The Geography of Pokémon GO: Beneficial and
Problematic Effects on Places and Movement
Ashley Colley1*, Jacob Thebault-Spieker2*, Allen Yilun Lin3*, Donald Degraen4, Benjamin
Fischman2, Jonna Häkkilä1, Kate Kuehl2, Valentina Nisi5, Nuno Jardim Nunes5, Nina Wenig6,
Dirk Wenig6, Brent Hecht3**, Johannes Schöning6**
*Indicates co-First Authors, **Indicates co-Principal Investigators
1University of Lapland (Finland), 2University of Minnesota (USA), 3Northwestern University (USA),
4Saarland University (Germany), 5Madeira-ITI (Portugal), 6University of Bremen (Germany)
Contact e-mails: email@example.com, firstname.lastname@example.org, email@example.com,
The widespread popularity of Pokémon GO presents the first
opportunity to observe the geographic effects of location-
based gaming at scale. This paper reports the results of a
mixed methods study of the geography of Pokémon GO that
includes a five-country field survey of 375 Pokémon GO
players and a large scale geostatistical analysis of game
elements. Focusing on the key geographic themes of places
and movement, we find that the design of Pokémon GO
reinforces existing geographically-linked biases (e.g. the
game advantages urban areas and neighborhoods with
smaller minority populations), that Pokémon GO may have
instigated a relatively rare large-scale shift in global human
mobility patterns, and that Pokémon GO has geographically-
linked safety risks, but not those typically emphasized by the
media. Our results point to geographic design implications
for future systems in this space such as a means through
which the geographic biases present in Pokémon GO may be
Location-based games, geography, Pokémon GO,
augmented reality, algorithmic bias, GeoHCI
ACM Classification Keywords
H.5.m. Information interfaces and presentation (e.g., HCI):
One of the most visible HCI developments of 2016 was the
widespread success of the location-based game Pokémon
GO. While the HCI community has studied location-based
gaming for over a decade, Pokémon GO represents the true
democratization of this domain. With a peak popularity
defined by more active users than Twitter and more
engagement than Facebook , it is likely that Pokémon Go
catalyzed the first meaningful experience with location-
based gaming for tens of millions of people around the world.
When a topic that is well-known in the literature undergoes
widespread popularization, it is frequently an exciting
opportunity to address open questions about the topic and its
broader implications. Indeed, the success of Pokémon GO
presents a number of compelling research opportunities in
location-based gaming and related areas (e.g. augmented
reality, computer vision, game mechanics). We expect that
researchers will rapidly begin to leverage these opportunities
in the near future.
In this paper, we focus on an important but targeted subset of
questions about location-based gaming that are raised by
Pokémon GO: those related to Pokémon GO’s geography.
The geographic HCI (“GeoHCI”) [24,25] literature and the
location-based gaming literature have both hypothesized that
the democratization of a technology like Pokémon GO would
have substantial geographic effects, particularly effects
related to movement and places [6,11,16,17,54] (two of the
“five themes of geography” [70,71]). In this research, we
utilized the unprecedented opportunity presented by
Pokémon GO to investigate these broad hypotheses.
Focusing on the themes of movement and places, we ask two
• RQ1: How has the movement of people changed as a
result of Pokémon GO?
• RQ2: Which types of places are advantaged and
disadvantaged by Pokémon GO?
A key component of the successful execution of this research
was its rapid mobilization. With the goal of seizing a
potentially rare opportunity to peek into a geographic future
in which location-based games are an everyday
phenomenon, we collected data at or near the peak of
Pokémon GO’s popularity. The race to study rapidly
emerging topics, however, can sometimes lead researchers to
sacrifice rigor for expediency. In an attempt to avoid making
such a sacrifice, we designed a mixed-methods approach that
was comprised of two studies that each addressed our
research questions from a different angle. First, coordinating
Permission to make digital or hard copies of part or all of this work for
personal or classroom use is granted without fee provided that copies are
not made or distributed for profit or commercial advantage and that copies
bear this notice and the full citation on the first page. Copyrights for third-
party components of this work must be honored. For all other uses, contact
Copyright is held by the owner/author(s).
CHI 2017, May 06-11, 2017, Denver, CO, USA
with a team in Europe and in the United States, we conducted
in-the-wild field surveys of 375 Pokémon GO players across
five countries. Second, focusing on the United States, we
executed a large-scale geostatistical analysis of the
distribution of a fundamental game element in Pokémon GO:
“PokéStops”. This analysis included the application of a
technique called spatial Durbin modeling, a recently
established best practice for controlling for spatial
autocorrelation (an essential concern when examining many
The combination of these two studies allowed us to gain both
a broader and a deeper understanding of the geography of
Pokémon GO than either would have alone, helping us to
answer our two research questions with significantly more
robustness. More specifically, we identified a set of five core
findings across both our field survey and spatial modeling
exercises. These findings point to a relatively cohesive story
about Pokémon GO’s effect on movement and places.
Namely, we found that Pokémon GO causes people to visit
new locations at a remarkable scale (and spend money while
they are there), although this movement is associated with
some degree of distraction-related risk. Critically, however,
while people visit novel places, these places tend to be in
areas with significant pre-existing advantages: our results
strongly suggest that the design of Pokémon GO heavily
advantages urban places with few minorities and the people
who live in these areas.
