What data should you preserve for the long term?
It is important to retain:
Data which support primary research findings (data needed to validate/reproduce the findings)
Data that can be reused or has long-term value (uniqueness of the data, reference data, meta-analyses)
Data subject to legal requirements such as clinical trials or engineering equipment test data (quality control) or journal/funder policies
Data that is socially or culturally relevant
Data that can be used for learning/teaching purposes
It can also be important to retain:
Raw data, pending on how complicate it is to generate this data anew
Outputs from models and simulations (would rerunning produce the same output anyway?)
Software used to process research data (scripts, simulations, packages etc)
Large datasets, if the importance outweighs the preservation costs
What is not helpful to preserve?
Data without sufficient metadata, that is not reusable, or data from experiments that were not carried out correctly.
If data is not preserved for the long term, these selection processes should be transparent and not dependent on solely one individual making these decisions.