Mathwords logoMathwords

German Tank Problem — Definition, Formula & Example

The German Tank Problem is the task of estimating how many items exist in total when the items are numbered 1, 2, 3, … and you see only a few of their serial numbers. It takes its name from World War II, when Allied statisticians used serial numbers on captured German tanks to estimate monthly tank production.

Suppose kk distinct integers are drawn at random, without replacement, from {1,2,…,N}\{1, 2, \dots, N\}, where NN is unknown, and let mm be the largest value drawn. The minimum-variance unbiased estimator of NN is N^=m+mk−1\hat{N} = m + \frac{m}{k} - 1. Its expected value equals NN exactly.

Key Formula

N^=m+mk−1=m+m−kk\hat{N} = m + \frac{m}{k} - 1 = m + \frac{m - k}{k}
Where:
  • N^\hat{N} = Estimated total number of items in the population
  • mm = Largest serial number in the sample (the sample maximum)
  • kk = Number of items in the sample

Worked Example

Problem: A city's bike-share bikes are numbered 1 through N. You check four bikes chosen at random, and their numbers are 14, 27, 41, and 60. Estimate N, the total number of bikes in the fleet.
Step 1: Identify the sample size and the sample maximum.
k=4,m=60k = 4, \quad m = 60
Step 2: Find the average gap. There are m−k=56m - k = 56 unseen numbers below 60, spread across 4 gaps. That leaves about 14 missing numbers per gap. Expect a gap of about this size above the maximum as well.
m−kk=60−44=14\frac{m - k}{k} = \frac{60 - 4}{4} = 14
Step 3: Add that average gap to the sample maximum.
N^=60+14=74\hat{N} = 60 + 14 = 74
Answer: The estimated fleet size is about 74 bikes.

Why It Matters

In AP Statistics, this problem shows what makes an estimator unbiased and why some unbiased estimators beat others. After the war, German production records confirmed that the serial-number estimates were far closer to the truth than intelligence reports from spies. The same method is used today to estimate quantities such as a manufacturer's total sales from product serial numbers.

Common Mistakes

Mistake: Using the largest serial number you saw, 60 here, as the estimate of N.
Correction: The true total can never be smaller than the largest number you see, so the sample maximum almost always falls below N. It is a biased estimator that underestimates on average. Add the average gap, m−kk\frac{m-k}{k}, to correct for this bias.

Related Terms