The Reuters/CNN/Zogby story you link to says Obama’s support has eroded by two points and Edwards has increased by two points – but the poll margin of error is 3.4 points. That margin of error means that the poll result is an estimate of what is true in the populaton, that the true support for each candidate is “somewhere near” the poll result. The 3.4 points tells you the “worst case” of how far the true population value might be away from the poll result.
A change of 2 points from one poll to another is most likely due to the sampling process. It does not tell us that Obama’s support has “eroded” or Edwards’ support increased. Given a two point change in support for a candidate in a poll with a 3.4 point margin of error, all you can say is that there is no evidence of any change in the true population values.
