Perhaps someday I will learn to read these things before hitting send, but since that day has not yet come, allow me to register my regret for that first, errant comma before Nate Silver's name and note that I meant "stated preference" not "state preference" in that last clause of the third paragraph. Like the mechanic whose car is always in disrepair, I am the editor who won't edit himself. -Wm. -----Original Message----- From: lit-ideas-bounce@xxxxxxxxxxxxx [mailto:lit-ideas-bounce@xxxxxxxxxxxxx]On Behalf Of William Dolphin Sent: Friday, November 07, 2008 8:23 AM To: lit-ideas@xxxxxxxxxxxxx Subject: [lit-ideas] Re: Polls: for future reference The most accurate prediction of the election's outcome appears to have been not a pollster at all but, Nate Silver, a young baseball fan whose models for projecting results are based on algorithms he developed for fantasy baseball. He projected the popular vote totals to within one-tenth for both McCain and Obama. Or at least that's what it looks like now, with many provisional and absentee ballots still to count. Silver's method takes polls as some of the raw data, but weights them based on their performance measuring previous races and incorporates historic voting patterns, demographic data and a secret special sauce. You can see his work at http://www.fivethirtyeight.com. But to address Julie's question on how they cook up their numbers: Pollsters make assumptions about the actual composition of the electorate on election day -- the proportion of younger voters, Latino voters, etc. -- that may or may not prove to be what actually happens. Those "likely voter" models that most use are largely based on prior elections and the stated behavior of those polled. So if you are a 56-year-old white woman with income between $75-100k who voted in 2004 and say you are sure to vote this time, the pollster is highly likely to count your stated candidate preference as a vote for that candidate on election day. Conversely, if you're a 19-year-old male high school dropout making $19k and say you'll probably vote, then your state preference may be discounted in the computing the polls. Even without likely voter models driving the math, any given pollster will tend to match their polling sample to the expected composition of the electorate, interviewing a predetermined percentage of young, old, rich, poor, etc. If those expectations are wrong, say because a particular candidate has energized a voting segment or because the demographics of the electorate have changed (e.g. increased registration of Latinos or college age folks), then the accuracy of the poll result will suffer. Personally, I'm still scratching my head over the vote totals and procedures in Alaska and Georgia. In Alaska, if the official tallies are to be believed, polls showing convicted felon Ted Stevens trailing in his reelection bid by substantial margins were wrong because voters would not admit they intended to vote for a criminal (making Stevens the new Bradley, I guess), and turnout was actually 20% BELOW 2004 levels, even though a native daughter was on the top ticket. Anyone find that believable? In Georgia, apparently the Sec. of State did not total or include or have counted (the details are sketchy at best) the early votes cast in the metro Atlanta counties, where 40% of voters cast ballots before election day and Obama leads. The difference may not swing it red to blue for the Presidential contest, and the 15 electoral votes are of little consequence given the rest of the national map, but since I picked Georgia as my election night Obama-shocker, I'm still holding out, yes, hope. William Dolphin Memphis -----Original Message----- From: lit-ideas-bounce@xxxxxxxxxxxxx [mailto:lit-ideas-bounce@xxxxxxxxxxxxx]On Behalf Of Julie Krueger Sent: Thursday, November 06, 2008 2:29 AM To: lit-ideas@xxxxxxxxxxxxx Subject: [lit-ideas] Re: Polls: for future reference To what (in terms of polling tactics, etc.) do attribute that? On Wed, Nov 5, 2008 at 10:42 PM, Eric Yost <mr.eric.yost@xxxxxxxxx> wrote: The List: Which presidential polls were most accurate? The Pew Research Center and Rasmussen Reports were the most accurate in predicting the results of the 2008 election, according to a new analysis by Fordham University political scientist Costas Panagopoulos. The Fordham analysis ranks 23 survey research organizations on their final, national pre-election polls, as reported on pollster.com. On average, the polls slightly overestimated Obama's strength. The final polls showed the Democratic ahead by an average of 7.52 percentage points -- 1.37 percentage points above his current 6.15-point popular vote lead. Seventeen of the 23 surveys overstated Obama's final victory level, while four underestimated it. Only two -- Rasmussen and Pew -- were spot on. http://blogs.chron.com/txpotomac/2008/11/the_list_which_presidential_po.html ------------------------------------------------------------------ To change your Lit-Ideas settings (subscribe/unsub, vacation on/off, digest on/off), visit www.andreas.com/faq-lit-ideas.html -- Julie Krueger No virus found in this outgoing message. Checked by AVG. Version: 7.5.549 / Virus Database: 270.8.6/1769 - Release Date: 11/5/2008 7:17 AM