Reply #140: You're Right--Oregon and Wisconsin (Paper Ballot States) Confirm it.

[jwmealy]

Here's my latest offering--Sent to Conyers via Barbara Lee today.

My study looks at the national pattern in election results, specifically focusing on the phenomenon of “red shift”. Red shift describes Bush getting a greater percentage of the vote in the tabulated results than in the exit poll; negative blue shift describes Kerry getting a smaller percentage of the vote in the tabulated results than in the exit poll. My studies of the pattern of "red shift" (see ) led me to the general hypothesis that there could have been large-scale election tampering. A sense that there were certain anomalies in the distribution of red shift in relation to other factors, such as the spread (the final percentage difference between Bush and Kerry), plus a speculation as to approximately how an automatic election rigging program might be written, led to the following test hypothesis:

For the class of states in which, either according to the last-thing Nov. 2 Mitofsky consortium exit polls or to the official tabulated results, the spread between Bush and Kerry is less than 5.6 percentage points, the following equation, which I have named Formula F, will be true:

Red Shift + Final Spread + 1 > Greatest Recorded Red Shift for the 51 States.

The following six points explain why this equation is very unlikely to be true in all cases unless election tampering has occurred.
Point 1. The master hypothesis being tested is that many of the 13 states in the class described above exceed the margin of error because there is an electronic intervention system in place to fix the result for Bush (a radical hypothesis, but testable). The precise mechanism through which this intervention is accomplished will have to be uncovered by others, but I have appended a general sketch (see Appendix 1) as an illustration.

Point 2. For the purposes of this test I grant, in addition to the master hypothesis in Point 1, the possibility that some or all of the red shift effect in some states may be traceable to Republicans walking out of the polls past the exit pollsters for some as-yet-unknown reason, AND I grant that for some reason this effect may be somewhat exaggerated for states in which the media war leading up to the election is the hottest (especially in the swing states with the most electoral college votes at stake).

Point 3. In the case that Point 2 is true, the red shift due to voter exit poll avoidance behavior will only constitute a kind of noise that partially obscures the pattern that must exist in the final results data if Point 1 is true. It will not be able to hide the pattern altogether.

Point 4. The formula "Red Shift + Final Spread + 1 > Greatest Recorded Red Shift for the 51 States" describes what the election results would look like in every state in which some piece of software at some stage in the system was incrementally altering percentages so as to keep Bush ahead by 1% until one of two conditions applied: (1) Bush was projected to win by 1% or more, or (2) the amount of intervention required to give Bush the win by 1% exceeded a certain parameter, hypothetically 5.6%, which is the amount of red shift of Vermont, the highest recorded for the 50 states and Washington DC.

Point 5. If red shift were solely based on voter behavior as hypothesized in Point 2, and there was not in addition an engineered result being produced by something like a software program or programs, then there would be a general tendency for the equation to be true (i.e. the slimmer the margin, the greater the red shift phenomenon would tend to be). However, there would inevitably be a fair proportion of randomness in the relationship between the two parameters (red shift, spread), because there would not be any simple straight-line relationship between the intensity of the media battle (leading to the red shift), plus whatever the random portion of the red shift (simple error traceable to sample size), on the one hand, and the spread, on the other hand. Thus, one could easily contemplate a red shift of 3 (a relatively high number) and a spread of 1 (a relatively low number) for state X. That would still fail the equation (3+1+1>5.6 is FALSE). One could also contemplate a red shift of 1 (implying a relatively low media battle, or whatever is the supposed mechanism driving voter behavior, lowered yet again by a random sampling error favoring Bush), together with a final spread of 3.5. In that sort of case the equation would also be false. And, of course, one could contemplate a red shift of, say, 2, for whatever reason, and a true spread of 2, which would be more or less in the middle. All of these sorts of combinations of the two parameters would fail the equation. If there is no engineering of the result, every one of the states in the class may individually be somewhat more than 50% likely to pass the equation, but the likelihood that all will pass without exception is very small. In particular, I estimate that the odds against New Hampshire, Pennsylvania, Ohio, Nevada, Michigan, Florida and Colorado all passing the Formula F equation as somewhere between 10 and 30 to 1. The result is equivalent to flipping four or five coins and all of them coming up heads.

