« More on "Eugenics" | Main | 15 Years »

July 31, 2007

New ARG Polls

The new American Research Group polls are out, and they show Edwards slipping behind Clinton in Iowa, and Obama tying for first in New Hamshire, and taking the lead in Soth Carolina.

July 31, 2007 | Permalink

Comments

You know ARG polls are bullshit, right?

Posted by: david mizner | Jul 31, 2007 1:52:15 PM

Either the ARG poll is dead on and everyone else is crazy (including the Des Moines Register poll that historically has been very accurate), or Clinton does not in fact have a lead in Iowa. I'm betting that Clinton does not in fact have a lead, given that this is the first year that ARG is polling Iowa.

Posted by: Clark | Jul 31, 2007 2:49:08 PM

On MYDD they just said that the sample size was perhaps not restricted enough to represent caucus participants.

Posted by: Paula | Jul 31, 2007 3:26:25 PM

ARG doesn't actually show Edwards slipping behind Clinton in Iowa -- they say that Clinton has led Edwards in Iowa almost all the way. Back in January, ARG had Clinton over Edwards by 17 in Iowa, when all the other pollsters were putting Edwards on top. I don't know anything about the methodology here -- it could be that ARG has Iowa right, Clinton has always been ahead in Iowa, and all the other pollsters have the state wrong. Not what I'd bet on, though.

One of the weird things about the Iowa polls is that partisan polls from every side show Edwards with a big lead. (The Richardson campaign recently released one that showed Edwards with a 34-24 advantage over Hillary.) ARG is the one that shows Hillary with a consistent lead, and everyone else is more or less in the middle.

Posted by: Neil the Ethical Werewolf | Jul 31, 2007 4:02:46 PM

This poll across the board, both for the GOP and Democrats, doesn't make much sense. I am going to go with something is wrong with the polling outfit over thinking that the numbers in say SC make much sense.

Posted by: akaison | Jul 31, 2007 4:32:39 PM

"ARG doesn't actually show Edwards slipping behind Clinton in Iowa -- they say that Clinton has led Edwards in Iowa almost all the way."

Yup. ARG has had Clinton ahead of Edwards 6 out of the last 7 months in Iowa, including the last 3 months straight.

"it could be that ARG has Iowa right, Clinton has always been ahead in Iowa, and all the other pollsters have the state wrong. Not what I'd bet on, though."

I believe the relevant word can be found here...

Posted by: Petey | Jul 31, 2007 6:04:36 PM

Petey, you linked to the wrong wiki page. What should've linked to is this, but of course, that begs the question, which system is producing the error...

Fact is, all the different polling firms have different voter screens and some are even measuring different things (Dems vs Dem voters vs Likely Dem Voters vs Primary voters vs Likely Primary Voters...). Its pretty hard to know why they're different specifically, and as such difficult to tell if those differences are meaningful and for whom do they favor.

That being said, ARG has been consistently the most Clinton-friendly of the firms, so there's no way to read the NH and SC data that isn't damaging to Clinton, except as literally coincidental error.

Posted by: Michael | Jul 31, 2007 11:12:37 PM

What everyone from mizner on down have said. ARG's polls are unlike everyone else's in all three states, and this isn't exactly the first time they've consistently given misleading readings.

Remember last year in CT where ARG twice showed Lamont and Lieberman in a statistical dead heat in the general election? Everyone else was showing that Joe was ten or more points ahead. Everyone else was right, and ARG was wrong.

Posted by: low-tech cyclist | Aug 1, 2007 4:52:36 AM

There are two distinct possibilities related to ARG's likely voter model:

  1. They have a likely voter model that differs regarding the importance of some demographic that tends to support Clinton
  2. They have a likely voter model that is unreliable for a caucus, but is reliable for a primary

In the first case, what is more important than the levels compared to other polls is the trendlines ... that is, if the trendlines over the past five months in NH shows a slow but steady slide for Hillary, a slide and then a small recovery this last two months for Edwards, and Obama sliding then surging, then with a biased but consistent likely voter model, that would tell us much of interest, whether or not they are biased up or down for a particular candidate:

  • Hillary: +0%, -3%, +0%, -3%
  • Obama: -9%, -11%, +10%, +8%
  • Edwards: +6%, -8%, -7%, +3%

In the second case, they could well be spot on in NH and SC, but unreliable in IA and NV.

Posted by: BruceMcF | Aug 1, 2007 9:25:40 AM

The comments to this entry are closed.