Crash reports usually yes (but how to prevent receiving personal data from stack dumps?), features selection probably not. The reason is that you don't get data from the people who disable telemetry and they could use different features. You must find them, poll them and merge the results with what telemetry gives you. If you don't do that, telemetry is detrimental to those people.
Yeah, such statistics will always be skewed to some extent.
One could argue that telemetry _is_ the poll. If a user decides to disable feature-usage-tracking telemetry, and then realize they’d prefer to have a voice regarding feature selection, one possible solution would be for them to re-enable it.
If they absolutely won’t opt-in, they’re always free to email developers to tell them about their use case – a pretty powerful way to make yourself heard.
