Ignore changes in variance

[Neverstop7]

Hello!
It seems that EnvCpt R package uses internally cpt.meanvar() function from the changepoint R package. The function cpt.meanvar() detects both changes in mean and variance. Suppose that I am only interested in changes in mean. How can I use the envcpt() function to ignore changes in variance?

library(EnvCpt)
set.seed(123)
y = c(rnorm(50,100,sd = 5),rnorm(50,100,sd = 50))
plot(y)
out = envcpt(y)
which.min(BIC(out))
plot(out$meancpt,type="p")

It would be nice if envcpt would detect no changepoints since the mean has not changed.
Thank you in advance!

[rkillick]

Ah, the age old question of what model we should use. You can directly use the cpt.mean() function from the changepoint package if you don't want variance changes. However, the reason why we use cpt.meanvar() is because cpt.mean() is sensitive to the range of the data and (currently) assumes a variance of 1. Thus if the variance is larger than 1, lots of mean changes will be detected when no mean shift is present, just larger variance. Conversely if the variance is considerably smaller than 1, large mean shifts may not be detected. To avoid the thorny issue of how to estimate the global variance in the presence of mean shifts and people blindly applying cpt.mean() to data with variance shifts (which will induce mean shifts which aren't really there), we just use cpt.meanvar().
I do have a method here:
https://onlinelibrary.wiley.com/doi/full/10.1002/sta4.351
which might do the trick but I haven't had chance to put it into a package yet. Another enhancement to add to the list!

Thank you so much for all your comments!