PERBANDINGAN ESTIMASI MODEL RESPON KUALITATIF MENGGUNAKAN METODE OLS, GMM DAN MAXIMUM LIKELIHOOD: PADA KASUS PROBABILITAS KEPEMILIKAN MOBIL RUMAH TANGGA DI KELURAHAN PAHLAWAN KOTA PALEMBANG

Abstract

This study aim to comparing accuracy in the analysis of qualitative response data especially to analyze the probability of car ownership households by using LPM models , GMM,  Probit and Logit Models. Primary data is taken from households in the Kelurahan Pahlawan Kota Palembang. The result of the coefficients and constants model by using OLS and GMM estimation  is the same, only slightly different in the standard error, where the GMM standard error is slightly smaller than the OLS standard errors, but of probability gives the same conclusion. Where as estimated by using Maximum Likelihood method such as  probit and logit models better than OLS and GMM estimate. In the case of car ownership results estimation methods of logit model give coefficient greater than the coefficient probit model. But the determination of the coefficient by McFadden ( R²MC_F ) probit models is higher than R²MC_F logit model. Based from Akaike information criterion (AIC) and Schward Criterion (SC) indicators,  probit model is better than the logit model. Thus, in the result of the model, probit model is better than the logit model. If income rises, as the prediction Probit models, households will have a probability buying a new car is rapid, otherwise if income drops, then the probability of the household will be quickly decided not to buy a car. In the logit model if income increases, then probability of buying car a smaller because it does not immediately decide to buy a car , so if revenues fall,  does not mean it will not buy the car, but did delay for the foreseeable future . Excellence both probit and logit models, can be predicted of the probability additional effects (marginal effect). However, marginal  effects on the model probit is higher than the logit model.

Keywords: Probability Linier Model, General Method of Moment, Maximum Likelihood- Probit, and Logit.