The effect sizes we identify with respect to race, ethnicity
and the urban/rural spectrum are substantial and troubling. In
the United States, people who live in predominately white
non-Hispanic urban areas have extensive advantages in the
game relative to people who live in urban areas with large
minority populations and to people who live in rural areas.
For instance, we find evidence that predominately white non-
Hispanic neighborhoods in urban areas in the United States
have 20 more PokéStops per square kilometer than urban
areas with very large minority populations, with 20
PokéStops per square kilometer being approximately 4 times
the overall mean density. The effect sizes across the rural and
urban spectrum are even larger: in core urban counties,
PokéStop density is over 95 times greater on average than in
entirely rural counties.
Our results lead to implications for the geographic design of
location-based games. For instance, we discuss below how
game designers can avoid introducing the racial, ethnic, and
urban bias present in Pokémon GO (and even work to
counteract them) through the use of alternative geographic
design strategies. Our results also point to ways to make
movement safer for players by helping them avoid
distraction-related safety risks.
BACKGROUND AND RELATED WORK
Geographic Human-Computer Interaction
This research was motivated by prior work in both
disciplines associated with the “geographic HCI” community
[24,25]: geography and HCI. While not focused specifically
on location-based gaming, geographers have examined the
geographic effects in the highly-related area of augmented
reality (Pokémon GO is often described as “augmented
reality” rather than a location-based game, e.g. ). These
geographers have argued from a critical perspective that data
and code that “augments” reality can remake place, often in
a fashion that reinforces preexisting power structures (e.g.
[14,16,17,34]). This work helped to motivate the decision to
put pre-existing advantages and disadvantages at the center
of our research question about place.
This paper was also motivated by work within the HCI
community that examines geographic crowdsourcing
processes like the efforts in Wikipedia to describe all notable
locations (e.g. [15,19,23,50]) and OpenStreetMap’s efforts
to map the world (e.g. [47,66,67]). Broadly speaking, across
nearly all this work, researchers have identified that these
processes lead to advantaged areas having better coverage
than disadvantaged areas. For instance, researchers found
that rural areas have lower-quality content than urban areas
in Wikipedia and OpenStreetMap (e.g. [30,37,66,67]); that
geotagged tweets and photos are more common in well-
educated areas (e.g. ); and that these biases have carried
over into the crowd processes of the sharing economy (e.g.
[49,56]). This paper adds location-based gaming to this
unfortunate list, but our work also suggests solutions that can
be employed in future location-based games, and perhaps
geographic crowdsourcing more generally.
As noted above, while Pokémon GO is the first blockbuster
success in the area, the HCI community has studied location-
based gaming for over a decade (e.g. [8,40,42,63,65]). Well-
known contributions include (but are certainly not limited to)
Bell et al.’s work on their Feeding Yoshi game, which
introduced the notion of seamful design to location-based
games and provided the first longitudinal, qualitative view
on location-based gaming , and Pirates! , which
highlighted the potential social impact of such games.
Overviews of research on location-based games – which are
also known by other names such as “pervasive games” and
“augmented reality games” – can be found in the survey
papers by Montola et al. , Avouris and Yiannoutsou 
and Magerkurth et al. .
While the vast majority of location-based games research
examines this domain through a lens other than geography
(e.g. technical implementation, game mechanics, narrative,
health benefits, social dimensions), several games have been
designed as geographic data collection tools (e.g. [39,46])
and others have taken a geographic perspective in their
analyses. For instance, Gentes et al. examines the
relationship between space, place, and several other
dimensions (e.g. infrastructure) in the context of location-
based gaming , a discussion that relates to that of the
work in geography mentioned above. Additionally,
geographic factors are often emergent themes in location-
based gaming research, and some of these factors relate to
our findings below. For instance, Bell et al.  noticed that
movement was more difficult while playing Feeding Yoshi
due to the distraction associated with playing the game. This
and a series of related results [5,22,32,57] motivated us to
inquire about this issue in Pokémon GO.
The location based game Ingress (also developed by
Pokémon GO’s developer Niantic) presents perhaps the
closest prior art to Pokémon GO. As is the case with location-
based gaming work more generally, research looking at
Ingress (e.g. [6,28,72]) has largely adopted non-geographic
perspectives. However, this research has led to a few relevant
geographic results, particularly related to movement. For
instance, Chess  highlights that Ingress players become
both more active in their local physical space, but at the same
time become part of the global virtual game space.
Moreover, a blog post  reporting on in a survey of
Ingress players found that 56% of players play within a
radius of 11-100 km and, importantly, 88% of players have
visited previously unvisited locations whilst playing. These
results helped to motivate our question related to movement.
Because a large body of literature tells us that people tend to
be sedentary and their mobility patterns are highly
predictable when they do move (e.g. ), if these patterns
extended to the broader audience Pokémon GO, it could
represent a significant change in human movement behavior.
Through our research question associated with movement,
we sought to see if this was the case, as well as to elucidate
more details about other changes to movement behavior
related to Pokémon GO.
