Google Trends and Little Brother's Obsession

Having spent the better part of the afternoon getting my 20 points so I could post this, I'd like to introduce my self to the MGoBlog community.

I am MGoData, a senior undergraduate at Michigan majoring in various technical fields. I am very interested in data and its ability to shed light on human behavior. Before embarking on a 5-6 year tour of PhD studies, I plan to take some time this summer to relax and do some "fun" projects. Given the emphasis this community places on actual facts and numbers over other more subjective metrics, I thought this would be a good place to look for a project or a two.
Jumping right into it then...

Given the success (luck?) Michigan State has had over the last few seasons, there has been a reasonable amount of discussion over Little Brother's "obsession" with everything Michigan. So I ask the question:

Is Michigan State obsessed with Michigan and if so, how can we quantify it?

To answer this question we need some data. We want this data to reasonably capture behaviors associated with constantly thinking about or wanting information on Michigan and we want there to some way to identify this behavior as being directed from Michigan State fans/students towards Michigan fans/students.

Google is a great place to start. People search google for millions of things, millions of times per day, and they search for things they are interested in. Google has had recent success predicting flu outbreaks 2 weeks ahead of the CDC, just be looking for places where people are searching for things like "flu symptoms".

Conveniently, Google has build a tool for called Google Trends with which you can compare the popularity of search terms, including where most searches are coming from. So what happens when we compare people searching for the "University of Michigan" to the people searching for "Michigan State University". Lets look at the data.

http://www.google.com/trends?q=University+of+michigan%2C+Michigan+State+University

The most immediate thing to notice is the obvious fact that more people are searching for The University of Michigan than Michigan State University. Sadly, it appears that "University of Michigan" is becoming less and less popular over the last few years.



This isnt surprising given the national (and world) recognition that Michigan recieves. Google scales this data as follows:

In relative mode, the data is scaled to the average search traffic for your term (represented as1.0) during the time period you’ve selected. For example, if you entered the term dogs, the graph you’d see would be scaled to the average of all search traffic for dogs from January 2004 to present. But if you chose a specific time frame – say 2006 – the data would then appear relative to the average of all search traffic for dogs in 2006. Then, let’s suppose that you notice a spike in the graph to 3.5; this spike means that traffic is 3.5 times the average for 2006.

Now lets look at the other information Google gives us. Conviently, we also get information on geographic regions that are searching for these two topics. The top cities break down as follows:



Again we look to Google's FAQ to understand how these results are calculated and interpreted.

To rank the top regions, cities, or languages, Google Trends first looks at a sample of all Google searches to determine the areas or languages from which we received the most searches for your first term. Then, for those top cities, Google Trends calculates the ratio of searches for your term coming from each city divided by total Google searches coming from the same city.

It's possible that Google uses the IP address of the searcher to tailor results to the geographic location of the user, but others can test this by performing the same Trends search and seeing what you get.

Interpreting these results is a little bit tricky so bear with me. First thing to notice is that everything is scaled to the number of relative search popularity of "University of Michigan" in Ann Arbor. Essentially they use the percentage of all Ann Arbor Google searches that are for "University of Michigan" as a baseline and compare everything else to this number. Again though, the search popularity is scaled by the total searches coming from Ann Arbor so population of other cities does not skew results. If they didn't do it this way, places like New York City would always be on top just because there are so many people Googling for everything.

What are some qualitative things we can learn just from the bar graph. Well first of all, people in Ann Arbor are searching for "University of Michigan" way more than people everywhere else, and people in East Lansing are searching for "Michigan State University" more than any place else. Furthermore, it seems that people in East Lansing are searching for "University of Michigan" more than people in Ann Arbor are searching for "Michigan State University". We can see this by comparing the the Blue Bar next to East Lansing, to the Red Bar next to Ann Arbor.

Even better, though, is the fact that Google will actually let you download text files with actual numbers. The following table are Google's measure for search popularity coming out these top cities. Again notice that Ann Arbor searching for "University of Michigan" is the baseline at 1.000. 

City	university of michigan	university of michigan (std error)	michigan state university	michigan state university (std error)
Ann Arbor (USA)	1	0%	0.1	3%
Ypsilanti (USA)	0.5	2%	0.125	5%
Birmingham (USA)	0.375	3%	0.15	3%
East Lansing (USA)	0.27	2%	0.52	2%
Southfield (USA)	0.215	2%	0.075	3%
Lansing (USA)	0.215	3%	0.29	2%
Bay City (USA)	0.175	2%	0.09	3%
Detroit (USA)	0.155	2%	0.095	3%
Kalamazoo (USA)	0.135	3%	0.09	3%
Troy (USA)	0.13	3%	0.07	3%

From these numbers we can see that "Michigan state University" is about ten times less popular than "University of Michigan" in Ann Arbor (not surprising) where as East Lansing is searching for "University of Michigan" a little more than a quarter as much as Ann Arbor is. Comparing the search popularity of opposing universities in each town, we see that East Lansing is searching for "University of Michigan" a whole 2.7X as much as Ann Arbor is searching for "Michigan State University". Furthermore search popularity of "Michigan State University" in East Lansing is about half that of "University of Michigan" in Ann Arbor, so maybe they just don't like searching for themselves (its no fun reading more "SPARTY NOOO!" articles). 

Finally some disclaimers about this analysis. The entire arguement hinges on the assumption that people's search behavior reflect something about things they are interested in and particularly something they are obsessing over. This is probably a long shot conclusion. Even if we accept that people might Google things they are obsessed with, there is no gaurentee that the search trends for people obsessing won't be washed out by the everyday searches of people who just need information. The fact is we have no idea WHY these people are searching for things, just that they are. Finally, there are probably tons of demographic differences between East Lansing and Ann Arbor that make these numbers really difficult to compare. If, for example, students make up a larger portion of the population if one city, it will skew the data because students use Google in a much different way than other demographics.

However, if you buy that people's search behavior is a reasonable proxy for things they are thinking about a lot, and that the demographic breakdown of two communities, largely driven by students and tech-savy individuals, is similar, then these results are kind of need. We can basically see that people in East Lansing are trying to get info on The Univerity of Michigan nearly 3 times as much as Ann Arbor is trying to find out whats going on in East Lansing. I assume Brain keeps info on who is visiting his site and from where so it might be possible to look at just people trying to get information on Michigan sports.

This is the first of many small projects I'd be interested in doing over the summer. If anyone has questions, comments, or ideas for the future, leave them in the comments.

MGoData.

**DISCLAIMER**
I should probably have post more about this in anticipation of the argument that doing this research is proof of obsession or whatever.  My answer to this is mainly in two parts.  First of all I am an out-of-state student so I don't have the complicated relationship with State that most people do.  I dont have friends or family there and I dont plan on getting a job in Michigan so I wont have to deal with co-workers.

My main point in this is that there are interesting ways to try to test some of the very subjective debates that go on in these parts.  Plus I just plain find it interesting that you can attempt to make sociological claims from behavioral data being generated and made available these days.