How it works
An instrument's resolution determines the terminal digit: for data reported to two decimals, the last digit is dominated by rounding noise and should be near-uniform. Any systematic terminal-digit preference deserves questioning:
- 0/5 preference — a classic fingerprint of invented data (“12.5” and “30.0” come to mind far more easily than “17.3” or “23.8”)
- Even-digit preference — a trace of some artificial generation mechanisms
- Certain digits entirely missing — data may have been truncated or filtered
The test: a chi-square uniformity test on the 0-9 distribution (df = 9), plus a report of the combined 0+5 frequency relative to expectation (20%).
A worked example
A paper's table contains 100 measurements. Real instrument data would put about 10 observations on each terminal digit; if 0 and 5 together appear 38 times (expected 20, 1.9×), the chi-square test is significant — consistent with someone inventing numbers by feel, and worth asking the authors for the original records.
Implementation details in this tool
- Terminal digits are read verbatim from the input: the last digit of “12.30” is 0, with no floating-point conversion — avoiding contamination from scientific notation or float representation.
- Supports 0.x decimals, negatives, and scientific notation (mantissa only).
- Small samples (< 100) give low expected frequencies — the tool advises caution.
Applicability
- Data produced by instruments/measurements (terminal digits driven by genuine noise)
- Count data (integers) have naturally non-uniform terminal digits — not applicable
- Summary statistics already rounded to the same precision don't apply (raw observations needed)
- Terminal-digit preference can also come from reading habits (e.g., favoring round tens) — a data-quality issue rather than fabrication
Sources
Mosimann JE, Wiseman CV, Edelman RE (1995). Data fabrication: Can people generate random digits? Quality and Quantity, 29(3), 281–299.
Related: Hartgerink CHJ et al. (2019), meta-analysis of statistical tests detecting data fabrication.
⚠️ Important: Terminal digit analysis and every method on this site are screening aids. No single anomaly is sufficient to determine data fabrication. Final determinations must be made by competent institutions through due process.