Benford’s Law — Why Numbers Prefer to Start With 1
Beyond the syllabus | GCSE and A-level Maths | Statistics
Put a penny on the table. Imagine its value doubles each day. Write down the daily balances in pence, keeping only the first digit of each balance. Which will you see more often, 1 or 9?
The opening few days hardly settle the argument. The balances are 1, 2, 4, 8, 16, 32, 64 and 128 pence. There are already three 1s, but perhaps the 9s will catch up. Give the penny 100 days, counting the starting penny as day one. Now there are 30 balances beginning with 1 and just five beginning with 9.
What makes this worth a maths lesson is that you can put the penny away, open a file of real measurements and sometimes see much the same pattern. Before looking at the formula, I’d ask a student to sketch what they originally expected. Nine bars of equal height would be a perfectly reasonable guess. The interesting question is what kind of experiment would actually give us those nine equal bars.
What counts as the first digit?
Read from the left until you reach the first non-zero digit. A river that is 407 km long goes in the 4 column. An electricity reading of 0.048 kWh also goes in the 4 column. The zeros tell us about the size of the measurement. The 4 is its first significant digit. An exact zero has no first significant digit, so leave it out of the tally.
If we picked at random from nine cards numbered 1 to 9, each digit would have a chance of 1/9, or about 11.1%. A river has acquired its length through a rather different process. Why assume that measuring it amounts to drawing one of those cards?
The prediction beside a result you can reproduce
Benford’s Law gives this prediction for a first significant digit d, where d is any whole number from 1 to 9.
P(d) = log10(1 + 1/d)
Here log10 means the base-10 logarithm. GCSE students can use the percentages below straight away. The final column gives the actual tally for the doubling penny, from 2^0 to 2^99 inclusive.
| First digit | Benford prediction | Observed days out of 100 |
|---|---|---|
| 1 | 30.1% | 30 |
| 2 | 17.6% | 17 |
| 3 | 12.5% | 13 |
| 4 | 9.7% | 10 |
| 5 | 7.9% | 7 |
| 6 | 6.7% | 7 |
| 7 | 5.8% | 6 |
| 8 | 5.1% | 5 |
| 9 | 4.6% | 5 |
The 1s and 2s together account for almost half the predicted observations. That makes a good question for a student who is still getting comfortable with probability. If you had to choose two first digits before seeing the next value in a dataset that follows Benford’s Law, which pair would you pick?
Mathematical background and formula — Wolfram MathWorld
Why does 1 get so much room?
Imagine watching a balance grow smoothly at a constant percentage rate. It has just reached £100. To lose its leading 1 and reach £200, it must grow by 100%. Later, when it reaches £900, it needs to grow by only about 11.1% to reach £1,000. At that point the leading 1 is back again.
If you checked the balance at regular time intervals over many such cycles, you would have far more opportunities to catch it showing a 1. The same argument works between £10 and £100 or between £1,000 and £10,000. This is one way a first-digit bias can arise. For a collection of rivers or towns, whether their measurements show the pattern is a question for the data.
A-level extension. Draw a logarithmic number line from 1 to 10. Label the ends 0 and 1 in log10 units. The interval belonging to first digit 1 starts at log10(1) and ends at log10(2). Its width is about 0.301. The strip belonging to 9 runs from log10(9) to log10(10) and is only about 0.046 wide. If observations are evenly spread along this logarithmic scale within a tenfold interval, the probability of digit d is the width of its strip.
P(d) = log10(d + 1) - log10(d) = log10(1 + 1/d)
Try adding the nine widths before rounding. Each right-hand edge is the next strip’s left-hand edge, so all the middle terms cancel. You are left with log10(10) - log10(1) = 1. The probabilities account for the whole line.
Build the chart, then put it to a harder test
First, reproduce the penny result. In Excel or Google Sheets, put 1 in A2 and enter =2*A2 in A3. Fill A3 down to A101. In B2 enter =VALUE(LEFT(TEXT(A2,"0.000000E+00"),1)) and fill down. Scientific notation keeps the first significant digit at the front even when the balance is too large to display in full.
