 Usage:  {m,b,const} = backfit(t,y,h,loc,kern{,opt})  
 
 Input:

  t                      n x p matrix, the observed continuous 
                         explanatory variable 
                         
  y                      n x 1 matrix, the observed response variable 
                         
  h                      p x 1 vector or scalar, the bandwidth if loc>-1, 
                         else the second parameter of knn. 
                         
  loc                    {-1,0,1,2}, if loc>-1, the degree of the chosen 
                         local polynomial smoother else the knn is chosen. 
                         
  kern                   string, the kernel to be used. 
                         If loc=-1 it has no meaning. 
                         
  opt.x                  n x d matrix, optional, the observed discrete 
                         explanatory variables (linear part) 
                         
  opt.miter              integer, maximal number of iterations. The default 
                         is 50. 
                         
  opt.cnv                integer, convergence criterion. The default is 1.0e-6. 
                         
  opt.shf                integer, (show-how-far) if exists and =1, an output 
                         is produced which indicates how the iteration 
                         is going on (additive function / point of estimation / 
                         number of iteration). 
                         
 Output:

  m                      n x pp matrix, where pp is p*(loc+1). The estimates 
                         of the additive components are given in column 1 to p, 
                         the first derivatives in column (p+1) to (2p) and 
                         the second derivatives in (2p+1) to (3p). 
                         
  b                      d x 1 vector, parameter estimate of the linear part 
                         
  const                  scalar, estimate of the constant 
                         
--------------------------------------------------------------
(C) MD*TECH Method and Data Technologies, 17.8.2000
