I have been so impressed this election cycle with the hard work and diligence of the "numbers guys" (guys used colloquially here to mean both sexes).
Sam Wang at Princeton Election Consortium has really gotten the attention he richly deserves this season for his no-nonsense "just the numbers" approach to analysis.
I wanted to use this diary to highlight another excellent blog which also continues to impress me with each additional piece of analysis that they provide. It is is votamatic.org (http://votamatic.org/).
From the site's "How it Works":
The site provides two different types of information. First, it estimates the trend in public opinion for Obama vs. Romney in each state up to the current day. But it also forecasts ahead to Election Day – given what we know now, what is the election outcome most likely to be? These forecasts are produced at both the state and national level, and are accompanied by a statement of uncertainty in each projection.I really enjoy having multiple serious analysts crunching the numbers from different rational perspectives. Votamatic currently has the final electoral count at 332 - 206 for President Obama.
All of the estimates are based on a single, unified statistical model. The forecasting component combines data from past presidential elections with results from the large number of state-level opinion surveys released during the 2012 campaign. As new polls become available, the forecasts update in real time, gradually increasing in both accuracy and precision as Election Day nears.
The model is specifically designed to overcome the data limitation that not every state is polled on every day of the campaign. To smooth out and fill in any gaps in the polls, the model looks for common trends in voter preferences across states. These are usually the result of major campaign or news events that affect voters in all states at the same time. By drawing upon these patterns, the model is able to keep the election forecasts completely up-to-date.
The model is described in greater detail in a research article, Dynamic Bayesian Forecasting of Presidential Elections in the States, forthcoming in the Journal of the American Statistical Association.
Please join me over the fold for some discussion of an excellent piece of analysis posted by Votamatic today.
The polling aggregators such as Votamatic, PEC and 538 rely on data from the big national polling organizations such as Rasmussen, Gravis, ARG, Quinnipiac, SurveyUSA, YouGov, and PPP. They also rely on polling data from much smaller firms which don't produce as many polls and also are local in scope.
Polling firms all use different approaches to crunching their raw data to present their polls. The different approaches may present itself as bias towards one candidate compared to other polls. Rasmussen, Gravis, and ARG appear biased towards candidate Romney, while Quinnipiac, SurveyUSA, YouGov, and PPP appear to be biased towards President Obama.
In an interesting piece of analysis Votamatic decided to aggregate all of the small, local polls together to see where they would fall compared to the split among the big national polls. If the aggregation of the small polls looks more like Quinnipiac, SurveyUSA, YouGov, and PPP then they probably are not biased but rather it is Rasmussen and its ilk that are showing bias. If on the other hand, the aggregation of the small guys looks more like the Rasmussen crowd, then it is the Quinnipiac/PPP crowd that is probably showing bias. Finally, if the aggregation of the small polls comes out smack dab in the middle of the Rasmussen crowd vs. the Quinnipiac/PPP crowd then it probably shows that there is bias on both sides and the answer is somewhere in-between.
The finding was that the small polls aggregation is in agreement with the Quinnipiac/PPP crowd and that the Rasmussen gang is probably biased.
I encourage you to go read the article titled, "Another Look at Survey Bias" posted today. It provides a much better description of what they have done along with some great graphs showing the results. Here is a link to the article: http://votamatic.org/...