In The American Campaign, James Campbell analyzes Presidential campaigns from 1872 to 1996 and finds that the campaigns have only a small chance of affecting the result. Voter choice is heavily constrained by Partisanship, the current economic situation, and incumbency, and is mostly set by the end of the conventions. Thus by Labor Day and the start of the Campaign (though one could argue the Kerry-Bush campaign started in Mar last year) the candidate leading in the polls goes on to win about 90% of the time, though the actual margin usually decreases by election day.
In fact Presidential campaigns appear to have only affected the outcome in 4 elections (only two in the 20th century: Wilson 1916 and Truman 1948).
2004 seems to fit the pattern, Bush was ahead after the conventions and went on to win. Indeed going back to Feb, Bush led Kerry in every month of the Rasmussen Poll, except Mar and Jul. Again suggesting the Campaigns reenforced voters opinions but didn’t really change them.
http://www.rasmussenreports.com/Monthly%20summary.htm
So maybe each candidate should simply donate their Federal Election budget to their favorite charity, stay home and answer reporter’s questions.
Allan Lichtman's
The Thirteen Keys to the Presidency makes the same argument. In his analysis, the campaigns don't matter much and most elections turn on factors that have locked in months before the election.
Lichtman would not list 1916 as one where the campaign mattered. The critical factors were sown up by the time of Wilson's nomination. He would count the 1948 election as one where the campaign might have mattered. A charismatic campaign by the GOP might have made a difference, but Dewey avoided clashes assuming he was ahead. OTOH, if Wallace had stepped back from the campaign, the race would have looked secure for Truman.