You're setting the bar pretty low for "reasoned and mature."
Ouch.
But, that things seem to track more or less is pretty nifty. The next step would be to actually automate the predictions...
No API, but you could hard-code a script to run the searches (really, all you need is a url to the image itself) and programmatically look at the trend line to find the index numbers/compare the candidates. Then you'd have a whole bunch of nice happy numbers, and you could control and find significance of and pivot allll you want.
I wanted to see if Google Trends could apply to political campaigns. So I went with the most obvious measure - primary/caucus results - and developed a model that worked very well, for the Dems anyway. I certainly did have to backtrapolate a bit.. the model failed until I removed memes (Ron Paul) and found that using first names sort of "normalized" Obama's massive popularity online with the younger generation. So predicting future polls may or may not work, although if anybody tried, I would again suggest excluding memes and, if the first couple don't work too well, consider asking why and seeing if using first names works better.
But the whole point was to see if politics and Google Trends correlate in the real world. It makes sense that they would, but I haven't seen any studies that really analyze it. This was just my attempt as a high schooler for my senior seminar paper. I think it at least opens the door to using search data to help candidates understand what's working and what isn't. My article isn't demanding that you try to predict elections - just that you consider using Google Trends results in a field outside of technology, in this case in politics.
Regardless, thanks for your reasoned and mature comments, Z.
Yeah, your feedback made me realize that I was being unclear, so I added a "disclaimer", preceded by the word "Update:", to ensure that nobody actually takes this as statistically valid. Sure, you can't conclude anything about this year's race based on my article. But my entire point with this is to show people that there's this tool out there, that can provide a lot of useful information to candidates, which they don't seem to be using. I'm not trying to push any kind of agenda or act a**holish.
Hey, guys,
You're all right. There are serious statistical flaws here. I chose the keywords "hillary" and "barack" and "edwards" because that was the model that showed the best success. If one were actually predicting this live, he or she would soon see that "hillary" and "barack" do produce better results and would switch to those terms, simply because it fits the model better.
Every election will have its outliers (Obama, Ron Paul). But hopefully this brief overview will provoke some campaign analysts to actually try to interpret Google Trends results for themselves, using more advanced models.
Sure, and this is part of the reason why many states in the South or states that like certain candidates don't work as well with predicting. Google Trends itself currently only uses a relatively small sample size, too, exacerbating the problem. But these charts certainly do not reflect *nothing* -- just be careful what conclusions you draw.
Slashdot Mirror
So he gave the key to the city to the mayor.
You're setting the bar pretty low for "reasoned and mature."
Ouch.
But, that things seem to track more or less is pretty nifty. The next step would be to actually automate the predictions...
No API, but you could hard-code a script to run the searches (really, all you need is a url to the image itself) and programmatically look at the trend line to find the index numbers/compare the candidates. Then you'd have a whole bunch of nice happy numbers, and you could control and find significance of and pivot allll you want.
Actually, Slashdot has been declining in terms of Google Trends search volume: http://google.com/trends?q=slashdot,digg
When you compare it to other things like Facebook or Google you realize how insanely small the percentage of Slashdot readers must be.
I wanted to see if Google Trends could apply to political campaigns. So I went with the most obvious measure - primary/caucus results - and developed a model that worked very well, for the Dems anyway. I certainly did have to backtrapolate a bit.. the model failed until I removed memes (Ron Paul) and found that using first names sort of "normalized" Obama's massive popularity online with the younger generation. So predicting future polls may or may not work, although if anybody tried, I would again suggest excluding memes and, if the first couple don't work too well, consider asking why and seeing if using first names works better.
But the whole point was to see if politics and Google Trends correlate in the real world. It makes sense that they would, but I haven't seen any studies that really analyze it. This was just my attempt as a high schooler for my senior seminar paper. I think it at least opens the door to using search data to help candidates understand what's working and what isn't. My article isn't demanding that you try to predict elections - just that you consider using Google Trends results in a field outside of technology, in this case in politics.
Regardless, thanks for your reasoned and mature comments, Z.
Haha, excellent. But I'm confused as to whether they've predicted anything besides Obama becoming the nominee.
Yeah, your feedback made me realize that I was being unclear, so I added a "disclaimer", preceded by the word "Update:", to ensure that nobody actually takes this as statistically valid. Sure, you can't conclude anything about this year's race based on my article. But my entire point with this is to show people that there's this tool out there, that can provide a lot of useful information to candidates, which they don't seem to be using. I'm not trying to push any kind of agenda or act a**holish.
Exactly - while this isn't real stats, it shows promise for future research.
Hey, guys, You're all right. There are serious statistical flaws here. I chose the keywords "hillary" and "barack" and "edwards" because that was the model that showed the best success. If one were actually predicting this live, he or she would soon see that "hillary" and "barack" do produce better results and would switch to those terms, simply because it fits the model better. Every election will have its outliers (Obama, Ron Paul). But hopefully this brief overview will provoke some campaign analysts to actually try to interpret Google Trends results for themselves, using more advanced models.
Sure, and this is part of the reason why many states in the South or states that like certain candidates don't work as well with predicting. Google Trends itself currently only uses a relatively small sample size, too, exacerbating the problem. But these charts certainly do not reflect *nothing* -- just be careful what conclusions you draw.