Via NZPundit via Professor Bunyip via Technorati, I’ve tracked down the study by the New Zealand Land Transport Safety Authority (PDF file) which formed the basis of the estimate that higher speed limits in the US cost 1900 lives.

The study looks at states which had a highway limit of 65mph (about 110 km/h) in 1995, at which time this was the maximum allowed under the National Maximum Speed Limit. The NMSL was repealed in December 1995 and a lot of states increased the limit to 70 or 75. Over the next few years, fatality rates rose, on average, in the states that raised their limits and fell in the states that did not. The LTSA uses a statistical model to check that this wasn’t merely a random fluctuation and to correct for the most important possible source of spurious correlation, an increase in vehicle miles travelled. They find that the increase in fatalities was statistically significant – given that their estimates imply a loss of 1900 lives, it was obviously significant in the ordinary usage of the term.

NZPundit admits to a bit of confusion about the paper and in particular, the divergence between statistical estimates and predictions. To clarify this point, the estimates are those based on the actual outcomes observed in the states in question. The predictions are derived from previous studies that estimated the impact of a given increase in average speeds, and are higher than those obtained here. The suggested explanation is that, since lots of people were speeding anyway, an increase in speed limits does not translate to an equal increase in average speeds, though average speeds do increase.