I heard Bill Schneider on CNN on the night of the Pennsylvania primary making some argument that there was very little evidence of campaign effects. He compared the exit polls from Ohio with those from Pennsylvania and noted that Obama's and Clinton's support from key demographic groups was pretty similar between the two states. And in the end, Clinton won both states by around 10 points. All the stuff that happened in the intervening six weeks, notably including Obama's "bitter" statement, appeared to have no effect.
A quick check of the exit polls largely supports Schneider's assertion. Note what's going on in this scatterplot; each data point marks how Obama did among a demographic group in both states. The line is x=y; if a point lies along that line, Obama did the same in one state as he did in the other. For most of the groups, there's not much movement. (The two votes correlate at .85) Actually, there's a bit of improvement for Obama between Ohio and Pennsylvania. But there are a few negative outliers. Obama did notably worse among union members, frequent churchgoers, and white Catholics in Pennsylvania than he did in Ohio six weeks earlier.
This is far from an exhaustive look at the demographics, but it does tend to suggest relative stability from one state to the next. Which would mean that you could probably predict the candidates' performances in the upcoming states simply by knowing the share of the electorate that each of these demographic subgroups comprises.
This is, in one sense, evidence against campaign effects. All the day-to-day noise about Jeremiah Wright and bitter voters and guns and bowling doesn't seem to be moving the vote much. On the other hand, the extent to which each of these demographic subgroups actually turns out to vote does matter and can be influenced by campaigns.
The tentative lesson: GOTV matters more than persuasion.