Pokémon GO Background
Pokémon GO is a “free-to-play…location-based game
developed by Niantic for iOS and Android devices”  that
was released in July 2016. Within Pokémon GO, “players
use a mobile device's [positioning] capability to locate,
capture, [and] battle…virtual creatures, called Pokémon,
who appear on the screen as if they were in the same real-
world location as the player” .
There are 151 different Pokémon in the game at the time of
writing, spread across 15 different types, such as normal,
water, ground, grass and ghost types. Individual Pokémon
appear (‘spawn’) temporarily at a location, during which
time they can be caught by players at that location. Catching
Pokémon is one primary way players progress in the game.
In this paper, we focus extensively on “PokéStops”, virtual
game features that are assigned to fixed locations in the
physical world. When players visit PokéStops, they receive
benefits in the game (e.g. Poké Balls which are used to catch
Pokémon, Potions which are used to heal Pokémon after
battles at “Gyms”, and experience points). Additionally, in
certain game conditions (e.g. using a lure), PokéStop
locations have a high frequency of spawning Pokémon.
PokéStops can be revisited, but players must wait at least five
minutes before doing so. In general, as we will discuss, the
higher the PokéStop density in a region, the better for the
Niantic established the locations of PokéStops by drawing
from the locations of “portals” in its earlier location-based
game Ingress . Portal locations were initially seeded with
crowdsourced historical markers , as well as with
churches, parks, monuments, and public art mined from
geotagged images . This dataset was then expanded
using a much larger crowdsourcing process that invited
Ingress players to submit portal locations . This
crowdsourcing system has since been closed, drawing the ire
of the community (this shuttering is considered to be the least
popular ‘game feature’ of Ingress ).
As part of Pokémon GO’s gameplay, players are provided
with a limited amount of information regarding the detailed
algorithms underlying the game. Additionally, the map view
in the game’s mobile app only gives players visibility of
PokéStop and gym locations within an approximately 3 km
radius of their current location.
In this work, we take a mixed methods approach, focusing
on two primary approaches for understanding the geography
of Pokémon GO. First, we deployed a multi-national field
survey and interviewed Pokémon GO players at or near the
peak of Pokémon GO’s launch-related popularity. Second,
we augment our findings in the field study with a
geostatistical analysis of the distribution of PokéStops in the
United States. As we hypothesized, the integration of these
two studies allowed us to gain a broader and deeper
understanding of the geography of Pokémon GO than either
alone. Below, we discuss each of these studies in turn.
We designed our survey to address both our research
question about place and our research question about
movement. After making basic inquiries as to demographics
and Pokémon GO experience, the survey contained a series
of questions targeted at better understanding the role of place
and movement in Pokémon GO. For instance, with regard to
place, we asked respondents to describe a place they found
“boring” (disadvantage) or “exciting” (advantage) with
respect to the game. At the time of survey development, a
series of news stories [75–77] had emerged describing
Pokémon GO players being victimized in “dangerous
places”, with the notion of such places having a long
literature in geography (e.g. ). As such, we also inquired
as to whether participants had experienced any related
incidents. For movement, we inquired as to whether
participants had visited any new locations as a result of the
game (and to describe these locations), as well as the means
by which they engaged in Pokémon GO movement (i.e.
mode of transportation, awareness of environment).
The field study took place during two weeks from July 22,
2016 to August 5, 2016 in five different countries (USA,
Germany, Portugal, Finland, Belgium). Background data
from the respondents is presented in Table 1. The time period
of the study roughly corresponded with a timeframe of 2-4
weeks after the launch of the game in each country, which
aligned well with Pokémon GO’s popularity peak .
All the interviewers (half male and half female) had local
knowledge and selected the locations for interviews as those
places where they had previously observed people playing
Pokémon GO. At each selected location, the interviewer
spent a minimum of one hour. Subsequent interview
locations were chosen to be at least 1km from previous
locations. Interviewers visually identified Pokémon GO
players based on their behavior and approached and
interviewed consenting players, resulting in 375 valid
interviews. The distribution of interview responses by
country was: Germany: 103, USA: 95, Belgium: 68,
Portugal: 59 and Finland: 50.
Respondents’ free text responses were analyzed using an
open coding approach: A single coder defined the codebook,
and two coders evaluated each response. A third researcher
then arbitrated disagreements between the coders. Answers
were coded such that an individual answer could produce
codes in multiple categories, however multiple mentions in
the same category were counted as only a single code.
While there are many geographic data streams in Pokémon
GO that could help augment our field survey, we focused on
the geographic distribution of PokéStops. Specifically, we
examined the metric PokéStop density (PokéStops per square
kilometer) as our core dependent variable.
