Data Dictionaries

ID 529: Data Management and Analytic Workflows in R

Dean Marengi | Tuesday, January 10^th 2023

Motivation

• We’ve learned a bit about:
  • Git and GitHub for code version control and management
  • Reproducible data analysis workflows

• But have not discussed data dictionaries and their role in reproducible research
  • Facilitate collaboration with other researchers
  • Provide clear and detailed information about the data being used
    • Data sources and considerations
    • Structure and format
    • Variable definitions and interpretation of their values

Learning objectives

• Understand the importance of data dictionaries for data analysis
  • Ensuring consistency of data, and how the data are used
  • Facilitating research collaboration

• Learn about the common components of data dictionaries
  • Variable names and definitions
  • Data types
  • Value labels, and other attributes
  • Missing data codes

• Discuss creating and maintaining data dictionaries
  • Well-described entries
  • Keeping data dictionaries up-to date

What are data dictionaries?

• Data dictionaries
  • Provide key information and metadata about variables included in a dataset
  • Serve as an important reference guide for data analysts
  • The structure, format, and contents will vary depending on the project

• Core elements typically include
  • A description of the study or data source(s)
  • Variable names, definitions (and units, where applicable)
  • Data types
  • Value labels
  • Calculations performed (where applicable)
  • Other details relevant to a specific project

Dataset descriptions

• Dataset descriptions help to contextualize the data they contain

• Descriptions should typically include
  • An overview of the project or data origins
  • Why the data were collected
  • What data was collected
  • Who collected the data
  • When the data were collected
  • Data collection methodology and related considerations
    • e.g., data quality issues
  • Other details relevant for specific projects

Variable entries

• Variable entries provide specific information pertaining to each variable in the data

• Entries should include
• Concise but descriptive variable names
  • Accurately reflect the underlying data
• Detailed descriptions of what the variables are
• Data type
• Characteristics such as:
  • Units
  • Factor levels
  • Range of values
  • Transformations

• You may also want to include:
  • Links to relevant documentation
  • Notes on data quality issues
  • Other relevant information helpful for use and interpretation
  • Data type and format of the variable, including any missing data codes

Examples of variable entries

Bad Entry

• Variable Name: Age
• Definition: Age of participant
• Data Type: Number
• Notes: N/A

Good Entry

• Variable Name: Age
• Definition: The age of the patient at the time of the study, measured in years
• Data Type: Integer
• Notes: Age data was collected via self-report on the baseline questionnaire and verified with a government-issued ID.

Bad Entry

• Variable Name: smk
• Definition: Smokes
• Data Type: Yes/No
• Notes: N/A

Good Entry

• Variable Name: Smoking status
• Definition: Whether or not the patient currently smokes cigarettes
• Data Type: Dichotomous (yes=1 | no=0)
• Notes: Smoking status was determined through self-report on the baseline questionnaire and verified at clinical study visits.

Common mistakes

• Insufficient detail
  • Inconsistent and poor variable naming conventions
  • Imprecise or vague definitions
  • Omission of pertinent information about data quality issues

• Inaccuracies
  • Failure to keep current with evolving research
  • Discrepancies with the dataset

• Poor standardization and formatting
  • Difficult for others to understand
  • Inconsistencies make it difficult to maintain

Discussion questions

• What do you look for in a data dictionary? What do you include?

• Have you ever had a negative experience working with poorly documented data?

• Do you follow a specific process when developing your own data dictionaries?

Key takeaways

• Data dictionaries are important tools in research
  • Provide a clear, standardized way of describing the data being used
    • Ensure consistency in how data should be interpreted by investigators
    • Overall, help to promote transparent and reproducible research!

• Data dictionaries should be detailed, consistent, and accurate
  • Include all variables relevant for data analysis and, for each, provide:
    • Variable names
      • Detailed variable descriptions
      • Data types
      • Notes relevant to data collection procedures, calculations, etc.
      • Any other details that will help to interpret the data

Key takeaways (cont.)

• Data dictionaries should be kept up to date over the course of a project
  • Include new and derived variables
  • Document changes in data collection and/or analysis methods