Some Research Notes on Using Geo-Tagged Cell Phone Data to Study Urban Economics Questions

 Several excellent urban economics research teams have access to U.S geocoded cell phone data.  Since I am not part of any of these teams, I will use this blog post to offer them some unsolicited advice.  For some examples of these teams; skim this  and this .  

Geotagged data provide researchers with a spatial and temporal high frequency database to know where each person spends her day.

I carry a cell phone with me 98% of the day when I am awake.  This suggests that I am geotagged.    An observer with access to my high frequency cell phone geotagged data would see the following patterns in my 2020 data;

1.  I am not in Baltimore even though I work at Johns Hopkins.  I have adapted to the crisis by going West.

2.  80% of the time I am within 1.5 miles of our Los Angeles house.

3.  20% of the time, I am elsewhere in California.

4.  I am rarely in a retail district in LA.

5.   I walk 2.5 hours  a day.

These five facts don’t add up to a QJE paper.   The real action begins if one could access my Amazon Prime data in 2018, 2019 and 2020 as well as my geotagged data.    The observer would see a more than $1 to $1 crowd out of face to face shopping in 2020 as we do almost all of our non-fruit shopping online.

So, my first point is that Amazon has access to its own data and knows each person’s home address who is already a member of Amazon Prime or has purchased anything on the platform. IF Amazon’s researchers can access anonymous cell phone data then they can geocode those data and match these “anonymous” people to people it can identify in its database.  In then can track which people are still engaging in more retail shopping and target them with special incentives to substitute to the Amazon platform.  The geocodes from our cell phone indicate which people originating at which homes make weekly trips to which retail outlets.  Food shopping patterns will quickly emerge even though no $ statements concerning actual expenditures are observed.

With a data merge of the Amazon platform data and the anonymous geotagged cell phone data, a researcher can study many interesting research questions.  For example, when wild fires occur or when other natural disasters occur;  how quickly do people start to adapt to the emerging crisis? How do they use the market products that Amazon sells to protect themselves?   

Do only the rich engage in this activity?  By scrapping Zillow data, one could know each person’s neighborhood home value.  One could also use monthly Amazon expenditures to infer a person’s income.  Armed with these wealth indicators, the researcher can study whether poor people are engaging in similar adaptation strategies as richer people during air pollution and natural disaster crises.

Returning to low frequency events, one could geocode the public schools in a city and see which adult cell phones commute to those schools during school days. This would indicate which people have school aged children.  This is an example of how to fleshout the set of X covariates about the cell phone owner even without surveying her. She reveals who she is by what she does all day long.

Once the observer knows who has school aged children (and the Amazon platform purchases can also be used to establish this), one can use the cell phone data to explore summer vacation patterns.  Are summer experiences a complement or substitute with the amenities where one lives?  Do LA people go somewhere sunny in summer?  

Returning to environmental economics, on highly polluted days — how much more time do people spend indoors? How much less time do they exercise on those days?  

In 2021 as the pandemic subsides, who will go back to work and how often will they go?  The geocoded cell data will be ideal for studying this.   Suppose a researcher has the universe of LA cell phone geocoded data, by calculating the distance between points (assuming this can be done in a 3 dimensional space such as an office building’s 17th floor), the researcher can infer which people work together! What new questions can then be answered?   

Turning to local macro-economics,  how does time outside of the home change as an economic boom begins?  Due to the rise of Working from Home, do people spend more time inside working during a boom?

As I think more about this topic I will add more points.  What I hope I have conveyed is that the cell phone data create a great natural experiment laboratory.   

A weakness here is that this yields a time allocation facts, it does not report “output”.  In Gary Becker’s time allocation model, you allocate time across tasks and you gain utility from these tasks.  To clarify, here is an example.   Suppose you are shopping for Christmas and your goal is to make your spouse happy.  The researcher observes how many minutes you spend at each store but we do not observe your spouse’s marginal utility gain if you shopped for one more minute at one store versus another.  The optimizer will think this through and the “bang per buck” condition will hold but the observer won’t know your personal gains from this optimizing behavior.  Instead, the observer will observe your optimal time allocation based on your geocoded data.