Pollster.com

Articles and Analysis

 

Poblano's Model

Topics: 2008

My latest NationalJournal.com column, about the remarkable success of the non-poll statistical model created by Poblano of FiveThirtyEight.com, is now online.

 

Comments
jsh1120:

I've been impressed (to say the least) with Poblano's work. I'm curious though as to whether the approach is more accurate and reliable when applied to primary elections where party identification is not a factor than in general elections where it plays a significant role.

Apologies if Poblano has addressed this particular issue. Or if the question is better directed at his/her website. Just curious how critical the most important attitudinal variable (i.e. party choice) is likely to be compared to the current "primary" election model.

____________________

tom brady:

I've also been impressed with Poblano's work, and enjoy reading his posts, particularly since - unlike many pollsters - he is upfront about his methodology. Ironically, perhaps, it's that transparency that allows me to take a slight issue with Mark's claim: in fact, Poblano hit the nail on the head in NC despite getting the numbers wrong in terms of the model's estimate. A big reason he predicted the double-digit margin was his belief that African-American turnout would be closer to 40%, rather than the 33-35% predicted by most pollsters. In fact, the pollsters were right, Poblano wrong. But he nailed it anyway in part because of the lopsided support for Obama by African-American voters. The larger point, however, that Poblano and others have grasped during this election cycle, is how important demographics are to driving the outcomes.

____________________

akamrearl:

Question:

After doing the best you can, what would the Poblano model predict for the november election? Which states would go demo, which would go repub?

____________________

Chris G:

Poblano deserves a lot of credit for introducing multiple regression into the discussion, but I think the only real take-home message from his model's accuracy in the primaries is that the Clinton and Obama coalitions were remarkably stable, as Mark B points out.

from what I can tell the model relies heavily on that stability because all of the inputs are static, they do not involve time-dependent patterns at all. so without coalition stability there'd be a lot more error in model estimates.

so as far as future predictions, I'd like to see something that also factors in how persuadable different groups are. plenty of data on that and methods that combine regression with time-series analysis

____________________

kingsbridge77:

Shame on Pollster.com for never showing interest in how Latinos have voted during these primaries. This site is living in a black and white world.

Sergio Bendixen, a pollster who worked for Clinton, was crucified by the race baiters for saying that Hispanics are less likely to vote for an African American.
Will pollster.com tell us whether he or the race-baiters were vindicated during these primaries?

Here is a nice article about how some faked outrage about Bendixen's well-intention analysis, which was just the answer to a question, and which may well be accurate:

http://www.dailyhowler.com/dh021908.shtml

____________________

Rasmus:

akamrearl,
Poblanos site is mainly about the General Election. His IN and NC predictions were just a sidekick. So just visit the site and you get detailed informations what he predicts for the General Election, including a non-poll regression model (that is affected by poll results, even from other states) combined with a poll weighting system- updated daily.

Roughly, the chances for both Democrats are about 50% to win the election. Obama 48% (was 50, but went down two points because of the WI poll), Clinton 47%.

Clinton is more the traditional Democrat, Obama is weaker in some main swing states (OH, PA, FL, MO), but has a good chance to win states where Clinton is the underdog (OR, WI, WA, MN, MI, IA, CO, NM, NV) and good outsider chances to win a lot of Republican states in a good election (VA, NC, AK, MT, ND, NE, even Utah on a good day).

____________________

jsh1120:

^^^ Not meant to be a criticism of Poblano, whose work I enjoy immensely, but it should be noted (I believe) that his poll-influenced predictions are weighted heavily (out of necessity) toward Rasmussen's state-by-state polls. Since I'm inclined to believe that Rasmussen's results have a GOP tilt, especially in state polls, I have a feeling that Poblano's Electoral College results are biased in that direction, as well.

____________________

StatsProf:

Mark,

Looking at Poblano's models show them to be off the mark (yes, I could not resist). The margins may fit, but the vote predictions are off which means the models are not a good representation of the actual vote.

You should insist that the modelers you promote show their work and not just the results. Many of my students stumble onto the correct answers, but their work proves their models were terribly flawed.

That, unfortunately, is the case here and you fell for it.

____________________

richard pollara:

Mark:
Over the last three or four weeks I have posted several times that I thought the polls (which had such large spreads) were not predictive and that a demographic overlay of the states would yield more reliable results. Pablano seems to have proven that theory. I would like to add one sentence to your statement in the National Journal: "Bad news for Obama, worse news for Clinton," Terrible News For the Democratic Party! As you pointed out, despite all the good press and bad press Obama's support amongst white working class Democrats has barely budged. Without securing a reasonable chunk of that voting bloc (along with women, seniors and Hispanics) his ceiling is very low. In North Carolina almost 60% of ALL Obama votes came from AA's. Such will not be the case in November. An Obama nomination will cause a fundamental shift in the core Democratic coalition. In 1972 the Democratic Party turned its back on the center with disastrous results. I am afraid we are about to repeat history.

____________________

BitJam:

If Poblano's work is correct then it raises the question:

What the heck is wrong with all the polls?

If demographics are better than polls are predicting election results (a big if) then the pollster need to either improve their performance or consider gracefully bowing out.

____________________



Post a comment




Please be patient while your comment posts - sometimes it takes a minute or two. To check your comment, please wait 60 seconds and click your browser's refresh button. Note that comments with three or more hyperlinks will be held for approval.

MAP - US, AL, AK, AZ, AR, CA, CO, CT, DE, FL, GA, HI, ID, IL, IN, IA, KS, KY, LA, ME, MD, MA, MI, MN, MS, MO, MT, NE, NV, NH, NJ, NM, NY, NC, ND, OH, OK, OR, PA, RI, SC, SD, TN, TX, UT, VT, VA, WA, WV, WI, WY, PR