Data management and sharing

Data Management

  • All research-related materials (including data and code) should be stored on a volume that is automatically backed up to a cloud storage server, such as Google Drive (which provides free storage via your Stanford login).

    • Stanford also provides access to Crashplan backups for faculty and staff which automatically backs up laptops and computers.

  • Neuroimaging datasets should be backed up on TACC Corral.

  • Every dataset should be organized with the goal that another researcher could take the dataset and immediately understand its content without the need to ask questions of the dataset owner.

  • All imaging data should be organized using the BIDS format.

  • For non-imaging data, files and folders should be named and organized using the Psych-DS format when possible.

  • Variable names should be as expressive as possible, with automated parsing in mind. The structure of variable names should follow the key-value schema used in BIDS (with key and value separated by a dash, and key-value pairs separated by underscores). All embedded numbers should be zero-padded.

    • Examples:

    • for item 5 on the Barratt Impulsiveness Scale, the variable name might be “survey-BIS_item-005”

  • All datasets should be accompanied by a data dictionary that specifies the meaning of each variable.

    • This should preferably be stored as a JSON file, with variable names as dictionary keys to support automated parsing.

Data sharing

  • All data generated within our lab will be shared upon publication at the latest, and preferably upon preprint submission.

  • In some cases, we may work with datasets that we are not able to share in full, due to data use agreements or legal restrictions. In this case, we will push to share at least the processed data necessary to run the primary statistical analyses.

  • Data will be shared through platforms that allow snapshotting and generation of a DOI:

    • OpenNeuro for neuroimaging data

    • For behavioral data, repositories include Zenodo (preferred), OSF, Stanford Digital Repository, or Dryad.

    • Large data files should not be committed to Github repositories. Small data files may be hosted on github, but it should not be the primary sharing platform.

  • A snapshot should be shared that matches exactly the analyses in the publication, and which can be analyzed directly using the shared analysis code.

  • All data will be shared with an explicit data use agreement (aka “license”)

    • We prefer the minimally restrictive license possible, preferably CC0

  • All data should be shared with an explicit description of how they should be cited.

  • When using shared data, lab members should ensure that they properly cite the data source in any publications using the language recommended by the data owner.