Always keep one untouched, original version of your dataset. Data cleaning and analysis processes can sometimes "break" data unintentionally. Having an original copy means you can go back, compare, or restart without losing your baseline.
Losing your data can be catastrophic for a dissertation, so robust storage and backup are essential:
Backup Strategy (3-2-1 Rule): Keep at least 3 copies of your data on 2 different types of media, with 1 copy stored off-site. Set up automated backups if you can.