Broadly speaking, PokéStop density can be thought of as a
proxy for advantage in Pokémon GO. That is, the game
advantages players who live in areas with high PokéStop
density over those who do not. This inequality manifests
itself in several ways. Most importantly, moving in search of
spawned Pokémon is a core element of gameplay (see
below), and in regions with high PokéStop density, there will
always be a PokéStop nearby, ensuring resource availability
as needed (PokéStops provide Poké Balls to catch Pokémon
and Potions to heal players’ Pokémon injured during Gym
battles). Additionally, people in regions with high PokéStop
densities are more likely to have a PokéStop closer to them
at all times than would be the case in regions with low
density (subject to the ecological fallacy). Finally, players in
high-density regions have an additional nuanced but
important capability: they can continuously loop between
PokéStops, substantially reducing the negative effect of the
five-minute revisit restriction on PokéStops (i.e. the benefits
of PokéStop density do not increase linearly).
There are two other geographic elements of the game that
could have been used as proxies for advantage, but both had
important drawbacks. First, we could have analyzed the
distribution of spawned Pokémon, but the geography of this
spawning is highly variable, and collecting these data would
not have been possible under the ethical constraints
described below. Second, Pokémon Gyms are an interesting
geographic element that, like PokéStops, are fixed in
physical space. However, Gyms are not nearly as
fundamental to the game. A player could play the game
without battling in Gyms, but the resources PokéStops
provide are necessary (unless the player wishes to spend their
own money to purchase resources, a possibility in the game,
but clearly a disadvantage). Below, we discuss (1) how we
collected PokéStop data, (2) the types of places we examined
and (3) our geostatistical methods.
We collected PokéStop data directly from Niantic using a
customized data collection program. The program is based
on two open source projects – pgoapi , a popular third-
party Pokémon GO python API, and PokémonGo-Map ,
a Pokémon GO visualization app. At a high level, our
program takes as input the minimum bounding rectangle of
a U.S. county and captures geographic locations of all
PokéStops present in that county.
At the time of our analysis, it was unclear whether our use of
PokéStop data was permitted under Niantic’s terms of
service. Because of this ambiguity and the fact that SIGCHI
is currently undergoing a review of its ethics protocol related
to terms of service and has not yet published its guidelines
, we took as conservative an approach as possible.
Specifically, we reduced our impact to the Pokémon GO
servers to an absolute minimum and collected only data
essential to our research questions. This ensured that the
benefits of our collection program (e.g. identifying racial and
ethnic bias in Pokémon GO) outweighed any costs.
Male 65%, Female 33%, Other 2%
M = 25.1 years, SD = 8.0. Min = 11, Max = 56
iOS 45%, Android 55%
Do not usually play games on smartphone 48%
Played a geospatial game 27%, heard of Ingress
52%, played Ingress 11%
Previous fans 79% (TV shows, Gameboy
games, trading cards)
Pokémon GO Gameplay
M = 20.8 days, SD = 8.2
Mdn = 2 hours (< 1 h = 11%, 1 - 2 h = 35%, 2 -
4 h = 37%, > 4 h = 17%)
M = 78 minutes, SD = 87
Mdn = 17 (1st quartile = 13, 3rd quartile = 21)
Friends 72%, family 29%, exclusively alone
12%, sometimes alone 30%
* Note that we were not allowed by IRB of the US university
participating in this research to approach people under 18 years.
** Typically, players started playing in the same week as the game
was launched in their country.
Table 1. Background data from field interviews (n = 375)
To minimize our impact on the server, we issued requests to
the server as infrequently as possible while still being able to
collect the minimum amount of data to achieve our goals.
This amounted to issuing a request once every ten seconds
and pausing the collector for one minute after every 15
requests. We also maximized the geographic extent of each
request to minimize the overall number of requests.
Because our access to PokéStop data was significantly
restricted by the speed of our data collector, we focused our
geostatistical analyses on specific regions. For our urban vs.
rural comparison, we randomly selected 20 U.S. counties in
each of six government-defined classes along the urban-rural
spectrum, with these classes explained in detail below. For
our race and ethnicity analyses, we focused on two
metropolitan areas: Chicago and Detroit. We also motivate
the choice of these cities below. More generally, these focus
regions mean that the conclusions of our geostatistical
analyses are restricted to the United States (and in some
cases, may be restricted to just Chicago and Detroit). While
the conclusions may apply more globally and the restriction
of focus to a single country (and even a single metropolitan
area) is common in related work in the GeoHCI space (e.g.
[66,67]), future work should investigate these phenomenon
using a more international perspective.
As noted above, the GeoHCI community has identified that
geographic systems can be prone to significant
geographically-linked demographic biases when they rely on
crowdsourced datasets like Pokémon GO does with
PokéStops. Two of the most significant biases that have been
observed occur along the urban-rural spectrum (e.g.
[12,13,26,30]) and across ethnic/racial lines (e.g.
[30,33,49]). As such, when examining places for advantage
and/or disadvantage, we do so through the lenses defined by
the urban-rural spectrum and race and ethnicity. That is, we
ask (1) Do places of a specific racial and ethnic make-up
have advantages in Pokémon GO? and, similarly, (2) Do
more urban areas have advantages over more rural areas?
Following prior work, we make use of specific U.S.
government sources for our demographic data. With regard
to race and ethnicity, we utilize the percentage of the
population that is white and non-Hispanic
, a variable from
the U.S. Census  that is commonly used to assess the
percentage of the population that identifies as a racial and/or
ethnical minority in the United States (e.g. [21,55]). For
urban/rural data, like prior work (e.g. ), we turn to the
National Center for Health Statistics’ (NCHS) urban-rural
ordinal classifications , which assigns each U.S. county
a rating from “1” (“large central metro”) to “6” (“noncore”,
or not part of any metro- or micropolitan area).
