For those who don't believe in polls...

[Kurt_and_Hunter]

For those who "don't believe" polls, or like to point out that "only" a few hundred people are sampled...

Two organizations polled about 400 NH Democrats at the same time last week, with a sampling error margin of about 5%. Looking at the results, the only sensible conclusions are that either polling is indeed a scientific process that reveals something actual about a population, or else Marist and the University of New Hampshire colluded to fake the polls in exactly the same way, to show the same roughly 5 point swing from Clinton to Obama, with Edwards flat.

Results:

Marist:
Hillary 36%, Obama 25%, Edwards 14%

UNH:
Hillary 35%, Obama 21%, Edwards 15%

[itsrobert]

1. I believe in polls; not national polls months before the primary

Now a state poll out of Iowa, NH, and SC I pay more attention too. What we are seeing is people from these states are starting to look at the candidates more closely. That's why Hillary is sputtering out and Obama along with Edwards is picking up steam.

[skipos]

2. The polls in 2005, 2006 and 2007 were pretty accurate. nt

[frogcycle]

4. Dean and Gephart were leading 4 years ago

[DemocratSinceBirth]

5. I Think He's Referring To Polls Taken In The Immediate Days Before Voting

Those polls are incredibly accurate...

[skipos]

6. That's exactly what I was referring to

Election polls the days before the 2005, 06, 07 elections were pretty accurate.
Primary polls are much harder to do because of the lower turnout.
It is wrong to think all polls are meaningless, and it is wrong to think they are all accurate.

[Kurt_and_Hunter]

8. I was shocked to be reminded that Kerry led Iowa in the last Zogby poll before the 2004 caucus

In that case the polling right before the vote was right, but seemed so unlikely that many people blanked it out.

[DemocratSinceBirth]

9. A Poll In November Can Not Predict What Will Happen In January

DSB

[frogcycle]

11. but the op is talking about polls taken a year before voting

apples and oranges

[Kurt_and_Hunter]

14. They can be accurate without being predictive

Polls cannot directly measure who people will actually vote for. All they measure is what percentage of people say a certain thing to a pollster on a certain day.

But they seem to be fantastically accurate at measuring that narrow thing. (Otherwise, they couldn't get such similar results... it would be like winning the lottery a few days in a row)

[frogcycle]

3. but how many declined to answer?

that is, "undecided"


If what I have heard re Iowa (over 80% still undecided) is valid and anywhere similar in NH, then these two polls may confirm each other very nicely, but still only tell you what <20% of the people think. I submit that the subset willing to commit is not necessarily a representative sampling of what all the undecideds eventually will do. The early-committers may well be the name-recognition or "vote for the woman/black come hell or high water" types, whereas the undecideds intend to study deeper. Whether that study will find a "diamond in the rough" is subject to speculation, but not a statistically predictable phenomenon.

That is where "dark horses" come from. Premature polling.

[Kurt_and_Hunter]

7. The point is that the two polls get about the same measure of undecideds and no responses

These two small-sample high-MOE had no trouble identifying something real... that about 35% of people who do answer the phone and identify themselves as Dem primary voters currently *say* they prefer Hillary.

That doesn't mean the polls tell us who will get the most votes in real life. The polls cannot speak to the correlation between what a person says and how that person will vote. We have to deduce that from coparing past polls to past voting results.

But whatever is being measured, the two polls are able to reach the same result with seemingly tiny samples. The correlation cannot be due to chance, because the odds against this level of agreement would be in the zillions.

[frogcycle]

13. agreed

the two polls are evidently measuring the same thing and getting the same result

we just don't know what that result means, if anything, at this point in time

So I stand by my position that polls taken now are meaningless - that is - not predictive, and not necessarily measuring what they mean to. I am not arguing with the statistical sampling validity - that is good science. But context and timing can make the interpretation bad science

[valerief]

10. Yeah! If polls are accurate, then Kerry won in 2004!!!! nt

[DemocratSinceBirth]

12. Ummm ..... No.....

http://www.pollingreport.com/2004.htm#Pollster