothesis: H 0 : β... = β 0 1 = p = H 1 : at least there is one β i 0 ; i = 1,2,..., explanatory variables. p, where i is the number of

36 19 The likelihood-ratio test uses G statistic, which is G = -2 ln(l 0 /L k ) where L 0 is likelihood function without variables and L k is likelihood function with variables (Hosmer & Lemeshow 2000). If H 0 is true, the G statistic will follow chi-square distribution with p degree of freedom and H 0 will be rejected if value of G > X 2 (p,α ) or p-value < α. A Wald test is used to test the statistical significance of each coefficient in the model. Hypothesis are H 0 : β = 0 i H 1 : β i 0 ; i = 1,..., p where i is the number of explanatory variables. A Wald test calculates a W statistic, which is formulated as W ˆ = β i ˆ βi SE( ˆ β ) Reject null hypothesis if W > Z α /2 or p-value < α (Hosmer & Lemeshow 2000). i β i Assumption of Logistic Regression Logistic regression is popular in part because it enables the researcher to overcome many of the restrictive assumptions of OLS (Ordinary Least Square) regression: 1. Logistic regression does not assume a linear relationship between the dependent and the independent variables. It can handle nonlinear effects even when exponential and polynomial terms are not explicitly added as additional independents because the logit link function on the left-hand side of the logistic regression equation is non-linear. 2. The dependent variable does not need to be normally distributed (but does assume that its distribution is within the range of the exponential family distributions, such as normal, Poisson, binomial, gamma). Solutions may be more stable if the predictors have a multivariate normal distribution. 3. The dependent variable does not need to be homoscedastic for each level of the independents; that is, there is no homogeneity of variance assumption: variance does not need to be the same within categories. 4. Normally distributed error terms are not assumed.

37 20 5. Logistic regression does not require that the independents be an interval scale variable. However, other assumptions still apply: 1. The data doesn t have any outliers. As in OLS regression, outliers can affect results significantly. The researcher should analyze standardized residuals for outliers and consider removing them or modelling them separately. 2. Between explanatory variables there should be no multicollinearity: to the extent that one independent is a linear function of another independent, the problem of multicollinearity will occur in logistic regression, as it does in OLS regression. As the correlation among each other increase, the standard errors of the logit (effect) coefficients will become inflated. Multicollinearity does not change the estimates of the coefficients, only their reliability. High standard errors flag possible multicollinearity ( Model Validation Model validation was carried out by using Correct Classification Rate (CCR). CCR indicates the percentage of true (suitable) prediction. CCR can be calculated by using the equation below CCR the number of true predictions the number of observations x100% The higher the percentage of CCR, the more accurate the model is (Hosmer and Lemeshow 2000)

38 21 III. METHODOLOGY 3.1. Source of Data The source of data that used in this research were National Social Economic Survey (Susenas 2005), data analyzed by Insan Hitawasana Sejahtera (IHS) on the percentage of job seekers (2005), and Potensi Desa (Podes 2005) conducted by CBS. The data used for hotspot detection consisted of household monthly consumption per capita (used for poverty hotspot detection), proportion of unemployment per municipality (used for unemployment hotspot detection) and household total daily calorie intake (used for food scarcity hotspot detection) for 111 municipalities (78 districts and 33 cities) in Java, Indonesia. There were several definitions on Poverty made by the NGOs and GoI Institutions. This research used a widely definition based on Central Bureau of Statistics (CBS). A household is categorized as poor if they have an income per capita below the poverty line. The poverty line is a measure of the amount of money a government or a society believes is necessary for a person to live at a minimum level of subsistence or standard of living. For food security, Ariani (2006) referred that World Bank since 1986 categorized food insecurity into chronic food insecurity and transitory/occasional food insecurity. This research focused on chronic food insecurity, where there is a condition of frequent food scarcity over a certain period of time. On a household level, chronic food insecurity indicates that the share of food owned is slight lower than needed. A household is considered to be food scarce if the total daily calorie intake is lower than <70% of the minimum calorie needed (±1400 kcal). Meanwhile unemployment occur when a person is available to work and seeking work but currently without work. The prevalence of unemployment is usually measured using the proportion of unemployment, which in this study was defined as the percentage of those in the labour force who are unemployed and aged y.o. The explanatory variables that used for ordinal logistic regression were from Podes. The 12 variables used were chosen from a wide variety of 23 variables that were not correlated and assumed to have a significant influence on poverty, unemployment, and food scarcity hotspots. These variables were related to

39 22 citizenship and labour, education and health, transportation, communication and information, economy, politics and security, housing and environment sectors, and also location. The variable and their specific sector can be seen in Table 2. Table 2 List Of Explanatory Variables Per District And Its Sector No Variable Sector Note 1. Economy Potentials of a Village: Farming, Industry, Trade, Services/Others Economy Ratio of Potential Villages/Villages in a Municipality 2. The amount of farmers Citizenship and Labour 3. The amount of farm labourers 4. The average amount of education facilities Citizenship and Labour Education and Health Ratio of Farmers/Villages in a Municipality Ratio of Farm Labours/Villages in a Municipality Ratio of School/Villages in a Municipality 5. The average distance between the village to the capital state/city 6. The amount of small and medium scale industries Location Economy Km Ratio of industry/villages in a Municipality 7. Credit facilities Economy Ratio of Credit/Villages in a Municipality 8. The presence of conflicts within the society Politics and Security Ratio of Conflicts recorded /Villages in a 9. The average amount of families without electricity (PLN) Housing and Environment Municipality Ratio of Families/Villages in a Municipality 10. Province Location Dummy Variable (Banten, West Java, Jakarta, Central Java, Yogyakarta, and East Java) 11. Slum areas Housing and Environment 12. Indonesian Labour Force Citizenship and Labour Ratio of Slum Areas/Villages in a Municipality Ratio of People/Villages in a Municipality

