Tuesday, April 22, 2008

Where did we go wrong?

The networks have called it for Hillary and I think after Florida in 2000 they're not going to make that mistake again, so obviously our polls were wrong.

First off, please do not call us or e-mail us and tell us we suck! We are well aware, and it does not feel good.

It's pretty easy, based on the exit polls, to see where we went wrong. We had the black vote at 18% when it turned out to be 14-15%, and we had the under 45 vote at 41% when it turned out to be 31%.

I reweighted the results from our final Pennsylvania poll to those figures for race and age, and the result of our poll was flipped- Hillary leading 49-46. Assume she pulled 60% of the undecideds and that gives her the 52-48 lead that the extrapolation from the original exit polls does.

So it's pretty easy to see where we went wrong. I'm glad that it was misweighting rather than some sort of systematic issue with our polls. Obviously we'll take some heat in the next few weeks for being so far off, and we certainly deserve it, but at the same time I don't think this should completely discount our credibility as a company- we've been pretty darn good a lot of other times this cycle.

Update: I posted this after Fox called the race while the exit polls still extrapolated to four points. Obviously the margin got greater as the night went along. Looks like the two areas where we were particularly off, as might have been expected, were with white women and senior citizens. We showed the race too close with both of those groups in our final poll, and Hillary ended up winning them by the kind of large margin that would have been expected. Live and learn...

We will just move on and try to do better in North Carolina.


Anonymous said...

Hilarious backtracking. Hopefully you all will lose the self-righteous attitude now.

Anonymous said...

shake it off.

are you doing Indiana?

Anonymous said...

You were only 13% off. Don't worry about it.

Will you stop manipulating models now?

John said...

You were the 2nd most accurate national pollster coming into this race. Even with this included, you will still be up there.

Ignore the idiotic anonymous' who are sure to continue posting.

Anonymous said...

"but at the same time I don't think this should completely discount our credibility as a company-"

You got a lot of nerves. You did everything you could to sow doubts in the minds of the voters in PA and give fodder to those Obama-worshipping so-called pundits in the media.

Do us a favor and just quit it.

Anonymous said...

I love it that you guys have a blog! I imagine you're wishing right now that you'd waited an hour or so before writing it? ;-)
13% off requires slightly more severe tinkering with the model!

Anonymous said...

Wow, you guys really suck.

Anonymous said...

Don't feel bad. I heard that Zogby cheated (one of my fraternity brothers knows someone who heard that supposedly Hillary gave them the answer key) and that's why they hit it right on 10% in Pennsylvania. She'll stop at nothing to win, so what are you going to do? Obama has integrity and it's better to stick with him. Dude, cheaters never prosper (in the long run) and Indiana will be a lot better, so hang in there. Obama Rulz, Hillary Droolz!!

Anonymous said...

"To err is human" ... make up for it in North Carolina and Indiana.

Anonymous said...


Question: I get where your error was, but my question is why? Why didn't African Americans and young voters come out in higher proportion? Given the enthusiasm and the nature of the candidacy, your original expectations were not crazy, so what happened? I would have expected dramatically higher than ever before turnout of these two groups, but it didn't seem to happen. Any insight into why not?


Anonymous said...

same question:

why wasn't there a massively historic and disproportionate turnout of young voters and african american voters? Same thing back in TX I never saw an answer to: black voters there voted at same turnout rate as 2004. Why? How is it that he easily pulls 2X or more people than Hillary to every rally he does, everywhere he goes, and yet we aren't seeing a disproportionate turnout of his demographic at the polls?


Anonymous said...

You can't help the people didn't show out and vote. I think most didn't because the polls showed Clinton winning and the Obama voters didn't think their votes matter. Also are you guys going to poll IN?


RS said...

I am really glad you have a blog - easier to contact you folks.
I see you did pretty well in WI, TX and OH on the overall results. Is it possible for you to post your final demographic compositions and splits and compare that to the exit polls? That will show if your boosting of the African-American/youth vote in those primaries was correct. Apparently, this was not the case in PA. I could do it, but I figure it's easier for you (and maybe you already have)!
Thanks - look forward to your results for IN and NC (SurveyUSA just came out with a low 9% Obama lead in NC.)

