Contents

function [cov, cor, sd, ss] = pregr2 (func, p, x, y, Z, W)

  %  created: 2002/02/05 by Bas Kooijman, modified 2010/05/08
  %

Description

Calculates covariance matrix and standard deviations of parameters in
regression models, like pregr, but for 2 independent variables.

Input

func: character string with name of user-defined function
  f = func (p, x, y) with x: n-vector; y: m-vector; p: k-vector
  f: (n,m)-matrix with model-predictions for dependent variable
  see regr2_NR
p: (2,k) matrix with
   p(1,:) parameter values
   p(2,:) binaries with yes or no conditional values
   all conditional parameters have zero (co)variance
x: (n,1)-vector with first independent variable
y: (m,2)-vector with second independent variable
Z: (n,m)-matrix with dependent variable
W: (n,m)-matrix with weight coefficients

Output

cov: (k,k) matrix with covariances
cor: (k,k) matrix with correlation coefficients
sd: (k,1) matrix with standard deviations
ss: scalar with weighted sum of squared deviations

Remarks

calls nrdregr2, and user-defined function 'func'
The elements in the covariance and correlation matrices equal zero
  for parameters that have code 0 in the second row of the parameter input matrix.
The values are the maximum likelihood estimates in the case of a identially normally distributed scatter distribution.
Therefore, no corrections for bias are made.

Example of use

assuming that function_name, pars, xvalues, yvalues, zdata, and weights are defined properly:
[cov, cor, sd, ss] = pregr2('function_name', pars, xvalues, yvalues, zdata, weights) .

Code

  global nxy n_pars index;

  [np, k] = size(p); % np is number of parameters
  index = 1:np;
  if k>1
    index = index(1 == p(:,2)); % indices of iterated parameters
  end
  n_pars = size(index, 2); %number of parameters that must be iterated
  if (n_pars == 0)
    return; % no parameters to iterate
  end

  [nx, ny] = size(Z); % nx,ny is number of values of dependent variables
  nxy = nx * ny; % total number of data points
  if exist('W') ~= 1
    W = ones(nx, ny);
  end

  W = W/ sum(sum(W)); WW = reshape(W, nxy, 1);

  [f, df] = nrdregr2(func, p(:,1), x, y);
  ss = sum(sum((W .* (f - Z) .^ 2))); % weighted sum of squares
  cov = zeros(np,np); cor = cov; % initiate cov and cor matrix
  cov(index, index) = inv(df' * (df .* WW(:,ones(1, n_pars))))/ nxy;
  cov = cov * ss;
  sd = sqrt(diag(cov)); % standard deviations
  cor(index, index) = cov(index, index) ./ (sd(index) * sd(index)');

Published with MATLAB® 7.14