No Special Sauce Needed For Electoral Projections

by: Chris Bowers

Tue Jun 10, 2008 at 15:48


For all you election forecasting geeks out there, in the extended entry I present a case, based on a study I conducted today, that my method for forecasting the Electoral College will be more accurate to those found at CNN  (like, way, way better than CNN) and fivethirtyeight.com. Granted, they are only marginally better than 538's, and even then only when more state polls are available).

Read on, if you dare to be that geeky.  

Chris Bowers :: No Special Sauce Needed For Electoral Projections
The latest fashion in electoral projections is to use data other than polls. For example, CNN unveiled their electoral map today, and declared it was based on a variety of factors, of which polling is only one:

The map is based on analysis of several factors, including polling, voting trends, ad spending, candidate visits, and guidance from the campaigns and political strategists.

Um, yeah. Good luck with that. Note to self: ignore CNN's electoral projections, if this is their bizarre methodology.

More interestingly and compellingly, Fivethirtyeight.com has also gained quite a bit of notoriety in recent weeks. Once again, no surprise, DIY election analysis trumps the stuff found on national news outlets. According to the 538's FAQ, it's projections also include a variety of other factors instead of just polling averages:

There are several ways that the FiveThityEight methodology differs from other poll compilations. Firstly, we assign each poll a weighting based on that pollster's historical track record, the poll's sample size, and the recentness of the poll.  More reliable polls are weighted more heavily in our averages.  Secondly, we include a regression estimate based on the demograhics in each state among our 'polls', which helps to account for outlier polls and to stabilize the results.

Now, I am a bit of a skeptic about any methodology that does not simply average polls. My basic reason for this is that, back in 2004, in my electoral forecasts at MyDD, I included 2000 results and "the incumbent rule" as counterweights to polling averages. While a simple poll averaging methodology would have resulted in forecasting the correct national popular vote and the winner of every state except Wisconsin, my methodology resulted in incorrectly forecasting Florida, Iowa and New Mexico, along with the national popular vote. I freely admit that my experience in this regard has biased me against any electoral forecasting methodology that includes data other than polls. Now, however, I have the data to back it up this guy feeling of mine.

In the FAQ for fivethirtyeight.com, Poblano writes the following:

Well, I still think you're making a mistake by using 'old' polls.  It is your right to think that, but I'd challenge you to present a case based on the evidence.  When I attempted to mimic the Real Clear Politics method -- including only the most recent poll from among pollsters who conducted surveys within 10 days of the election -- I found that the average error in my state-by-state projections would have increased by about half a point (from 2.4 points of error to 2.9) over 2000-2006.

This afternoon, I compiled a case based on the evidence, and developed a methodology based purely on poll averaging that produced an average error of 2.0, which is 0.4 lower than Poblano's (and a zillion times superior to CNN's insane methodology). Here are the parameters of the polling data that I included in my study:

  1. Only use polls that were conducted entirely during the final week of the 2004 and 2006 elections. For 2004, this means October 25th, 2004 through November 1st, 2004. for 2006, this means October 31st, 2006 through November 6th, 2006.
  2. The nine Senate races in 2006 that were decided by 10% or less.
  3. The eight Governors races in 2006 that were decided by 10% or less (does not include Idado, since no polls were conducted for that campaign entirely during the final week of the election).
  4. The nineteen states that were decided by 10.0% or less in the 2004 Presidential election. This does not include Delaware or Hawaii, since no polls were conducted in those states entirely during the final week of the election.
  5. The six Senate campaigns in 2004 that were decided by 10% or less. This does not include South Carolina, where there were no final week polls, and it also does not include Louisiana, where Vitter won by 1% or 22%, depending on the way one counts.

For all 42 of these statewide campaigns, I subtracted the simple mean of the polls conducted entirely during the final week from the final result of the election. The absolute value of these 42 numbers were then added up, and divided by 42. The result was an average error of 2.0. Election results were taken from CNN.com and Dave Leip's Election Atlas. Polls were taken from Real Clear Politics and Pollster.com. If one polling organization conducted multiple polls during the final week of the election, then only the final poll from that organization was included in the average.

The data that I used for this quick study can be viewed here.

