Swing states and polls

[fyo]

National polls don't mean a whole lot. Elections aren't decided by the national result and for those who may feel bitter when a president wins one but not the other, all I can say is that because local results are what counts, the local polls strongly influence voter turnout. Strongly contested areas generally have greater turnouts.

Anyway, I was looking at polls for swing states in the presidential election and it got me thinking about how these polls are made. The problem, as I see it, is that turnout across different "segments" can vary quite significantly and there's just no precedent for a black presidential candidate.

Take New Mexico, which many electoral maps now place in Obama's column. One poll has Obama up, 52% to 45%. However, that's some weird weighted value, where the pollsters have tried to estimate the turnout among different "segments". Obama has greater support among minorities and young people, whereas McCain has greater support among whites (especially men) and the elderly.

If this were Bush v Kerry, I can see how they could reasonable estimate turnout. However, with the age disparity between Obama and McCain - and the fact that Obama is black - would seem to make any estimate completely unreliable.

Bottom line, I have ZERO faith in the polls this year - even less than usual.

[YoFuggedaboutit]

Good points, fyo.  You're right, you can't really trust polls.  These same polls predicted that Gov.  Alfred Landon would defeat incumbent Pres. Franklin Roosevelt in the 1936 Presidential Election.  The actual result was that Roosevelt won in a landslide. 

In 1948, the polls screwed up again, predicting Thomas Dewey would defeat incumbent Pres. Harry Truman.  For a brief moment, it looked as if the polls were accurate because Dewey had a lead after the first few votes were counted.  This prompted the Chicaco Tribune to print the first edition of their newspaper with the headline "Dewey Defeats Truman" 

The next morning, Truman was on live TV gleefully waving a copy of that newspaper as he celebrated his victory. 

[Dave Gray]

I think that polls are actually pretty accurate.  They move so much, because people move so much.  It's just using statistics to make an estimate.  Any one poll can be garbage, but when you look at a large group of the data looks similar, I think you're pretty safe to expect what it shows.

I don't know that you can cite examples from an election in 1936 and apply it to today.  Polls happen quickly, with new ones updating daily, and the variables are different than they were in the 30s. 

[fyo]

Dave, I don't necessarily have a problem with polls in general - although I do think the uncertainty is commonly underestimated - but how do you even begin to predict how the presidential lineup will affect traditional "segment" turnouts?

[Dave Gray]

^^^
I don't know how they weigh that stuff.  There's a margin or error of something like 3.5 percent, which is pretty large.



There is a site I like called fivethirtyeight.com.

It's run by an economist, who used his system to predict baseball statistics, and the results were much more accurate than other systems.  He's been doing it for the last few elections, too.  His predictions are based on a variety of polls and other means and are broken down pretty far.  He's got all kinds of stuff.  It's a completely non-partisan site.

[Dphins4me]

   I really wish they would stop this polling & providing news on the election until all the voting is done for the night on the west coast.

[MyGodWearsAHoodie]

^^^ While there would be some benefit to that it runs smack against the 1st amendment. 


Plus I enjoy Presidential election night parties.  Gather around the TV with others (some times only fans of the same team, some times a group of fans from both teams), eat food and drink beer and root for your team.  Event normally takes about 4 hours (unless you go into insane overtime like 2000) Kinda like watching the superbowl. 

[bsfins]

I kind of feel if you look at enough of them...The data is fairly close...I feel if most of them are about the same...they tend to be pretty accurate..Now will they ring true Election day..I'm not sure...people tend to sway with the wind....one day they're for this,the next they're not....

[Defense54]

Good points, fyo.  You're right, you can't really trust polls.  These same polls predicted that Gov.  Alfred Landon would defeat incumbent Pres. Franklin Roosevelt in the 1936 Presidential Election.  The actual result was that Roosevelt won in a landslide. 

In 1948, the polls screwed up again, predicting Thomas Dewey would defeat incumbent Pres. Harry Truman.  For a brief moment, it looked as if the polls were accurate because Dewey had a lead after the first few votes were counted.  This prompted the Chicaco Tribune to print the first edition of their newspaper with the headline "Dewey Defeats Truman" 

The next morning, Truman was on live TV gleefully waving a copy of that newspaper as he celebrated his victory. 

dude.......that way back when the Refrigerator was still a New Modern convenience.  This is the Modern information age. Collecting data is way more easier and accurate. That said........if it was predicting say a 55%-45% Race I would say, OK. But any landslide predictions are gonna be pretty much in the leading candidates favor.  Anything could happen, but I see a record number of Dems and Liberal minorities out to vote this time around.

[YoFuggedaboutit]

Anything could happen, but I see a record number of Dems and Liberal minorities out to vote this time around.

And that's what Obama has been calling for based on the radio ads I've been hearing. 

[fyo]

Anything could happen, but I see a record number of Dems and Liberal minorities out to vote this time around.

My only question is this: What are they using in their poll results?

Are they ALREADY factoring in record minority turnout? Are they using numbers from previous elections? Are they asking people, in essence, "are you going to vote?"?

[Defense54]

My only question is this: What are they using in their poll results?

Are they ALREADY factoring in record minority turnout? Are they using numbers from previous elections? Are they asking people, in essence, "are you going to vote?"?

I see your point. It could be much worse or much closer if they are assuming the record turnout in the polls and it doesn't materialize. But I know it my area they already have the buses booked and there are actually be taxing people to and from the the polls. They are NOT fucking around this time.

[fyo]

It would be nice if they published their assumptions, not just the "end result".

[Dave Gray]

When you're estimating large numbers, with many variables, generally, you make many, many small estimates.  So, you have to estimate turnout in each area, polling in each area, etc.  You will be wrong on a whole lot of things, but generally, you'll overestimate in one area as much as you underestimate in another, and you'll come out pretty close.

That's the idea.

Also, like we've said before, when you keep polling over a period of time, you develop trends.  Sure, that any one day could be off, but if you're seeing similar results over a long run, you have more reliable data from which to draw.