DEAN GAINING GROUND?

January 26, 2004

Comments

LATEST NEW HAMPSHIRE POLLS CONFLICT. Howard Dean is coming back strong in New Hampshire … or John Kerry is heading to huge win in the state. Which is correct? That depends on which polls you trust. The latest polls out of New Hampshire are offering sharply different forecasts for the likely vote there Tuesday. The new MSNBC/Zogby poll shows that Howard Dean has bounced back in the state over the past three days, and now trails John Kerry by only three points. The MSNBC/Zogby numbers: Kerry-28%, Dean-25%, Clark-11%, Edwards-10%, Lieberman-9%. Factoring in the 4-point margin of error, the Kerry-Dean race is a now statistical tie. Contrast the close race numbers with the numbers out Monday from the WHDH-TV/Suffolk University poll: Kerry-38%, Dean-17%, Clark-10%, Edwards-9%, Lieberman-5%. This polls shows Kerry slightly up, and Dean slightly down — the opposite of the MSNBC poll. Also see the numbers from the new Boston Globe poll: Kerry-37%, Dean-17%, Edwards-12%, Clark-11%, and Lieberman-7%. Sunday’s WMUR-TV/UNH poll showed similar numbers: Kerry-37%, Dean-22%, Edwards-12%, Clark-11%, Lieberman-10%, and Kucinich-2%. The forecast for several inches of snow on election day is expected to also skew the outcome, as it traditionally supresses the turnnout of some demographic groups — particularly seniors — more than others.

My instincts conflict here, in that I simultaneously think Kerry is going to win New Hampshire going away and yet have learned to disregard the Zogby poll at my peril. Interesting.

7 responses to “DEAN GAINING GROUND?”

melvin toast

January 26, 2004 at 11:04 am

Doesn’t anyone read history books? Iowa doesn’t mean anything and New Hampshire polls change with the wind. Dean wins the candidacy even if he doesn’t win New Hampshire.

He’s alive; he’s a straight shooter. Clark, Kerry and Edwards are frauds. I voted for Dubya and I’m voting for him again… but I’m tellin ya that if Gore had half Dean’s charisma, he’d be President.
Moe Lane

January 26, 2004 at 11:29 am

Zogby is extremely good starting at about 48 hours before any given election and fairly useless before that point. Go ask a dKos alumni about Zogby and Nov 2002 if you don’t believe me. 🙂
Ron

January 26, 2004 at 11:37 am

Polls do not control for multiple treatment interaction. The result is that polls have no internal validity (i.e. the results are likely not valid for the sample), without internal validity there is no external validity (i.e. the results are likely not valid for the population).

The published error margins assume internal validity, and are only meant to cover the inference to the larger population. Without internal validity, the error margin is unknowable.

Polls are better than nothing, but not by much.
James Joyner

January 26, 2004 at 12:00 pm

Ron,

Multiple treatment interaction is a threat to EXternal validity, not internal. Public opinion have all manner of problems but those taken this close to an election are markedly better than nothing.
James Joyner

January 26, 2004 at 12:02 pm

Moe,

All polls have problems predicting very, very close elections, as both the 2000 and 2002 elections were. Even if everything was perfect, 3-4% is a wide margin if the results are within it!
Moe Lane

January 26, 2004 at 1:04 pm

“All polls have problems predicting very, very close elections, as both the 2000 and 2002 elections were. Even if everything was perfect, 3-4% is a wide margin if the results are within it!”

Point, but I was thinking of the way everything suddenly shifted in ’02, just before the election. Granted, that particular blindsiding had other factors – I remember a constant reiteration that superior Democrat GOTV procedures in and of themselves would be enough to gain seats, and never mind why – but the last-second shift surprised people.

Something for the Republicans to bear in mind for 2004, actually.
Ron

January 26, 2004 at 1:40 pm

It is possible that my memory is flawed (especially since it’s been a dog’s age since experimental design). The Campbell & Stanley list of internal and external threats at this link does not include it under either. Some links recently Googled had it under external, and some under internal.

It makes more sense to me that multiple treatment interaction would be a threat to internal validity, since the interaction would affect the results of my sample. I’ll check it out tonight.

—