The posterior probability of each instant being a change point was proposed by Loschi & Cruz (2005) as a measure of evidence that a change occurred in that instant. In this paper, the efficiency of such a measure is evaluated and it is compared with the measure proposed by Hartigan (1990). Poisson data sequences, with and without changes, are considered. The method is evaluated in the presence of both structural changes and atypical observations. The proposed method over performs Hartigan's proposal. A series of violent crimes data from Belo Horizonte is analyzed by the proposed method. It was observed that the rate of violent crimes changes frequently throughout the time.
Gibbs sampling; product partition model; structural change; atypical observation