40 Method This study will be divided into three main phases (Figure 5). The first phase was data preparation and exploration that was done by using Microsoft Access and Microsoft Excel The second phase of the research was evaluating ULS and Satscan performance on poverty and food scarcity hotspots of 78 districts in Java. The evaluation included stability and accuracy performance of these tools in detecting hotspots. The most suitable hotspot detection method was used to detect poverty, unemployment, and food scarcity hotspots and also used to develop joint hotspots of 111 municipalities in Java. Afterwards, the research determined the most influencing factors causing poverty, unemployment, and food scarcity hotspots by using Ordinal Logistic Regression. 1 Data Preparation and Exploration: Susenas IHS Survey Data Podes ULS Poverty Hotspot Unemployment Hotspot Food Scarcity Hotspot Satscan Joint Hotspots 2 3 Ordinal Logistic Regression Determining factors that influence poverty, unemployment, and Identify effectiveness of reduction efforts food scarcity hotspots Figure 5 Geoinformatics Research On Poverty, Unemployment, And Food Security Roadmap

41 24 IV. RESULTS AND DISCUSSION 4.1. Data Description The number of households in SUSENAS 2005 (National Socioeconomic Survey) for Java was households. Based on the poverty line issued by CBS, 16.36% of the households were categorized as poor. A poor household was a household having an income per capita below Rp for urban areas and Rp for rural areas. There were 14.57% families in urban areas of Java categorized as poor and 21.25% household in rural areas of Java categorized as poor. The average per capita expenditure of poor household in urban areas in Java was Rp.120,188 while the average per capita expenditure of poor families in rural areas in Java was Rp There were 5 municipalities that had more than 35% poor households. Based on Food Security Agency (2005), an area that has a level of poverty exceeding 35% would be considered as the first priority zone, areas with a poverty level of % would be considered as the second priority zone, and areas with a poverty level of % would be considered as the third priority zone. The average percentage of poor in Java was 17.88%. From 78 regencies and 32 cities studied, Trenggalek and Batang (> 40%) had the highest percentage of poor families in the Java. While Depok City, East Jakarta, Central Jakarta, West Jakarta and South Jakarta had the lowest percentage of poor families (<1%) in Java. Appendix 1 showed municipalities that had the highest percentage of poor families in Java. From Figure 6 it can be seen that there were six main source of income of the poor households. Most of the poor households worked in the field of agriculture and husbandry (46.18%). This indicated that people who worked in the field of agriculture and husbandry were more reluctant towards poverty. Agriculture and husbandry was followed by small scale retailers, building constructions, public transportation, money transfer, and housekeeping. Meanwhile other sources of income include industry, education, trade, forestry, government employees, and etc.

42 25 Others, % House Keeping, 3.07% Money Transfer, 4.07% Agriculture and Husbandry, % Public Transportation, 6.8% Small Scale Retail Sellers, 12.02% Building Construction, 8. 10% Figure 6 Main Source of Fix Income of Poor Households It is stated in Ariani (2004) that chronic food insecurity conditions are often associated with poverty issues. As mentioned in Simatupang (1999) and AusAID (2004), chronic food insecurity is increasing caused by poverty (chronic food insecurity as a food insecurity poverty gap). Meanwhile, according to UNDP China (2001) the causess of food insecurity at the household level is very complex, such as the socio-political situation of agriculture and farmers, lack of productive agricultural land area per capita, low productivity and fertility of the land, climate anomalies, the low application of modern agricultural techniques the impact on the low food production, low purchasing power of households as a result of the limited off-farm income. Meanwhile, according to Witoro (2003) The main causes of food insecurity in these countries is the weakness of developing access to land to produce food. In other cases, lack of food and poverty can be caused by trade policies (domestic and international) as well as disasters such as war or economic crisis. Java is very reluctant to food insecurity. Based on analysis of Susenas (2005), the average percentage of food insecure household in Java was 67.40%. There were 37 municipalities in Java predicted to have more than 70% food insecure households. Kudus, Batang, Pati, Pemalang, Temanggung, and Jepara had the highest percentage of food insecure families in Java (>80%). Meanwhile Sukabumi, Ciamis, Cianjur, Indramayu, Pandeglang, and Sumedang had the

43 26 lowest percentage of food insecure families in the Java (<50%). Appendix 2 showed the percentage of food insecure households and the national ranks used by Food Security Agency and a ranked based on the quartiles. Based on a benchmark research conducted by Food Security Agency (2005), Java had 7 food insecure municipalities, which included Bondowoso, Probolinggo, Situbondo, Jember, Brebes, Serang, and Lebak. This research also indicated that in Java there were 33 municipalities considered as the second priority food insecure municipalities. To decide whether an area is considered a food insecure area the Food Security Agency used a composite indicator using ten indicators which included food availability, food and livelihood access, and health nutrition indicators. There are differences between the priority based on the Food Security Agency and priority based on a quartile method on the calorie intake data. It can be seen that the priority based on composite indicators first priority had a relatively lower percentage of insecure households compared to other municipalities that are considered as the second and third priority. Hence using the food insecure map should be done cautiously. The reasons was food insecurity varied among areas, hence the interpretation of food security also differed from a place to another. An area could be in the group of Priority 1 mainly because this district had very low food and livelihood access, female literacy, high infant mortality, low life expectancy, and poor health infrastructure. Meanwhile another area was food insecure mainly because of infrastructure deficiencies and had very low level of self sufficiency in cereal productions. The list of indicators used in the Food Security Map can be seen in the Appendix 9. The Food Security Composite Indicator (FSCI) was calculated by using Principal Component Analysis to calculate the weights for the indicators used for deriving FSCI final score for each municipality. The 10 indicators were first converted to z-score as a standardization pre requisite for conducting the analysis. PCA was used to assign weight to all 10 indicators and not eliminate indicators, because all indicators were considered to be important. Five principal components were extracted and the final FSCI were a result of multiplying weights from PCA with the corresponding z-scores of indicators which resulted :