The U.S. Census treats race (e.g. “White”, “Black” “American
Indian and Alaskan Native”) as orthogonal to ethnicity
For our urban/rural analyses, we randomly selected 20
counties from each NCHS class. For our race and ethnicity
analyses, we focused on Chicago and Detroit. We selected
Chicago because it has been used in prior related work on
geographic crowdsourced systems (e.g. ). We added
Detroit because it is a poorer metropolitan area with a large
The nature of our datasets required that we use different
approaches for our urbanness question and our race and
ethnicity question. With regard to the former, due to the lack
of spatial autocorrelation (see below) in our random sets of
20 counties, we were able to use straightforward descriptive
statistics to analyze PokéStop density across each NCHS
class on the rural and urban spectrum.
Looking at race and ethnicity within urban areas, however,
requires significantly different methods because of the
presence of spatial autocorrelation. Spatial autocorrelation is
a complex topic and is discussed in an HCI context in several
recent papers (e.g. [30,36]). However, in our particular study,
the presence of spatial autocorrelation means that the
demographics of one area of a city might affect both the
PokéStop density in that area and in neighboring areas
(among other spatial relationships). Indeed, as described
below, people in our field study reported traveling non-trivial
distances in search of PokéStops, which makes accounting
for these spatial dependence relationships critical to our
While autocorrelation was ignored in HCI and related fields
for many years, this is increasingly no longer the case.
However, the methods that have been used to control for
autocorrelation in HCI thus far – spatial error and spatial lag
models – do not capture the spatial relationship between the
demographics in one area and the PokéStop density nearby.
Spatial Durbin models, an emerging best practice in the
geostatistics literature, do capture this type of dependence,
which is fundamental to our analysis. As cross-region
relationships between dependent and independent variables
like those in our analysis (more formally, “exogenous spatial
relationships”) are quite common in spatial data studied in
HCI, spatial Durbin models will likely prove useful for HCI
research questions outside the context of this paper.
Overviews of spatial Durbin modeling can be found in Yang
et al.  and Elhorst .
We applied spatial Durbin models to census tracts (a
standard U.S. Census spatial unit) within Chicago and
Detroit. Our primary independent variable was the percent of
each tract’s population that identifies as non-Hispanic white.
We also included as a control the population density of each
tract, an important consideration given prior work on the
urban-rural spectrum. We log-scaled this variable to account
(“Hispanic”). Most people of Hispanic ethnicity report their race
white, hence the need for a “non-Hispanic white” variable. 
for a long-tail distribution of population densities. Our
dependent variable was PokéStop density measured in
PokéStops per square kilometer (note that spatial Durbin
models also include a “lag” term for each independent and
Spatial Durbin models are interpreted somewhat differently
than standard regression models. Interpretation of the model
hinges on the direct effects and indirect effects. Thus, we do
not report the coefficients fit by the model, as they are not
commonly interpreted (e.g. Yang et al. ). Direct effects
describe the relationship between an independent variable
(e.g. % non-Hispanic white) and the dependent variable (e.g.
PokéStop density) within a tract. Indirect effects describe the
relationship between the average independent variable value
of a tract’s neighbors and the dependent variable in that tract.
More generally, like in a traditional regression, a positive
effect (either direct or indirect) between our race and
ethnicity variable (% non-Hispanic white) and PokéStop
density would indicate that white non-Hispanic regions have
an advantage in the game. Conversely, if no significant direct
or indirect effect is found, no relationship between PokéStop
Density and race or ethnicity would have been identified.
In this section, we present the results of our field survey and
geostatistical analyses. We organize this section into 5 high-
level findings that emerge across both analyses, with results
from the survey supporting the geostatistics and vice versa.
Finding #1: Existing geographic advantages are
The results of both our survey and our geostatistical analyses
suggest that the design of Pokémon GO follows and
reinforces existing geographic contours of advantage and
disadvantage. More specifically, we find that people who
live in urban places with small minority populations (and to
a lesser extent richer places) have distinct advantages over
people who live in other areas, where PokéStop density is
substantially lower. Moreover, the game incentivizes
movement towards these advantaged areas and away from
rural places and places with larger minority populations, a
problem that we will see has important financial
implications. Below, we discuss these findings in detail.
The Urban-Rural Spectrum
Our findings suggest that rural places and the people who
live in them are substantially disadvantaged in Pokémon GO.
The effect sizes in this respect are considerable. Figure 1
shows the results of our randomized county-level analysis.
The figure shows a dramatic decrease in PokéStops per
square kilometer as counties become more rural. While there
are approximately 2.9 PokéStops per square kilometer in
core urban counties, the equivalent number in rural “class 6”
counties is 0.03 PokéStops per square kilometer. This
difference is significant (t(19)=4.2, p < 0.001). Put another
way, the most urban counties have, on average,
approximately 97 times more PokéStops per square
kilometer than the most rural counties. Moreover, this result
also means that Pokémon GO incentivizes people to move
away from rural areas and towards urban areas, where they
can much more easily find dense regions of PokéStops. As
we will see below, this has an effect on travel patterns,
money flows, and other factors.
