help scls 



scls -- Symmetrically censored least squares estimator


scls depvar [indepvars] [if] [in] [, scls_options]

options Description -------------------------------------------------------------------------

start(estimator) method used to compute the starting values

powell use Powell's (1986) original optimization algorithm; the default is Santos Silva's (2001) algorithm

tolerance(#) tolerance for the coefficient vector; default is tolerance(1e-6)

iterate(#) perform a maximum of # iterations; default is iterate(16000)

nolog suppresses the iterations log


scls does not allow weights.


scls implements Powell's (1986) symmetrically censored least squares estimator and reports standard errors and t-statistics that are asymptotically valid under heteroskedasticity. The robust covariance matrix is computed following Powell (1986). Additionally, scls reports the value of the objective function (see equation 2.10 in Powell, 1986, p. 1439), the size of the sample and the number of observations effectively used. Two optimization algorithms are available. By default, estimation is performed using Santos Silva's (2001) algorithm, but Powell's (1986) original algorithm can be chosen with the option powell. It is also possible to choose how the starting values are computed using start(estimator).


start(estimator) specifies how the initial values are computed. The following methods are available:

start(ols) specifies that the starting values are obtained by OLS as in Powell (1986).

start(tobit) specifies that the starting values are obtained by Tobit.

start(clad) specifies that the starting values are obtained by performing the first two steps of the Chernozhukov and Hong (2002) estimator for censored median regression.

The default is start(clad) because these starting values are consistent under the maintained assumptions and this method appears to provide faster convergence.

powell specifies that Powell's (1986) algorithm is used in the optimization. By default, the algorithm proposed by Santos Silva's (2001) is used because it tends to be much faster. Using powell can be useful if convergence cannot be achieved with the default options or as a check (especially useful if censoring is severe).

tolerance(#) specifies the tolerance for the coefficient vector. Convergence is achieved when the change in the coefficient vector is smaller than #.

iterate(#) specifies the maximum number of iterations that is performed; default is iterate(16000).

nolog suppresses the iterations log.


scls was written by J.M.C. Santos Silva and it is not an official Stata command. For further help and support, please contact Please notice that this software is provided as is, without warranty of any kind, express or implied, including but not limited to the warranties of merchantability, fitness for a particular purpose and noninfringement. In no event shall the author be liable for any claim, damages or other liability, whether in an action of contract, tort or otherwise, arising from, out of or in connection with the software or the use or other dealings in the software.


--------------------------------------------------------------------------- Setup . webuse womenwk . replace wage=0 if wage==.

SCLS with default options . scls wage educ age married children

SCLS using Powell's algorithm . scls wage educ age married children, p

SCLS using Powell's algorithm and Tobit starting values . scls wage educ age married children, p s(tobit)


Saved results

scls saves the following in e():

Scalars e(N) sample size e(N) number of observations effectively used e(obj_func) value of the objective function e(crit) convergence criterion e(rank) number of linearly independent regressors e(iter) number of iterations performed e(converged) 1 if converged; 0 otherwise

Macros e(vcetype) Robust e(title) Symmetrically Censored Least Squares e(marginsok) predictions allowed by margins e(cmd) scls e(predict) program used to implement predict e(properties) b V e(depvar) name of dependent variable

Matrices e(b) coefficient vector e(V) variance-covariance matrix

Functions e(sample) marks estimation sample


Chernozhukov, V. and Hong, H. (2002), Three-Step Censored Quantile Regression and Extramarital Affairs, Journal of American Statistical Association, 97, 872-882. Powell, J. L. (1986), Symmetrically Trimmed Least Squares Estimation for Tobit Models, Econometrica, 54, 1235-1460. Santos Silva, J.M.C. (2001), Influence Diagnostics and Estimation Algorithms for Powell's SCLS, Journal of Business and Economics Statistics, 19, 55-62.