Reweighting Newsweek’s ‘08 Poll
by Patrick Ruffini :: January 20th, 2007 6:42 pmIt doesn’t take a genius to figure out that Newsweek’s polling leaves much to be desired. Nevertheless, a poll showing Hillary essentially tied with the leading Republican candidates will probably get a lot of attention on a weekend where the former First Lady took the plunge with her exploratory committee.
Newsweek shows Hillary leading John McCain by 1 and trailing Rudy Giuliani by 1. John Edwards’ advantage over McCain and Giuliani is 5 and 3 points respectively. Obama leads McCain by 2 and trails Giuliani by 2.
In polling against Hillary, both leading Republicans capture more of their base than any Democrat does their base, while generally getting more of the independent vote. So, the Party breakdown of registered voters in this poll comes as no surprise:
Republican 26%
Democrat 37%
Independent 33%
No party/Not interested 2%
Other party *
Don’t know 2%
That’s right, the poll is +11D, and there are suddenly 30% less Republicans than there were last November.
How would these results look if the responses were re-weighted to reflect the makeup of the electorate in 2006, an overwhelmingly Democratic year?
Giuliani 51.8% - Clinton 44.6%
McCain 50.5% - Clinton 45.2%Giuliani 50.6% - Obama 42.5%
McCain 47.4% - Obama 44.4%Giuliani 49.3% - Edwards 45.2%
McCain 47.2% - Edwards 45.9%
Tagged: 2008 
hillary clinton john mccain rudy giuliani
Comments (16) 
Trackbacks (9) 
del.icio.us 
digg it 
subscribe
Both comments and pings are currently closed.
It’s not a good idea to reweight samples the way you do. The number of self-declared republicans and democrats varies a lot over time. In 20044 it was liberals who kept claiming that the polls overweighed Republicans. Pollster.com has several posts on this topic.
The Newsweek poll results are very bad for the GOP but they in line with several other major polls released recently, as a quick look at pollingreport.com shows.
Gabriel,
Actually, party affiliation stays pretty constant over time, at least by the only measuring stick that matters: elections. In 2006, the electorate was +2 Dem despite a horrible environment for the GOP. Before the election, I think it was Michael Barone who showed that in the last 15 years, party affiliation in exit polls has fluctuated within a very narrow band of +5D to +2R. The question was whether a good Democratic year would lead to the +8-10D number the polls were projecting. It didn’t. The only polls that got the final generic number right were the ones the weekend before the election that resolved on likely voters.
Now, party affiliation in *polling* can fluctuate wildly over time. It did during the post-Convention bounce in ‘04 for the GOP (what you’re referring to) and it’s probably doing that now for the newly minted majority Democrats (even ABC/WaPo had Dems up +12 in party ID). But such surges never show up in elections. The standard is not to project what an RDD phone universe thinks, but to project what the likely electorate thinks. There’s a difference.
The current political environment has been horrible for the GOP! In fact, I’d argue that the past 6 months (maybe even longer) have been as much pro-Democrat and anti-Republican as I’ve seen in at least 8 years. Bush’s popularity is now abysmal (much to my disappointment–but it’s true). The fact that both Giuliani and McCain are essentially tied with the 3 top Dems (or perhaps are even leading if the poll were more accurate, as Patrick points out) is reassuring in this negative current environment. And it shows how strong Rudy and McCain would be despite the misgivings about them that some Republicans hold.
Patrick,
I agree with that. My point is that at this point we don’t know what the party affiliation will look like in 2008 and there is no reason to reweigh the Newsweek poll. It is favorable to Dems but it is roughly similar to other polls in recent weeks. As you point out even the Wapo poll shows a similar uptick in dem self identification.
We are not going to predict today what will happen in 2008. In a couple of weeks even these polls will be old as new information comes out. I understand that for GOP supporters there is a lot of psychological benefits to coming up with reasons to dismiss polls they dislike. Happens with Democrats as well. I still remember trying to convince myself in 2004 that the polls did not say what they in fact were saying.
The polls generally show the GOP is in a bad position right now. The last month with the new Democratic Congress and the public’s growing rejection of Iraq, have only made things worse. Novak is reporting that GOP insiders think 2008 will be worse than 2006.
Isn’t everyone tired of this re-weighting game? Democrats were upset in ‘04 when it showed Bush had the lead, and Republicans were certain of a vast conspiracy in 2006 when Democrats had the lead. In both cases, the polls turned out to have more right than wrong. If Bush could only beat Kerry by 3%, it isn’t likely that either party is going to beat each other by the 5-7% your re-weighting has found here.