So, while my previous experience makes me biased toward only using the simple mean of recent polls in order to forecast statewide general elections, I believe the evidence also supports me. There are also good, deductive reasons for believing that only using polls conducted in the final week of the campaign is the most accurate method for predicting election results:

  1. Polls are the only scientific measurement of public opinion in an election before the election takes place. Introducing any other element renders the study deductive and based on assumptions, rather than inductive and scientific.
  2. Polls measure a snapshot of public opinion only during the time period when they were taken. As such, including polls more than one week old in a predictive analysis simply does not make sense. The goal is to predict the final result, not what the result would have been if the election had been held a week or two earlier.
  3. Before 2004, there simply were not many polls taken during the final week in statewide general elections. In most cases, only one or two polls were available for statewide campaigns in 2000 and 2002. As such, the lack of data would increase the potential for error in the final polling averages, resulting in a higher margin of error for any study that includes pre-2004 campaigns.
  4. I did not look at elections where the final margin was greater than 10.0%. For my purposes, polls only need to be be highly accurate when they are looking at close elections where the final result could go either way. As such, polls for campaigns that were decided by double-digits are not useful for forecasting the winner of elections.
  5. Early voting does not impact these results, since polls conducted in the final week of an election always have subsets of people who already voted.
  6. It is not necessary to correct for demographic imbalance, since polls are also measuring the projected demographics of the electorate. If there are enough recent polls, collectively they will produce a very accurate projection of the make-up of the electorate.
  7. Favoring differing polling firms based on the accuracy of their past results is not always wise, given that the future performance of a polling organization can be different from their previous performance. For example, in 2000, Rasmussen was the least accurate organization for the national popular vote, while Zogby was the most accurate. My, how times change.

And so, for all of these reasons, and because of the data backing it up, I will stick with taking the simple mean of the most recent polls in a state for my electoral forecasts. While it isn't as sexy as adding a lot of special sauce, ala CNN and 538, I believe, and the evidence shows, that it is the most accurate methodology. Sometimes, the simplest solution really is the most accurate one. Until someone can present me evidence indicating otherwise, this certainly appears to be one of those cases.

My current Presidential and Senate forecasts are based on pure poll averaging methodologies. As the election moves forward and more state by state poll data becomes available, I believe these will be the most accurate projections around. Right now, in many cases there isn't enough state-level polling data for them to be highly reliable, but that will change the closer we move to the election.  


Tags: , , , (All Tags)
Print Friendly View Send As Email

Visual information (4.00 / 4)
There needs to be a better way to present electoral information. Using a map based upon land area gives a misleading impression. I can understand why this might be useful for the MSM since they have to keep the horse race story going to keep their audience coming back (and their advertisers happy). A big red blob in the middle of the country looks very scary.

The map as shown makes the red states appear too important. North Dakota with 3 votes looks bigger than NY with 31. The article on wikipedia tries to compensate for this with one of those maps which distorts the geography to represent the voting weight, but then people don't recognize the states.

I think what is needed is either some use of the third dimension (higher states?) or the use of color or some other visual way to convey a third variable. Perhaps someone knows of such a representation or has ideas on how to do this. Visual aids need to clarify things, not make them more obscure.

Policies not Politics


I agree (0.00 / 0)
My other main goal, besides accuracy, is to present what is really a huge, huge amount of info into an easy to read and understand format. I'm not satisfied with current formats, but I will keep looking.

[ Parent ]
Political Cartograms (0.00 / 0)
Investigate the wide world of cartograms-

http://www.infovis.info/search...

The Gastner, Shalizi, and Newman cartograms were an internet sensation in 2004. They helped Democrats lick their wounds-

http://www-personal.umich.edu/...

For your purposes a non-contiguous cartogram of states sized by their number of electoral votes is what I like. Animated version shows how that looks-

http://www.geog.ucsb.edu/%7Esa...

This guy blogged about free cartogram software. Maybe you should get in touch with him, his expertise is political cartograms-

http://indiemaps.com/blog/2008...


[ Parent ]
Well (0.00 / 0)
cartograms are pretty common to use. You can think of the electoral map on mydd.com as a kind of cartogram where each state is represented by the number of electoral votes it has.

But I actually think the bigger problem with the MSM is calling some states "swing states" just to make McCain look like he has a chance in those states. Like, OR and WA. Yeah right! While completely excluding states like NC from consideration.  


[ Parent ]
i agree with almost all of this (0.00 / 0)
but why not use [de-weighted] older polling for early voting states?

