If Pollster.com deserves the 2006 award for the Best Polling Website of the year, the 2008 award has to go to Poblano's FiveThirtyEight.com. This is by far the most interesting website for us statistics and polling wonks.
Poblano, a frequent Daily Kos contributor, is predicting the state-by-state primary results for Clinton and Obama based only on demographics and regressions...Look ma! No opinion polls. Given the long primary season, this is giving us wonks an unparalleled in-depth look at who makes up the Democratic voting populace, and how they are voting in the primaries.
Today's 538 prediction: Obama wins North Carolina by 17 points, resulting in a district-by-district delegate count of 66-49 Obama over Clinton. This compares with Chris Bowers' 7 point Obama margin and 62-53 Delegate count based on polling averages. Five districts are close to delegate thresholds. Why does poblano think the pollsters are wrong? It's mostly about their inability to predict the African American primary turnout:
By the way -- I think I know what the pollsters are doing wrong too. They're calibrating black turnout to a proportion of the population, but not to a proportion of the Kerry vote. This is a significant mistake, because in some states, the vast majority of the available white voters will vote as Republicans -- meaning that black voters make up a larger share of what remains in the Democratic electorate. I noticed for instance that SurveyUSA underestimated black turnout in Republican-leaning South Carolina (55% against their estimate of 41%) but overestimated it in Democrat-leaning New York (16% against their estimate of 21%), which would be consistent with this sort of error.
538 also keeps a running tally of the General Election matchup predictions on the front page. This requires an explanation, because the state-to-state results are displayed as probabilities. So, Massachussetts at Obama 88% and Clinton 99% shows the likelihood of winning the state, not the voting percentage.
What else makes 538 so cool? Let me count the ways:
Check out how well 538.com stacks up against the the professional houses for Clinton's margin over Obama in Pennsylvania. 538 was even singled out for some flack from American Research Group polling head Dick Bennett, despite the fact that 538 comes in just 2 points under the final vote, much better than ARG's 6.7 point miss.
(4) Rating the Pollsters: Sampling Errors and Pollster Induced Error (PIE errors): Rasmussan, Selzer, SUSA good, Zogby Interactive bad.
(5) General Election Projections (McCain-Obama vs McCain-Clinton.) Statistical approach using regressions, PVI, and polling, so that both demographic and regional effects are taken into account. Poblano readily admits we still have a lot of uncertainty this far out from the General. Present results show McCain over Obama with a popular vote advantage of 50.2%, and electoral vote 269.7 to 268.3. McCain bests Clinton in the model by 50.6% and 284.3 to 253.7.
STATE BLOGS
(3) How good is this statistical approach?