Well, unless the GOP nominates Tancredo or Gingrich. (Please, please, please)
Gabriel,
If your sample doesn’t accurately represent the voting population, you’re polling a parallel universe. Sure, it’s theoretically possible that the national electorate could skew 55/45 male, or 20% African American, but highly unlikely given their makeup of the population and their propensity to vote. Any sample showing numbers like that should be reweighted.
I understand that party ID isn’t exactly the same. Someone can change their party at any time. But historically, polling has drastically overestimated these shifts. This problem isn’t endemic to polling an election two years from now. Most polls weeks out from the ‘06 election got it wrong — showing a Dem Party ID margin somewhere on the order of 8 to 10.
I personally wish that we’d settle for some sort of rolling average of party ID — take the average Likely Voter party ID from the last 10 polls and average it — so we’re not flying blind with just exit polling data and we can spot the party with momentum a point or two.
Patrick,
I don’t think there is a single easy way to solve this and different pollsters use different techniques. IIRC Newsweek is one of those that lets the chips fall were they do, meaning they don’t use any specific weighs. Plus it’s January 2007, way to early to start worrying about that.
The poll fits well with other recent polls. As you say even the WaPo poll shows a Dem surge so there’s nothing special about this one.
Also you say that most polls were showing much higher party ID than was the case a few weeks before the election. Yet the vote numbers were pretty much in line with what was expected. It seems to me that the Party ID tends to converge to its ‘natural’ state as the election draws closer but the vote intention does not change as much.
The bottom line in all of this is that the Newsweek poll simply confirms several others showing the GOP in serious and growing troubles.
“Yet the vote numbers were pretty much in line with what was expected.”
Actually, they weren’t. Democrats won the House vote by something like 4.5%. The polls a week out were calling it for the Dems by 12-16 points. The difference can all be explained by the last-minute party-ID “shift.” Midterms for whatever reason tend to resolve late in terms of getting a good read on likely voters — we didn’t know how well the GOP would do in ‘02 until that Monday.
What was in line with expectations was the race by race polling and the number of seats lost. (I shudder to think what it would have been like had that 12 point polling margin held.) This is not an unknown phenomenon — local polling can be a good diagnostic tool when the national polls are wrong. In 1996 and 2000, state polling weeks out accurately predicted closer elections than the national polling indicated (on either side), possibly because of the larger number of state polls and an over-reliance on a few outlets on the national level that “let the chips fall where they may.”
Likewise, if you want to pick a fight on ‘06 polling, I’d say you have a much better argument against those of us who were arguing against the screwiness of CD-based polling. Yeah, any given poll was more likely to be off because of low sample sizes, but except for a slight Dem lean in the Majority Watch polling, if you average them together they pointed towards a 25-30 seat loss. And in general, they were conducted by partisan pollsters or outlets like SurveyUSA that tend to weight their samples for Rs and Ds.
Hotline calculated that the Dems won the two party congressional vote by more than 8% last November.
http://hotlineblog.nationaljournal.com/archives/2007/01/just_how_bad_wa_1.html
A quick look at http://www.pollingreport.com/2006.htm shows that you are right that most pre-election polls showed a bigger spread. But Fox News, hardly a left-wing operation, had the Dems up by 13 right before the election. It doesn’t appear that mistakes have anything to do with ideology.
Still, as you point out that was a midterm and we are talking about a presidential election. I don’t remember how much party ID variation the 2004 and 200 election polls had. Do you?
Finally, I have come to focus much more on poll averages and trends. In that sense the new WaPo poll confirms the Newsweek poll.
lest we soon forget. It seems that americans are quick to forget what types of unanswered questions came up back during the Clinton administration involving both Bill and Hillary.
A better comparision polling wise would be to look at the reweighed samples done by Patrick and compare them to recent resultes from Rasmussen (who by now has to be considered the gold standard of election polling).
What you will find is that Patrick’s reweighed projections are very close to what Rasmussen is polling… I am guessing that further investigations into the numbers from the other pollsters from Pollingreport.com will show a similar bias. Probably to build up a story that 2008 is up for grabs… when in fact even with things as bad as they may seem for the GOP… The Democrats still have an uphill battle.
The Hotline calculation referenced above is based on candidates, not voters. Even with regard to candidates it draws questionable conclusions, stating that Democrats won Tennessee based on House votes, but ignoring the statewide Republican victory in the Senate race.
I honestly hope those percentages hold up to the 2008 elections and a Republican is elected President but that’s an awfully long time between now and the 2008 elections. I still see it as a Clinton/Obama ticket for the Democrats. I just wonder how nasty the campaigning will get for the Democratic nomination.
BTW, where’s Bill in all of this? I wonder if Bill & Hillary are pretty much divorced in all manner except on paper?