44 27 FSCI= Availability Road Poor People Electricity Female Literacy Inverse Life Expectancy Nutrition Status Under Five IMR Clean Water Health Centre Further research on the affectivity of these indicators should be analyzed to enhance the accuracy of the indicator. Food insecurity map based on these indicators can be seen in Appendix 10. In 2005, the average share of income used for purchasing food in urban areas was relatively smaller than in rural areas for all households. Meanwhile the average amount of money spent for food was higher in urban areas compared to rural areas. The averagee expenditure used to purchase food for poor households was Rp , which was around 67.15% of the total income. For urban poor 64.79% (Rp ) of income was used to purchase food, while for rural poor 69.24% (Rp ) of income was used to purchase food. According to Engel law, increasing the proportion of food expenditure indicates declining welfare. That's because there assuming the budget constraint, increasing the proportion of household food expenditure will lead to further decline in the proportion of the budget to buy other than food such as: for clothing, housing, education, health, and others (Ariani 2004) ). all urban rural poor not poor poor not poor poor not poor 0% 20% 40% 60% 80% Figure 7 Share of Java Food Expenditure 2005 The results of this research was also similar with a research conducted by Ariani (2004). Ariani studied the trend of food-insecure households in

45 28 using restriction of less than 1680 kcals/capita/day (80% from 2100 kcal/capita/day). Susenas data analysis of 2005 also showed that: (1) the proportion of vulnerable food households in Java was larger than in Outer Java, (2) the proportion of vulnerable food households in urban areas were lower than in rural areas, (3) the higher the average household income the lower the amount of household food insecurity, (4) the proportion of food-insecure households with livelihoods in the agricultural sector was lower than non-agricultural sectors. Meanwhile for Unemployment, the percentage of job-seekers in Indonesia was quite high. Based on a research conducted by Survey IHS (2005) the average percentage of unemployment in urban and rural areas in Java was 38.6% and 33.8%. Therefore it has been estimated that from a population of 122,406,000 people in Java there are unemployed people. Hence, overall the percentage of unemployment in Java was around 35%. Appendix 3 showed municipalities that had the highest unemployment percentage in Java. Sukabumi City, Pandeglang, Banjar, Kerawang, Lebak had the highest proportion of unemployment (> 50%) in Java. While Pekalongan, Semarang, Jepara, Wonosobo, Temanggung, and Pacitan has the lowest unemployment percentage (<20%) in Java island Satscan and ULS Evaluation We have noted above that Satscan has made the spatial scan statistic widely accessible, substantially impacting numerous domains in which spatial clusters are of interest. However, Satscan is known to have limitations. Circles that have been used for the scanning window, had low power for detection of irregularly shaped clusters (Patil 2006). Therefore, Patil (2006) acknowledged a new version of the spatial scan statistic designed for detection of arbitrary shaped hotspots and for data defined either on a tessellation or a network. This version looks for hotspots from among all connected components of upper level sets of the response rate and is therefore called the upper level set (ULS) scan statistic. The method is adaptive with respect to hotspot shape since candidate hotspots have their shapes determined by the data rather than by some a priori prescription like circles or

46 29 ellipses. Hence, these two methods were compared to see which scan method is most suitable for detecting hotspots in Java Island, that has an arbitrary shape. The evaluation used in this study was based on the stages presented in Figure 4. First, scan statistics methods were run multiple times, starting from a small maximum-size (5%) and systematically increased to the 50% default value. Second, the results were visualized in a map matrix for side-by-side comparison of different maximum-sizes presented in Appendix 4 and Appendix 7. Third, the reliability of a region in a map was calculated and interpreted. Fourth, core clusters would be discriminated from heterogeneous clusters through interpretation of the reliability. Fifth, the interpretation of core clusters has been confirmed by comparing the results to other independent techniques and consultation with domain experts. In this study for poverty and food scarcity, the results of Satscan and ULS were compared with the Food Security Map and Poverty Map accomplished by CBS and FSA Poverty The evaluation of poverty hotspots based on a map matrix for side-by-side comparison of different maximum-sizes was presented in Appendix 4 and summary table was presented in Appendix 5, which showed that the sensitivity to the maximum-size parameter. It has been known that sensitivity was related to 1) clusters tend to contain heterogeneous contents, particularly when using large maximum-size values and 2) components relate to stability of clusters in terms of location and size as the maximum-size value is varied. For an example clusters that tend to contain heterogeneous contents, particularly when using large maximum-size values, Satcan detected Gresik and Lamongan (while ULS detected Demak) that had a poverty level below 22% as a significant hotspot at a 50% maximum spatial cluster size. Gresik and Lamongan were aside Bojonegoro and Tuban meanwhile Demak was aside Batang that had high poverty levels. These areas were not detected in lower maximum spatial cluster size of 30%, 20%, 10%, and 5%. Hence, these kinds of regions were described as heterogeneous clusters. Such clusters were composed of not only the high-risk

47 30 locations that are of interest in hotspot detection, but also many low-risk locations that are not of interest. After taking notice of these heterogeneous clusters, the discrimination of stable/core clusters from heterogeneous and/or unstable ones were done. A core cluster is considered as an often smaller, homogeneous subsets within heterogeneous clusters that exhibit values high enough to reject the null hypothesis on their own strength. Or in other words a core cluster is a cluster that contains homogeneous, high reliability scores and high percentage of cases. Both ULS and Satscan had 20 municipalities having a value of '1' meaning that the location was within a significant cluster in all scans. Between these 20 municipalities 16 of them were the same. All of the 6 first priority poverty areas and 10 second priority areas (FSA) had a reliability of 1, hence it was certain that Trenggalek, Ponorogo, Pacitan and Wonogiri; Lumajang and Jember; Bojonegoro, Ngawi and Kab Blora; Batang Magelang, Temanggung, Banjarnegara; and Garut were reliable core clusters. The average reliability of ULS (0.724) was slightly higher that Satscan (0.708). Both ULS and Satscan had 20 municipalities having a value of '1' meaning that the location is within a significant cluster in all scans. Satscan had 3 municipalities having a value 0.17 (the municipality was only detected once at a 10% maximum spatial cluster size). While ULS detected 9 municipalities having a value 0.17 (the municipality was only detected once at a 50% maximum spatial cluster size). These various values of reliability indicates that one should be precautious in choosing the most suitable maximum spatial cluster size. This study has also conducted an evaluation of Satscan and ULS s poverty hotspots based on the priority criteria of FSA shown in Figure 8. By using a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5% ULS and Satscan were able to detect all of the first priority areas. When detecting the second priority areas ULS performed better than Satscan. ULS detected an average of 90.71% second priority areas, while Satscan detected an average of 77.14% second priority areas. The average false detection, which is the average percentage of having detected a non critical area as a hotspot, of ULS (3.52%) is lower compared to Satscan (10.38%). In detecting first and second priority areas,

