Small area in Geoinformatics and its applications to detect poverty pockets in Jember.

Abstract

Scan statistic is a method in Geinformatics considering spatial unsure to detect high or low potential area geographically and examine an area proportion parameter compare with others proportion statistically. This method detects extreme area (maximum or minimum) called hotspot. Hotspot means something unusual, anomalies, or critical condition area. Scan statistic is used to detect and evaluate area called Most Likely Cluster (MLC). It needs complete information about detected area with large sample size, while survey usually uses small sample size. One solution to overcome small sample size is by increasing sample size, but it is cost, time, and energy consuming. On the other side, there is a developing method in statistics to use survey data with small sample size, i.e. Small Area Estimation (SAE). Although this method is not used to compare among area, the biggest benefit from SAE is that it can be used to estimate area parameter complete, both surveyed and un-surveyed area. With complete estimation data, scan statistic can be used to detect hotspot or Most Likely Cluster (MLC). Based on the need of complete estimator value, SAE application is necessary in scan statistic. Quality of SAE application result in scan statistic depends on (i) sample size of hotspot, (ii) SAE estimation quality of model with some auxiliary variables and estimation methods, so SAE estimation methods need to be examined. SAE quality in handling small sample size case and scan statistic have been studied separately. This dissertation studies scan statistic in small sample size case, as an application of SAE on scan statistic. Estimation result gained through survey is called as Direct Estimate (DE), but usually with low accuracy and precision. There is a direct estimation proportion in MLC problem in Geoinformatics that can be replaced by estimation through SAE. This study’s purposes are to replace DE role with SAE estimation in Geoinformatics problem (in search for hotspot), to do cluster detection through scan statistic using small sample size by using SAE model into scan statistic, and use small area based Geoinformatics to detect poverty area pockets based on surveyed data and estimation of all village in Jember, East Java. This study is an application of SAE based scan statistic that is examined theoretically, by simulation, and by application on real data. SAE application on scan statistic can handle problems in detecting highly potential and statistically significant by using only small sample size. Simulation and application examination on real data are evaluated using statistic properties. Analysis continues with evaluation of correct classification on result gained. 100% correct classification through simulation examination on surveyed area is gained using Bayes and non-Bayes SAE. Bayes methods used in this study are spatial Hierarchical Bayes (HB) and non-spatial method, while non-Bayes methods used are EBLUP and SEBLUP. In un-surveyed area (large area), MLC will be bigger. HB1 (spatial correlation weighted) with hotspot proportion 0.95 is capable to estimate 80.97% correctly, while HB2 (nearest neighbor) is 75.62%. On the other side, EBLUP is only capable to estimate 65.99% of correct classification. From estimator property side, both methods (HB and EBLUP) have stable on result and unbiased property. HB method gives better classification result than EBLUP, shown in high correct classification result, but its weakness is the need to choose prior variance and match spatial weight. Application on poverty data shows that based on calorie consumption as dependent variable, SAE based scan statistic can classify 100% correct using HB2 and SEBLUP on 35 villages of Susenas 2008 result. Based on household expenditure variable, SAE based scan statistic produce 88.57% correct classification using HB2 and 62.85% with EBLUP. Results of SAE based scan statistic in all villages in Jember district on calorie consumption variable yield 70.45% correct classification with HB2 and 57.09% with SEBLUP. Estimation of the poorest village proportion based on household expenditure using HB2 yields 93.93% and SEBLUP yields 97.95%. Amount of auxiliary variables does not affect analysis result both on SAE and scan statistic. Application on poor village detection in Jember district indicates better result than without SAE method in detecting high potential of poverty level. Calorie consumption is more stable than monthly household expenditure. Based on calorie consumption with HB2 and SEBLUP method of 35 surveyed village, poverty pockets are found in Wringin telu, Ampel, Sidodadi, Garahan, Wringin Agung, Pringgowirawan, Sumber Pinang, Suren, and Sumber Sari village. On the other hand, based on household expenditure, HB2 method yields poverty pockets in Serut, Jatiroto, Sukorejo, Sumberjambe, Gumukmas, Pringgowirawan, Sempolan, Suren, and Randu Agung. Pringgowirawan and Suren village are the two poorest village based on calorie consumption and household expenditure in surveyed village. These two poorest villages can be said as poverty pocket in order to be improvement target in minimizing poverty in Jember district. Poverty pockets in all Jember’s villages based on calorie consumption using HB2 and SEBLUP methods are Sumber Kejayan, Tegalrejo, Tegalwaru, Pakusari, Gambiran, Plalangan, Ajung, Glagahwero, Sumberjeruk, Sumber Ketempah, Lembengan, Sumber anget, and Karang Paiton village. This study finds 119 villages as poverty pockets in all villages of Jember district based on household expenditure by HB2 and EBLUP method. From 13 villages’ poverty pockets from calorie consumption and 119 villages from household expenditure in all villages of Jember district, there are 10 poorest villages, such as Sumber Kejayan, Tegalrejo, Tegalwaru, Pakusari, Gambiran, Plalangan, Ajung, Glagahwero, Sumberjeruk, and Sumber Ketempah. Problem of small sample size in Geoinformatics will not disturb statistic property of estimator. With SAE application in Geoinformatics, there is no need to study all population members. Available information is enough to study poverty pockets in order to save cost, time, and energy in surveying MLC.