The authors base their analysis on the assumption that humans are very bad at creating random numbers. They examined the last two digits of the official vote count from each province, looking for evidence of two human foibles concerning how people pick random numbers.
First, when picking random numbers, people tend to pick some digits more than others. For example, humans tend to pick the number 5 less often than the number 7. In the election results, the last digit is a 5 only 4% of the time, and 7 17% of the time. With completely random numbers, each 5 and 7 would appear as the last digit about 10% of the time.
Second, people have difficulty creating random numbers with non-consecutive digits. This pattern also shows up in the results.
They authors conclude that the chances of the election numbers being completely clean are 1 in 200.