One confounding variable here might be the fact that conservatives are actually less likely to use Google. There has been a strong backlash that I've seen personally, where conservative members of my family actively encourage me to stop using Google because it's in some way "liberally biased" and that it has a "hidden political agenda" and is trying to "impose its radical views on the world." I'm not making this up, there are even books on this very subject. One example is "Search and Destroy."
Google has an advantage in mindshare in that it's effectively the "default" search engine on the web, but its population is still, to a degree, self-selecting. Whether this effect is pronounced enough to skew the data, I can't say.
I guess that implicit in the article is the assumption that this self-selection effect hasn't changed much in the last 4-8 years. The author is careful to stick with like-to-like comparisons.
If on the other hand there is a trend away from Google among certain groups then that would seem to be a problem for Google's branding team! Perhaps there is a market here for a left-wary search engine, say DDG hooked up to conservapedia instead of wikipedia at the backend?
Tom Ptacek certainly made a good case that Google's politics (more precisely, the politics of a few individuals on the relevant team) are informing their search results on occasion:
All this energy to figure out what people will think on a particular day. And so much energy spent trying to manipulate their opinion as that day approaches. We should be figuring out how to make the system better respond to what people think all the time, just as entrepreneurs do in any industry.
Yes, people try to influence us when we are at decision points. But many things are about long term relationships and trust. Shouldn't we be striving for that?
Its a bit absurd that we cannot change our minds later. Wouldn't having that ability make politicians more honest?
I'd prefer to see proportional representation in both the legislative and executive branches (Switzerland does it) PLUS the ability to change who represents me whenever I want. We have much better technology now, so shouldn't we use it to make our electoral decisions more effectively?
I think he hints at the concept that, because of the emergence of the internet, certain statistical analyses may now be redundant, or worse yet, not actually accurate. If the average voters' feelings can be adequately captured using aggregate search data, what other types of polling can be augmented or replaced with data from Google, or Facebook, or Twitter?
Pure statistical analysis and true polling will probably never be replaced, due to their sheer importance and applicability, so I'm not trying to suggest that people start using Google data rather than creating their own surveys. But I'm certain that there exist some academic researchers who would leap at the idea of using [Aggregate Facebook Statuses, 2010-2011] as a source for their next piece of research.
What would be the point of manipulating the data used to make a prediction? How would that influence the final event - the number of votes cast?
Seems like if you have the ability or money to influence the search traffic, you could find plenty of other areas related to the election where you would get a higher rate of return.
Political entities try to create an air of inevitability for their eventual victory. You can see this at work right now in the spinning of various polls.
Manipulating the data used to create a prediction is a variation on that theme. If they could reliably do it, they would.
Influencing search traffic statistics could be done pretty low-cost, so the return wouldn't have to be that high.
It's all part of the gestalt they try to build anyway.
Polls are pretty useless. I'm a fan of using prediction markets like betfair to figure out what's going on. When people stake money on an opinion, it gives it a lot more weight.
Google has an advantage in mindshare in that it's effectively the "default" search engine on the web, but its population is still, to a degree, self-selecting. Whether this effect is pronounced enough to skew the data, I can't say.