Research Paper: Mobility patterns are associated with experienced income segregation in large US cities

Hi all. We’re happy to share our paper published in Nature Communications on the relationship between mobility in urban areas and the diversity of people we encounter: 55% of experienced income segregation is due to mobility patterns. Using a large-scale mobility dataset of 4.5 million mobile phone users in 11 large American cities, we show that income segregation experienced in places and by individuals can differ greatly even within close spatial proximity. Even “across the street”. Places have different income segregation. On average local shops are more segregated by income than museums, art venues, or airports. Factories are more segregated than offices & some types of restaurants more segregated than others. But there is large spatial variability so the experienced income segregation depends not only on where we live but also on their mobility patterns and the places we visit. We use well-know mobility + segregation models to understand that dependence. Basically, people that are explorers (place and social) are the ones that are less segregated. Our results show that we should complement the understanding of urban residential segregation with studies of how mobility happens beyond neighborhoods.

Paper -> Mobility patterns are associated with experienced income segregation in large US cities | Nature Communications
Article about it -> Mobility patterns influence how we experience income segregation in our cities — MIT Media Lab

Check the diversity of your favorite place in our “Atlas of Inequality” -> inequality.media.mit.edu

We would love to know your opinion/comments!


This topic was automatically generated from Slack. You can find the original thread here.

Thanks so much for sharing ! This is fantastic - congratulations on the publication. Very interesting perspective on income segregation. Would not have thought that income segregation hinges so much on the places people visit opposed to where they reside.

We’re always trying to learn more about non-SafeGraph datasets as these resources can be very helpful for our Community members. I’ll have to keep the Data for Good and Public Search API tools you mentioned in mind. We have lots of Community members ask about similar datasets in general-discussion, and it’ll be good to point them to these. Would love to hear about your experience working with these as it could help our other Community members.

For members reading this, I also wanted to highlight that the code used to produce Esteban’s main results can be found below. It’s tucked away at the end of the paper so I wanted to be sure to drop it here:

Out of curiosity, how far did this conclusion stray from your original assumptions?

I’m also interested in how you settled on the two dimensions of place exploration (an individual’s tendency to explore new places) and social exploration (the tendency to explore places having visitors from different income groups) as modalities in which to understand income segregation patterns. What was tried and discarded before that?

Heads up @Jeremy_Ney - any interest in this for your “American Inequality” blog?

@Ruowei_Yang_UM_Baltimore I know you’ve studied mobility + inequality. Have you come across this paper, and any thoughts about their methodology (detailed here)?

Totally forgot to tag - thanks Karissa! Would be very interested in Jeremy’s thoughts on this!!

Hey @Esteban_Moro_MIT - to give you some background, Jeremy is another one of our fantastic Community members! He actually was a guest on one of our recent events (check out the recording here). He runs a series called American Inequality that looks at aspects of inequality that tend to be overlooked. Check out his work here!

One more thing re: your comment on comparing your results to other datasets to check against bias…

I’d recommend checking out this paper by Amanda Coston, Alexandra Chouldechova, https://dl.acm.org/doi/pdf/10.1145/3442188.3445881

They gave a seminar about it a few months back. It’s a really good analysis of potential biases in mobility datasets like SafeGraph’s, and recommendations to avoid it.

We also published out own whitepaper about bias in the SafeGraph dataset, if it’s helpful

@Derek_Ouyang_Stanford @Daniel_Ho_Stanford_University - I’ve seen you both active in the past on here- any thoughts about bias potential in @Esteban_Moro_MIT paper?

Couldn’t find Amanda to tag-- she’s here: @Amanda_Stanford_RegLab Here’s a recording of that seminar: Mobility Data Used to Respond to COVID19 Could Be Biased - YouTube

@Amanda_Stanford_RegLab - any thoughts on areas of potential bias in @Esteban_Moro_MIT 's paper (thread above)

this is amazing! Let’s definitely chat. Would love to feature your paper on the American Inequality substack if you’re interested? DM me and let’s figure out a way to collaborate and get the word out!

@Amanda_Stanford_RegLab Let me know if you want to collaborate too and can share some of your work as well!

As any other observational data, location data is not free of demographic, temporal and behavioral biases.

thanks for your response! Bias is definitely a perennial problem, and based on that additional context, it seems one you’ve done extensive work to pre-empt.

Random thought, but would you be open to sharing the Twitter dataset in the general-discussion channel? (assuming it’s a public one). I’m not familiar with that one, and others might be able to benefit from it.

Again, congrats on the paper!

This is an incredibly cool study; as someone in political science, I think this data set of experienced income segregation could be a really good explanatory variable for policy choices, and I am probably going to send this on to my coauthors. I did have a few lingering questions, though, particularly about the data.

  1. When was the data collected? I assume that the place visited data was pre-pandemic, given the categories that are predictive of being a place explorer (e.g. particular types of restaurants – ramen, tapas, dim sum), but I couldn’t find in the paper when the data was collected, and if the data collection period included any of the pandemic (and if so, how experienced income segregation changed as mobility decreased).
  2. Do you have any demographic information about the individuals included in the sample? Are children included in the data? (Since it’s smartphone-based, likely only high school students would be allowed to have smartphones.). Public schools seem like they might have some of the highest experienced income segregation, since (unlike other POIs, as the paper demonstrates), schools do only contain people drawn from a small geographic area. Does including/excluding schools change your results?
  3. Did you try to validate these results against other POI data? The paper mentions it uses Cuebiq and Foursquare data; how similar are the results if you use, for instance, SafeGraph mobility data?
  4. I found your point on overall individual experience of income segregation being fairly similar to the income segregation at any given location visited by an individual very interesting. However, I was not sure how to square that with the distinction between explorers and returners. Does this mean that most place explorers choose to explore places of similar integration/segregation (as determined by their \sigma_s)?
  5. I was very also interested to see your result that overall income segregation is similar across cities, because Census-block-based measures of income segregation are not particularly similar across cities. Do you think this is a universal property of cities worldwide, or just of US cities? I’d really love to see some comparison across different countries, if that data is available.
  6. A very small comment on the associated website: the dark color scheme makes it really hard to read street names. In a large city like LA, it’s hard to find your favorite places without being able to read the street names.

Let me try to answer them here:

  1. The data was collected between 2016 and 2017, so it is pre-pandemic. We are now working on extending the analysis to the data during the pandemic and after the pandemic to see the change in experience segregation. More information about the data can be found in the Supplementary Material.
  2. We proxy demographic traits by the census data of the area where people live. This is how we get a proxy for the socio economic status (SES) of the individuals in the sample. No, we don’t have children included in the data. Schools are indeed more segregated than the average place in the city. However, they don’t play a major role in defining individual segregation for most of the people. As you can see in our Supplementary Material, Shopping, other Working places, etc. are were most of experience segregation is built.
  3. We haven’t tried Safegraph data yet but it something in our radar.
  4. Yes, that is precisely what we found! People explore places and, on average, those places have all of them similar integration/segregation. We believe this is reflecting your lifestyle + where you live.
  5. We also found this result very interesting. As you said, cities have very different residential segregation, but it seems that when it comes to work, shop, transportation, we see very similar patterns of experience segregation. This could be a manifestation of common patterns on how we shop, eat, move around in all this big US urban areas. In any case, we do see some individual differences between cities (we have another paper coming up), specially in cities in which there is more public transportation. Also we have data from Mexico where mobility is highly affected by income, so we expect to see different results there.
  6. Thank you for this comment. We tried not to make the visualization too messy and this is why we removed the names of the streets. But you are right, we probably have to include a “day mode” in white and show the names of the streets.