Help System (web edition)

REGDIAG <data>,L
estimates the parameters of a linear regression model by orthogonalizing
the design matrix X.
These procedure is more accurate than the classical approach of LINREG
through moment matrix (correlations).
The model is specified by activating variables with Y and X's. The
constant term is omitted by setting CONSTANT=0.
Furthermore the following masks for special output variables may be used:
  R is the residual variable,
  P is the variable for predicted values,
  H is the diagonal of the hat matrix X*INV(X'*X)*X',
  S is the variable for (externally) Studentized residuals,
  C is the variable for (signed square roots of) Cook's distances.

These output variables provide a basis for regression diagnostics.
See Belsley,Kuh&Welsch (1980): Regression Diagnostics, for example.
  M = Matrix files created by REGDIAG 
  R = More information on regression analysis 
  S = MAT SOLVE (is a similar procedure in the matrix interpreter) 
  C = Confidence intervals in simple linear regression (in PLOT) 
  D = Durbin-Watson test statistics and its P-values