Regression and Uncertainty of a Straight-Line for Measurements of x and y Variables with All Correlations

The work continues the series of publications on the estimation of the parameters of the equation and the limits of the uncertainty band of the straight-line y = ax + b fitted to the measurement results of both coordinates of the tested points with the use of the linear regression method. A general case was considered when these coordinates have different uncertainties and there are all possible autocorrelations and cross-correlations. Description of matrix equations was used. The results of the coordinate measurements are presented as elements of the X and Y vectors. The propagation of their uncertainty was described by the UZ covariance matrix with four component matrices, i.e., UX and UY – for the uncertainties and autocorrelations of X and of Y, and UXY and its transposition U – for the cross-correlations. The equation of a straight line and of the borders of its uncertainty band are given. Obtained them for the function of parameters a and b satisfying the so-called total criterion WTLS, i.e., the minimum sum of squared distances of points from the straight line weighted by the reciprocal of the coordinate uncertainty. When the coordinates of different points are not correlated, the simplified criterion WLS is used. The directions of projecting the points result from the minimization of the function describing the criterion. In the general case, there is only a numerical solution. This is illustrated by an example, in which the parameters a and b of the straight line were determined numerically from the enlarged fragments of the graph of the criterion function around its minimum. The conditions for the uncertainty and correlation of coordinates of points required to obtain an analytical solution and its example are also given.