48 31 ULS also indicated more stable performance compared to Satscan when the maximum spatial cluster size is changed. Therefore, based on this research ULS indicated more precise and stable performance. A more detail poverty hotspots output and comparison table can be seen in Appendix % 40% 30% 20% 10% 5% First Priority-ULS First Priority-Satscan Second Priority-ULS Second Priority-Satscan Third Priority-ULS Third Priority-Satscan OK-ULS OK-Satscan Figure 8 ULS and Satscan Performance with a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5% Based on the results above, a comparison of hotspots was also conducted to see whether these two methods detected similar hotspots. For the maximum spatial cluster size of 50%, there were 23 (82%) similar second priority municipalities and also 7 (54%) similar third priority municipalities detected by ULS and Satcan. For the second and third priority all of the hotspots detected by Satscan were detected by ULS. As an example, there were four second priority areas that was detected by ULS and not detected by Satscan, which were Brebes, Cilacap, Pemalang, and Purbalingga. It can be seen that Figure 9, if Brebes, Cilacap, Pemalang, Purbalingga, Tegal, Banyumas, Banjarnegara, Kebumen, Wonosobo, and Pekalongan were to be scanned by the circle window A, compared to the surrounding areas, Banjarnegara, Batang, Purworejo, and

49 32 Wonosobo had higher occurrence of poor (see Table 3). Hence, only Banjarnegara, Batang, Purworejo, and Wonosobo became candidate hotspots. Table 3 The Percentage of Poor in Scan Window A and B Scan Window A Scan Window B Municipality % of Poor Municipality % of Poor Batang 43.07% Cilacap 28.61% Banjarnegara 32.57% Banjar 26.91% Wonosobo 29.15% Brebes 25.89% Cilacap 28.61% Tegal 21.05% Pemalang 28.29% Banyumas 20.07% Kebumen 27.36% Cirebon 18.52% Purbalingga 26.28% Kuningan 14.11% Brebes 25.89% Ciamis 12.97% Tegal 21.05% Banyumas 20.07% Pekalongan 15.35% But if the circular window B consisted of only Banyumas, Brebes, Ciamis, Cilacap, Cirebon, Kuningan, Tegal, and Banjar, compared to the surrounding areas, Brebes and Cilacap had higher occurrence of being poor, hence it should have become a candidate hotspot. See the picture below and Table 3 to get a better understanding of this illustration. It could be seen that the circle scanning window used had limitations. ULS method was more adaptive in this case. B A Figure 9 Circle Scanning Windows and a Comparison Of ULS and Satscan Hotspots

50 Food Scarcity The evaluation of poverty hotspots based on a map matrix for side-by-side comparison of different maximum-sizes was also presented in Appendix 7 and summary table presented in Appendix 8 showed the sensitivity to the maximumsize parameter indicated the same trend as the previous case of poverty. Clusters tend to contain heterogeneous contents, especially in Satscan. Satcan detected areas such as Kendal, Trenggalek, Sragen, and Wonosobo that had a lower food scarcity percentage as a significant hotspot at 50%, 40% and 30% maximum spatial cluster sizes. Meanwhile ULS only detected areas that had more than 65% of food scarcity cases. Kendal and Wonosobo were aside Batang and Temangung, Sragen was aside Sukoharjo, Boyolali, and Grobogan, meanwhile Trenggalek was aside Ponorogo that had high food scarcity levels. These areas were not detected in lower maximum spatial cluster size of 10%, and 5%. Hence, we described these kinds of regions as heterogeneous clusters. For reliability, ULS detected 16 municipalities having a value of '1', while Stascan detected 12. ULS detected most of the first priority areas that was developed based on a quartile method and Satscan detected a more diverse range of area. The average reliability of ULS (0.620) is also slightly higher that Satscan (0.595). Most of these areas detected by ULS also had very high proportion of food scarcity (>75%) and had a reliability of 1, hence it was certain that Kudus, Batang, Pati Pemalang, Jepara, Magetan, and Kab. Madiun are reliable core clusters. This study had also conducted an evaluation of Satscan and ULS s hotspots based on the priority criteria based on a quartiles and a comparison between the hotspots of ULS/Satscan and FSA shown in Figure 10. The results were similar with the evaluation of Satscan and ULS of Poverty. By using a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5%, ULS was able to detect an average of 94.74% of the first priority areas. While Satscan detected 83.33% of the first priority areas. For detecting the second priority areas ULS performed better than Satscan. For a maximum spatial cluster size of 50%, 40%, 30%, 20%, and 10% ULS detected an average of 58% second priority areas, while Satscan detected an average of 56% second priority areas. The average false detection,

51 34 which is the average percentage of detecting a non critical area as a hotspot, of ULS (0%) is lower compared to Satscan (9.65%). Therefore, based on this research ULS indicated more precise and stable performance % 40% 30% 20% 10% 5% Maximum Cluster Size First Priority-ULS First Priority-Satscan Second Priority-ULS Second Priority-Satscan Third Priority-ULS Third Priority-Satscan OK-ULS OK-Satscan Figure 10 Food Scarsity s ULS and Satscan Performance with a maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5% In this study of food insecurity we have also compared the hotspots with the Food Security Map. To decide whether an area is considered a food insecure area the Food Security Agency used a composite indicator using ten indicators which include food availability, food and livelihood access, and healthh nutrition indicators. Principal Component Analysis was used to calculate the weights for the final score for each municipality. This benchmark research was also used to evaluate the performance of Satscan and ULS Hotspots. Keeping into mind that ULS and Satscan only needs the amount of cases. Therefore, if ULS or Satscan can achieve similar results, hence it would be much more efficient for decision makers to allocate a critical area by using Satscan or ULS. By using the maximum spatial cluster size of 50% ULS indicated more precise hotspots shown in Table 4. ULS detected 5 (71.43%) municipalities and

