Documenting Data

What are the critical components of data documentation?

• Who collected the data
• Why the data was collected
• What the data is about
• When the data was collected
• Where the data was collected
• How the data was generated/collected
• Structure of the data
• Formatting decisions in the data
• Data validation/quality control
• How the data can be reused/license
• Suggested data analysis methods
• Measurement instruments used

Documentation is Project Dependent

Project 1: Building a Shoe Print Wear Database

• 150 pairs of shoes
• 2 brands of shoes
• several sizes for each brand
• step counters used on the shoes
• questionnaires measuring activities
• wearer weight/height/gait

Shoe Measurements

Initial measurement period + 2-3 additional measurement periods (~6 weeks between)

• Photos of shoe soles
• Digital shoe sole prints
• Powder prints
  • Film
  • Paper
  • Vinyl flooring
• 3d scans of shoe soles

Measurements taken in the lab by research assistants.

Important Documentation?

• See all of the metadata here

• Probably should have included which research assistant was wearing the shoe, how much they weighed, their gait/height/etc., and so on.

Whoops.

Documentation is Project Dependent

Project 2: Wire cuts

• Goal is to estimate the length of sharp surfaces on all wire-cutting tool in peoples’ homes
• General survey with instructions for measuring each type of tool
• Collected data is a list of
  • tool types,
  • # blades on the tool,
  • # cutting surfaces,
  • blade length, and
  • # of that type of tool
• Estimates are generated by adding up total length of cutting surfaces

Important Documentation?

• What tools would the participant use to cut a wire typically?

• What tools would they use in a pinch?

• Is participant a professional craftsperson? hobbyist? not handy at all?

• How was blade length measured?

• … I haven’t run this study yet. But I got over 980 cm of cutting surface in my garage/house and my dad got over 2200 cm.

Codebooks

Basic documentation that contains:

• Variable name in the code
• Long-form description of what was measured
• Units of measurement
• Acceptable values
• Values used to indicate missingness, refusal to respond, etc.
• Additional notes that may be relevant

Very common for government data - CDC codebooks are intense.

Data Doc Influences Analysis

• Experimental design
• Randomization
• Sampling strategy
• Random effects
• Transformations of collected data
• Sources of measurement error

Data Documentation

Documentation is a love letter that you write to your future self

Damian Conway

Additional Resources

• DDI Alliance - probably overkill but in a good way

• Data Librarians are amazing to work with