| dc.description.abstract | Classical regression model assumes that no interaction between objects. However, in some cases these conditions are not hold so we have to find an alternative solution that can accommodate the interaction effect. Spatial model is a model that can accommodate the interaction effect by adding a spatial component in the right hand side that represented by a spatial weights matrix. Generally, spatial weighted matrix uses closeness concept among the units, without include the proximity of interest variables. Another way to construct a spatial weights matrix is to use AMOEBA procedure introduced by Aldstadt and Getis. In this procedure, each element of the matrix, beside is determined by the closeness relationship, also determined by proximity among variables using local Getis statistics. Getis and Ord have claimed that the statistic is normally distributed. However, empirically, normal curve test shows that normality of local Getis statistic is influenced by variable of interest (original variable). Due to this reason, we propose a modification of local Getis statistic, namely Gnew, which gives a robust result. A modification of local Getis statistics by transforming to ( ) gives a robust result, that is, it has normal distribution for any distribution of Xj. To assess local Getis and modified local Getis statistics, they are applied on AMOEBA procedure to create weights matrix, namely WG and WGnew. We compare them with contiguity matrix (WC). The model that used for comparison is spatial dynamic panel model with respect to Cizek. By using simulated data and root mean square error relative criteria, it is found that WGnew gives the best performance. In the next stage, we evaluate spatial matrix performance in model to real data, whereas in this case, we take poverty issue in Center Java Province. Poverty is one of the issues of concern at both the central and local government. Central Java was the province with the second largest number of poor people in Indonesia. There are many factors that influence poverty, some of them are GDP per capita, population, education, share of agriculture labor, share of industry labor, share of trading labor and share of services labor (four dominant sectors). Focus of this research is to study influence of labor of junior high school graduates, population, GDP per capita and the share of labor in four dominant sectors to the number of poor people using spatial panel model. Sources of data for modeling obtained from central BPS and Center Java BPS from 2008 to 2012. The research results show that an increase in population, labor of junior high school graduates and share of agricultural labor can increase the number of poor people. Meanwhile the increase in labor share of industry, trade and service sectors and GDP per capita can decrease the number of poor people. Based on these findings, there are several ways to reduce the number of poor people, such as to control population by family planning program (KB program), improving quality of human resources, expand some business, especially in sectors of industry, trading and services. | en |
