Invisible primary aside, though, what does the search volume look like for each of the top five candidates once primary and caucus contests begin providing tangible results in January 2008?
Well, for starters, tacking on that extra year and the election day spike really dwarfs some of the earlier data. What's important, though, is that red line (Obama) spikes in early 2008 and stays above the yellow line (Clinton) through mid-June when Clinton's volume trails off. Sure this gives us some scale, but that election day jump for Obama is deterring us from seeing some of the changes from the invisible primary.
If we cut off the data at September 2008 -- just after the Democratic convention -- we lose some of the skew from that election day spike. Instead of one multi-colored line running across the bottom of the time series from January 2005 to November of 2006, we can actually see Hillary Clinton hovering above the other potential candidates instead of appearing to be one in the crowd (while still appearing to be at the top).
The picture is clearer still when just the 2008 data is isolated. Obama still is clearly ahead of Clinton throughout the time series, but there are certainly some fluctuations given the events on the ground. The space between Obama and Clinton is widest following Super Tuesday in early February and Obama's streak of wins to close out the month. Wright, bitter-gate and losses to Clinton in Texas and Ohio in March, however, closed that gap. Obama, though, maintains that lead, diminished though it is until the nomination contests end and Clinton withdraws.
Again though, much of this tracks with the events that were happening in the contest at the time, but Glenn does raise an interesting point. What does polling look like during this period? That's obviously another layer to add into this; one that Cohen, et al. (cited in original post) considered. In the original study, (actually Karol et al. 2003), they found that endorsements had three times the impact on polls than polls had on endorsements during the invisible primary period. No, that doesn't answer the question in the context of the primaries and caucuses, but it does indicate that polls (like the data here) are likely next in line on the causal chain behind events in the nomination race.
From One Contest to Another: Ohio Redistricting Competition
Newt on 2012: "We'll See"
2008 Democratic Presidential Candidate Emergence: The View Through Google Trends