1
$\begingroup$

I am bit confused on the algorithm of Benford's Law using more than the first leading digit.

I have read a thesis here that states the law for leading 2 digits is:

     $\log_{10} \left( 1 + {\small\dfrac 1 {10d_1 \! + d_2}} \right)$ where $d = d_1 d_2$ and $d$ is within the bounds of 10 to 99.

Would that not be equivalent to the original formula of:

     $\log_{10} \left( 1 + {\small\dfrac 1 d} \right)$ where $d$ is within the bounds of 10 to 99.

Is that correct? I have tried googling for a good night, but do not seem to be able to find a formula which applies to leading digits > 1.

Appreciate any help, below is the sample code I have used to generate the results in Python.

expected = [log10(1+1/d) for d in range(1,10)]
expected2 = [log10(1+1/(10 *d1+d2)) for d1 in range(1,10)
                                    for d2 in range(0,10)]
expected22 = [log10(1+1/(d1)) for d1 in range(10,100)]

print(expected22,expected2)
$\endgroup$
2
$\begingroup$

Yes, under Benford’s law the expected frequency of a specific leading pair of digits $ d_1d_2 $ $\big( = 10d_1 \! + d_2 = d $, with $ d \in \{10,11,12,\ldots,99\} ~\big) $ is just as your program calculated.

\begin{align} P ( \text {first two digits are} ~ d_1 d_2 ) & = P ( \text {first two digits are} ~ 10d_1 \! + d_2 ) \\[1ex] & = P ( \text{first two digits are} ~ d ) \\[1ex] & = \log_{10} \left( 1 + {1 \over d}\right) \end{align}

Agreeing with your implicit conjecture, this holds in general for $n$ leading digits where $d = d_1 d_2 d_3 ... d_n$ with $d_1 \in \{ 1,2,3,...,9 \}$, as gathered from An Introduction to Benford's Law, Chapter 1, Princeton University Press (no author or year given) — stated slightly differently here in a subtractive format that facilitates many applications.

\begin{align} P ( \text {first} ~ n ~ \text{digits are} ~ d_1 d_2 d_3 ... d_n ) & = P ( \text{first} ~ \text{digits are} ~ d ) \\[1ex] & = \log_{10} (d{+}1) - \log_{10} d \kern 7.5em \end{align}

An example for the original question gives the same result as the familiar format.

\begin{align} P ( \text{first} ~ 2 ~ \text{digits are} ~ 76 ) & = \log_{10} 77 - \log_{10} 76 \\[1ex] & = \log_{10} {\small { 77 \over 76 }} \\[1ex] & = \log_{10} \left( 1 + { \small{ 1 \over 76 }} \right) \kern 6em \end{align}

The subtractive format shows its flexibility on other conceivable questions about the first two digits.

\begin{align} P ( \text {first 2 digits are equal} ) & = P ( \text {first are 11} ) + P ( \text{first are 22} ) + \cdots + P ( \text{first are 99} ) \\[2ex] & = ( \log_{10} 12 - \log_{10} 11 ) + ( \log_{10} 23 - \log_{10} 22 ) \\[.5ex] & \kern 12em + \cdots + ( \log_{10} 100 - \log_{10} 99 ) \\ & = \log_ {10} {\small{ 12 \cdot 23 \cdots 100 \over 11 \cdot 22 \cdots 99 }} \\[1.5ex] & = 10.9 \% \end{align}

\begin{align} \kern 3.5em \raise 2ex \strut P ( \text {2nd digit is} ~ 6 ) & = P ( \text {first are 16} ) + P ( \text{first are 26} ) + \cdots + P ( \text{first are 96} ) \\[2ex] & = ( \log_{10} 17 - \log_{10} 16 ) + ( \log_{10} 27 - \log_{10} 26 ) \\[.5ex] & \kern 12em + \cdots + ( \log_{10} 97 - \log_{10} 96 ) \\ & = \log_ {10} {\small{ 17 \cdot 27 \cdots 97 \over 16 \cdot 26 \cdots 96 }} \\[1.5ex] & = 9.3 \% \end{align}

This format also telescopes nicely when verified against the most famous prediction of Benford’s law, that 1 is the first significant digit in approximately 30.1% of suitably random numbers. The first 3 digits are employed this time for variety.

\begin{align} \kern 4.5em P ( \text {first digit is} ~ 1 ) & = P ( \text {first are 199} ) + P ( \text{first are 198} ) + \cdots + P ( \text{first are 100} ) \\[2ex] & = ( \log_{10} 200 - \log_{10} 199 ) + ( \log_{10} 199 - \log_{10} 198 ) \\[.5ex] & \kern 12em + \cdots + ( \log_{10} 101 - \log_{10} 100 ) \\ & = \log_{10} 200 - \log_{10} 100 \\[1.5ex] & \equiv 30.1 \% \end{align}

(The material above was learned while applying Benford’s law to Roman numbers in a puzzle.)

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.