I agree with opt-in policies. The only ethical case for opt-out is if you obtain informed consent ahead of the participation of the user which is de facto opt-in.

I don't think opt-in is necessarily required ethically but having clear information available about the entire extent of the use of the data and the ability to remove it should be practiced. In the same way I might conduct a human trial on someone who agrees to engage as a participant, but I'm not giving them all of the information at the onset so as to not bias their response. Though I do provide a full summary once I'm finished collecting the data (i.e. the study is over).