The people who already voted may have since changed their preferences, after all.  You could weight by the fraction of early/absentee votes from 2004, or something like that.


but why? (4.00 / 1)
Early voters are still included in the polls.


New Jersey politics at Blue Jersey.

[ Parent ]
Their poll responses can differ from their votes (0.00 / 0)


[ Parent ]
McCain Strategic Search Stuff (0.00 / 0)
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain
McCain McCain McCain McCain McCain McCain McCain McCain McCain McCain

I don't think spamming it like that works (4.00 / 3)
I think you have to only have a few links, with surrounding text, for it to actually help the rank.

[ Parent ]
I believe (4.00 / 1)
I've read that Google will only count 1 link per page anyway.  Linking to the same thing multiple times probably doesn't count for more.


[ Parent ]
Darn Google... (0.00 / 0)
Well that sucks.

[ Parent ]
So, No Need to Check in Until October? (0.00 / 0)
If you can't get a good and accurate forecast until the week or so before the election, why are you bothering with a daily forecast five months ahead of time? Polling numbers will bounce around a lot between now and then, and you're not making any effort to take the factors that may cause those bounces into account.

Now, that's probably not a bad idea, given how hard it often is to predict the future. But it seems to me that right now an excessive focus on polling is just so much noise.


Obama will probably be tracking a large number of things (0.00 / 0)
with fairly good accuracy. Looking at response to issues, to strengths and looking to weaknesses. I think Obama can take west Virginia if he is allowed to present his case strongly: being reintroduced by Edwards and Hillary for example. Being the individualistic leftists we are however, and inveterate advice mongers, we will need the next best thing to million dollar (next best?) polling firms, Chris Bowers with a spreadsheet.

--

The government has a defect: it's potentially democratic. Corporations have no defect: they're pure tyrannies. -Chomsky


[ Parent ]
pretty much (4.00 / 1)
It is actually true that polls will only give an accurate snapshot of public opinion at the time when they were taken. As such, the only ones that will predict the final result will be the ones conducted in the final week, and then released in the final weekend or day before the election.

If there was a different way to predict the future, I'd jump all over it. But there isn't.

Current polls only tell you where the campaign stands now. And really, that is the best I can do for you. I still think it is useful to monitor the election as it progresses, so you know where we are doing well, where we need to do better, and where we need to focus our resources.  


[ Parent ]
Best Kevin Neilan (sp?) SNL joke... (4.00 / 1)
when he used to do weekend update,

"A new poll finds that if the election were held today, 84% of Americans would be really surprised."


[ Parent ]
Projections (4.00 / 1)
Yes, let me second Dave Thomer's concern. You have shown that your prediction a week before the election is quite accurate. But right now you are preparing maps using the same methodology. Is there any indication that these projections provide much of an indication for what will actually happen in November? Is it possible that some other methodology might do a better job of predicting what will happen in November?

A crystal ball? (0.00 / 0)
There is no scientific means of predicting what public opinion will look like in the future. The best we can do is the present and the past. Otherwise, it is all so much astrology.  

[ Parent ]
But Before You Focused on Demographics (0.00 / 0)
I was impressed (and heartened)  by your demographic analysis a few months ago of how Obama could win -- a grand coalition of white educated upper middle-class progressives, non-whites, and young people (do I remember that correctly?). That prediction was based entirely on the demographic shift in the population and those groups' current high-level of voting -- not at all on polls. So, I guess, I am surprised that you have now shifted completely to polls for your projections.

But overall, I think it is great that you have your methodology, poblano has his, etc. It gives a lot of different perspectives to fill out the picture.


[ Parent ]
Those are hints (0.00 / 0)
And they provide a strategic roadmap. However, strategy and predicting the future are different things. I like the demographics, because it tells us what we can do and what we need to do. However, what we want and what we get are not always the same thing.

I like demographics. I like polls. But they don't predict the future with precision.  


[ Parent ]
Not Static (4.00 / 1)
I think that's just the reality of the campaign.  Chris is predicting what would likely be the result if the election happened today, but that's really the only prediction there is.  Sure, we can look at demographics and past behavior and all that and it can give us a good handle on what we--in a very general sense--might expect out of a state, but that outcome is going to be tweaked (maybe a little, maybe a lot) by what happens over the next five months of campaigning.  And we don't know that.  We can't predict that.  Thus, there's no way to give any indication of what will happen in November and I assume Chris isn't trying to do that.

