.svg)
Foot Traffic Data for Economics Research
Foot traffic data records visits to individual establishments, where those visitors travelled from, and how long they stayed, giving economists a weekly panel of local activity at a grain no official statistic reaches. The datasets described here, from Advan, SafeGraph and PassBy, are available to academics through a Dewey subscription.
How economists use foot traffic data
Place based interventions and nearby business activity
The standard design is a difference in differences on weekly visit counts, comparing establishments within a short radius of a new piece of infrastructure against matched establishments further away, before and after it opens. One study on Dewey used exactly this structure to estimate what happens to visits at local businesses once public electric vehicle charging stations are installed nearby, with the point of interest week as the unit of observation. The variable that carries the estimate is the visit count, and the thing that breaks is that this count is not observed directly: it is scaled up from a panel of mobile devices whose composition shifts as apps enter and leave the underlying SDK network. That makes year over year level comparisons unreliable, so the design has to lean on a contemporaneous control group rather than on the raw trend, and researchers working across long horizons usually normalise each point of interest against its own pre period baseline.
Sectoral and neighbourhood heterogeneity in recovery from shocks
Economists studying how local economies absorb a shock build an event study on visits aggregated by NAICS category and census tract, tracking visit volume, travel distance and time spent against a pre shock baseline to see which sectors and which neighbourhoods return first. Because the panel is weekly rather than annual, it resolves the sequencing of a recovery, which matters when the question is whether high income tracts rebounded ahead of low income ones or whether the gap opened only later. The limitation is definitional rather than statistical: a point of interest that stops appearing in the panel cannot be distinguished from one that closed permanently, so establishment exit and simple dormancy are confounded unless the panel is matched against a business registry or permit record. Category labels also sit at the point of interest, not the transaction, so a mixed use site is assigned to one sector when the visits it generates belong to several.
Retail competition, entry, and demand estimation
In industrial organisation, store level visits stand in for quantity demanded when transaction data at the establishment level is unavailable, and visitor home origins at block group resolution allow catchment areas and distance based choice models to be estimated directly rather than assumed from a radius. Researchers use this to study how an incumbent's visits respond to a competitor entering the same trade area, and how much of the loss is diverted rather than destroyed. The caveat is that a visit is not a transaction: it carries no basket size, no price, and no indication of whether the visitor bought anything, so revenue has to be imputed or the panel joined to spend or receipt data at the brand level. Devices are also not people, and households with multiple phones or none are represented unevenly, which biases catchment estimates toward device rich demographics. Where the stakes are high, two independently derived foot traffic panels can be run side by side to check whether they agree on the direction and size of the effect.
Foot traffic datasets available on Dewey
Advan
Weekly visit counts, dwell time and visitor home origins at the point of interest level, which economists use to build high frequency panels of local activity in the years between annual official statistics.
View datasetSafeGraph
Point of interest records with brand, NAICS category, geometry and spend, giving a stable place identifier that lets visit panels be aggregated to sector or matched against establishment level administrative data.
View datasetPassBy
Branded consumer foot traffic at individual locations, derived independently of the other panels here, which makes it useful as a second source when an estimate needs checking rather than assuming.
View datasetPublished work using foot traffic data
Yunhan Zheng, David R. Keith, Shenhao Wang, Mi Diao & Jinhua Zhao · Massachusetts Institute of Technology · 2024
Jiaxin Du, et. al · Texas A&M, University of South Florida, Grand Valley State University, and Emory University · 2024
Mohamed Amine Bouzaghrane, et. al · University of California, Berkeley · 2024
.svg)