As you can see, the results are rather erratic. Don't hesitate to press "Apply" multiple times to observe the variation between each generation. We only have 83 numbers to analyse for each candidate, so you will see a much greater variation than you did on the previous data-set, which generated 1 000 000 numbers (Fitting a table with 1 000 000 counties on a web-page would be rather complicated, as you can probably understand). Nonetheless, the result is very clear, we end up with a rather homogeneous distribution of first digits that does not follow Benford's Law's predictions. But that isen't all, let's go a little deeper, try to change the number of voters from 20 000 to 10 000, and observe the percentages. Done? Odd isen't it, now we have a distribution of first digits that favors digits between 1 and 5. Why is that? Well, we are splitting these votes between two candidates, just about 50-50, So if we get 10 000 voters in a county, our candidates will get about 5000 votes each, it is therefore much less likelly that we would have numbers that start with a 6,7,8 or 9, than it is to have numbers that start with a 1,2,3,4 or 5, as these possibilities only represent 500 numbers (from 500 to 999) out of the average maximum of 5000 avaliable. This is another major aspect of why Benford's law cannot be relied upon to detect electoral fraud, as the number of voter in a county and the percentage of distribution between candidates is the one aspect that will dictate the first digit of the resulting number.