Pollster.com

Shouldn't this be 46-41?

Also reposting a question from the Zogby thread in case that's dead.

Mark, A question about these tracking polls like Gallup and Rasmussen. Do they just do a simple arithmetic mean of the relevant days e.g. Gallup adds up candidate preferences for last three days and divide by 3 or something more complicated (I read somewhere about smoothing but not sure what that means)? Thanks

@zebedee:

Gallup has individual responses from each day, so to produce a 3-day average, Gallup just has to add up, say, Clinton supporters from each day, and divide that by the total respondents during that 3-day period. For the next 3-day period, Gallup just replaces the responses from Day 1 with the Day 4 responses, and you get the result for Days 2-4. And so on.
I really doubt Gallup calculates candidate preferences for each day and then produces an average of the daily values - that would just increase the uncertainty.

Mark B. does smoothing on his "poll of polls" - which is essentially a curve-fit of different poll results to produce the overall trend (unlike RCP, which I think does an simple arithmetic mean). Depending on the smoothing factor applied (this is an indicator of the sensitivity of the curve to individual polls - not entirely sure off-hand), the result can change. There was a post elsewhere on pollster.com that showed the effect of different smoothing factors on Pollster.com's "poll of polls."

RS, thanks for the reply. Not sure I understand the difference though. If the samples are the same size doesn't it come to the same thing - whether you average the candidate individual day percentages or add the number of voter over the 3 days and divide by 3? Or am I misunderstanding.

Also I was trying to work out something about the individual days from the tracking poll data. If HRC gains 4 points since yesterday's tracking results, it should mean the individual day was 12 pts better than 4 days ago. But if these are simple averaging tracking polls, they don't seem to fit unless you allow for variations af about 15% between adjacent days polling. Is that likely?