A Tale of Cities: Urban Biases in Social Media

I am one of the rare computer scientists who likes country music. Not just “old” country music – e.g. Johnny Cash – but the newer stuff as well. Think Eric Church, The Dixie Chicks, Brad Paisley, and Toby Keith.

When I get a chance to jam away to America’s most popular music genre in my office, I am frequently presented with a stark contrast: the latest papers in my research area tend to be focused on advances related to urban computing, “smarter cities” and “connected cities” but the music to which I am rocking out has a spiritual home in rural areas and has an overwhelmingly negative view of technological change. Consider Miranda Lambert’s recent hit, “Automatic”, which is a harsh critique of automation featuring a protagonist who grew up in rural Texas.

This contrast highlights a much larger, ages-old trend: Rural areas have had a different relationship with technology than cities for a long, long time. In the United States, for instance, the adoption of the telephone and electricity followed a unique  – and often delayed – trajectory in rural areas compared to the deployment of these technologies in urban areas.

In the “big data” era of crowdsourcing and social media, differences in how people use technology have implications far beyond the individual technology user and her immediate social network. A person who does not tweet or use Facebook loses more than just the chance to, for instance, broadcast information to her friends. She also effectively removes herself from numerous studies and omits her points-of-view from technologies like enterprise sentiment monitoring. Heck, a person who does not use social media probably removes herself from a good 50 percent of the studies and technologies written about on this blog.

In order to understand the extent to which differences in technology use between urban and rural areas have led to urban biases in social media “big data”, my colleague Monica Stephens (of Floating Sheep fame) and I executed a series of studies on Twitter, Foursquare, and Flickr, focusing on the United States (we hope to expand our research to other countries in the future).

Examining large datasets of tweets, check-in, and photos, we found a surprisingly extensive urban bias in all of these social media communities. For instance, there are 3.5 times more Twitter users per capita in core urban counties in the United States than rural counties. Moreover, urban users tweet more than their rural counterparts, resulting in there being 5.3 times more tweets per capita from city-dwelling Twitter users than rural ones.

It was with Foursquare, however, that we saw the largest urban biases: there are over 24 times more Foursquare users per capita in urban areas than rural ones. In other words, a rural venue – say The Pink Pig BBQ in Hardeeville, South Carolina – is likely to get many fewer check-ins per visitor than a hip urban one — say, HipCityVeg in Philadelphia. Moreover, many of the check-ins (and subsequent reviews) for restaurants like the Pink Pig are likely to come from urbanites travelling through, rather than people who can express local perspectives.

The enormous discrepancy in Foursquare usage across urban and rural areas begs the question: Has the location-sharing model in Foursquare fundamentally failed to appeal to rural users? If so, what model might appeal more? At the turn of the 20th century, the telephone had to be adapted to rural areas’ needs. Perhaps the same is true for location-sharing.

Many social media datasets are biased towards urban points of view.
Our research suggests that many social media datasets are biased towards urban points of view.

More generally, our results add to the growing body of literature that shows that the views and observations presented in social media do not perfectly represent those of the overall population. Whether you’re a social media researcher or the person at your company in charge of social media, it’s important to remember that you’re dealing with a heavily biased sample, especially when considering the urban/rural dimension.

We also hope that our work can add to our colleague Eric Gilbert’s call for more research on technology use in rural areas (and the development of technology specifically designed for these areas). In the United States, almost 60 million people live in rural areas (over 19 percent of the population), yet rural users’ relationships with technology receive almost no attention from many sub-fields of computer science, including my sub-field, human-computer interaction.

I’ll be presenting the paper Monica and I wrote about this work at this year’s International Conference on Weblogs and Social Media (ICWSM), a venue for research on social media and related topics that is, like Miranda Lambert, quickly rising up the A-list. For those of you who can’t make the conference, you can read the paper here.

About the author

Brent Hecht

I'm a computer science professor at the University of Minnesota. My research interests lie at the intersection of human–computer interaction, geography, and big data, and my research centers on the relationship between big data and human factors such as culture. A major focus of my work involves volunteered geographic information and its application in location-aware technologies.

View all posts


  • Is it possible that the Yelp issue could be explained as a critical mass problem? e.g., that with fewer people on these networks, there is less incentive to tweet or rate, and thus fewer ratings/tweets?

  • Interesting study … Do the results depend on rural/urban dichotomy or is it a continuum? There are known scaling laws for cities: e.g., GDP, intellectual output, etc. grow as some power of city size. Your result could be an expression of the same, and my prediction is that 4square activity will vary monotonically with city size (Twitter variation probably too small).

  • Yup, cold start / critical mass phenomena could be playing a role. That said, although their customers live further away, a rural business still has to have a sufficient number of customers “within range” in order to support the sale of the good. We’re looking into some of these issues now 🙂

  • Great question! It’s definitely a continuum (we have Spearman’s correlations in the paper), but we’re working on understanding the relationship in more detail along the lines of what you have mentioned!

  • Great post! I’m thrilled to see more HCI research on rural areas. Part of what I think is happening with location-based media like Foursquare and Yelp is that when the space is constrained (say in a small town) location information isn’t useful enough to warrant using an app. I don’t need to broadcast my location because you can literally see me, or if I’m getting coffee you know I’m at the only place to buy coffee. Anonymity and exposure work differently in rural areas too – critical mass isn’t the only issue but is related to the social consequences of reviews/comments in a small social space.

  • Hi Libby!

    These are definitely possible causes for the phenomena we observed. The latter one is one that has come up in a number of our recent conversations about this work. I think it’s particularly interesting because it would mean that even when a rural “venue” gets a review, it could be even more likely to come from an outside perspective rather than a local one. Investigating this is on the active project list for this summer 🙂 We’re actually looking at a related phenomenon on Wikipedia first (e.g. who gets to define a place?)

    Thanks for the wonderful feedback!

    p.s. Sorry for not seeing this earlier! I was swept up by the ICWSM + CSCW madness 🙂