How important is election fraud?
That depends heavily on what country we are talking about and during what time period. We'll focus on the USA in the last 100 years, which appears to have been comparatively unaffected by fraud.
(a) It appears during 1900-2004 (27 US presidential elections) that 2 of them were altered by fraud/manipulation, namely Bush-Gore 2000 and I'm giving half credit to each of Bush-Kerry 2004 and Kennedy-Nixon 1960. That's (7.4±5.2)% of presidential races (where I put in a 1σ error bar; this since based only 2 datapoints is a very noisy estimate).
(b) In Florida 2000 and Ohio 2004, the fraud & manipulation appears to have been approximately of order 120,000 votes in Ohio out of 5.5 million cast, and 40,000 votes in Florida out of 5.9 million cast. In other words, current fraud/manipulation levels in the USA under circumstances likely to generate such, are of order 0.7% to 2% of the total number of voters.That's pretty small compared to other effects like gerrymandering and the contemporary "gender gap." It is fairly rare that US elections have a victory margin of 2% or less. Typical very rough estimates are that this happens in less than 10% of US congress races. [So you can't fraud 90% of US elections even if you try, using these techniques, and presumably trying happens only some of (presumably less than half) the time.]
US News & World Report 5/7/06: The Cook Political Report rates just 35 of this fall's 435 House races as competitive. Other analysts put the number at 25, or fewer than 6 percent of races. That's down from more than 100 competitive races in 1992... with the introduction of sophisticated redistricting software in the 1991 round, the number of competitive races – where the winner took less than 52 percent of the vote – began to drop. In Pennsylvania, home to half a million more Democrats than Republicans, GOP-led redistricting turned a congressional delegation comprising 11 Republicans and 10 Democrats to [in 2000] a 12-to-7 Republican advantage (the state lost two seats because of slow population growth).
"Gerrymanders in Ohio, Michigan, Pennsylvania, and Florida pretty much account for the Republican advantage in Congress," says George Mason University Prof. Michael McDonald. "If we had neutral maps, we'd be talking about a much different nation."
Lee Mortimer, Raleigh News and Observer: fewer than one in ten congressional seats were won by less than 10% in 1998... More than two in five state legislative elections weren't even contested by one or the other major parties in 1998.
However, you are warned that both fraud and vote-buying have been considerably more important in US past than it is now; and also that computerization poses unprecedented new security threats.
While presidential races are probably less fraudable than other races since they are more scrutinized, they also are more fraudable thanks to an interplay with the electoral college. You can usually count on some state to have a small enough margin to be fraudable, and then swinging it can swing the election.
Because "fraud" seems to trigger people's internal "morality watchdog" whereas, for some reason, Spoiler Phenomenon does not, a lot of people get very upset about fraud and not about spoiler phenomenon. Probably about 95% of voter-reform effort is fraud-related with only maybe 5% about changing voting methods to get rid of spoiler and similar effects. But really I think it should be about 70-30 the other way. Morality is emotional, imprecise, and unclear. (I'm not opposed to morality, I'm just saying that a quick thoughtless appeal to emotion is a dangerous thing.) Importance measurements are clear and unemotional. The cold hard numbers say that Nader had a comparable or bigger effect on the 2000 election than all the fraudlike-manipulation in Florida that year (which was clearly a fraud hot spot). Poundstone's book Gaming the Vote documents that US parties are now funding spoilers like Nader because they think that is easier and cheaper and less risky than fraud. Poundstone claims 7 US presidential races went the wrong way due to spoilers. But only 2 went the wrong way because of fraud (far as I know).