The results of our survey indicate that rural disadvantage is
so significant as to make the game somewhat unplayable in
rural areas. When asked if there were any places that they
had been in which playing Pokémon Go was boring, 15
percent of respondents reported rural areas, e.g.
“Countryside, outside the cities, no game content there”
(#44, Belgium) and “In the woods; nothing is happening
[there]” (#271, USA). A number of participants responded to
this question by explicitly saying that “rural areas” or the
“countryside” were boring places to play Pokémon GO.
Race and Ethnicity
Our results also strongly suggest that the geographic
distribution of PokéStops substantially advantages areas
with large white (non-Hispanic) populations. Consider Table
2, which shows the outcome of our spatial Durbin modeling
analyses in the cities of Chicago and Detroit. In both cases,
we see a significant and substantial positive direct or indirect
Figure 1. Average PokéStops per square kilometer for
counties across the urban-rural spectrum. Density in entirely
rural counties (class 6) is orders-of-magnitude smaller than
density in urban core counties (class 1).
Table 2. Results of spatial Durbin models
effect for the percentage of the population that is white non-
Hispanic on PokéStop density. Put simply, this means that as
the share of the population that is African American,
Hispanic, and other minorities increases, the number of
PokéStops per square kilometer decreases, often by a
Unpacking Table 2 in more detail, we see in the “Direct”
column that if a census tract in Detroit were to go from 0%
to 100% white non-Latino, the PokéStop density would
increase by 26.8 PokéStops/km2. For context, the mean
overall PokéStop density for tracts in Detroit is 5.7. A similar
trend can be seen for Chicago’s “Indirect” column: the value
here means that if a census tract’s neighbors were to go from
0% to 100% white non-Latino, the census tract would see an
increase in PokéStop density of 21.6 PokéStops/km2. The
relative size of this effect is smaller, though: Chicago has a
mean density of 17.6.
The trends in Table 2 can be seen cartographically in Figure
2, which depicts PokéStop density in Chicago next to a map
of the percent of the population that is non-Hispanic white.
The mostly non-minority northeastern areas are replete with
PokéStops, as is the central business district and nearby
touristic areas. However, the largely African-American and
Hispanic “South Side” and “West Side” have much lower
densities, usually between 0 and 11 PokéStops/km2.
Many coverage bias studies of GeoHCI systems consider
income in addition to or instead of race/ethnicity as the two
are dismayingly correlated in the U.S. and in many other
countries. In keeping with this trend, we examined our
results with an income lens and found a somewhat surprising
result: PokéStop density seems much more linked to race and
ethnicity than income, even though income and race and
ethnicity are strongly associated in our study areas. At a city-
wide-scale, there is some indication that poorer places have
fewer PokéStops: Chicago has a median household income
that is almost twice as high as that of Detroit  and the
mean PokéStop density in Chicago is over three times higher
than in Detroit. Detroit is also significantly less white non-
Hispanic (7.8%) than Chicago (31.7%) , so, as is typical
in coverage bias work, the disadvantage experienced by low-
income areas is one-and-the-same with the disadvantage
experienced by areas predominately populated by minorities.
However, looking at a more local scale within cities, we see
a decoupling of this disadvantage, with PokéStop density
lower in minority neighborhoods but not necessarily in low-
income neighborhoods. We re-ran our Durbin models using
income instead of percent white non-Hispanic and found a
surprising result: despite strong associations between income
and race/ethnicity in our study areas, we did not detect the
same effects for income as we did for race. In fact, we
detected no significant results for income in either city, and
the trends were much smaller (e.g. around 4 PokéStops per
sq. km.). Examining our data in more detail, we observed a
few interesting examples of middle-class, minority
neighborhoods that experience very low PokéStop density.
For instance, this is the case for census tracts in the far south
of Chicago, which tend to be higher income, but unlike areas
further north, are almost exclusively African American.
Finding #2: Pokémon GO can be a rare catalyst for large-
scale destination choice change (Movement)
Humans rarely change their movement patterns. A large
body of work (e.g.[31,45,61]), including recent research in
Figure 2. PokéStop density in Chicago (a) and the % of the population that is non-Hispanic white (b). There is a substantial visual
correlation that bears out in our Durbin models. Data classification (colors) were defined by QGIS’s natural breaks algorithm.
Science , has established that human mobility is highly
predictable, with most people moving between home, work,
and a few other fixed locations (e.g. coffee shop, grocery
store, daycare, religious institutions). However, prior work
in the location-based gaming space suggests that Pokémon
GO might be successfully incenting people to do something
they rarely do: substantially change where they choose to go
(i.e. alter their “destination choice” or “trip distribution” in
transportation science parlance ). Moreover, given the
popularity of Pokémon GO, the game may be encouraging
people to go to new places at a tremendous scale.
Our results suggest that this hypothesis is supported. Two
data points from our field survey stand out in this respect.
