« Steve Strogatz writes about Math, Science, and the World |
Main
| Do You Have Physics Envy? »
12 March 2010
Cell Phone Data Collection - New Experiments and Existing Resources
I've been asked a lot by many different researchers how they can get their hands on behavioral data logging programs that work on cell phones, such as in Nathan Eagle and Sandy Pentland's landmark Reality Mining study. That study was back in 2004, and they were using old Nokia phones with the Symbian OS, which presented a host of problems. Below I'll go through the currently available data logging applications for phones, and I'll describe a new system being built on top of Android that will allow for an incredibly enhanced platform for social scientists. All of these applications log Bluetooth proximity information, call logs, and cell tower IDs, but some log additional information such as WiFi access points, SMS messages, and accelerometer data. Here are many of the dominant data logging applications available today:
Nokia
Only 6600 phones are officially supported, but the Context Group at the University of Helsinki has developed a number of behavior logging applications for these phones, available for download here (use mitv2).
iPhone
The iPhone is nice because a lot of people have them, but it's a poor choice for data logging because it does not allow processes to run in the background. This means you have to have jailbroken iPhones to run these applications, and it also means you can't offer them for download on the official app store. Anmol Madan from our group has made an iPhone app available for download here, and he also wrote a short tutorial on how to get this application running. Your iPhones have to have older versions of the firmware, however, and it doesn't work with the new 3G iPhones.
Windows Mobile
This is still a widely used phone OS, and Anmol has written a fairly robust data logging application that eclipses all of the previous versions in functionality with WiFi access point logging, survey launcher, and automatic updating tool. He hasn't made it available for download quite yet, but it should be appearing in the next few weeks on the Human Dynamics Social Evolution website. Unfortunately, this version will not be useful for new phones in a few months because Microsoft is releasing Windows Mobile 7.0, which is not compatible with the old 6.x version that this application is written for.
Android
Now the good news: Android phones are becoming increasingly popular and will most likely eclipse all other platforms as the dominant phone OS. Almost every cell phone manufacturer is producing Android phones and with a unified and unrestricted app store there is an opportunity to easily reach millions of people after a short development period. Android also allows easily for automatic updates.
Nadav Aharony from our group is spearheading the project for creating an Android data logging application, and he has already deployed it on over 50 phones in a new study of consumption patterns among family groups (rather than the normal college students in dorms study). This application logs most of the usual suspects (Bluetooth, WiFi access points, call logs), but it also hashes the contents of text messages, allowing researchers to see not just who texts who, but get an idea about how topics spread (not the actual content, since the words are hashed, but just that topic A passed from person 1 to person 2). Actually this application has been running on my phone for over a month with no real problems. The platform also comes with a special app store that allows researchers to log what applications people install, allowing you to look at how application usage spreads among friends. Soon Nadav is also planning to allow researchers to deploy their own apps over this new app store so that researchers can push surveys or more sophisticated logging tools to study participants. Instead of paying for apps, though, users will get paid to download apps so that they will participate (sort of like Mechanical Turk).
The Android Reality Mining platform promises to be extremely powerful, and the results from the current study should further push the boundaries of computational social science.
Posted by Ben Waber at March 12, 2010 10:33 AM
