The Florida recount provided one cautionary note about popular presidential elections: When implementing such a reform, we should be very careful about how the votes are counted, lest we wind up with a national recount that makes Bush vs. Gore look like a case in small claims court.
Obviously better machines should be the #1 priority, but I assume that even the best machines will have unforeseen errors. Besides, we can't (and shouldn't) make every voting machine in the US identical. How then to innovate and see if new technologies are better? And, if we ever have an exceptionally close race, even the tiniest voting machine error rate could matter. I know that in 2000 the _national_ margin was a statistically significant 500,000, but we can't assume that we'll never have a squeaker of a race under some popular election system. Here's one idea, applicable to any voting system I can think of: For each vote total (be it approval votes, preference orders in IRV, pairwise matrices in Condorcet, etc.), have each state round all numbers to the nearest multiple of 2,000 or some similar number (I'm not wedded to any particular number, and I'm open to the possibility of rounding to the nearest 1% of the number of voters or whatever number). With rounding, it would be very difficult to change a state's count of votes. If the national race were a dead-heat, without rounding candidates would have an incentive to scrounge for votes anywhere and everywhere. However, with rounding recounts would only makes sense in states with either (a) unreliable election procedures (cough, Florida...) or (b) totals close to the rounding threshold. Problem (a) is solvable, and the frequency of problem (b) is predictable, and unimportant except in super-close _national_ races. In our current system, of course, a national race with a statistically significant margin can still be affected by a single state with a statistically insignificant margin. One other note: Although rounding would prevent most recounts, it makes a tie more likely. Small non-zero margins could still produce a tie with rounding. However, I can think of a few reasonable ways of resolving the tie without resorting to a coin toss or a 5-4 Supreme Court ruling. Maybe toss the decision to the Congress, or maybe count the number of states in which one candidate won more votes than the other (count DC as a state if the # of states won is a 25-25 tie). I'm sure people can come up with other equally reasonable methods. What do people think of some sort of rounding scheme to prevent a national recount in the event of a popular presidential election? (And yes, Demorep, I agree that the EC is evil, I'm just pondering the best replacement for it.)
