Charles Franklin | November 1, 2008
Topics: House Effects
(Click once or twice on chart for full size image, which may be the only way to read this!)
There has been some demand for an estimate of house effects beyond those for just the tracking polls, so here you go.
These are ALL national polls in our database for McCain v Obama. The house effect is estimated relative to the Pollster.com Trend estimate at the time of the survey. That means the results are comparable over time despite movement in the polls overall.
Two notes. The "zero point" is the mean of all pollster effects. That is the best estimate, in my view, but it should be recognized that this may not be the "most accurate". For that, wait until Wednesday. Still, there is a strong reason to take the middle of the pollsters as the natural "neutral" point, as I do here.
Second, The tracking polls are included here, but are mixed with other polls by the organization. I don't have a convenient way to distinguish ABC/WP tracker from ABC/WP "regular" surveys. So they get mixed together here. If you want to compare the trackers, go to yesterday's post on trackers only. Also, the trackers here are the "independent polls", meaning non-overlapping tracker results, while the previous tracker post included ALL the tracker results. Finally, since the zero-point here is the mean of ALL these polls, while the zero-point in the tracker post was the mean of just the trackers, you'll find that these results differ from the previous post.
The good news here from my perspective is that the vast majority of polls are within a -5 to +5 range. Those outside that are almost all single polls, with wide confidence intervals. So there ARE significant house effects, but the magnitude is not as large as some seem to think.
Confidence intervals are wider for pollsters with either few polls OR pollsters who vary widely in their OWN results (i.e. they jump around a lot relative to other polls) or a combination of both these.
Also, the population sampled plays an important role in these house effects, so I've distinguished organizations by the type of sample used in their polls. Hence ABC/WP-RV and ABC/WP-LV.
Enjoy finding your favorite pollster.