In an earlier blog, I discussed e-voting systems in some detail. Here I focus on some specific issues such as why there is so little public discussion of hand counting of ballots, tactics used by e-voting system vendors, the Holt Bill, more about costs, and some possible solutions to a major problem associated with hand counting. I also elaborate further on why the use of machines increases vulnerability to fraud.
On the other hand, relatively few people have been motivated to advocate publicly the use of manual counting, and there has been minimal discussion of the relative merits. Based on their general experience, most technical people implicitly assume that computer-based devices are much more cost effective. Surprisingly, they are wrong. A fundamental reason for this is that, since e-voting machines are only used once or twice annually, the duty cycles of computers used for other data processing purposes are two orders of magnitude greater than those for e-voting machines. This is why manual vote counting is in most, if not all, cases significantly cheaper. It is hard to be precise here because e-voting costs are composed of many components. The costs for some of these, e.g., machine storage, transportation, servicing, and testing, vary considerably from place to place. But even using only basic costs, the calculations in my earlier article clearly establish that, with the possible (and the word possible here is emphasized due to real uncertainly) exception of ballots with well over 20 races, manual counting is cheaper.
With respect to hand counting, the average cost of an election is determined by a modest fixed overhead cost plus an amount proportional to the averages of the number of expected voters and number of races on the ballot. In contrast, when e-voting systems are used, the overhead is somewhat larger and the variable amount is proportional to the number of machines, which is determined by maximum numbers, principally the number of expected voters. So if a state holds hand-counted elections annually, and if every fourth year the ballot is much larger than those for the other elections, this election will be relatively costly, but it does not inflate the costs of the other elections. If the same state used e-voting systems, it must purchase enough machines to handle the worst-case election and incur the associated costs even when many of the machines are idle. (That is, the duty factor referred to above is further reduced.)
So far, I have data on 15 states. Texas is the "leader" with 45 races in 2006 (30 judicial contests!), with Colorado next with 43 and California third with an estimated 38, (obtained by adding an estimated 10 local candidates to the 28 federal and state races for 2006) both also in 2006. The other 12 states are all below 20, with Virginia having only 8 races on its biggest ballot (2006). The overall median is 15. I would appreciate any help in correcting or adding to this information.
When hand counting is used, evidence of fraud is usually easy to obtain and to understand, and is generally unambiguous (e.g., people testifying that they were offered bribes, or signatures of dead people at polling places). In the case of e-voting systems, matters are seldom so clear. The evidence is embodied in complex hardware and software, and, because security is most often lax, it is easy for cheaters to obliterate their footprints before forensic examinations are undertaken, perhaps weeks after election day.
Defenders of e-voting systems are very resourceful when it comes to explaining away such phenomena as large numbers of voters in one particular district not voting in a hotly contested race, such as occurred in Sarasota in 2006 [Herald Tribune], or gross differences between exit poll and voting machine results [Morris]. Even where some machine output is acknowledged to be invalid, proving that the cause is fraud rather than malfunction is extremely difficult. This greatly reduces the risks associated with the commission of fraud. From the point of view of the cheater, even if the effort fails, the output might be corrected, but there is little likelihood of criminal prosecution (the recent Ohio case [Plain Dealer] is exceptional). The point here is that, even if a manual count of the votes tallied by some randomly selected machine differs significantly from the machine-reported votes, the claim will be made that this was just one bad machine and there is a good chance that no real action will follow to expose a possibly massive cheating operation. State election agencies, often under the control of political partisans, can't be relied on to enforce the rules. We have seen that courts are not helpful here.
In its present form, the Holt Bill, HR 811, has many good provisions, addressing important problems. These include requiring that all e-voting software be open for public inspection, that there be no wireless access or internet access to e-voting machines, and that system certification be by organizations not paid by or affiliated with e-voting vendors.
There is, however, one very important omission. While HR 811 requires that all e-voting systems have voter verified paper trails, it does not ban DRE (touch-screen) systems, provided that they print voter verified ballots. As was discussed in my earlier article, there is good evidence, and it is plausible, that most people are not likely to verify the accuracy of the paper ballot printed by the machine. The requirement in the bill for a placard in the voting place urging voters to do this is not an adequate remedy (it reminds me of the warning required on cigarette packs). I doubt that there is any way to get most people in most places to do such checking properly. Another serious problem with these systems is the high probability (based on experience to date) of malfunction of the printers--jams, ink clogs, etc. Such failures could be due to ordinary breakdowns or they could be deliberately programmed.
My guess is that many of the people I referred to earlier are not happy with this, and probably objected to allowing the use of such systems. Most likely there were arguments made that so many DREs are now in use, that state officials responsible for their purchase would be very upset if scrapping were to be mandated. Perhaps concerns that this might jeopardize passage of the bill led Dr. Holt and his staff to conclude that they could not include such a prohibition. I believe this is most unfortunate. In fact, several states are, on their own, abandoning DRE's in favor of optical scan systems.
A less serious concern, and one that might be easier to remedy, is the method specified in the bill for organizing Election Audit Boards (EAB's) in each state. These are to include a representative of each significant party AND an unbounded number of other people appointed by the Chief Auditor of the State. This official could easily pack the EAB with partisans of one side. Having seen the unscrupulous behavior of the State Secretaries of Florida and Ohio in recent elections (two prominent, but not unique, examples), I believe this is a very dangerous situation since the EABs play key roles with respect to manual recounts.
Experience since 2000 indicates how hard it is to get even gross election errors corrected. It appears that, analogous to the legal concept that possession is nine tenths of the law, there is an election concept that in nine cases out of ten, the initially declared winner will be declared to be the final winner. Other than when the ostensible margin of victory is very small, protests by "losing" candidates are treated as "whining" by a "sore loser". This is why I think that post-election procedures including random recounts cannot compensate adequately for fraud and malfunction.
In the real world there are numerous opportunities to circumvent machine certification and pre-election testing procedures. For reasons spelled out in my earlier article, I believe it is nearly impossible to screen out corrupt hardware and software features, so that we need effective post-election procedures, such as random hand recounts, to detect and correct cheating or malfunction. But if, as stated above, detection-correction mechanisms cannot be relied on, then, given the high stakes involved, it follows that we should not use such systems if there is a viable alternative.
Apart from the cost issue, dealt with above, many people have argued that manual systems are unsafe, as evidenced by the fact that fraud was a big problem with manual voting systems in the past. They point to such legendary figures as Boss Tweed and Richard Daley. At the risk of being repetitive, I will address this point again briefly. There are three parts to the argument. One is that, if a polling place is controlled by a partisan group, then cheating will occur regardless of what level of technology is used. All cheating methods used in conjunction with manual voting, can also be used with e-voting systems. This point is illustrated by the recent conviction for old-fashioned vote buying of politicians in East St. Louis [Suhr] (where e-voting is well established [Fitzgerald]). The only difference is that the Tweeds and Daleys would derive additional graft from the machine vendors, and might need fewer thugs to control elections.
The second point is that if polling places using hand counting are properly staffed with poll watchers from rival factions, and run according to well understood rules, opportunities for cheating would be reduced to such individual acts as people voting in more than one district (which could also occur where e-voting is used). The third point is that, even in a properly staffed and operated polling place, if an e-voting system is used, significant cheating could still occur thru the use of clandestine features of the hardware and/or software with minimal participation of on-site people. Therefore, it is clear that, all other things being equal, changing over to a manual system would almost certainly reduce the amount of cheating.
Comments can be sent to me at unger(at)cs(dot)columbia(dot)edu