First, we asked respondents if they had ever previously
visited the survey location prior to their present visit. Only
83% had visited the location before, meaning that for 17% of
players, Pokémon GO caused them to visit the survey
location for the very first time.
This finding is substantiated by a second finding from our
survey: almost 60% of respondents indicated that they had
visited at least one new place while playing Pokémon GO.
The types of newly-visited locations were highly
heterogeneous and defined by the types of POIs at which
PokéStops were placed. This included parks (mentioned by
22% of respondents who had been to a new location), POIs
like soccer/football stadiums and castles (14%) and water
features (11%). Our data suggests that most of these new
locations are likely within a moderate (but not small)
distance from respondents’ homes or workplaces: the median
distance respondents reported travelling to the survey
location was 3km. However, 9% of respondents did indicate
visiting an entirely new town/city because of Pokémon GO.
It is interesting to note that the 3km finding is concordant
with Figure 2, in which areas of high PokéStop density tend
to be both in and near areas with a very large non-Hispanic
white population. This is captured in the indirect effects in
the Chicago spatial Durbin model, where we see that the
average non-Hispanic white population of a tract’s neighbors
has a substantial effect on the PokéStop density in the tract.
We did not see a significant indirect effect in Detroit,
however, and it may be that Pokémon GO movement is more
concentrated there owing to the fact that areas with large
non-minority populations are very concentrated.
More generally, given the tremendous regularity in trip
destination choice  under standard conditions, if almost
two-thirds of Pokémon GO’s tens of millions of players 
visited at least one new location as a result of the game, this
would represent a substantial and unusual shift in where
humans choose to go. Although more work needs to be done
to understand location-based gaming-related movement in
more detail, this finding may have interesting
interdisciplinary implications. If location-based gaming
continues to grow and if our mobility-related findings apply
in other location-based gaming contexts, it will be important
to consider game-incentivized movement in the many
models in domains ranging from urban planning to
epidemiology that require movement data.
It is also important to consider this higher-level finding in the
context of this paper’s other findings. Most notably, given
our results related to PokéStop distribution, to the extent that
demographic contours are crossed in Pokémon GO-related
movement, the movement likely involves mostly people
from disadvantaged areas going to advantaged areas, rather
than the other way around. As we shift into discussing the
economic geography of Pokémon GO below, the flow from
disadvantaged to advantaged becomes even more notable.
Finding #3: Pokémon GO plays a role in where people
spend money (Movement)
Geographic studies related to movement are interested not
only in the movement of people, but also in the movement of
goods and resources (e.g. [9,10,20,52]). As such, in our field
survey, we inquired as to whether people had spent money at
locations they had visited while playing Pokémon GO. This
question also has important implications related to the
monetization of location-based games, around which there
has been much discussion [82–84].
Almost half of interviewees (46%) had purchased something
at a venue they were near because of Pokémon GO-related
movement. Typically, these were foodstuffs (25% mentioned
purchasing drinks and 23% food). We also found evidence
that, for some players, Pokémon GO was a driver for a day’s
outing, e.g. visiting the cinema after playing was mentioned
by several participants (e.g. “A bar to have a drink and
cinema to watch a movie”; #46, Belgium)). The purchase of
alcohol was specifically mentioned by 11% of participants
(terms such as alcohol, beer, pub, bar, liquor), e.g. “Fast food
and drinks in a beer garden” (#110, Germany).
Finding #4: Pokémon GO is associated with group, not
individual, movement (Movement)
One clear finding from our field survey that has implications
the social computing community as well as the GeoHCI and
location-based gaming communities is that the vast majority
of Pokémon GO players appear to play (and move) in pairs
or groups. 70% of respondents said that they never play alone
and only 12% indicated that they always play alone.
The respondents who indicated that they at least sometimes
played Pokémon GO with others mostly did so with friends
(72% percent of overall respondents) and family (29%). We
also asked respondents who were playing Pokémon GO with
a group at the time of the survey to report the current group
size. The mean group size was relatively small at 2.7 (SD =
1.9), but a non-trivial portion (7%) of respondents were
playing with groups larger than five.
Finding #5: Playing Pokémon Go can be somewhat
dangerous, but the primary issue is movement not
places (Places and Movement)
As Pokémon GO surged in popularity following its launch,
there were many reports in the press about risks to health and
safety associated with the game. These reports fell into two
categories, each associated with one of the two geographic
themes that are the focus of this paper: places and movement.
The bulk of the press reports (e.g. [82–84]) related to places
and revolved around players wandering into “areas that they
should be avoiding”  a type of report that has been shown
to have an important negative effect on people’s “platial”
mental maps in their home regions . The reports
associated with movement described incidents in which
Pokémon GO players, distracted from their surroundings and
immersed in the game on their smartphones, encountered an
environmental hazard (e.g. a car [60,77], a cliff ).
We asked respondents several questions that inquired as to
any risks to their health or safety associated with movement
or places they had experienced while playing Pokémon GO.
Our results suggest that the danger associated with
movement is much more widespread than that associated
with places. Over one-third of respondents (33%) reported
some form of near miss or actual collision with an object in
their environment. Players mostly reported bumping into
signs, poles and other people (as in Bell et al. ).
