This is the last part of a three-part series of posts (part 1, part 2). We explained the method used for this experiment, i.e., analyzing "toss-up" states in the US presidential election by comparing Google Insights for Search and Pollster.com graphs in part 1. We continue with Ohio:
For the first couple of months of this year, the Democratic Party (DP) was ahead in Google-popularity in this state while McCain was ahead in the regression trend line of the polls. For the rest of the year, the DP and Republican Party (RP) lines related to one another in the same way as the Obama and McCain trend lines: first the DP/Obama in the lead, then the RP/McCain. However, in September, Obama again gained the upper hand. The RP stayed in the lead though. I wonder if there isn't some kind of delayed effect in these case, or maybe the high-sensitivity trend lines for the candidates overreacted to recent polls? Now we turn our attention to Pennsylvania.
This state too was kind of puzzling: the parties' graph showed a DP advantage for most of the year except in September when the RP overtook them, only to end in a tie. On the candidates' graph, McCain was ahead till Obama overtook him in April; the latter then put a lot of distance between him and his opponent after some tightening in the beginning of September. This seems related to the Republican-convention a.k.a. Palin-announcement bounce for McCain and the current financial-crisis bounce for Obama (see also, for instance, Ohio above). Next is Virginia:
Similarly to Pennsylvania, the DP had the edge till the RP bested them in September. McCain however was ahead but lost more and more till the race turned into a dead heat in June. In September, the same Palin bounce and the financial-crisis bounces gave the advantage to Obama. Finally, I looked into Missouri, a state that has gone in and out of the "toss-up" category quite a few times. It was in McCain's camp when I started this series of posts but is again in contention.
The DP dominated till the RP took over in September. McCain was in the lead the whole year but it's now a tie. One more thing: Nader, Barr and McKinney are an afterthought in the polls. This is partially due to them often not being included in polls in the first place and, when they are, they don't appear in enough of them to survive Pollster.com's high-sensitivity smoothing filter. To be fair, let me give you briefly what their poll standings were today when using normal smoothing, as well as their parties' Google-popularity for 2008 so far (shorter period didn't yield good data):
I quickly tested the correlation between the two measures by producing scatter graphs with a linear regression trend line added (without the Constitution Party as I had no poll data for their candidate).
Quick conclusion: it looked like there was no correlation, in other words the Google-popularity of third parties couldn't be used to try to predict the third-party candidates Pollster.com scores or vice versa. Which probably means that we didn't miss out by not having them on all but one state poll graph after all.
12 hours ago