Which means that in a way, these predictions are pointless.  But, really, they're not.  Because how one campaigns depends on what one believes the current state of the race is.  And the overall trends matter, as well.  If the projections shows Obama up 8 points and then, over the course of two weeks, it swings to McCain being up 8 points, then we know something is killing Obama or, alternately, helping McCain (or, likely, both.)  And then we know there needs to be a correction.  I think that's where much of the usefulness of these early predictions lie.


[ Parent ]
what about early predictions? (4.00 / 1)
You make a good case that your method will give the best prediction of the electoral outcome in the days before the election. But does it follow that it's the best method months out, when the polling data is much spottier, and the electorate less-informed, and presumably less settled? It seems to me that at this point in the election cycle, something like Poblano's regression method could be useful because it accounts for the dearth of a given state's polling data by, essentially, incorporating relevant data from other states. (Of course, that's just intuition; maybe the data would show that to be wrong.)

Could be (0.00 / 0)
Without question, Poblano's method is very useful when there is little polling data on a state. This also made it highly valuable in primary states, given that polling in primaries is not very good and that many states were virtually without polls.

No doubt, I need more data to make my method work. Lacking more data, I rely on Poblano as well. And anyway, we are only talking about 2.0% to 2.4% (and 2.9% for RCP). It is a difference, but both are pretty accurate.  


[ Parent ]
Adding Poblano's special sauce to your cheeseburger (0.00 / 0)
I trust Poblano's weighting of poll reliability. I think you could use your system (use the last week, etc.) and instead of taking a simple mean average take a weighted average of the polls based on reliability. That's a minor tweak that I think would refine your system at the edges.

John McCain

Trust (0.00 / 0)
Feel free to trust what you like, but I'm pretty sure this system is more accurate. At least, that is what the research I conducted today suggests.

Just because a pollster has been better in the past does not mean that pollster will be better in the future. As such, I do not weigh polls, at all.  


[ Parent ]
I trust you too! (0.00 / 0)

Nate's methodology is pollster reliability standard is very close to your own.

I presently have a database of 171 different electoral contests since 2000 that have been surveyed by at least three of the 32 pollsters that I include in my study. These include contests for President, Senate and Governor, as well as polls from the presidential primaries. A poll is included if (i) it is the last poll that an agency put out before the election; (ii) it was released no later than two weeks from the election date.

 

You say 1 week, Nate says 2. You say pure average, Nate says weighted average. The weight is tiny (usually +/- 2% for any given pollster) and it's not arbitrary, it's based on the same principles you espouse in your own method (polls close to actual elections are the ones that count).  You may say it's unnecessary (probably true, it makes very little difference and your system is easier to understand) but I don't think Nate is wrong to try to come up with metrics to hold pollsters accountable. He's not using any secret sauce in that regard, his methodology is there for everybody to see.

 

Either way it's a tiny tweak.



John McCain

[ Parent ]
What's the variability in your prediction? (0.00 / 0)
Is your standard deviation lower as well? And what's the median of your error versus the other guys?

[ Parent ]
Although (4.00 / 1)
While I wouldn't be opposed to weighting based on past performance in the abstract, the problem comes with determining the weights you use. Any weighting system is arbitrary.

So, while in theory I am not opposed to weighting based on past performance, I don't think any weighting measurements would really work.  


[ Parent ]
Statistics don't lie; interpretations of statistics might lie. (4.00 / 1)
There are opinions, hunches, intuition, good guesses, experience, but none of those are mathematical the way statistics is. Statistics may not always lead to intuitive results, you do have to pay attention to certain issues because they are logical and mathematically correct. If not, then you can be wrong, not  just interpreting things differently

For example, one poll uses 500 another 2,000. You must to place more weight on the 2,000 sample poll. In addition, the correct weighting of the error terms is not an average, rather the square root of the sum of the squares.

But, as Poblano explains at some point sampling error is swamped by procedural errors, such as lack of cell-phone respondents, night-job workers, demographic shifts as people move in, likely voter models.

And, as we've been discussing here, polling is a snapshot in time, news cycles progress, and people come to new decisions, and presumably they finally make up their mind as the actual election approaches.

