D. Dataset Integrity
Dataset integrity means the data can be trusted, checked, and reproduced.
Note
A weak dataset can make a technically correct regression meaningless.
Source and Collection
Raw and Cleaned Files
Variable Documentation
Data Quality Checks
Constructed Variables
Warning Signs of Weak or Fake Datasets
Warning
These warning signs do not automatically prove a dataset is fake, but they require explanation.
- too perfect data
- no missing values when missingness is expected
- no source link
- unrealistic variable distributions
- repeated values without explanation
- no collection date
- no raw data file
- numbers that cannot be verified
- variable names with no definitions
- cleaned data that cannot be reproduced
- categories that appear inconsistent or misspelled
- values outside realistic economic ranges
Minimum Integrity Statement
Every empirical project should include a short statement like this:
The raw dataset was preserved separately from the cleaned dataset. The unit of observation is [unit]. Missing values, duplicates, categorical levels, and outliers were checked before estimation. Constructed variables were created using documented formulas. The data source is acknowledged in the report.