Counting Tanks From Serial Numbers

  • Hard
  • Probability puzzles
  • Tests: Estimating a total from a sample maximum

The puzzle

Tanks are numbered 1 to N as they leave the factory. Five are captured, and the highest serial number among them is 60.

Estimate N. "At least 60" is true but useless; "60" is clearly too low.

Stuck?

One at a time, smallest first. Open only as many as you need — the whole point of the puzzle is the part you work out yourself.

Hint 1

If the five numbers were spread evenly over 1 to N, how far short of N would the largest fall?

Hint 2

On average the gap above the maximum is about the same as each gap below it.

Show the answer Hide the answer

About 71. The estimator is m(1 + 1/k) − 1 with m = 60 and k = 5.

How to get there

Five samples cut the range 1 to N into six gaps on average - one above the maximum and five between the samples. So the maximum sits about 5/6 of the way up, and N is roughly m × 6/5.

The standard unbiased estimator is m(1 + 1/k) − 1, which here gives 60 × 1.2 − 1 = 71.

This was used against real German production figures in the Second World War. Allied intelligence estimated 246 tanks a month from serial numbers; conventional intelligence said 1,400. Post-war records gave 245. The method works because factories number sequentially, and it fails the moment they stop - which is why serial numbers are now routinely randomised.

The technique worth keeping

A sample maximum underestimates a population maximum by a predictable amount: with k samples the gap above is about one of k+1 equal gaps. Correct for it rather than reporting the raw maximum.