52 35 Satscan did not detect any of the 7 municipalities considered as the most food insecure municipalities in Java. Two of the municipalities considered as an unsecure area were not significant in ULS. These municipalities were Lebak and Serang that had relatively lower percentage of food insecure household. In the quartile priority, these two areas were in ranged below the first quartile Table 4 Performance of Satscan and ULS Food Insecurity Hotspots Priority First Priority Second Priority Third Priority Food Insecure Areas ULS/Satscan Food Insecure Areas (Food Security Map) Percentage of Significant Insecure Areas ULS Satscan ULS Satscan ULS Satscan It can also be seen that from 37 municipalities considered as a third priority or a relatively secure area, ULS detected 30 of these areas as a hotspot meanwhile Satscan detected 31 municipalities as a hotspot. This pointed out that from 37 municipalities considered as a relatively secure area, it turns out that 30 of these areas were predicted as a hotspot. These areas include Kudus, Jepara, Temanggung, dan Pati that had high food unsecure households (>80%). The advantage of using the hotspot map of ULS/Satscan was that it could not only see whether an area was a critical area or not, but it also conducted hypothesis testing to see whether the area was significantly different or not compared to surrounding areas. Beside that the geographical information in our data set could be used for more precise results. Hence, precautions in using thematic maps based on descriptive statistics or indicator maps, such as the food insecurity map, should be highlighted. From the evaluations above on poverty and food insecurity, it can be concluded that ULS had better performance for detecting hotspots of poverty and food insecurity in municipalities of Java. These results strengthend previous research that have found limitations in Satscan, especially for arbitrary areas such as Java Island. Satscan had lower precision in detecting hotspots in irregularly

53 36 shaped clusters. Satscan s hotspots were also less stable when the maximum spatial cluster size was changed. Therefore ULS with a maximum cluster size of 50% was used to detectt hotspots of poverty, food insecurity, and unemployment in all areas (municipalities and cities) of Java. The maximum cluster size of 50% was used to get a globall description of these issues and this studied also allocated the core cluster of poverty, food scarcity and unemployment. Hotspot maps based on the maximum cluster size of 50% can be seen in Appendix Joint Poverty, Food Scarcity, and Unemployment Hotspots Poverty, food insecurity, and unemployment are multidimensional problems faced by the society nowadays. Therefore joint hotspots of poverty, food scarcity, and unemployment weree built based on the criteria given in Table 1. It can be seen that most cities in Java faced unemployment problems, meanwhile most municipality areas facedd poverty and food insecurity problems. Yogyakarta and Central Java had the highest percentage of municipalities that were considered most reluctant towards poverty, food insecurity, and unemployment. Kota Pekalongan was the only city in Central Jave that was reluctant towards poverty, food insecurity, and unemployment. Most of the municipalities in West Java were facing unemployment or food insecurity problems. More detail information can be seen in Appendix 12. % City District City District City City District City District City District Banten DIY Jakarta West Java Central Java East Java Province Figure 11 Joint Hotspots Distribution in Java

54 37 Table 5 Joint Hotspots in Java Joint Hotsp ots Total District/ City District/City 0 3 Kota Batu, Kota Salatiga, Kota Serang - Core Cluster Bekasi, Bogor*, Ciamis, Cimahi*, Indramayu, Karawang*, Kota Bekasi*, Kota Bogor*, Kota Cilegon*, Kota.Cirebon*, Kota.Depok, Jakarta Barat, Jakarta Pusat, Jakarta Selatan*, Jakarta Timur*, Jakarta Utara, Kota Sukabumi*, Kota Tangerang, Kota Tegal, Kota Tasik, Majalengka, Pandeglang*, Purwakarta, Serang*, Subang*, Sumedang, Tangerang Gresik*, Jepara*, Klaten, Kota.Belitar, Kota Malang*, Kota Probolinggo, Kota Surabaya, Kota Yogyakarta*, Kudus*, Lamongan, Pati*, Semarang, Sidoarjo, Sleman* Kota Magelang, Kota Semarang, Kuningan*, Sragen*, Trenggalek*, Tuban*, Tulungagung, Wonosobo* Bandung*, Banjar*, Banjarnegara, Cianjur*, Cirebon, Garut*, Kebumen, Kendal, Kota Bandung, Kota Madiun, Lebak*, Sukabumi, Tasikmalaya, Tegal Bantul, Banyuwangi, Bojonegoro, Bondowoso, Boyolali*, Brebes, Grobogan, Gunung Kidul*, Jember*, Jombang, Kab.Blitar, Kab.Blora*, Kab.Kediri, Kab.Malang*, Kab.Mojokerto, Kab.Pasuruan, Kab.Probolinggo, Karanganyar, Kotakediri, Kota Mojokerto, Kota Pasuruan, Kota Surakarta, Lumajang*, Magelang*, Magetan, Nganjuk, Ngawi, Pacitan, Pekalongan*, Ponorogo*, Purbalingga*, Rembang, Situbondo, Sukoharjo*, Temanggung, Wonogiri Banyumas, Batang, Cilacap*, Demak, Kab.Madiun, Kota Pekalongan, Kulon Progo, Pemalang, Purworejo, Note:* is the core cluster having a reliability score above 0.6 There were 14 core clusters in the unemployment category. These core cluster had a proportion of unemployment above 40% and an average of 50% unemployed people There were 7 core clusters in the food scarcity category. These core cluster had a proportion of food scarcity above 70% and an average of 78% food scarce household There were 4 core clusters in poverty and 1 cluster in food scarce and unemployed (Kuningan) category. There were 5core clusters in the poor and unemployed category.these core cluster had an average proportion of poverty of 23% and an average proportion of unemployment of 49% There were 13 core clusters in the poor and food scarce category. These core cluster had an average poverty of 30% and an average proportion of unemployment of 74% Cilacap was the core cluster for this category and had a proportion of poverty of 30%, food scarcity of 74%, and of unemployment of 46%