I'm pretty much on poblano's side here, as I have enough statistics to understand where he's coming from (even if that makes me possibly be dangerous on my own). You could several make simple adjustments to deal with some of the more obvious issues:
- average polls, but
- give higher weight to polls with more numbers, and
- give lower weight to older polls.

As for the demographic model that poblano uses, we can at least rely on that for additional information and understanding.


[ Parent ]
RCP tried the Poll weighting (4.00 / 1)
idea in 2000.  It was a complete disaster.

I would watch a couple of matrics, though.
1.  Watch the undecided in the state polls versus the national polls.

I tracked this metric in 1996 and found that while the State Polls were showing the same margin, they were showing a signficantly higher number of undecided than the national numbers were.  When I generalized the state results into a popular vote projection and allocated the undecided to Dole on a 3 - 1 basis (which is typical for a race involving a challenger) I came up with a predicted margin of 7.7 percent, far below the last national polls.

This projection was also more accurate.

2.  Be wary of newly competitive states.
At this point a number of states looked in play (notably NC and Ariz) that were not even close in the Fall.

3. Be wary of polling from Universities.
There are a number of University pollsters that were terrible in both '04 and '00.  Most prominent are:
*The badger poll from the University of Wisconsin
*Arizona polling from Northern Arizona University
*Tennessee polling from Middle Tennessee State
*Several NJ universities poll, and were just awful

4.  Watch the undecided in Rassmussen and SurveyUSA
Both of these pollsters tend to have very low undecideds versus other pollsters.  I think this is because they are robo-pollsters, but it is something to watch for.

Anyway, some of the things I would suggest.


Worth noting (0.00 / 0)
I'll make sure to keep it in mind.

Also, why did you allocate undecided to Dole at a 3-1 margin? That seems to have been the key to your success. In fact, had you followed the Incumbent Rule margins, 2-1 would have given you the final margin almost precisely. Were you aware of the incumbent rule back then?


[ Parent ]
Yes (0.00 / 0)
And my 3-1 allocation was based on the work I did in '92, which was the first year I made an effort to really track state polls.

The last Gallup poll in '92 allocated all of the undecided to Clinton.  Had they allocated based on 3-1 (and split the votes against the incumbent equally between Perot and Clinton) they would have been right.  As it ways they were quite a bit off.

Hence my 3-1 split in 1996.    


[ Parent ]
These comments seem odd to me (4.00 / 1)
For all the people running around the intertubes about how we need to fight like we're ten points down until safely into December, I'm kinda surprised that so many people are demanding that a projection today magically reflect what the final map will look like.

Tracking these numbers now is a guide for all of us as to where to focus our energies and our messaging. Whatever the polls say a week before election day, they're going to say that because of the decisions made based on information like this.

For example, it's this sort of information that tells me that probably Obama's resources are better spent in Ohio and Virginia than in Rhode Island or Alabama. That may be the sort of thing that many scoff at as obvious common knowledge, but it's only obvious common knowledge because we don't wait until a week before the election to check.

John McCain opposes the GI Bill.


Also (0.00 / 0)
It's why we don't panic when Karl Rove plays mind games with "the math" or has his candidate do a swing through Democratic states just before the election.  Knowing empirically that Rove is full of shit gives our side the confidence to ignore his shenanigans and not panic and assume he knows something we don't.


[ Parent ]
Track vs. Project (0.00 / 0)
I may very well be guilty of being caught up in semantics, but  tracking and projecting seem like two very different projects to me. I understand the value of tracking in order to understand the current situation, but in order to use that snapshot of the moment in order to make decisions, it seems to me like you need to factor in the kinds of "special sauce" that Chris wants to avoid. Are there enough of "our" voters in a particular state to close the gap? How much can we afford to spend on media there? What way do undecideds typically break? Tracking tells you where things are right now and how they'd be if nothing changes. But the one thing you can count on is that somewhere something IS going to change. So in order to make a forecast or a projection, you would have to take that into account.

Another question is, why do I need to make any kind of decision about where Obama's resources are best spent? I'm no spending any of them, and I sure as heck hope that the people who do make those decisions have access to more information than we do - including information about their own plans and resources.

Now, in lower-ticket races where netroots activists have to make their own decisions about who to endorse and who to try to raise money for, I can see why tracking and predicting can be important. But again, I think you need to mix the polling information this far out with demographic and historical information in order to get a strong sense of possibilities.


