In the USA at present (2007) about 5-6 million voters are disenfranchised by state laws that prevent felons and/or ex-felons from voting. Those laws differ in different states.
That can exert undemocratic effects. For an example (which I am making up), suppose 51% of the US public wanted to decriminalize marijauna. Then, pretty clearly, the democratic thing to do would be to decriminalize it. But suppose a goodly chunk of those 51% were felons because of marijauna convictions! Well then, if they couldn't vote, marijauna would stay criminal, and so on forever.
But even in areas where they are legally permitted to vote, felons have smaller turnout than non-felons. (Ex-felons currently vote heavily democratic, especially in the South.)
The net vote-shift for Bush caused by these effects (versus what would have happened with 100% voting by all) was about 3% of the number of votes cast.
There are other turnout biases.
According to the CNN exit poll in 2004 (USA-wide) Bush-Kerry splits were 36-63 for under-$15K/year household-income voters and 63-35 the other way for over-$200K/year voters. That helped Bush because income correlates with turnout. (Based on US Census surveys & CNN exit polls) we find in the 2004 election that those with household income≥$75K turned out at an 80% rate (voting 57% for Bush) while those with income<$50K turned out at a 59% rate (voting 44% for Bush). The net vote-shift for Bush, caused by this income-based turnout bias (versus what it would have been with 100% turnout) then, was about 3%. Incidentally, the decision to make election day in the USA be Tuesday has presumably had the effect of biasing the vote, preventing working-class people from voting more than it prevents richer people with more flexible schedules. Election day is a weekend or holiday in most other countries.
In 2002, 52% of college-degree registered voters voted versus 27% for those with no college degree. That also helped Bush versus Kerry since the higher-education voters preferred Bush more – except that the voters with the highest (post-graduate-degree) education levels actually supported Bush the least of all education-level-groups. The net effect was probably about a 1% shift for Bush.
The turnout gap was similar for age≥30 versus age<30 voters. That also helped Bush versus Kerry since the older voters both had higher turnout and preferred Bush more; probably the net shift Bush's way was about 1%.
Blacks and Whites supposedly had fairly close turnout in 2004, but Whites had higher turnout by about 10% and favored Bush more, with the net effect favoring Bush again by about 1%.
Women had higher turnout than men in 2004. This (for a change) favored Kerry since women currently vote more Democratic than men. The net effect was probably about a 1% shift Kerry's way.
Registration biases, not just turnout biases, have further effects.
The total effect of all these turnout biases in 2004 favored Bush by roughly 3+3+1+1+1-1=8%, versus what would have happened with universal adult suffrage and no turnout biases.
Warning: adding up the effects in this way is not correct because some of the voter-sets overlap. So this was not quite right. To fix that would in principle be simple. If we had raw data for polls of a random sample of people about Bush & Kerry, then we'd know the true Bush/Kerry support fractions. The trouble is... who is going to provide this raw, unadjusted data to us?
In a CNN poll (20 April 2004) registered voters gave the president a 50-to-46 percent lead over Kerry in a two-man race. And among all adults, Bush led Kerry 49-to-46 percent.
From Robert Moran "A snapshot in time," 4 Feb 2004 National Review: "CNN/Gallup/USA Today/Democratic National Committee (kidding on that last part... maybe) survey of 1,001 American adults January 29-February 1, 2004 showed John Kerry defeating Bush 53 to 46 percent. A Newsweek survey of 1,022 registered voters taken at the same time showed Kerry defeating Bush 48 to 46 percent." [My italics.]
The NEP exit poll indicated Republican and Democratic self-identified-party-ID voters were equal (both 37%) in 2004 while a CBS/NYT poll of 885 adults [18-21 November 2004, discussed in NYT 23 Nov] showed there were 36% Democrat-self-identifying adults versus 29% for Republicans.
These figures do seem to support the hypothesis Bush benefitted from non-universal turnout, but do not find that benefit to be as large as our 8% estimate, they seem to think it is 1-7%.
In conclusion, I estimate that the Bush-over-Kerry margin in 2004 would have been lowered by (5±3)% of the number of total voters (note the central estimate would easily be enough to make Kerry win) if the USA had had universal adult voting.