SEMIPARAMETRIC ESTIMATOR CENSORED REGRESSION ON ZERO OBSERVATION

Zero observation on the response variable in the socio-economic field are often found in household demand models. The data structure is called a data censored. Thus will imply on the method to estimate parameters of the model used. One model to address the problems of zero observations on the response variable is using censored regression model is also know as tobit model. But Maximum likelihood estimators (MLE) in the standard tobit model is inconsisten if errors have non normality or heteroskedastic. Another estimator alternative is semiparametric estimator censor least absolute deviations (CLAD) . CLAD estimator has the advantages of not sensitive to outlier data, is able to produce robust estimates, and can be used for non-normal data and data with homoskedastisitas are not fulfilled This study aims to applying CLAD and MLE in censored regression model where data used is about households energy demand of LPG. The data used is the LPG expenditure in rural areas in the provinces of West Java that the number zero observations is 37 percent of the sample. The result shows that CLAD and ML estimators are consistent estimators. It can be seen from the average variance of all coeficient parameter estimators that exist. But along with increasing the number of samples, the CLAD estimators performance is getting better than MLE.    Key words :  Zero observation, Censored regression, Censored Least Absolute Deviations (CLAD), Consistent estimator, LPG demand