Critique of successive differencing method of regression analysis
Various regression techniques have been proposed to calculate barometric coefficients from time-varying cosmic rays recorded by neutron monitors. One of these, the successive differencing method, was carefully examined by applying the technique to simulated time-dependent and time-independent cosmic-ray data and also by investigating the mathematical foundations of the technique. When applying the successive differencing method to time-independent data, the slope and slope error values were found to be dependent on the order in which the data appeared. By ordering the data from highest to lowest pressure, the slope error increased markedly. This falsification of the slope can be shown to be due to improper weighting of the differenced data. Calculations have also been performed to show that the successive differencing method, when applied to intensities containing a linear time dependence, produces a regression slope which retains a time dependence proportional to the difference between the first and last pressure values. The falsification of the slope due to improper weighting of the differenced data can be very much larger than the maximum effect of any reasonable time variation of the nucleonic component.