Variation in reading error in P times for explosions with body-wave magnitude

The differences between true travel-times of P and times predicted from travel-time tables (path effects) can be estimated for groups of closely spaced explosions with known hypocentres and origin times, if the onsets are observed at large signal-to-noise ratios (SNR) and read by analysts. Reading error can also be estimated and is usually assumed to be normally distributed with zero mean. Two experiments have been carried out to look at how reading error in P times from explosions varies with magnitude - taken as a measure of SNR - when read by analysts and by automatic systems. Although at low magnitudes there is some evidence of analyst readings being biased late, the largest variation in reading error with magnitude is found for automatic systems. The results show just how difficult it can be to estimate path effects free from observational bias, at least using bulletin data. The current programme to estimate path effects to improve epicentre location for verification of the Comprehensive Test Ban needs to include checks to ensure that apparent variations in path effects with location, are not due to bias from systematic reading error.