Neural Sync Active
Learning Objectives
Registry Synced
Learning Objectives
424 words
2 min read
Reading compass
Now · 1. Benford's Law
Learning Objectives
- Understand Benford's Law and its application
- Apply Benford's analysis to detect anomalies in financial data
- Use statistical tests for fraud detection
- Basic statistics
- Understanding of financial data
1. Benford's Law
Intuition: In naturally occurring numerical datasets, the first digit follows a predictable distribution rather than being uniform. Small digits (1, 2, 3) appear more often than large digits (7, 8, 9).
Expected Distribution of First Digit d: P(d) = log10(1 + 1/d)
| Digit | Probability |
|---|---|
| 1 | 30.1% |
| 2 | 17.6% |
| 3 | 12.5% |
| 4 | 9.7% |
| 5 | 7.9% |
| 6 | 6.7% |
| 7 | 5.8% |
| 8 | 5.1% |
| 9 | 4.6% |
2. When Benford's Law Applies
Works best for: Large datasets with numbers spanning multiple orders of magnitude. Transaction-level data (invoices, payments, journal entries).
Does NOT apply: Assigned numbers (phone numbers, ZIP codes), constrained ranges (heights, ages), prices set by humans ($9.99).
3. Applying Benford's Analysis
- Extract first digit of each transaction
- Count frequency of each digit
- Compare to Benford distribution (chi-square or Z-test)
- Significant deviation = potential anomaly (investigate further)
4. Other Statistical Tests
- Z-test for individual digits: Detect specific digit anomalies
- Chi-square goodness of fit: Overall conformity to Benford
- Mean absolute deviation (MAD): Measure of deviation magnitude
- Second digit test: More sensitive for fraud detection
Q1: According to Benford's Law, what percentage of numbers should start with digit 1?30.1%. Nearly one-third of all natural numbers start with 1. This is because log(1+1/1) = log(2) = 0.301. Q2: Why does Benford's Law work for fraud detection?When people fabricate numbers, they tend to create balanced digit distributions (equal frequency of all digits). Benford reveals this unnatural pattern. Fraudulent numbers deviate from the expected distribution. Q3: When should you NOT apply Benford's Law?Small datasets (< 100 numbers), data constrained to a narrow range, assigned numbers, prices set by humans (e.g., $9.99), data with built-in minimums. Q4: What statistical test is used to test conformity to Benford?Chi-square goodness of fit test compares observed vs expected frequencies. Also Z-statistic for individual digits, and Mean Absolute Deviation (MAD) for overall fit. Q5: A dataset shows digit 1 = 18%, digit 9 = 12%. Is this suspicious?Yes, both deviate from Benford's expectation (1=30.1%, 9=4.6%). Digit 1 appears too rarely, digit 9 too often. May indicate fabricated data where creator tried to make digits "random" and overcorrected. Join Discord PreviousFraud Schemes & Red FlagsNextForensic Accounting & Financial Analysis