There's always a great deal of talk about how close an election is. To myself, this can mean several different things, as I touched upon in my earlier discussion of the 2000 US presidential election.
An election can be considered close in several senses. First, that the final tallies are close. Second, that a small number of voters could have changed the result. Third, that alternative methods of tallying the election would have altered the election result.
The 2000 example is a good one because the final tally was close (271 to 266), the number of voters needed to change the election result was small (a few hundred out of a hundred million), and a change in election procedure would have likely resulted in a change in the result, e.g., going by popular margin instead of electoral votes, counting one or more states via a non-plurality method, et cetera.
Even very small alterations in the electoral vote mechanics, such as proportional allocation in some states that do not currently split electors, or the number of electors allocated, i.e., the size of the US House, could have altered the 2000 election.
The 2004 election was also a close election - but only in the sense that had a little over 50,000 votes been changed from Bush to Kerry in that state. Given that Bush enjoyed majority support, his position in 2004 was much more secure against alterations in the electoral mechanics, such as shifting to a national popular vote.
While some have charged the 2004 election was stolen, the amount of alteration that such charges must allege in order to secure Kerry's victory against, say, a national plurality vote - is truly staggering. It would take a total of around 1.5 million ballots altered or 3 million added (or subtracted) ballots to account for such a change.
And it is in discussing such metrics that the weakness of the electoral college comes out. The electoral college has most of the pros and cons of the plurality system it is based on - except that it is much more strongly vulnerable to local shifts, whether how easy it is for a third party candidate to make it onto the ballot, voter suppression or fraud, or - in the case of the 2000 election - simple counting error to shift the overall result. Electoral votes are necessarily going to be closer than national popular votes by the metric of how many voters need to change their mind to change the election.
For example, the 2008 election was not considered a particularly close one. Barack Obama's total plurality vote margin was 7.2%, and the electoral vote margin 365-173. Yet to change the result of the total election, it suffices to barely flip North Carolina, Indiana, Florida, Ohio, New Hampshire, Virginia, and Iowa - which he won by a total of 987,000 votes, 0.75%. The 2004 election could have been changed with 0.097% of the votes, in spite of a 1.5% popular margin.
I'm of the opinion a comprehensive listing (e.g., Bush/Dukakis would have required 1.23% of votes to be added/subtracted, or 0.62% changed) would give a good idea as to the typical quantitative relationship between how sensitive a national plurality vote is, as opposed to an electoral college ballot.