How graphs can lie to you

I want to discuss graphs for a bit. When we were wee first years at college, we spent a fair amount of time talking about how to make graphs well. Anyone can make a graph in Excel, but you need some knowledge to choose the right graphs for the right sets of data. You need to know when you need error bars and how to choose the right size error bars. Scale is an important aspect as well. Have a look at these two graphs.

This screen capture was taken Friday night, little before 7 pm central time.

The two graphs set the front runner as the whole and the other candidates as portions of that whole. No matter how many people are added, Obama and McCain stay as 100% of the available graph. But the more important issue is that it makes it look like Obama and McCain have the same amount of support, and that Hillary is way behind McCain. Thats part of the reason I started making my own graphs.

This graph is linked to the data on my google spreadsheet, it will update.

This graph puts all six of the candidates that ABC/Facebook is tracking on the same scale. Its pretty clear how huge of a lead Obama has among the Facebook crowd. Its also clear that Hillary is actually ahead of McCain by about 35,000 people, but thats rather hard to see since the difference is much smaller relative to Obama's lead.

So when you see a graph, look for the scale and just because two bars look the same doesn't mean that they're the same value.

Just in case you weren't sure...

Its true, I'm a huge nerd. This graph I made tracks support of the major candidates based upon the Facebook/ABC US Politics application. I record the data using Google documents over my lunch break, but I also have an Excel file on my computer that I update in the evening.

There are a couple things that I find interesting. The first few are in the raw numbers; Obama has a commanding lead, Paul has a strong contingent, and McCain only broke one hundred thousand supporters last week. The next few interesting things are in the change in support from one day to the next. McCain gains more people per day than Clinton. The day after Gravel announced he would seek the Libertarian Party's nomination, he started gaining supporters. For awhile there, the net gain of people for all three Republican candidates was less than McCain's because Huckabee was loosing so many supporters. The next interesting thing is on sheet two, each candidate's daily gain of support as a percentage of the net gain. Obama's daily gain is consistently sixty to seventy five percent of the daily gain. He is gaining more people than everyone else combined every day.

Now the graph of Change in Support will, I think, be more interesting as the sample size increases. But for now I think it just shows that less happens on Sundays.

The thing to keep in mind is that this is far from a representative sample of Americans, and probably not even a great sample of young Americans. Facebook started as a site for college students with a valid .edu email address. While it has become an open site, I don't think it would be unfair to guess that it still skews toward the college student. Plus, the people who have this application and have used it to show support for a particular candidate probably care about national politics a bit more than the general population.

So if this isn't a representative sample, is it really that interesting? I think so and not just because I made it. It does show that there are at least a million people who have gone out of their way (but not terribly far) to show their support for a candidate for president. Will this translate into votes? Votes by young people? I don't think you can tell that from this data. Especially since there really isn't anything like it to compare it to.