Shor answer: no, you don't. We recommend it as a best practice.
Still, some customers are afraid we might accidentaly remove a dataset when we shouldn't—which is understandable. We're sharing this article to provide the necessary transparency so you can decide what's the best approach for your case.
Here's how we select which datasets to remove when it's necessary
We basicallly filter only datasets that starts with "earXX" (XX being the sequential ID of your data source). There are contigencies in the code to prevent any other dataset from being deleted.
Why do delete datasets sometimes
One of our integrations (App Store, Google Play, etc) changed their schema so we need to change ours too.
A customer asks for their data to be redownloaded from scratch
In rare cases, to remove duplicates or gaps in the data