Fitria Virgantari, Tjut Awaliyah, I Wayan Mangku, Siswadi

Judul	:	Evaluation of some methods for estimating parameters of regression model with various zero observation by Monte Carlo Simulation
Abstrak	:	This research aim to study the nature of parameter estimate of linear regression model containing zero observations at the dependent variable. This data structure is often met at consumption/expenditure household data. Estimator studied is estimator of Ordinary Least Square/OLS, Weighted Least Square/WLS and Maximum Likelihood/ML method. The regression model to study is linear regression model with five dependent variables, that is Y = β0 + β1X1 + β2X2 + β3X3 + β4X4 + β5X5 + ε, where ε drawn from normal distribution with mean 0 and variance 0.1. While value of β1, β2, β3, β4 and β5 and also value X1, X2, X3, X4, and X5 based on the research by Virgantari ( 2005). Proportion of zero observation studied is 4%, 6%, 8%, 10%, 20%, 30%, 40%, 50%, 60% and 70%; each repeated 100 times of 960 random sample. The result show that bias of β obtained from ML method is biggest compared to two other methods. Bias of OLS and WLS is almost the same at all of proportion of zero observations. Bias of β0, β1, and β5 is negativite, which mean that the estimator is smaller than the real value or downward bias. Meanwhile bias of β2 , β3 dan β4 is positive, which mean that the estimator is bigger than the real value or upward bias. Variance of OLS and WLS estimator seem the same and tend to stabilize at all of zero proportions tried. Meanwhile variance of ML estimator show the greater value with the greater of zero proportion, even sharply downward after proportion of zero 60%. MSE of β obtained from ML method is biggest compared to two other method. It indicate that ML estimator is not consistent. MSE of OLS and WLS method is the same at all of estimate values. All of methods have the same pattern value at estimate of β5. According to the bias, variance, and MSE estimator obtained at various zero proportion tried, hence can be said that in general OLS and WLS method is the better and both have same nature of efficiency compared to the ML method. Keywords: OLS/WLS/ML estimator, zero observation, bias, variance estimator, MSE
Tahun	:	2008	Media Publikasi	:	Seminar Internasional
Kategori	:	Prosiding	No/Vol/Tahun	:	0 / 0 / 2008
ISSN/ISBN	:	978-979-19256-0-0
PTN/S	:	Universitas Pakuan	Program Studi	:	MATEMATIKA
Bibliography	:	Dey, M. M. 2000. Analysis of demand for fish in Bangladesh. Journal of Aquaculture Economics and Management 4(1/2):63-81. International Center for Living Aquatic Resources Management, Penang. Malaysia. Draper, N. R. and Smith H. 1981. Applied regression analysis. Second Edition. John Wiley & Sons, Inc. New York. Kennedy, P. A guide to econometrics. 1996. The MIT Press. Cambridge, Massachusetts. Kshirsagar, A. M. 1983. A course in linear models. Marcel Dekker, Inc. New York. Maddala, G. S. 1983. Limited dependent and qualitative variables in econometrics. Cambridge University Press. New York. Myers, R. H. 1990. Classical and modern regression with applications. Second Edition. PWS-Kent Publishing Company, Boston. Tobin, J. 1958. Estimation of relationships for limited dependent variables. Econometrica Vol 26:24-36. Virgantari, F. 2005. Perbandingan model Tobit, regresi terpotong dan regresi biasa pada data konsumsi rumah tangga. Tesis Pascasarjana. Institut Pertanian Bogor. Tidak dipublikasikan.
URL	: