This is a graph of data from Kathy Dopp plotted in the same manner as Charlie Strauss . For a current summary of the problem see The Wikipedia article You can click on the plot to get a bigger version.
For each Florida county, the expected vote based on registration is plotted against actual vote for both Republicans and Democrats. A different symbol is used for Electronic Touch Screen Machines versus Paper Scan Machines.
If the process was normal, you would see the expected votes closely match the actual votes, creating a scatter of points that falls on the blue 45 degree line. In the actual data, however, there is an extraordinary split between Republican votes and Democratic votes in only the small counties that used the Paper Scan Machines.
This graph plots expected trend lines that very closely match the actual data. In the upper trend line, the Republican vote is boosted by 20% and an offset of 2000 fictional votes was added to the tally. The 20% offset is quite reasonable due to shifting voter preference, and only has the effect of shifting the line upwards slightly. The adding of 2000 votes creates the more troubling "hockey-stick" shape to the graph. The votes tend to flatten out at 2000 for all tiny counties, because the fixed offset dominates.
There is a similiar defect in the Democratic portion of the graph. Here I've modeled the data by assuming the actual is only 80% of the expected vote. Again, this is reasonable if there was a 20 shift in voter preferences after registration. However, the distinct turn-down of the Democratic vote in small counties is fairly well explained if we make a fixed subtraction of 1000 votes from the actual totals.
As a practicing engineer, I find these graphs extremely indicative of some untoward activity. At the minimum, there is a defect in the vote tabulation software. In the worst case, it suggests that every OPTSCAN machine result had about 2-3000 extra votes added for Bush. This could have been done by simply hacking into the Windows vote tabulation machine through the unprotected modem lines that were used in the election.
If we care about democracy, we must investigate this anomoly. Whether the vote was manipulated by a teenage hacker, a terrorist, or by one or another political party is irrelevant. What matters is that we have a true reckoning of the vote of every American.
Since this particular anomoly only occurred on the Paper Scan machines, and the expected failure is in the modem tabulation security, this issue is very easily addressed. Firstly, just do a simple rescan of all the ballots and manually tally the totals without using the GEM system. Secondly, sub-poena the telephone company logs for all the county office modem lines. Identify all non authorized numbers and trace them back.