Totally, it looks as if the nation is breathing (i kinda like that for aesthetic reasons). If we bin the counts by day, a lot of the temporal resolution will be lost. What do you suggest?
We mainly want to filter this keyword to see how the discussion of "gun" is affected by the events following Sandy Hook tragedy. The opinions classification is not done over the same dataset.
You made a good point. But it makes more sense to divide by the number of tweets (within the given time interval) from the state instead of the user count which would not normalize it.