Variances of regression estimates of expected accident frequencies.

Multiple regression techniques have been used at the RRL to relate accident rates on various classes of road in Britain in more than 307-day periods to a number of variables descriptive of weather and traffic conditions. It is shown that if certain plausible assumptions of statistical independence are made and a suitable model is assumed for errors in the vehicle-mileage estimates, the variance of the estimated number of accidents can be expressed in terms of available data.