[ Parent ]
Well (0.00 / 0)
The semantic point is a fair one. I'm not titling Chris' posts nor am I speaking for him, but I agree that it certainly seems as though they're more tracking and informational than predictive. Nevertheless, that doesn't entirely justify the reactions from some which dismiss the information as irrelevant.

You don't necessarily have a personal need to know where Obama should be spending his resources, I was using that as an illustrative example more than a personal one. However, this sort of information has already been guiding me as I look towards October and start considering where I might want to travel in order to maximize my canvassing time (I'm in CA where knocking on doors won't likely change the electoral map much).

In a broader sense, I could ask the same about the relevance of delegate tracking or VP speculation or any number of other subjects that get a disproportionate amount of attention among political junkies relative to the impact readers actually have. While there might be incidental impact from hundreds of thousands of people keeping tabs on exact superdelegate counts or VP straw polls, the larger value is that it informs a broader understanding and deepens inherent grasp of political nuance to simply have the conversations.

John McCain opposes the GI Bill.


[ Parent ]
Heisenberg Uncertainty applies to elections too (4.00 / 2)
No system of polling, will ever be 100% accurate, and in fact, I think as soon as a system is perceived to be "most" accurate, it will soon cease to be.

The reason is that the act of measuring public opinion inevitably alters it.  Even just conducting the poll means 1000 voters get reminded that there is an election, and the nature of the questions asked remind them of specific issues to think about.  When the voter hangs up the phone, how many of them go online and search for new info on the candidates and issues and change their minds?

Then once the polling is known, it changes the behaviour of the campaigns accessing it.  So even if a snapshot could predict the outcome weeks or months from the election, it could only do so if the campaigns do not alter course, and no outside events intervene to alter the relevant issues to the election.  No poll can control for an Obama video tape, or a DUI revelation.  

Voters respond to polls too.  If you live in a landslide state for one candidate, polling may tell you not to bother voting, as the victor is certain anyway.  Who knows if there might be enough Democrats in Wyoming to win the state if they all showed up, and Republican turn out was lower because they didn't feel any big need to bother.

It's a chaotic system, and polling both measures and affects the system.  


This is the heart of my concern (0.00 / 0)
Tracking polls is fine, but if it means you write off a state like Wyoming when you might actually have been able to win it, then it seems counterproductive. I've been very frustrated that so many people in the netroots seem to make very early decisions about which races are winnable and which not. Many would probably have said not to bother with the three special elections in strongly Republican districts that Democrats ended up winning. In the same way, writing off Wyoming right now might be a mistake even though the polls show Wyoming to be a disaster.

It is important to be strategic -- and not spend resources on unwinnable races -- but it is also important to be smart and recognize that some races that seem unwinnable might actually be winnable.


[ Parent ]
CNN needs to fire its consultants (0.00 / 0)
They don't have New Mexico as a swing state, but as Lean McCain.  Obama has led in numerous polls this year, and Gore won it.  

New Jersey politics at Blue Jersey.

Inputs from a statistician (4.00 / 1)
Just a few comments here from someone who works as a statistician, though not in polling.  Your assumptions are generally good ones, particularly the timing of the polls close to the election.  The biggest bias will likely be a late shift that is not captured by the polling.  The 10-point rule (#4) shouldn't matter, in fact, as results move away from a 50-50 split, the statistical variance gets smaller, not larger.  I'd like to see the evidence if the polls are actually less accurate on races with a more decisive split since it would work against the pure statistics.  Since you are focusing on gaps between actual and poll results, why should it matter what the size of the difference is anyway?

Your base data has a mean gap of actual results of about +1.0, with a sigma of 2.37.  This sigma is an estimate of the variation about the true results, and would give a rough +/-4.74% confidence interval.  This would equate to a typical poll size of n=500 persons, and I would expect this is about your minimum sample size in the individual polls used.  One thing that might help you is the averaging of multiple poll results within some states, which should drive the results to tighter levels.  You might try weighting the results by the number of available polls, giving more weight to the races that have the most available information.  This would tend to lower the impact of single poll results, wherever they might occur.


USER MENU

Open Left Campaigns

SEARCH

   

Advanced Search

QUICK HITS
STATE BLOGS
Powered by: SoapBlox