Jeffrey -

It seems to me - not just as a 'packrat' myself - that saving the raw data forever, if possible, is highly desirable.   If this is not possible, then one may need to prioritize the importance of your studies, to see which raw data to save.  If there are no confidentiality problems - and I hope there are fewer such obstacles in the future - then perhaps putting data on ResearchGate as 'datasets,' which would definitely fit under that category, would be helpful for everyone. There have been multiple occasions on ResearchGate where people have asked questions about summary data they have found and they want to know more about the data that produced that report. But generally speaking, if you have statistics on a dataset, but not the original dataset, then there is no way to know many details which may be desired about the original dataset, as the possible datasets that could have produced those results would be numerous or even  infinite.

If you or anyone else wants to re-analyse a dataset in the future, the data need to be saved. The decision not to save it may be based on cost and/or other resource-based criteria,  but when making such a decision, one should also consider the cost of performing the study again.  It may often be a good idea to repeat a study for confirmation, but if results differ, and you do not have the original data, how will you determine why that happened?

For Official Statistics from a statistical agency, there are establishment and other surveys which - unlike experiments indicated above - are repeated periodically.  But in the case of Official Statistics, there are times when one may desire a reanalysis or confirmation of what occurred at a given point in history, and again the microdata are needed, and other information - metadata - may need to be saved as well.  - I assume that experiments should have similar additional information that needs to be saved.   

- All data need to be saved for as long as possible, depending upon resources and importance and/or potential importance.   

Cheers - Jim

PS - As I see on your profile you are an archivist, let me tell you that I believe that activity to be very important.  

Thank you very much Jim for your thoughtful response.  The danger (as I see it) with retaining project raw data forever, is the potential Pandora Box effect if the company is ever audited by the FDA (not to mention the cost issue of offsite storage).  For other companies I have worked for in the past, having an established Archive Retention Schedule seems to be the best course of action for compliance.  Over the past several years, the FDA has imposed citations for pharma companies and laboratories who maintain a "keep everything" approach, noting that this does not constitute a valid Records Management policy.  I do agree wholeheartedly with you that every project must be evaluated for importance, not to mention the cost factor of having to recreate project data to be reanalyzed in the future, however this does not negate the fact that retaining project raw data for an extended period of time (beyond it's usefulness) opens the company up to potential  "discovery" issues when the FDA shows up on your doorstep.  Of all my job functions as an Archivist, one of the greatest (and one that I take very seriously) is the necessity to advise management regarding potential compliance issues.