Point 6. The equation proved true in 9 of 13 members of the class. Two conspicuous instances of failure were Oregon and Wisconsin, which, despite having close final spreads, had exceedingly low red/blue shift. Under the master hypothesis being tested, the reason for this is not far to seek. Those two states are the only two in the entire Union that still use paper ballots. The current hypothesis concerns an electronic vote rigging system. These two exceptions therefore reinforce the result and tend to disprove the most discussed alternative hypothesis—that in states where the race was close and hard fought there would be a lot of red shift. The equation was notably true without exception among the 4 states of the class that went for Kerry in the final tabulation. I emphasize the 4 states that went for Kerry, because the equation does not actually have to prove true for states that Bush won because it could fail in the case that there was a slim true final spread of 1% or so. Therefore, it only takes a small amount of intervention to put Bush ahead. For example, if Kerry is ahead and has a margin of 2%, the program only needs to swap 2.5% votes to Bush, and now Bush is ahead by 1% rather than behind by 2%, so the software is happy. In that case, 2.5+1+1<5.6. I have implicitly predicated my model on the idea that Bush was in trouble in the swing states, and that a fair amount of intervention will have been needed. But not necessarily in every one of the swing states. For example, there are two swing states, namely Nevada and New Mexico, whose exit poll results show Kerry ahead by between 2% and 3%, but the tabulated result shows Bush winning by a bare 1% in each state. That result is entirely consistent with the hypothesis (Point 1), but each state fails the equation, and each state is within the margin of error. So it might just be chance and Bush won, or it might be that those states really went for Kerry, but the software swung the result to Bush without much intervention because the result was very close to start with. Without a different kind of forensic evidence, there is no way to determine which is the case.

See

Points 1 - 6 explain why I regard the results of the Formula F equation test as revealing the equivalent of a statistical fingerprint of a several-states-wide election tampering scheme.

Seeking a Simple and Irrefutable Proof of the Hypothesis
The hypothesized election rigging scheme, in order to be employed to the full without excessive fear of discovery, should be performed on individual voting machines that have no voter verifiable paper trail. The state of Delaware has (1) a statewide uniformity in using such a machine, and (2) a strong win by Kerry. On the current master hypothesis, the predicted result is that the red shift for Delaware would approximate the maximum (cutoff) parameter. Delaware, indeed, had the second-highest red shift in the nation, with 5.1. If the results of the Delaware race could be subjected to the same type of statistical analysis as shown above, the pattern of manipulation would come out in bold relief. Beyond that, if the actual order of votes cast could be reconstructed in even one machine (keeping blind from the order of voters, to preserve constitutionality), the hypothesis of an engineered result could be decisively proven or ruled out.

Until such proof is accomplished, the current analyses should be treated as provisional and suggestive.

Appendix: Hypothetical Method for Fixing an Election Using an Electronic Voting Machine (EVM) Together with Election Management Software (EMS). Based on the known specifications of a widely used EVM.
1. The firmware for the EVM itself (the EPROMS) is written in machine code so no one can ever figure out what makes it tick except by experience.

2. The firmware is designed to work with proprietary Election Management Software (EMS).

3. Written into the firmware is a program designed to switch presidential votes from one candidate to the other on November 2, 2004 according to a certain algorithm. This program knows to look for the names of the presidential candidates and their positions on the ballot, when that information is plugged into each machine by means of the Memory Cartridge supplied by the local election boards. The EMS, together with the instructions for election board officials and the suggested conventions for ballot layout, is designed in such a way as invisibly to give the firmware election tampering program the information it needs to alter votes under certain conditions.

4. Each machine has a hefty 12-volt rechargeable battery that keeps EPROMS board alive at all times, for years on end. Barring some accident, the EPROMS knows what date and time it is. Any time besides November 2, 2004 between certain hours, the firmware will act normal and will not switch votes.

5. After the end of the day on November 2, or after a certain sequence of events that signals to the machine that this particular election is over, each machine’s firmware is programmed to rewrite itself (reprogram its own EPROMS) to erase all trace of the vote switching instructions. That way, if the machine is tested at any earlier or later date than November 2, 2004, tests will come out normal.

6. There are two ways of switching votes: incrementally, and upon completion of a cartridge.

a. Incrementally. When a vote for candidate A is cast, the program does nothing. When a vote for candidate B is cast, the program asks, “Is candidate A ahead by 1% or more?”. If so, then the program does nothing. If not, the program records a vote for candidate A. The program keeps track of how many votes it switches as it goes along. Generally, before switching a vote, the program also asks a second question. “Have more than 5% (or some similar pre-set parameter, which may be partially randomized) of the so-far-recorded votes been switched?” If so, the program does nothing. Otherwise, if Candidate B is in fact far ahead, so many votes will be switched in order to keep Candidate A appearing to be ahead that it will be obvious that tampering has taken place. So the program has a cutoff parameter that essentially instructs it to do as much vote switching as required to keep Candidate A ahead, but tells it to stop when the situation is hopeless, and further intervention will only lead to inevitable detection.

b. Once per Memory Cartridge. When the cartridge is being prepared to be swapped out, the calculation mentioned in 6.a. is made.

7. The EMS scrambles the order of votes cast in its own records, erasing the inevitable pattern that would arise from manipulation.