Seeing as the midterm elections are tomorrow, and since Nate Silver’s notoriety stems, at least in part, due to his ability to predict elections, I feel like I wold be doing the statistic gods a disservice by not talking a little bit about elections for this post.
One of the big story lines surrounding tomorrow’s vote is whether or not the republicans will take control of congress. One interesting detail I noticed in my wandering around on the internet, is that the senate seats that are up for grabs tomorrow were last voted on in 2008, since senate seat elections occur every 6 years. Nate Silver discusses that in 2008, when Obama was elected, it was an extraordinarily strong year for the democrats, and if this year is even politically neutral, there will be a drop off from the results of 2008. This trend will be even more accentuated if there is even a modest lean in the republican direction – not a good sign for the democrats. On the other hand, Silver mentions that based on historical data, incumbents still win an overwhelming majority of races even if they are unpopular at the time of an election. Being the incumbent still represents an advantage – perhaps a good sign for democrats. Looking at polls this late in the campaign can pose some risks too. There can be a big difference if the sample taken by a poll is of “likely voters” or “registered voters”, and there can also be issues if a candidate lacks name recognition. There are also firms that conduct polling with bad track records or bad practices.The data acquired from these polls all tell a slightly different story and conclusions from them must be carefully made.
As we have discussed in our class, the fallacy of the obese N, commentators (and politicians for that matter) can really find any pattern that they want in predicting political results or election data. By manipulating the weight of certain criteria, or excluding outliers based on bias reasons, the potential for cherry picking data you want to see can be very easy. There is also the possibility of “overfitting” your model, which Silver describes as “one that is engineered to match idiosyncratic circumstances in past data, but which is not an accurate picture and makes poor predictions as a result.” Regardless of what happens, it’s always fun to try and see into the future using stats!