So far, in Iowa and New Hampshire, CI’s method of predicting the primaries has produced approximately the same results as traditional polls. However, we are working on different ways to improve this. We tried something different for Michigan.

Our original method takes the average sentiment times the total of posts on each candidate, when mentioned near the primary state in the blogosphere. Again, we didn’t measure Ron Paul because of the high amount of spam associated with him which so far hasn’t transferred to the voting booth. We also did not measure Duncan Hunter. Combined, these two candidates took about 6.6% of the vote in Michigan. So in our measurements, we used the actual primary results for them, and adjusted the rest of the candidates’ totals to get the following:
We got most of the order right, but Romney’s lead was exaggerated over the actual primary results, and we showed Thompson and Giuliani flipped. So, we tried a different method as well. We studied a sample of 40 Michigan-based political blogs (20 left-leaning, 20 right-leaning), and applied the same formula to that data set:

The difference between Romney and McCain greatly decreased, to practically a tie. This could be because our data set was so small and thus potentially less accurate. This is offset though, by the fact that all the bloggers are in Michigan, and likely actually primary voters. We tried averaging the two data sets–Michigan bloggers and national bloggers–and the percentages were much closer to the actual final numbers:

There was a lot of talk the day before the election about Dem voters crossing over to vote for Mitt Romney, because they want their candidate to run against him. An MSNBC exit poll showed that only around 7% of the GOP-primary voters were registered Democrats. This number has a lot of potential for inaccuracy though, especially considering that many Dems voting for GOP candidates may not necessarily want to admit that. It’s interesting to note the discrepancy between national left-leaning and right-leaning bloggers when discussing the various candidates in Michigan:

In the left-leaning blogosphere, “Romney & Michigan” was discussed almost twice as much as Hillary Clinton & Michigan, and more than three times more than Obama & Michigan. Granted, the Democratic field was practically a non-factor in Michigan. You’d think, however, that there’d be a lot more discussion on the Left about the potential implications of the Dems’ decisions in Michigan. Instead, Romney completely dominated the discussion, which leads me to believe that there were maybe a few more stealth Dem voters than those who copped to it. Even if the MSNBC exit poll was right, most of the Dem 7% likely voted for Romney, and could be responsible for the higher-than-predicted margin.

Going forward, for our Nevada projection, we will implement the technique of combining local Nevada blogs with national ones. We’ll grab a larger sample of Nevada blogs than we did for Michigan. With each new adjustment, we’re coming closer and closer to a truly predictive model.

Thanks to Mehrshad Setayesh for creating the database queries; and to Brandon Line, John Knowles, and Brian Heisler for assistance with the formulas

Tags: Analyzing Consumer Generated Media, Mass Media/Culture //