npmc                  package:npmc                  R Documentation

_N_o_n_p_a_r_a_m_e_t_r_i_c _M_u_l_t_i_p_l_e _C_o_m_p_a_r_i_s_o_n_s

_D_e_s_c_r_i_p_t_i_o_n:

     npmc implements the nonparametrical multiple testprocedures
     (Behrens-Fisher- and Steel-type for the all-pairs and many-to-one
     situations) described in the paper "A unified approach to
     Simultaneous Rank Test Procedures in the  Unbalanced One-way
     Layout"  written by Ullrich Munzel and Ludwig Hothorn.

_U_s_a_g_e:

     npmc(dataset, control=NULL, df=2, alpha=0.05)

     ## S3 method for class 'npmc':
     summary(object, type="both", info=TRUE, short=TRUE, corr=FALSE, ...)

_A_r_g_u_m_e_n_t_s:

 dataset: a data-frame with variables  'var' (the response-variable)and
           'class' (containing the class-level). The data may be
          unsorted and unbalanced. The 'summary'-function supports
          'name' and 'description' attributes 

 control: level of control-group for the many-to-one situation or NULL
          for the all-pairs situation (default) 

      df: determines the BF-teststatistics' asymptotic distribution
          function (0 = standard-normal approximation, 1 = simple
          t-approximation, 2 = Satterthwaite t-approximation (default)) 

   alpha: Level for the (1-alpha) confidence-intervals 

  object: An object of type 'npmc' 

    type: You may select either "BF" or "Steel". All other values
          extract information for both types 

    info: Prints info about the procedure and the data 

   short: Only prints the most relevant items of the test-results 

    corr: Prints the correlation-matrices 

     ...: further arguments to be passed to or from methods.

_D_e_t_a_i_l_s:

     npmc performs nonparametrical multiple testprocedures
     (Behrens-Fisher- and Steel-type for the all-pairs and many-to-one
     situations) and computes the simultaneous (1-alpha) confidence
     limits for the relative effects.

     The one-sided tests reject if group with smaller index has larger
     values due to the calculation of the relative effect-estimators.

_V_a_l_u_e:

     A list of several other values and structures with subvalues 

    info: group.index: an integer number to identify the group

          class.level: the class-level of this group

          nobs: the number of observations in this group 

    corr: BF: The Behrens-Fisher-type correlation-matrix

          Steel: The Steel-type correlation-matrix

          The matrices have attributes 'adjusted' which indicate if 
          negative eigenvalues were changed to zero (see
          'Notes'-section). 

    test: BF / Steel: Two lists containing the test-results and some
          other characteristics of the multiple nonparametric
          Behrens-Fisher- and  Steel-type testprocedures

          cmp: names the compared groups ('a-b')

          gn: the sum of both sample-sizes

          effect: the relative effect-estimator

          variance: the variance-estimator

          std: the standard-deviation

          statistic: the test-statistic

          p-value 1s: the 1-sided p-value

          p-value 2s: the 1-sided p-value

          zero: TRUE if zero variances occured and were substituted by
          0.0000001 

 control: The factor-level of the control group (NULL for
          allpairs-comparison) 

df.method: The original df-paramter 

      df: The estimated degrees of freedom for the multivariate
          t-distribution or 0 if the standard normal distribution was
          used for approximating the teststatistics' distribution
          (depending on the df-parameter). 

_N_o_t_e:

     This function requires the 'mvtnorm' package to calculate the
     p-values for the test-statistics. If this package is not available
     on your system, the result will contain NA's as p-values.

     The functions in the 'mvtnorm' package seem to use randomized
     values for  integral calculations, so the results from npmc
     concerning p-values and confidence-intervals differ from call to
     call and can be recognized only as an approximative solution.

     If a correlation-matrix has negative eigenvalues, they are
     replaced by 0 as recommended by the paper. The matrix's 'adjusted'
     attribute indicates  whether or not the matrix was forced to be
     positive semidefinite.

_A_u_t_h_o_r(_s):

     Joerg Helms

_R_e_f_e_r_e_n_c_e_s:

     "A unified approach to Simultaneous Rank Test Procedures in the
     Unbalanced One-way Layout" written by Ullrich Munzel and Ludwig
     Hothorn.

_E_x_a_m_p_l_e_s:

     data(brain)
     summary(npmc(brain), type="BF")