55 38 In Table 5 above it can be seen that there were 9 areas that were considered as the most critical areas that needed more attention from the government. These areas were located in the northern Central and East Java. There were only three cities that were not either poverty, food scarcity, nor unemployment hotspots, which were Kota Batu, Kota Salatiga, and Kota Serang. Areas which were considered more secure to poverty were located in West Java. A joint hotspot map of poverty, food insecurity, and unemployment can be seen in Appendix 13. After locating these areas, the core cluster of each case was determined. These core clusters gave indications of which cluster should be given prioritization. The core clusters that had a reliability score above 0.6 or it is at least detected four times when using the maximum spatial cluster size of 50%, 40%, 30%, 20%, 10%, and 5%. The stability of clusters can also be seen in Appendix 11. Afterwards an ordinal logistic model was built in order to identify the main factors causing these joint hotspots. By having knowledge on the factors causing these hotspots, hopefully precise actions can be done to alleviate critical areas in Java Determining Factors Causing Poverty, Food Insecurity, and Unemployment In a model to determine the main factors causing poverty, food insecurity and unemployment joint hotspots, there were 12 variables (Table 2) from various sectors analyzed. These sectors included Citizenship and Labour, Education, Economy, Politics and Security, Location, and also Housing and Environment. Before selecting these 12 variables 23 indicators were used (Appendix 15), but there were 11 indicators that were highly correlated with other variables. Therefore to prevent multicolinearity in the model these variables were excluded. The correlation table between of these variables can be seen in Appendix 16. At first an ordinal logistic model for the 6 categories stated in Table 1 was built. It can be seen from Table 6 that there were 7 significant variables at a 15% level of significance, which were School facilities, credit facilities, the percentage of trade village, industrial village, and services village, ratio of farm industry, and proportion of villages without electricity. Hence these factors should be given

56 39 prioritization in alleviating poverty and eradicating food scarcity and unemployment. Table 6. Ordinal Logistic Regression Table Predictor Coef SE Coef Wald P Const(1) Const(2) Const (3) Odds Ratio Const(4) Const(5) Ratio of School per Village %of Credit Facilities % of Industry Village % of Trade Village % of Service Village % of Villages Without Electricity Ratio of Farm Industry Based on the result above, the government should take notice that the increase of school facilities, stimulating economical potential of a village in industry and services, decreased the possibility of an area to become critical areas. The government should give more attention to credit facilities, economical potential of a village in trade, villages without electricity, and small scale farm industry. It turned out that the increase of these factors increased the possibility of a municipality to become a critical area. From this study it was pointed out that credit facilities, farm Industry and trade in a village did not show indication that it could improve the welfare of people living in critical areas. Hence, these factors should be revitalized. Areas that had a high ratio of families living without electricity were also critical points in solving the problem of poverty, unemployment, and food scarcity. Therefore the government should have given more attention to people who lived in these areas. Evaluation towards the model was also done by conducting the likelihoodratio test which uses G statistic. Based on the results given, G = with a p- value=0.000 indicated that H 0 will be rejected or there is at least one explanatory variable had a significant influence on the joint hotspots. Further evaluation was

57 40 carried out by using measures of association, Correct Classification Rate (CCR), and Goodness of fit test that would be explained in more detail below. For measure of association, concordant and discordant pairs indicated how well your model predicted data. The more concordant pairs, the better the model's predictive ability. In the model above there were 77.2% concordant pairs, which was a good indicator. Goodness of fit test intended to test whether the observed data were inconsistent with the fitted model. If they were not (indicated by the significance values that are larger-then α) it can be concluded that the data and the model predictions were similar and that the model was good. Another evaluation used was the Correct Classification Rate (CCR). The CCR of the model above for all response categories were 52% while the CCR for responses categories 1, 5, and 6 was 78.87%. The CCR for response categories 2, 3, and 6 were 0. This indicated that the categories with low response had very low precision. Hence it was suggested to try other link functions or reorder the response value. Other link functions have been used such as the complementary log-log suggested for skewed distribution. The results of the CCR was still low. While reordering the response value in to three categories established higher CCR but it will be difficult to interpret. Further results on the model can be seen in Appendix 16.

58 41 V. CONCLUSION AND RECOMMENDATION 5.1. Conclusion By using Geoinfarmatics techniques the research concluded that: a. By comparing ULS and Satscan on poverty and food insecurity cases in Java, this research pointed out that ULS had a more precise and stable performance compared to Satscan. ULS was suggested as an alternative to thematic maps often used by government institutions. Maps based on Satscan/ULS were more precise compared to thematic maps because spatial scanning methods could not only detect whether an area was a critical area or not, but also conducted hypothesis testing whether the area was significantly different or not compared to surrounding areas and used the geographical information data to enhance the accuracy of results. b. Based on the joint hotspots of poverty, unemployment, and food scarcity there were nine areas (Banyumas, Batang, Cilacap, Demak, Kab. Madiun, Kota Pekalongan, Kulonprogo, Pemalang, Purworejo) considered as the most critical area that needed more attention from the government. Most of these areas were located in Central Java. There were only three cities that were not either a poverty, food scarcity, nor an unemployment hotspots, which were Kota Batu, Kota Salatiga, Kota Serang. c. Main factors causing the joint hotspots were identified by using Ordinal Logistic Regression Model. Factors related to the hotspot were school facilities, village trade, village industry, village services, slum areas, and proportion of families without electricity, and proportion of credit facilities Recommendation Further research on other methods used for hotspot detection should be done. In this research, ULS has better performance than Satscan, it should be simulated/applied not only in Java but also in Indonesia where there is also a large body of oceans separating the islands. Development towards tools that can be used to enhance the practicality in Satscan/ULS outputs is also needed. This study hopefully would become a pioneer in further studies at a national level and

59 42 improvements in results can be done by exploring the possibilities of other covariates and using other sufficient models.

