A method is presented which, using as input time series of data that come approximately out of Poisson distributions, tests whether the last number compared to the penultimate one differs so much that one can speak of a real change. The method also judges whether the period before the last number shows a linear trend (increasing, horizontal or decreasing). Only if such a trend is found, the most recent observed value will be compared to the expected value based on the trend. All comparisons take into account the degree of noteworthiness as well as statistical significance. A noteworthy change is defined as one greater than a chosen percentage. The method has been developed to aid the analysis of time series of road accidents. (Author/publisher)
Abstract