Anang Kurnia

Cluster Information of Non-Sampled Area In Small Area Estimation

Empirical Best Linear Unbiased Predictor (EBLUP) has been widely used to predict parameters in area with small or even zero sample size. The problem is when this model should be used to predict the parameters of non-sampled area. Ordinary EBLUP predicted the parameters using synthetic model which ignore the area random effects because lack of non-sampled area information. Thus, those prediction will be distorted based on a single line of the synthetic model. One of idea that developed in this paper is to modify the prediction model by adding cluster information by assuming that there are similiarities among particular areas. These information will be added into the model to modify the intercept of prediction model. Another approach is by adding random effects of auxiliary variable into the previous model in order to modify both intercept and slope of the prediction model. In this paper, simulation process is carried out to study the performance of the proposed models compared with ordinary EBLUP. All models evaluated based on the value of Relative Bias (RB) and Relative Root Mean Squares Error (RRMSE). The results show that the addition of cluster information can improve the ability of the model to predict on non-sampled areas.

A study of area clustering using factor analysis in small area estimation (An analysis of per capita expenditures of subdistricts level in regency and municipality of Bogor)

Empirical Best Linear Unbiased Prediction (EBLUP) is one of indirect estimating methods which used to estimate parameters of small areas. EBLUP methods works in using auxiliary variables of area while adding the area random effects. In estimating non-sampled area, the standard EBLUP can no longer be used due to no information of area random effects. To obtain more proper estimation methods for non sampled area, the standard EBLUP model has to be modified by adding cluster information. The aim of this research was to study clustering methods using factor analysis by means of simulation, provide better cluster information. The criteria used to evaluate the goodness of fit of the methods in the simulation study were the mean percentage of clustering accuracy. The results of the simulation study showed the use of factor analysis in clustering has increased the average percentage of accuracy particularly when using Ward method. The method was taken into account to estimate the per capita expenditures based on ...

Influence of presidential candidates e-campaign towards voters in 2014 presidential election in Bogor City

The use of social media (facebook, Twitter and youtube) is an activity that is most in demand by Internet users in Indonesia. The media can be used in the election campaign to shape public opinion in order to influence voters and increase participation of voters. This paper analyzes the effect of e-campaign on Indonesian Presidential election 2014 study conducted in West Java (Bogor City). Respondents were asked to complete a questionnaire containing the activities undertaken in the use of the website, facebook, Twitter and youtube about the presidential candidates and campaign statements to the effect of e-voting decision. The relationship between the use of websites and social media by voters with the voting decision are examined using Spearman correlation. Based on the analysis found score relative overall respondents agreed argue that campaign through the website, Twitter, facebook page, and youtube channel is able to influence their voting decision to choose one of the Candidates. Correlation test shows there are significant relationship between website utilization, social media towards voting decision.

Cluster information of non-sampled area in small area estimation of poverty indicators using Empirical Bayes

Survey is one of data collection method which sampling of individual units from a population. However, national survey only provides limited information which impacts on low precision in small area level. In fact, when the area is not selected as sample unit, estimation cannot be made. Therefore, small area estimation method is required to solve this problem. One of model-based estimation methods is empirical Bayes which has been widely used to estimate parameter in small area, even in non-sampled area. Yet, problems occur when this method is used to estimate parameter of non-sampled area which is solely based on synthetic model which ignore the area effects. This paper proposed an approach to cluster area effects of auxiliary variable by assuming that there are similar among particular area. Direct estimates in several sub-districts in regency and city of Bogor are zero because no household which are under poverty in the sample that selected from these sub-districts. Empirical Bayes method is used to get the estimates are not zero. Empirical Bayes method on FGT poverty measures both Molina & Rao and information clusters have the same estimates in the sub-districts selected as samples, but have different estimates on non-sampled sub-districts. Empirical Bayes methods with information cluster has smaller coefficient of variation. Empirical Bayes method with cluster information is better than empirical Bayes methods without cluster information on non-sampled sub-districts in regency and city of Bogor in terms of coefficient of variation.Survey is one of data collection method which sampling of individual units from a population. However, national survey only provides limited information which impacts on low precision in small area level. In fact, when the area is not selected as sample unit, estimation cannot be made. Therefore, small area estimation method is required to solve this problem. One of model-based estimation methods is empirical Bayes which has been widely used to estimate parameter in small area, even in non-sampled area. Yet, problems occur when this method is used to estimate parameter of non-sampled area which is solely based on synthetic model which ignore the area effects. This paper proposed an approach to cluster area effects of auxiliary variable by assuming that there are similar among particular area. Direct estimates in several sub-districts in regency and city of Bogor are zero because no household which are under poverty in the sample that selected from these sub-districts. Empirical Bayes method is used to get...

