thanks for the detailed analysis. One additional thing to consider is that in the changepoint package cpt.mean assumes that the variance is constant and equal to 1. If this is not the case then it will end up detecting many many changepoints as there will likely be many instances where a datapoint is more than the previous by 3 stddevn or more. Therefore it is best to use cpt.meanvar instead which makes no assumption about variance. or if you must use cpt.mean you could use it with a scaled version of the data that has variance =1.

I am the first one to comment here?

Highly useful comparison study. Thanks a lot! -- I started with strucchange and then got a hint to changepoint (which seems to be considerably less well documented). This blog saved me loads of work comparing these two and possible more options.

Have you ever thought about how to approach things if you want to allow for a break point in one of the coefficients of a regressions, but not in all of them at the same time? The latter seems to be unavoidable in strucchange, but I would like to have more flexibility.

please i need to ask u if there is any reference to know more about change point with simple way , 2nd question is this comparison in timeseries only right..?