60 43 REFERENCES Adithmc Simulasi Monte Carlo. [Link] Agresti A Categorical Data Analysis Second Edition. John Wilew&Sons, Inc :New Jersey. Anderson NH, Titterington DM Some Methods for Investigating Spatial Clustering with Epidemiological Application, Journal of the Royal Statistical Society, Series A, 160, [Link] [May 7 th 2008] Ariani M, Gatoet SH Karakteristik dan Perubahan Pola Konsumsi Pangan Rumah Tangga Rawan Pangan. Pusat Sosial Ekonomi: Bogor Betti G, Ballini F, Neri L Hotspot Detection and Mapping of Poverty. DGO 2006 Conference. [BKP] Badan Ketahanan Pangan A Food Insecurity Atlas of Indonesia. Dewan Ketahanan Pangan, Departemen Pertanian RI: Jakarta [BKKBN] Badan Koordinasi Keluarga Berencana Nasional Pendataan Keluarga; Selayang Pandang. [link]. [August 2 nd 2006] Bungsu AP The Spatial Scan Statistic For Detecting Dengue Fever Hotspots in Bogor Municipality. Department of Statistics. Mathematics and Natural Science Faculty, Bogor Agricultural University. [Thesis] Chen J, Roth RE, Naito AT, Lengerich EJ, MacEachren AM Geovisual Analytics to Enhance Spatial Scan Statistic Interpretation: An Analysis of U.S. Cervical Cancer Mortality. [link] Freeman DH Applied Categorical Data Analysis. Marcel Dekker, Inc : New York Haran M, Molineros J, Patil GP Large Scale Plant Disease Forecasting: Case Study of Fusarium Head Blight. DGO 2006 Conference. Hosmer DW, Lemeshow S Applied Logistic Regression. John Wilew&Sons, Inc :New York.

61 44 Kardi T Monte Carlo Simulation. [link]. Kulldorff M A Spatial Scan Statistic. Communications in Statistics: Theory and Methods 26: [link]. [June 14 th 2006] Kulldorff M Spatial Scan Statistics: Models, Calculations, and Applications. In Balakrishnan and Glaz (eds), Recent Advances on Scan Statistics and Applications. Boston, USA: Birkhäuser. [link]. [June 14 th 2006] Kulldorff M SaTScanTM User Guide for version 6.1. [link] Naus J Clustering of Random Points in Two Dimensions, Biometrika, 52, [link]. Patil GP, Taillie C Upper Level Set Scan Statistic for Detecting Arbitrarily Shaped Hotspots. Environmental and Ecological Statistics 11: [link]. [May 30 rd 2006] Patil GP, et all Hotspot Detection and Prioritization GeoInformatics for Digital Governance Song C, Kulldroff M Power Evaluation of Disease Clustering Tests. International Journal of Health Geographics. [link]. content /2/1/9 [June 14 th 2006] The Smeru Research Institute What Is Poverty and What Are Its Causes? [link]. [June 5 th 2006] The Smeru Research Institute The Measurement and Trends of Unemployment in Indonesia: The Issue of Discouraged Workers. [link]. [July 30 rd 2009] The United Nations Millennium Development Goals [link]. [ May 12 th 2009]

62 APPENDENCIES 45

63 46 Appendix 1. Municipalities with the Highest Proportion of Poverty in Java Rank Municipality % Priority Rank Municipality % Priority 1 Trenggalek First 25 Pemalang Second 2 Batang First 26 Kebumen Second 3 Bojonegoro First 27 Tasikmalaya Second 4 Lumajang First 28 Magetan Second 5 Jember First 29 Banjar Second 6 Pacitan Second 30 Kab.Kediri Second 7 Situbondo Second 31 Purbalingga Second 8 Garut Second 32 Brebes Second 9 Temanggung Second 33 Gunung Kidul Second 10 Magelang Second 34 Kab.Madiun Third 11 Kab.Blora Second 35 Grobogan Third 12 Kab.Malang Second 36 Jombang Third 13 Banjarnegara Second 37 Kab.Blitar Third 14 Bondowoso Second 38 Sragen Third 15 Kab.Probolinggo Second 39 Sukoharjo Third 16 Purworejo Second 40 Nganjuk Third 17 Ngawi Second 41 Kulon Progo Third 18 Wonosobo Second 42 Kota.Belitar Third 19 Rembang Second 43 Kota Pasuruan Third 20 Boyolali Second 44 Tegal Third 21 Ponorogo Second 45 Demak Third 22 Tuban Second 46 Kota Mojokerto Third 23 Wonogiri Second 47 Kota Pekalongan Third 24 Cilacap Second 48 Banyumas Third

64 47 Appendix 2. Municipalities with the Highest Proportion of Food Insecurity Household in Java Municipality Priority Priority % Priority % Priority (Composite Municipality (Composite (Quartile) (Quartile) Indicators) Indicators) Kudus Third First Grobogan Third Second Batang Second First Kab.Mojokerto Third Second Pati Third First Situbondo First Second Pemalang Second First Tangerang Second Second Temanggung Third First Demak Second Second Jepara Third First Banyuwangi Second Second Magetan Third First Nganjuk Third Second Kuningan Second First Lamongan Second Second Kab.Blora Second First Kab.malang Second Second Kab.Madiun Third First Garut Second Third Gresik Third First Sidoarjo Third Third Kab.Kediri Third First Klaten Third Third Wonogiri Third First Semarang Third Third Purbalingga Second First Karanganyar Third Third Sukoharjo Third First Brebes First Third Bantul Third First Kab.probolinggo First Third Jombang Third First Kab.Pasuruan Second Third Kulon Progo Third First Banjarnegara Second Third Pekalongan Second First Bojonegoro Second Third Banyumas Third Second Bondowoso First Third Magelang Third Second Bandung Third Third Cilacap Second Second Pacitan Third Third Jember First Second Bogor Third Third Kab.Blitar Third Second Rembang Third Third Sleman Third Second Cirebon Second Third Gunung Kidul Third Second Ngawi Second Third Purworejo Third Second Kebumen Second Third Boyolali Third Second Tuban Second Third Lumajang Second Second Tulungagung Third Third Ponorogo Third Second

65 48 Appendix 3. Municipalities with High Proportion of Unemployment in Java Rank Municipality % Estimated Estimated Rank Municipality % Unemployment Unemployment 1 Kota Sukabumi , Serang , Pandeglang , Kota Cilegon , Banjar , Cilacap , Karawang , 003, Kota Jakarta Timur , Lebak , Cimahi , Tasikmalaya , Kota Tasik , Subang , Sukabumi , Bandung , 965, Garut , Kota Bogor , Cianjur , Kota Madiun , Kota Belitar , 883

66 49 Appendix 4. Side-by-side comparison map of different Poverty Hotspots Using a. Satscan 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Maximum Cluster Size 5% Maximum Cluster Size 30% Maximum Cluster Size 10% Maximum Cluster Size 40% Maximum Cluster Size 20% Maximum Cluster Size 50

