robustSE¶

Purpose¶

Procedure to compute the Huber-White heteroscedastic robust standard errors. The procedure uses the “sandwich” variance-covariance estimator with a small sample correction of \((n)/(n-1)\).

Format¶

vce_robust = robustSE(x, resid[, const, verbose, var_names])¶

vce_robust = robustSE(dataset, formula, resid[, const, verbose, var_names])

vce_robust = robustSE(dataframe, formula, resid[, const, verbose, var_names])

Parameters:

x (NxK matrix) – independent regression variables, should not include a constant.
dataset (string) – name of dataset.
formula (string) – formula string of the independent variables. E.g "X1 + X2", X1 and X2 are names of independent variables;
resid (Nx1 vector) – regression residuals.
const (scalar) – Optional input, indicator variable for including a constant. 1 for including a constant, 0 for no constant. Default = 1.
verbose (scalar) – Optional input, 1 to print results, 0 for no printing. Default = 1.
var_names (string array) – Optional input, variable names. Default = X1, X2, ..., XK.

Returns:

vce_robust (KxK matrix) – Huber-White heteroscedastic robust variance-covariance matrix.

Examples¶

new;

// Load data from 'auto' dataset
fname = getGAUSSHome("/examples/auto.dat");
data = loadd(fname);

// Transform data
mpg = data[., 3];
weight = data[., 7];
foreign = data[., 12];

// Set independent and dependent variables
y = ((1/mpg) ./ weight) * 100 * 1000;
x = foreign;

// Control structure
struct olsmtControl o_ctl;
o_ctl = olsmtControlCreate();

// Turn on to estimate residuals
o_ctl.res = 1;

// Declare output structure
struct olsmtOut o_out;

// Run initial ols
o_out = olsmt("", y, x, o_ctl);

This estimates the OLS regression and finds the i.i.d. standard errors:

Valid cases:                    74      Dependent variable:                   Y
Missing cases:                   0      Deletion method:                   None
Total SS:                    4.298      Degrees of freedom:                  72
R-squared:                   0.218      Rbar-squared:                     0.207
Residual SS:                 3.361      Std error of est:                 0.216
F(1,72):                    20.068      Probability of F:                 0.000
Durbin-Watson:               2.455

                         Standard                 Prob   Standardized  Cor with
Variable     Estimate      Error      t-value     >|t|     Estimate    Dep Var
-------------------------------------------------------------------------------

CONSTANT     1.609004    0.029961   53.703680     0.000       ---         ---
X1           0.246153    0.054949    4.479678     0.000    0.466867    0.466867

Calling robustSE() estimates the heteroscedastic-robust standard errors:

// Find robust standard errors
vce_robust = robustSE(x, o_out.resid);

The results:

Total observations:                                           74
Number of variables:                                           2

        VARIABLE        Robust SE
-------------------------------------

        CONSTANT         0.023453
              X1         0.067924
-------------------------------------