Download LIMDEP Student User Manual
Transcript
Censoring and Sample Selection 16-3 and so on. Sample clustering for the estimated asymptotic covariance matrix may be requested with ; Cluster = specification as usual. 16.2.2 Results for the Tobit Model You may request the display of ordinary least squares results by adding ; OLS to the command. These will be suppressed if you do not include this request. The OLS values will be used as the starting values for the iterations. Maximum likelihood estimates are presented in full. Note that unlike most of the discrete choice models, there is no restricted log likelihood presented. The maximum likelihood estimates for a model that contains only a constant term are no less complicated than one with covariates, and there is no closed form solution for the (β,σ) parameter pair for this model. For a general test of the joint significance of all the variables in the model, we suggest the standard trio of Neyman-Pearson tests, which can be carried out as follows: First set up the Rhs variables in the model. NAMELIST CALC TOBIT CALC ; xvars = the x variables in the model, without the constant term $ ; kx = Col(xvars) $ ; Lhs = y ; Rhs = one $ ; l0 = logl $ This command will produce the Lagrange multiplier statistic. TOBIT TOBIT ; Lhs = y ; Rhs = xvars,one ; Start = kx_0,b,s ; Maxit = 0 $ ; Lhs = y ; Rhs = xvars,one $ Compute the likelihood ratio statistic. CALC ; List ; lr = 2*(logl - l0) ; 1 - Chi(lr,kx) $ This computes a Wald statistic. MATRIX CALC ; beta = b(1:kx) ; vb = varb(1:kx,1:kx) ; List ; Wald = beta’<vb>beta $ ; List ; 1 - Chi(wald,kx) $ Retained output from the model includes Matrices: b, varb Scalars: s ybar, sy, kreg nreg nonlimts = estimated σ = number of coefficients, = number of observations = number of nonlimit observations in estimating sample