Princeton Election Consortium

A first draft of electoral history. Since 2004

To poll obsessives everywhere: welcome (and welcome back)

August 14th, 2008, 11:25am by Sam Wang


If you’re a reader from 2004, welcome back. For everyone old and new, take a good look around. There’s lots to see.

At first glance, there are two reasons you might be less interested in the Meta-Analysis this year:

  1. In 2004, the suspense was greater. Although Bush was in trouble for most of the summer, the success of the August Swift Boat attacks tightened the race considerably. In 2008, many people have been expecting a less eventful campaign.
  2. The Meta-Analysis has a lot of competition. In 2008, polling sites have grown in number and in sophistication. Four years ago the stars were the data accumulators: RealClearPolitics, electoral-vote.com, Race2004.net, and others. There was modeling, but it had a certain cult-ish quality. This year, the advent of the highly popular predictive site FiveThirtyEight.com has made poll geeks a mainstream phenomenon.

But think again.

The race is currently closer than I was expecting. McCain and his campaign are pursuing their best option, to attack Obama repeatedly and relentlessly. The result is clear, as you can see in the graph on the right: a decline in Obama’s electoral vote expectations. To state the obvious, your degree of certainty in the fall outcome has almost certainly decreased in the last few weeks.

This brings me to the second reason. Whichever way you care about the outcome, you are probably watching polls, both national and in key states. What you may wish for is a way to look at all current polls at once to get a general picture. I believe that this site is a good stop for you.

But first let me describe some of the better sites, and why you might want to come back here.

  • For relatively raw data, Pollster.com is an excellent source for all US races, and provide us with a direct data feed. They have excellent display tools and expert commentary on the ins and outs of polls. Mark Blumenthal and Charles Franklin are, in particular, incomparable. There is also RealClearPolitics, which has been around longer and is less polished. They are right-leaning but when it comes to data they shoot straight.
  • Another site worth watching is Electoral-vote.com. Its host, Andrew Tanenbaum, author of the MINIX operating system and former teacher of Linus Torvalds. You can’t get better geek credentials than that. On his site he gives the most recent nonpartisan state polls using a simple averaging algorithm. His site provides a form of today’s snapshot. He also provides past data, a useful geek resource.
  • For modeling, you have many choices. The best known this year is FiveThirtyEight.com, run by fantasy baseball guy Nate Silver. His approach to polling analysis is heavy on numerical analysis and simulation with lots of trend fitting. In particular, he makes a projection into the future. The projection is done plausibly, but is very uncertain since the range of possible movement is large. Nonetheless his grasp of polling data is excellent, as is his commentary. There are many other hobbyists, just a few of which I list here: [1] [2] [3] [4] [5]

What you get here at election.princeton.edu is a pure snapshot of the freshest available state polls. Using simple statistical estimation tools, I give you the median outcome of an election held today using the populations sampled by the pollsters. The methods take into account inter-pollster and sampling variability and calculate a probability distribution of all possible outcomes. Finally, I give the Meta-Margin, an estimate of how much the margin would have to change to flip the outcome.

Simplicity is important for several reasons. It allows me to make the code open-source so that you can play around with it yourself. More importantly, the more elaborate adjustments done by hobbyists elsewhere reduce transparency – and might not increase accuracy. In my view, complexity is only warranted if it adds significantly to accuracy and is justified by both logic and data. The Meta-Analysis isn’t perfect (see my earlier post on its biggest flaw). But I think it’s the best available tool for synthesizing currently available state polls at a single glance.

My bottom-line recommendation to you for following the national race is what I do myself. First, I look at the national average margin between Obama and McCain, for which new data appear frequently. Second, I look at the median EV estimator at the top of this page, which is slower but more precise. Together, these two measures give you an excellent picture of what’s happening.

Tags: Site News

3 Comments so far ↓