I still don't understand why the movement in Rasmussen is always just a few points one way or the other. Never much fluctuation.
Rasmussen does a hard weight by party id.
This damps out changes.
In a "likely" voter poll, people are usually asked questions about how "excited" they are about the candidate, how much attention they are paying to the race, etc.
Right after the Dem convention, many democrats were more "energized" and hence a greater % of Dems were deemed "likely" - hence the % of dems in the sample went up.
Right after the GOP convention, the opposite happens.
This makes the result bounce around a bit more as the ratio of Dems to GOPers changes
If you have bad sample design and don't do callbacks etc, the problem gets much worse (CBS, LA Times, etc)
Folks who actually know how to poll (Gallup) ensure that their samples do not get skewed by these factors, which is why Gallup changed so little the last few weeks.
Rasmussen essentially runs three polls at once - one for Indys, One for GOPers, and a third for Dems. he then combines the three polls into one big poll by weighting
Dems ALWAYS count for 39% of Rasmussens poll
GOPers ALWAYS count for 35% of his poll
Indys ALWAYS count for 26% of his poll
so if he gets. say 42% dems in a poll, we weights them down by only counting the Dem results as 39% of his poll
Many polls do not do this "hard weight"
Doing this is a bit controversial actually.
Do we really "know" that the ratio of Dems/Reps.Indys is 39/35/26...?
In 1994 a lot of polls weighed congressional races something like +5 to the Dem side, but on election day the ratio was actually +2 to the GOP side - hence a lot of polls utterly missed the GOP tidal wave of 1994
To the degree that Rasmussen's +4 Dem weight is accurate, his poll will be too.
To the degree he is wrong, his poll will have a structural flaw.