Speaking of Obama…apparently he has the backing of George Soros and his ilk…especially the Soros-backed and influenced slander blog/web sites. I wonder how long it will take before a) they start attacking Hillary and/or b) like rats from a sinking ship, they jump/latch onto Hillary’s platform?
CH,
In fact most independent analysts ((think Rothenberg) have stated that its the GOP that has the uphill battle. Which makes a lot of sense.
Since WW2 the two parties have traded the WH every 8 years like clockwork, with the exception of 76-92 when instead of 8 years each it was 12 GOP and 4 dems. It’s not an iron law of politics but it’s close as it gets and based on solid reasons. After a while Americans get tired of whoever is in power and want a change.
Today 2008 doesn’t look bad for the GOP. It looks horrible, something recognized by GOP insiders as Novak points out. Right now the most likely outcoume if Dems sweeping all three in ‘08.
Of course a lot can change but the big, huge problem the GOP has is Iraq and that doesn’t seem to be going anywhere. It’s not just that Pelosi is now more popular than Newt ever was as Speaker or that Bush is now in the 20s. The GOP has even lost its national security advanatge and more Americans prefer Dems over Bush on Iraq.
[…] Patrick Ruffini slaps down Newsweek magazine’s latest presidential poll. […]
[…] Patrick Ruffini did it with the latest Newsweek poll. Now it’s time for Yours Truly to re-weight the latest Diageo/Hotline ‘08 poll in order to reach a more accurate assessment of the race for the White House. Here’s the poll as it was released, with a 40D/34R/26I split: WH ‘08 General Election Matchup […]
[…] Patrick Ruffini did it with the latest Newsweek poll. Now it’s time for Yours Truly to re-weight the latest Diageo/Hotline ‘08 poll in order to reach a more accurate assessment of the race for the White House. Here’s the poll as it was released, with a 40D/34R/26I split: WH ‘08 General Election Matchup […]
Newsweeks Poll Needs Some Work…
Patrick Ruffini shows us the error of Newsweek’s ways ……
Newsweek’s Bombshell: Democrats Prefer Democrats…
Newsweek has published a poll gauging voter sentiment about a number of prospective 2008 Presidential match-ups. At first glance, it seems to bode well for Democrats, with Edwards being unbeatable, and McCain being hopeless. All the match-ups fell wi…
[…] Patrick Ruffini notes the RV sample on the Newsweek poll is R 26%, D 37%, and I 33%. When adjusted to reflect the composition of the electorate in the 2006 midterms, Ruffini comes up with the following totals: […]
[…] My post last weekend on the Newsweek poll’s severe Democratic lean was a bit of a hit, so I made a beeline over to this week’s results, with blazing headlines about Presidential approval reaching a new low. […]
[…] The results of the head-to-head general election match-ups are worthless as Newsweek, true to their liberal ideological bent, have once again skewed their own survey’s data by deliberately polling a disproportionate number of Democrats. The survey’s breakdown was 52 percent Democrat/lean Democratic and 37 percent Republican/lean Republican. By party affiliation, it was 40 percent Democrat, 29 percent Republican and 27 percent Independent. Using the National Election Pool’s General Exit Poll from the 2006 midterm elections as a frame of reference, this Newsweek poll oversampled Democrats by nine percentage points. A Rasmussen poll released at the end of May showed there to be 5.5 percent more Democrats than Republicans. Therefore, it appears Newsweek has built in somewhere between a 5.5 and 9 point cushion for their Democratic peers; a heavy finger on the scale most certain to be ignored when reported with glee tomorrow morning by Tim Russert, George Stephanopoulos and Bob Schieffer that Hillary Clinton, Barack Obama and John Edwards all defeat each of the top Republican contenders. […]
[…] The results of the head-to-head general election match-ups are worthless as Newsweek, true to their liberal ideological bent, have once again skewed their own survey’s data by deliberately polling a disproportionate number of Democrats. The survey’s breakdown was 52 percent Democrat/lean Democratic and 37 percent Republican/lean Republican. By party affiliation, it was 40 percent Democrat, 29 percent Republican and 27 percent Independent. Using the National Election Pool’s General Exit Poll from the 2006 midterm elections as a frame of reference, this Newsweek poll oversampled Democrats by nine percentage points. A Rasmussen poll released at the end of May showed there to be 5.5 percent more Democrats than Republicans. Therefore, it appears Newsweek has built in somewhere between a 5.5 and 9 point cushion for their Democratic peers; a heavy finger on the scale most certain to be ignored when reported with glee tomorrow morning by Tim Russert, George Stephanopoulos and Bob Schieffer that Hillary Clinton, Barack Obama and John Edwards all defeat each of the top Republican contenders. […]