I want to perform a changepoint analysis with the help of the PELT method: Killick & al, Killick & Eckley I assume that the data are normally distributed before and after the changepoint. After reading these two papers I still do not know whether the assumption of equal variance before and after changepoint is required when using the PELT method ?
Rebecca Killick herself just answers the question (I sent an email to her). I copy her answer below:
When using the cpt.mean function with dist='Normal', it is assumed that the variance is constant across the process. If the variances before and after the changepoint are different then the cpt.meanvar function should be used. This fits a model with changing mean and variance and also has the dist='Normal' functionality.