Well for starters, if you are NIH or NSF funded, they have data storage requirements you must meet. So usually this involves something like tape backups in two

Well for starters, if you are NIH or NSF funded, they have data storage requirements you must meet. So usually this involves something like tape backups in two locations. The other is for reproducibility - typically you need to preserve lots of in-between steps for peer review and proving that you aren't making things up. Some intermediary data is wiped out, but usually only if it can be quickly and easily regenerated.