67 50 b. ULS Maximum Cluster Size 5% Maximum Cluster Size 30% Maximum Cluster Size 10% Maximum Cluster Size 40% Maximum Cluster Size 20% Maximum Cluster Size 50%

68 51 Appendix 5. Satscan and ULS Poverty Hotspots Comparison Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Municipality Priority % ULS Reli abili ty Satscan Reli abili ty Trenggalek First Batang First Bojonegoro First Lumajang First Jember First Pacitan Second Situbondo Second Garut Second Temanggung Second Magelang Second Kab.Blora Second Kab.Malang Second Banjarnegara Second Bondowoso Second Kab. Probolinggo Second Purworejo Second Ngawi Second Wonosobo Second Rembang Second Boyolali Second Ponorogo Second Tuban Second Wonogiri Second Cilacap Second Pemalang Second Kebumen Second Tasikmalaya Second Magetan Second Kab.Kediri Second Purbalingga Second Brebes Second Gunungkidul Second Kab.Madiun Third Grobogan Third

69 52 Appendix 5. Satscan and ULS Poverty Hotspots Comparison Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size (continued) Municipality Priority % ULS Relia bility Satscan Jombang Third Kab.Blitar Third Sragen Third Sukoharjo Third Nganjuk Third Kulonprogo Third Tegal Third Demak Third Kab.Pasuruan Ok Cianjur Ok Sukabumi Ok Tulungagung Ok Karanganyar Ok Banyuwangi Ok Sidoarjo Ok Lamongan Ok Gresik Ok Reli abil ity

70 53 Appendix 6. Comparison of ULS and Satscan Poverty Hotspots with Critical Maximum Spatial Cluster Size Poverty Areas based on the Food Security Map (Food Security Agency) Priority First Priority Second Priority Third Priority OK First Priority Second Priority Third Priority OK First Priority Second Priority Third Priority OK First Priority Second Priority Third Priority OK First Priority Second Priority Third Priority OK Method Significant Hotspots Critical Areas Based on FSA Percent ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan ULS Satscan Total of Similar Hotspots

71 54 Appendix 7. Map matrix for side-by-side comparison of different Food Scarcity Hotspots Using 50%, 40%, 30%, 20%, 10% and 5% Maximum Cluster Size a. Satscan Maximum Cluster Size 5% Maximum Cluster Size 30% Maximum Cluster Size 10% Maximum Cluster Size 40% Maximum Cluster Size 20% Maximum Cluster Size 50%

72 55 b. ULS Maximum Cluster Size 5% Maximum Cluster Size 30% Maximum Cluster Size 10% Maximum Cluster Size 40% Maximum Cluster Size 20% Maximum Cluster Size 50%

73 56 Appendix 8. Satscan and ULS Food Scarcity Hotspots Comparison Using 50%, Municipality 40%, 30%, 20%, 10% and 5% Maximum Cluster Size Priority Composite % ULS Satscan Rel* Rel* Kudus Third Batang Second Pati Third Pemalang Second Temanggung Third Jepara Third Magetan Third Kuningan Second Kab.blora Second Kab.madiun Third Gresik Third Kab.kediri Third Wonogiri Third Purbalingga Second Sukoharjo Third Bantul Third Jombang Third Kulonprogo Third Pekalongan Second Banyumas Third Magelang Third Cilacap Second Jember First Kab.blitar Third Sleman Third Gunungkidul Third Purworejo Third Boyolali Third Lumajang Second Ponorogo Third Grobogan Third Kab.Mojokerto Third Situbondo First Demak Second Banyuwangi Second Nganjuk Third Lamongan Second

74 57 Appendix 8. Satscan and ULS Food Scarcity Hotspots Comparison Using 50%, Municipality 40%, 30%, 20%, 10% and 5% Maximum Cluster Size (continued) Priority Composite % ULS Satscan Rel* Rel* Kab.malang Second Sidoarjo Third Klaten Third Semarang Third Karanganyar Third Brebes First Kab.Probolinggo First Kab.Pasuruan Second Banjarnegara Second Bojonegoro Second Bondowoso First Pacitan Third Rembang Third Cirebon Second Ngawi Second Kebumen Second Tulungagung Third Kendal Second Trenggalek Third Sragen Third Wonosobo Second Note: Rel*=Reliability Value

75 58 Appendix 9. Indicator used for Food Insecurity Atlas No Category Indicator Data Source 1 Food Availability 2 Food and Livelihoods Access Per capita normative consumption to net rice, maize, cassava, and sweet potatoes availability ratio Percentage of People below the poverty line Percentage of Villages without inadequate connectivity Percentage of people without access to electricity Provincial and District Food Security Agencies ( ) Data dan Informasi Kemiskinan, BPS PODES 2003, BPS Data dan Informasi Kemiskinan, BPS 3 Health and Nutrition Life Expectancy at Birth Children underweight Female Illiteracy Infant Mortality Rate (IMR) Population without access to safe drinking water Data dan Informasi Kemiskinan, BPS Data dan Informasi Kemiskinan, BPS Data dan Informasi Kemiskinan, BPS BPS and UNDP (computed for the Indonesia Human Development Report) Data dan Informasi Kemiskinan, BPS Percent of people living more than 5 km away from Puskesmas Data dan Informasi Kemiskinan, BPS 4 Transient Food Insecurity Percentages of areas without forests Percent of Puso areas Percentages of villages affected by flood and land slide Dinas Kehutanan, 2003 Departemen PU Provincial Food Security Agencies, 2003 Badan Meteorology Rainfall deviation Geofisika, 2004 Source of data: Food Insecurity Atlas 2005 (Food Security Agency of Indonesia and WFP)

76 59 Appendix 10. Poverty Map of Poverty, Food Insecurity, and Unemployment in Java (2005) Using ULS with a Maximum Cluster Size of 50% a. ULS Poverty Map of Java Island (2005) b. ULS Food Security Map of Java Island (2005)

77 60 Appendix 10. Poverty Map of Poverty, Food Insecurity, and Unemployment in Java (2005) Using ULS with a Maximum Cluster Size of 50% (continued) c. Food Security Agent s Food Security Map (2005) d. ULS Unemployment Map of Java Island (2005)