Enter the digits 1 to 9 in D2 to D10. In E2 enter =COUNTIF($B$2:$B$101,D2)/COUNT($B$2:$B$101). In F2 enter =LOG10(1+1/D2). Fill both down to row 10 and format E and F as percentages. Select D1 to F10, including your headings, and insert a grouped column chart. Put first digit on the horizontal axis and percentage of observations on the vertical axis. You should see two bars per digit. Give a student the chart without its legend. How easily can they tell which bars came from the penny?
Next, give the prediction a fair chance to fail. Download the ONS Census 2021 population workbook for England and Wales. Before opening it, write down whether you expect local authority populations to show a strong Benford pattern, a weak one or none at all. Give a reason. ONS population and household estimates workbook
Select the total resident population for each local authority, all ages and both sexes combined. Keep one value per authority. Exclude regional and national totals so that the same people do not appear twice at different scales.
Copy the positive numeric totals into a fresh sheet. For these whole-number counts, extract the first digit with
=VALUE(LEFT(TEXT(A2,"0"),1)). Repeat the tally and chart, extending the ranges to the final data row.Read the smallest and largest populations. Find the digit with the largest gap between observed and predicted percentages. Report the size of that gap in percentage points.
Examine a few places contributing to the tallest bar. What do their populations have in common? Consider how authority boundaries and the choice of geographical unit might have shaped your sample. The ONS workbook also uses rounded estimates. Record that detail alongside your findings.
In my teaching, the point where I’d put the calculator down is just after that second chart appears. A student who says, “Several authorities have between 100,000 and 200,000 residents, so I’d like to see what happens with towns instead,” has found a reason to run another investigation. I’d much rather hear that than see the axes adjusted until the two sets of bars look alike.
Can you make Benford’s Law fail on purpose?
Write down every whole number from 100 to 999 inclusive. Exactly 100 of them start with each digit. Choose a number uniformly at random from that list and each first digit really does have probability 1/9. We’ve found the experiment those nine equal bars belong to.
Now try a few familiar alternatives.
| Dataset | What shapes the first digit? |
|---|---|
| Adult heights in centimetres | A sample concentrated between 150 and 199 cm will produce mostly 1s. |
| Telephone numbers | Numbering plans assign prefixes. The digits identify a connection. |
| Lottery balls numbered 1 to 59 | Eleven labels start with 1 (1 and 10 to 19), but only one starts with 9. Fair draws inherit that pattern from the labels. |
| Asking prices on one local property search | Set the filters to £300,000–£499,999 and every result starts with 3 or 4. |
For a home-energy version, I’d try daily electricity use in kWh, then repeat the exercise using monthly totals from the same meter. Write down your prediction before grouping the readings. The house and meter are the same. The question is whether changing the period has compressed the range enough to change the chart. A large number of readings alone cannot guarantee a Benford pattern.
What would an accountant do with an unexpected bar?
Suppose an auditor finds a surprising number of payments beginning with 4. Looking at the entries reveals a monthly equipment hire charge of £480 across several sites. That one contract could explain much of the excess. A sensible next step is to inspect the contract and invoices and check whether the payments match the agreed terms.
Benford analysis can help someone decide where to look in a suitable collection of transaction amounts. A departure alone cannot establish fraud, and a close fit cannot certify that the accounts are accurate. A first-digit chart has discarded almost everything about each payment, including who was paid and what was bought. Those details matter when following up. Journal of Accountancy — using Benford’s Law to examine journal entries
The question I’d leave on the board
Give a student two unlabelled charts. One comes from the doubling penny and slopes down towards 9. The other comes from the whole numbers 100 to 999 and is perfectly flat. Both are correct. Ask the student to match each chart to its source and defend their choice.
That is what I like about Benford’s Law as a topic beyond the syllabus. It starts with a wager about a single digit. By the end, a student can look at a column of figures and ask the question that gives the graph its meaning. What had to happen for these numbers to end up on this list?

No comments:
Post a Comment