 Usage:  myfit = gplmbootstraptest(code,x,t,y,h,nboot{,opt})  

 

 Input:



  code                   text string, the short code for the model (e.g. 

                         "bilo" for logit or "noid" for ordinary PLM). 

                         

  x                      n x p matrix, the discrete predictor variables. 

                         

  t                      n x q matrix, the continuous predictor variables. 

                         

  y                      n x 1 vector, the response variables. 

                         

  h                      q x 1 vector, the bandwidth vector. 

                         

  nboot                  integer, number of bootstrap replications. 

                         If nboot<=0, the test is performed using the 

                         asymptotic normal distribution of the test 

                         statistics. 

                         

  opt                    optional, a list with optional input. The macro 

                         "gplmopt" can be used to set up this parameter. 

                         The order of the list elements is not important. 

                         Parameters which are not given are replaced by 

                         defaults (see below). 

                         

  opt.b0                 p x 1 vector, the initial coefficients. If not 

                         given, all coefficients are put =0 initially. 

                         

  opt.m0                 n x 1 vector, the initial values for the nonparametric 

                         part. If not given, a default is used. 

                         

  opt.tg                 ng x 1 vector, a grid for continuous part. If tg is 

                         given, the nonparametric function will also be 

                         computed on this grid. 

                         

  opt.tdesign            n x r matrix, design for parametric fit for 

                         m(t). This allows to test e.g. quadratic or 

                         cubic functions against m(t). 

                         If not given a linear function (incl. constant) 

                         will be tested by using the design matrix(n)~t. 

                         

  opt.weights            string, type of observation weights. Can be 

                         "frequency" for replication counts, or "prior" 

                         (default) for prior weights in weighted regression. 

                         

  opt.wx                 scalar or n x 1 vector, frequency or prior 

                         weights. If not given, set to 1. 

                         

  opt.wt                 n x 1 vector, weights for t (trimming factors). 

                         If not given, set to 1. 

                         

  opt.wc                 n x 1 vector, weights for convergence criterion, 

                         w.r.t. m(t) only. If not given, opt.wt is used. 

                         

  opt.wr                 n x 1 vector, weights for test statistics. 

                         If not given, set to 1. 

                         

  opt.off                scalar or n x 1 vector, offset. Can be used for 

                         constrained estimation. If not given, set to 0. 

                         

  opt.meth               integer, if -1, a backfitting is performed, 

                         if 1 a profile likelihood method is used, and 

                         0 a simple profile likelihood is used. 

                         The default is 0. 

                         

  opt.fscor              integer, if exists and =1, a Fisher scoring is 

                         performed (instead of the default Newton-Raphson 

                         procedure). This parameter is ignored for 

                         canonical links. 

                         

  opt.shf                integer, if exists and =1, some output is produced 

                         which indicates how the iteration is going on. 

                         

  opt.nosort             integer, if exists and =1, the continuous variables 

                         t and the grid tg are assumed to be sorted by the 

                         1st column. Sorting is required by the algorithm, 

                         hence you should switch if off only when the data 

                         are already sorted. 

                         

  opt.miter              integer, maximal number of iterations. The default 

                         is 10. 

                         

  opt.cnv                integer, convergence criterion. The default is 0.0001. 

                         

  opt.pow                scalar, power for power link. If not given, 

                         set to 0. 

                         

  opt.nbk                scalar, extra parameter k for negative binomial 

                         distribution. If not given, set to 1 (geometric 

                         distribution). 

                         

 Output:



  myfit.b                k x 1 vector, estimated coefficients 

                         

  myfit.bv               k x k matrix, estimated covariance matrix for b 

                         

  myfit.m                n x 1 vector, estimated nonparametric part 

                         

  myfit.mg               ngx 1 vector, estimated nonparametric part on grid 

                         

  myfit.rr               3 x 1 vector, 3 test statistics according 

                         to Haerdle/Mammen/Mueller 

                         

  myfit.alpha            3 x 1 vector, significance level for rejection of 

                         the parametric hypothesis (for each of the three test 

                         statisctics). 

                         

  myfit.stat             list with the following statistics: 

                         

  myfit.stat.deviance    deviance, 

                         

  myfit.stat.pearson     generalized pearson's chi^2, 

                         

  myfit.stat.loglik      log-likelihood, 

                         

  myfit.stat.r2          pseudo R^2, 

                         

  myfit.stat.it          2 x 1 vector, number of iterations needed in 

                         semiparametric and biased parametric fit 

                         

  myfit.stat.ret         scalar, return code: 

                         0 o.k., 

                         1 maximal number of iterations reached 

                         in estimation (if applicable), 

                         -1 missing values have been encountered 

                         in estimation, 

                         -2 missing values in test statistics encountered. 

                         -3 missing values in bootstrap encountered. 

                         

  myfit.stat.rrboot      nboot x 3 matrix, values of the bootstrap test 

                         statistics (if applicable). 

                         

--------------------------------------------------------------

(C) MD*TECH Method and Data Technologies, 21.9.2000

