I used to teach a course in satellite tracking, which involves a fair amount of statistics. For your poll debating pleasure, I offer this from the first year of Obama’s first term. Poll Tracker, now defunct I believe, let you do all kinds of manipulations on polls. Each red (disapproval) and black (approval) dot is a single poll. The solid red and black lines follow the running averages of the polls. Notice the scatter in both the red and black swarms. As most of these were 1000 person telephone polls, I could point out to my students that the uncertainty would be about
sigma ~ 1/sqrt( 1000 ) ~ 0.03
or about 3 percent. That’s exactly what the graph shows. But one feature now lost in the image is you could isolate an individual pollster. Just below the blue dotted line (which I added, the only modification) is a line of red dots from one single pollster. They nicely follow the average curve of Obama’s disapproval. In fact, they follow the average extremely closely. They follow the average without the expected plus or minus 3 percent error. The only way you get polls without statistical error, is to fabricate the data yourself. Looking closely, this one pollster’s results pull the red average disapproval curve up by a full percent.
Lies, Damned lies… and statistics.