I read the Meredith and Maholtra paper cited by MIke, and while it’s a nice methodological effort, I think the authors significantly overreach on the conclusions.

In essence, what the authors show is that the candidate totals differ between VBM precincts and non VBM precincts.  They have an elegant design comparison, choosing precincts just over and just under 250 voters, thus addressing some of the selection problems that bedevil other research that relies on this “natural experiment.”  Table 1 does a very nice job assessing the differences between the precincts.

The paper is solid methodologically, bu the conclusions overreach.  It is no secret that candidate totals differed between absentee/by mail and precinct place ballots.  I reported in an earlier posting here that 80% of John Edwards’s votes in Contra Costa County were absentee, while 40% of Obama’s and Clinton’s votes came in via that mode (in line with the state average).

But does this say anything generalizable about “information loss” and “campaign momentum”?  The authors say yes:

We predict that late campaign information loss will cause John Edwards, Rudy Giuliani, and Fred Thompson to perform better among VBM voters.  This is not to say that some voters do not prefer to vote for withdrawn candidates under full information as a protest or expressive vote; indeed John Edwards got more than seven percent of the vote in the West Virginia primary on May 13, 2008, months after he withdrew from the race.  Instead, we attribute any significant increase in support for withdrawn candidates when using VBM instead of polling place as evidence of late campaign information loss.

I would need to be convinced that this more complicated explanation is valid rather than the much simpler one: voters won’t cast a vote for a candidate who has withdrawn.

The analysis of Huckabee is a case where theory fits data much better–Huckabee’s momentum gains during this period are reflected (2-3 percentage points worth) in the vote by mail vs. precinct place votes.