In statistics, Benford’s Law describes the frequency distribution of leading digits of numbers in a dataset.
This law explains that in many real-life datasets, the leading digit is more likely to be small than large.
In fact, it has been found that the probability of the leading digit for a given number in a dataset being between 1 and 9 is as follows:
- 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%
This phenomenon has been observed in a wide variety of fields including house prices, stock prices, street addresses, death rates, and many others.
Note: Benford’s Law is also sometimes referred to as the Law of Anomalous Numbers.
It turns out that this law has several real-life applications, the most common being fraud-detection.
If it is known that the numbers in a dataset should follow Benford’s Law, then it can raise a red flag for spam or fraud if it is seen that larger numbers tend to occur more frequently than they should.
This can be useful when banks or other institutions are attempting to identify fraudulent account records or transactions.
When Does Benford’s Law Apply to Datasets?
Benford’s Law doesn’t apply to all datasets, but in general it does apply to any dataset that meets the following requirements:
- There is no artificial minimum or maximum value on a dataset.
- The dataset ranges over an order of magnitudes.
- The values in the dataset are measured rather than assigned or bucketed.
- The dataset only includes quantitative data.
If each of these conditions are met, then we would expect Benford’s Law to be able to describe the relative frequency of leading digits in the dataset.
Some examples of when a real dataset would not meet these requirements include:
- A dataset that describes the height of individuals (has a minimum and maximum height)
- A dataset that describes IQ values (does not cover an order of magnitude)
- A dataset that describes people’s movie ratings (values are assigned or bucketed)
- A dataset that describes political preferences (values are not quantiative)
In each of these examples, we would not be able to use Benford’s Law to describe the relative frequency distribution of leading digits in the values in the dataset.
Example: Using Benford’s Law in the Real World
Benford’s Law has been used to detect fraud related to socio-economic data in the real world.
For example, it’s known that population sizes of cities and towns tend to follow Benford’s Law.
Suppose a government official received a report that showed the following distribution of leading digits in the census of various cities in his state:
- 1: 10%
- 2: 15%
- 3: 12%
- 4: 8%
- 5: 9%
- 6: 10%
- 7: 11%
- 8: 10%
- 9: 15%
Notice that each leading digit occurs at roughly the same rate.
This would raise a potential red flag that the data could be fraudulent because scammers tend to use a uniform distribution when generating fake numbers.
If the data were accurate, we would expect the smaller digits to occur much more frequently, just as Benford’s Law would predict.
Additional Resources
The following tutorials provide additional information about topics related to probability:
Law of Total Probability: Definition & Examples
Law of Large Numbers: Definition & Examples
Slovin’s Formula: Definition & Example
Hi Zach Bobbitt,
It would be great if you can elaborate more on the example used in the Benford Analysis. I also have a doubt why it is mandatory that why it is necessary that the small digits repeat more frequently.
Thank you
Hi Albert…Benford’s Law arises due to the logarithmic distribution of numbers. Here’s a simple way to understand why smaller digits occur more often:
Logarithmic Scale Distribution:
Numbers in real-world datasets (like populations, financial figures, or scientific data) often span several orders of magnitude. For example, financial transactions can range from cents to millions of dollars.
When you think of numbers on a logarithmic scale (where each step corresponds to multiplying by a constant factor), there’s more “room” for numbers with lower leading digits. This is because the logarithmic distribution causes the intervals between powers of 10 to cover more numbers starting with smaller digits.
Mathematical Explanation:
The logarithmic formula explains why lower digits like 1 and 2 occur more frequently than higher digits like 8 and 9.
Scaling and Multiplication:
Many real-world processes involve multiplicative growth or scaling, such as interest rates, population growth, or market prices. When numbers grow multiplicatively, they cover more cycles starting with smaller digits before reaching larger ones. For example, as a number doubles, it spends more time in ranges starting with 1, 2, or 3 before moving into ranges starting with higher digits.
Example of Benford’s Law in Practice
Imagine you have a dataset of electricity consumption across a city, measured in kilowatt-hours (kWh). When you analyze the first digits of these measurements:
You would likely find that the digit 1 appears as the first digit in around 30% of the readings.
Digits 2 through 9 appear progressively less frequently.
Now, let’s take a simple practical example:
Suppose you’re tracking daily sales figures in dollars across different stores in a retail chain.
Store A: On day 1, the store earns $102, on day 2, $154, on day 3, $301, and so on.
Store B: Day 1: $250, Day 2: $470, Day 3: $600, etc.
When you compile a large set of these sales figures, Benford’s Law predicts that around 30% of the time, the sales figures will start with 1, and only about 5% of the time will they start with 9. This pattern emerges naturally because smaller leading digits cover more of the logarithmic scale.
Is there room in the law of first digits to cover zero as well as one to nine?
Hi Rick…Benford’s Law specifically applies to the **first significant digit** of a number, which cannot be zero. Here’s why:
### Explanation
1. **First Significant Digit Definition:**
– The first significant digit is the first non-zero digit in a number. For example:
– For \(0.045\), the first significant digit is \(4\).
– For \(450\), the first significant digit is \(4\).
2. **Why Zero Is Excluded:**
– Zero cannot be a first significant digit because it does not contribute to the magnitude of the number. For example, in numbers like \(0.045\) or \(0450\), zeros are place-holders, not significant digits.
– Benford’s Law describes the distribution of non-zero digits \(1\) through \(9\) in datasets that span several orders of magnitude.
3. **Mathematical Basis:**
– The probability of the first digit \(d\) (where \(d = 1, 2, …, 9\)) is given by:
\[
P(d) = \log_{10}(1 + \frac{1}{d})
\]
– This formula inherently applies only to digits \(1\) through \(9\).
—
### Room for Zero in Other Contexts
While zero is excluded from the first significant digit analysis, it plays a role in:
– **Leading Zeros:** In certain representations (e.g., numbers with fixed-width formatting like “007”), leading zeros are present but not analyzed under Benford’s Law.
– **Zeros in Other Positions:** Zeros may occur in the second or subsequent significant digits and are analyzed separately if needed.
—
### Summary
There is no room for zero as a first significant digit in Benford’s Law. The law focuses exclusively on the non-zero digits \(1\) through \(9\), reflecting their logarithmic distribution in many naturally occurring datasets. If you want to analyze zeros in other contexts, a different approach or metric would be required.
Let me know if you have more questions! 😊