Anonymous said...



Anonymous said...

PS Colleges are out before May 6th. Just a hint for your voter turn out models in IN and NC and black vote will be LESS than last time because of the interst in the two contests.

Anonymous said...

I have been impressed with your results until recently. Both you and Insider Advantage had very good final polls as they related to voting results. I was, however, concerned when I saw your expanded remodelling announcement for N.C. and then your immediate divergence from the IA results in PA. Given that your remodelling helped to dramatically change the CW with regard to N.C., (one point to 21 points in seven days) would you be willing to make the same adjustments to your N.C. data as you did for PA and report those results?

Anonymous said...

Ironically PPP polls helped Hillary far more than Obama. Showing him up by 3 made this race look a lot closer than it was. Without PPP factored in the expectations would have been even higher for Hillary.

So Hillary supporters should be thanking PPP and Obama supporters should be mad. Bad polling always helps the person they wrongly predicted would lose as it makes their win more meaningful and unexpected.

Come on people, think before you post.

Anonymous said...

For me, this missed result makes me question your latest results for North Carolina. You've polled him 10 points higher than everyone else in your most recent results... pretty much the same margin you missed PA by.

What is your methodology for North Carolina? Are you making the same assumptions in regards to weighting? Perhaps you've escaped systemic error while falling into incorrect weighting.

I'd like to see the next PPP numbers end up closer to the real results.

Anonymous said...

It was weighting so it wasn't a systematic issue of your polls?? That makes no sense. How you weight the data is a key component of a poll, and you do it with every poll, so how is that not a big deal?

And where did your weights come from anyway? Why should anyone believe that your weights will be any better on future polls?

Matt Willard said...

Yeah, so why don't you guys actually report your results using different weighting assumptions? Could it be that you're seduced by the idea of reporting your results with metaphysical certainty? How much of your error is due to your modeling assumptions, vs. sampling error? (This criticism applies equally to other polling firms.)

Anonymous said...

A couple of the earlier commenters were asking: Why didn't Obama's demographic base (Blacks and younger voters) turn out in bigger numbers and be a bigger proportion of all voters? I think the most important thing to appreciate here is that Hillary's demographic base showed up on the day. Specifically 2.3 million of Pennsylvania's 4.2 million registered Democrats voted. That's 54% of eligible voters voting. That's certainly one of the highest turnout percentages of eligible voters of any state in this year's primaries. It may very well be the tippy-top highest percent, although of course most other states have Open (or semi-Open) primaries, meaning lots more people are eligible to vote. New York State has a Closed primary. 32% of registered Democrats in New York State voted in New York's primary on Super Tuesday. (Looking at a sample of the states with Open primaries, if I only count the registered Democrats based on the exit polls, I get the impression turnout was in the vicinity of 30%).

In Pennsylvania Blacks were only 14% of the turnout because the whites showed up in big numbers, not because the blacks didn't show up. The same goes for the young people versus the old people. The people who preferred Hillary showed up.

Here's a paraphrase of the wise commenter "Poblano" (at fivethirtyeight.com) that he wrote the day before the election was held: For Obama to win or keep it very close he both has to win his own enthusiasm & get-out-the-vote game, and have Clinton lose hers -- and if anything, the latter is probably more important than the former. The largest single variable in the Pennsylvania primary is probably Clinton's ability to turn out voters who are really making a decision between voting for Clinton, and not voting at all.

It is extremely difficult and hazardous for any polling company to figure out this enthusiasm & turnout aspect in advance by using rational analysis. The approach taken by the SUSA polling company doesn't use rational analysis but has the virtues of simplicity, clinicalness, good empirical performance.

Anonymous said...

No problem. Plenty of polls make mistakes, and I'm willing to overlook this. It would be wise of you, however, to apologize to DailyKos for saying Markos was "full of it".

Web Statistics