A "poll of polls of polls"? Seriously?
Yes, seriously--sort of. The idea did start out as a joke. More specifically, it started out as a sarcastic voicemail left after hearing Wolf Blitzer promote yet another edition of CNN's "Poll of Polls".
After hanging up, though, I realized it made a twisted sort of sense. This election, there are a multitude of election trackers--sites which combine poll results to predict an overall outcome--using a wide range of methods for combining polls and producing vastly different results. Some of these, like RealClearPolitics and the New York Times Electoral Map, simply average poll results and award states to the candidate who leads in each. Others use more complicated methods, such as regression estimates (Pollster.com) and Monte Carlo simulations (FiveThirtyEight.com). Each tracker also differs in which polls they include in their models and which they leave out. Because each method produces different results, it's worth seeing what happens when we average over the methods to produce a new total result.
Though it all might sound ridiculously meta, there is a logic behind this. Trackers combine polls from a variety of pollsters because each pollster uses different techniques. They vary in their question wordings, their survey methods (live phone vs. automated phone vs. internet vs. ?), their response weightings (to produce representative samples), their likely voter models (if any), and many other factors. Such factors can have a severe impact (called a "house effect") on the pollster's results. By combining myriad pollsters together, trackers hope to compensate for house effects, because on average pollsters will cancel each other out. In the same way, this site seeks to get closer to the "true" state of the race by letting the effects of different tracker methodologies cancel each other out. We know that each tracker puts out a number that is their best guess of where the race stands, and that it is incredibly unlikely that any of them will get it exactly right. More likely, the true will result will be off to some degree in one direction or another. By using all of the trackers, we can make an educated guess as to which direction each tracker will be off by, and therefore (hopefully) come up with a better result than individual trackers can on their own.
Or, for the more practically minded, there is another, simpler justification for this site. I'm a bit obsessive, and spend my time each day going back and forth between a collection of many different sites. I know a lot of you do as well. By giving one place for you to go instead of many, this site will save you some time, time that you could perhaps be spending out campaigning, trying to save the world. Or sleeping--that's important too.
HOW THE RESULTS ARE CALCULATED: Each day (or so), I'll be pulling the most recent results from an assortment of trackers. They'll be broken up into three categories: electoral votes with tossups allocated to the leading candidate, electoral votes without tossups allocated, and popular vote estimates. In the the first category, each tracker's electoral votes will add up to 538 (the number of votes in the electoral college). The second category is necessary because some trackers do not allocate "tossup" states--states that are too close to call--and thus produce numbers that sum to less than 538. These numbers should not be read as a prediction of the outcome of the election, but instead as the minimum number of electoral votes each candidate should be expected to receive. The final category is popular vote estimates, estimated from either national poll numbers (for RealClearPolitics and Pollster.com) or from simulations of overall vote patterns (in the case of FiveThirtyEight.com). For each category, average results will be computed. These averages are unweighted, though this may change at some point (at which point I'll post an entry describing the changes and update this entry).
I also hope to be able to contribute some insight on the state and direction of the race, to give some analysis of events and polls as they happen, and to provide some perspective as to the advantages and disadvantages of the different methods of polling and poll-tracking between now and November 4.
Lastly, I thought I should give a bit of information about myself, for those of you wondering who's behind this and where my "insights" and "analysis" come from. I am a 3rd-year PhD student, studying American and Comparative Politics at a major research university in the northeast. My specific focus includes voter behavior, public opinion, and electoral strategies. What this means, in practical terms, is that I spend my days neck-deep in survey data and statistical analysis methods and programs. My goal with this site, then, is to take that training and use it to make the competing methodologies of professional polling comprehensible to the non-professional.
If you have any suggestions for topics you'd like to see covered, know of any poll trackers I'm missing (there are many, I'm sure), or just want to give feedback, send me an email: andrew [at] poll3.com. Otherwise, enjoy the site!
After hanging up, though, I realized it made a twisted sort of sense. This election, there are a multitude of election trackers--sites which combine poll results to predict an overall outcome--using a wide range of methods for combining polls and producing vastly different results. Some of these, like RealClearPolitics and the New York Times Electoral Map, simply average poll results and award states to the candidate who leads in each. Others use more complicated methods, such as regression estimates (Pollster.com) and Monte Carlo simulations (FiveThirtyEight.com). Each tracker also differs in which polls they include in their models and which they leave out. Because each method produces different results, it's worth seeing what happens when we average over the methods to produce a new total result.
Though it all might sound ridiculously meta, there is a logic behind this. Trackers combine polls from a variety of pollsters because each pollster uses different techniques. They vary in their question wordings, their survey methods (live phone vs. automated phone vs. internet vs. ?), their response weightings (to produce representative samples), their likely voter models (if any), and many other factors. Such factors can have a severe impact (called a "house effect") on the pollster's results. By combining myriad pollsters together, trackers hope to compensate for house effects, because on average pollsters will cancel each other out. In the same way, this site seeks to get closer to the "true" state of the race by letting the effects of different tracker methodologies cancel each other out. We know that each tracker puts out a number that is their best guess of where the race stands, and that it is incredibly unlikely that any of them will get it exactly right. More likely, the true will result will be off to some degree in one direction or another. By using all of the trackers, we can make an educated guess as to which direction each tracker will be off by, and therefore (hopefully) come up with a better result than individual trackers can on their own.
Or, for the more practically minded, there is another, simpler justification for this site. I'm a bit obsessive, and spend my time each day going back and forth between a collection of many different sites. I know a lot of you do as well. By giving one place for you to go instead of many, this site will save you some time, time that you could perhaps be spending out campaigning, trying to save the world. Or sleeping--that's important too.
HOW THE RESULTS ARE CALCULATED: Each day (or so), I'll be pulling the most recent results from an assortment of trackers. They'll be broken up into three categories: electoral votes with tossups allocated to the leading candidate, electoral votes without tossups allocated, and popular vote estimates. In the the first category, each tracker's electoral votes will add up to 538 (the number of votes in the electoral college). The second category is necessary because some trackers do not allocate "tossup" states--states that are too close to call--and thus produce numbers that sum to less than 538. These numbers should not be read as a prediction of the outcome of the election, but instead as the minimum number of electoral votes each candidate should be expected to receive. The final category is popular vote estimates, estimated from either national poll numbers (for RealClearPolitics and Pollster.com) or from simulations of overall vote patterns (in the case of FiveThirtyEight.com). For each category, average results will be computed. These averages are unweighted, though this may change at some point (at which point I'll post an entry describing the changes and update this entry).
I also hope to be able to contribute some insight on the state and direction of the race, to give some analysis of events and polls as they happen, and to provide some perspective as to the advantages and disadvantages of the different methods of polling and poll-tracking between now and November 4.
Lastly, I thought I should give a bit of information about myself, for those of you wondering who's behind this and where my "insights" and "analysis" come from. I am a 3rd-year PhD student, studying American and Comparative Politics at a major research university in the northeast. My specific focus includes voter behavior, public opinion, and electoral strategies. What this means, in practical terms, is that I spend my days neck-deep in survey data and statistical analysis methods and programs. My goal with this site, then, is to take that training and use it to make the competing methodologies of professional polling comprehensible to the non-professional.
If you have any suggestions for topics you'd like to see covered, know of any poll trackers I'm missing (there are many, I'm sure), or just want to give feedback, send me an email: andrew [at] poll3.com. Otherwise, enjoy the site!
posted by Andrew at
12:38 PM
0 Comments:
Post a Comment
<< Home