A comparative study of approximation methods for maximum likelihood estimation in generalized linear mixed models (GLMM)

Maximum likelihood estimates in GLMM are often difficult to be obtained since the calculation involves high dimensional integrals. It is not easy to find analytical solutions for the integral so that the approximation approach is needed. In this paper, we discuss several approximation methods to solve high dimension integrals including the Laplace, Penalized Quasi likelihood (PQL) and Adaptive Gaussian Quadrature (AGQ) approximations. The performance of these methods was evaluated through simulation studies. The ‘true’ parameter in the simulation was set to be similar with parameter estimates obtained by analyzing a real data, particularly salamander data (McCullagh & Nelder, 1989). The simulation results showed that the Laplace approximation produced better estimates when compared to PQL and AGQ approximations in terms of their relative biases and mean square errors.

Multilevel modeling and panel data analysis in educational research (Case study: National examination data senior high school in West Java)

Individual and environment are a hierarchical structure consist of units grouped at different levels. Hierarchical data structures are analyzed based on several levels, with the lowest level nested in the highest level. This modeling is commonly call multilevel modeling. Multilevel modeling is widely used in education research, for example, the average score of National Examination (UN). While in Indonesia UN for high school student is divided into natural science and social science. The purpose of this research is to develop multilevel and panel data modeling using linear mixed model on educational data. The first step is data exploration and identification relationships between independent and dependent variable by checking correlation coefficient and variance inflation factor (VIF). Furthermore, we use a simple model approach with highest level of the hierarchy (level-2) is regency/city while school is the lowest of hierarchy (level-1). The best model was determined by comparing goodness-of-fit and che...

A Comparative Study of Imputation Methods for Estimation of Missing Values of Per Capita Expenditure in Central Java

Missing values in repeated measurements have attracted concerns from researchers in the last few years. For many years, the standard statistical methods for repeated measurements have been developed assuming that the data was complete. The standard statistical methods cannot produce good estimates if the data suffered substantially by missing values. To overcome this problem the imputation methods could be used. This paper discusses three imputation methods namely the Yates method, expectation-maximization (EM) algorithm, and Markov Chain Monte Carlo (MCMC) method. These methods were used to estimate the missing values of per-capita expenditure data at sub-districts level in Central Java. The performance of these imputation methods is evaluated by comparing the mean square error (MSE) and mean absolute error (MAE) of the resulting estimates using linear mixed models. It is showed that MSE and MAE produced by the Yates method are lower than the MSE and MAE resulted from both the EM algorithm and the MCMC method. Therefore, the Yates method is recommended to impute the missing values of per capita expenditure at sub-district level.

A robustness study of student-t distributions in regression models with application to infant birth weight data in Indonesia

In regression models, the use of least squares method may not appropriate in modelling the data containing outliers. Many robust statistical methods have been developed to handle such a problem. Lange et al. [1] developed robust models based on t distributions and using M-estimation approaches. In this recent article we evaluate the performance of M-estimation as well as investigated the robustness of t distribution models in linear regression by means of simulation. The models are then applied to infant birth-weight data in Indonesia. We show that the t distribution models with small degrees of freedoms have produced better estimates from perspectives of their performance and robustness when compared to other estimates.

Small area estimation models with time factor effects for repeated measurement data

There has been an increasing demand for small area statistics based on sample survey in last few years. Generally, sample survey is designed for a specific area with large sample size. Since the sample size of a specific area is small, a direct estimation for sub population fails to provide enough precision. On the other hand, the per capita expenditure as a measure of prosperity and well-being is becoming an important issue in developing countries. In Indonesia, the available data related to this issue is usually measured repeatedly in quarterly using the national social-economic survey (Susenas). There are two problems to analyze Susenas data in district level. Firstly, the sample size in each quarter is relatively small. Secondly, the auxiliary data usually is available yearly only. To overcome this problem, a proper estimation technique was exercised using modeling systems. An extended of small area estimation (SAE) technique, based on both Fay-Herriot and Rao-Yu models in repeated measurement using Susenas data was discussed. Especially, SAE models with time factor effects motivated by Rao and Yu [7] were proposed. The performances of these models were evaluated by comparing the root mean square error (RMSE) and coefficient of variation (CV) of the estimators. The results showed that Rao-Yu with time factor effect models produced smaller RMSE which led to a significant reduction in a CV relative to other models.

Post-stratification sampling in small area estimation (SAE) model for unemployment rate estimation by Bayes approach

This research implemented unemployment rate in Indonesia that based on Poisson distribution. It would be estimated by modified the post-stratification and Small Area Estimation (SAE) model. Post-stratification was one of technique sampling that stratified after collected survey data. It’s used when the survey data didn’t serve for estimating the interest area. Interest area here was the education of unemployment which separated in seven category. The data was obtained by Labour Employment National survey (Sakernas) that’s collected by company survey in Indonesia, BPS, Statistic Indonesia. This company served the national survey that gave too small sample for level district. Model of SAE was one of alternative to solved it. According the problem above, we combined this post-stratification sampling and SAE model. This research gave two main model of post-stratification sampling. Model I defined the category of education was the dummy variable and model II defined the category of education was the area rando...