The most serious implication of players’ reduced
environmental awareness is when they come into conflict
with road traffic. In this respect, 11% of participants recalled
situations in which they had put their personal safety at risk
by, for example, crossing the street without looking. e.g., “I
wasn't paying attention and my boyfriend had to prevent me
from stepping into the street” (#330, USA). While such risks
are also present in other uses of smartphones (e.g. [5,57]),
the excitement of gameplay may intensify the risk.
With regard to place-related danger, only 1 of our 375
respondents reported an incident similar to those reported in
the media (though a degree less serious). This respondent
(#182, Finland) reported being threatened with a knife.
Thirteen percent of our respondents did, however, report
feeling unsafe in a place while playing Pokémon GO. In
some cases, these participants specifically referred to their
mental maps as the reason for their discomfort, e.g. “[I was
in the] inner city” (#45, Germany) and “Being in Dinkytown
late at night with my cell phone out” (#36, USA).
In this section, we first explicate several design implications
that emerge from the five findings above. We then continue
with more general discussion about our results.
Implications for Design
“Geotechnical Design” for Location-based Gaming
Our findings related to the relatively severe bias present in
Pokémon GO are very likely not endemic to location-based
gaming in general. Instead, they are likely emergent from
Pokémon GO’s geographic design (i.e. “geotechnical
design”), specifically the manner in which PokéStops were
As discussed in the Related Work section, the GeoHCI
research community has established through a large
literature that datasets that are the product of organic
geographic crowdsourcing processes tend to have significant
coverage biases. These biases are nearly always
demographically linked, providing advantages for already-
advantaged demographics (e.g. [18,29,30,48]). By relying
heavily on data submitted from Ingress players, Niantic used
an organic geographic crowdsourcing process to distribute
PokéStops. As such, it is not a surprise that Pokémon GO is
a game that advantages urban, white, non-Hispanic people.
Fortunately, alternative geographic design approaches can
likely lead to much more desirable outcomes. For instance,
some reasonable approaches might be:
• Making the reuse of game elements in undercovered areas
more advantageous than in more heavily covered areas
(e.g. reducing the cooldown time for PokéStops or
increasing spawn rates for rare Pokémon).
• Supplementing crowdsourced data in underrepresented
areas with the locations of all public spaces. This can be
done with OpenStreetMap data, among other techniques.
• Using non-geographic crowdsourcing (i.e. Mechanical
Turk) to search through Street View imagery to identify
adequate locations for game elements. Computer vision
approaches can likely be used to partially automate this
process once a training set has been developed.
• Identifying new types of suitable game element locations
for rural areas (e.g. road pull-outs) and dramatically
increasing the density of game elements in the small
populated places that exist in these areas.
• Adding features for groups of co-located players that have
lower geographical dependence, e.g. if five players are co-
located, a PokéStop type element is created dynamically.
Interestingly, bias in location-based games is probably an
easier problem to address than bias in other geographic
datasets important to the GeoHCI literature. For instance, to
resolve the urban biases in Wikipedia and OpenStreetMap,
the corresponding communities will likely have to engage in
content creation and recruiting efforts at a massive scale. For
location-based game designers, once a solution is identified,
it can likely be scaled with minimal effort. It takes a lot of
work to write a good Wikipedia article about an
undercovered place; adding a PokéStop is quite a bit simpler.
Finally, and more generally, our results suggest that location-
based game designers should at minimum audit the
geographic distributions of important game elements. To do
so, they can employ the exact same geostatistical approaches
that we have in this paper (e.g. spatial Durbin modeling).
Reducing Movement-associated Risks
Although the general risks of using a smartphone while
walking in urban environments has been widely reported
, few actual solutions to address the problem have been
proposed. In the scope of location-based gaming and
Pokémon GO, we believe the following approaches to
improving player safety could be explored:
• Avoid game content appearing across a road from the
player’s location, reducing the desire to rush to cross a
potentially busy street (although doing so would involve
interesting challenges at the intersection of spatial
computing and game mechanics).
• Whilst the game already requires players travelling at high
speed to acknowledge that they are passengers rather than
drivers, this feature could be extended to prevent aspects
of gameplay in a moving vehicle that could result in rapid
• Have the system notify the user, e.g. by freezing the UI,
when they are in dangerous areas, e.g. near busy roads.
Capturing a Moment in Time Using Mixed Methods
For those that study location-based gaming and geographic
technologies, Pokémon GO’s dramatic rise to prominence
was a fascinating phenomenon to observe. When the game
became a global blockbuster, we were struck by the
democratization of location-based gaming that was occurring
but, like others in the field [40,63,65], we anticipated that
that its mass popularity would be short-lived.
As such, in order to understand as much about this
phenomenon as quickly possible, we developed a mixed
methods approach that folded a research agenda that would
likely occur in serial under normal conditions into a single
project conducted in parallel. Our hypothesis was that our
two methods would reinforce each other in the same fashion
as if the projects were conducted in serial.
This hypothesis turned out to be supported. The results of our
field study substantially helped to shape our geostatistical
analyses, e.g. contributing to the motivation to use spatial
Durbin models (given the movement range from Finding #2)
and to use PokéStop density as our dependent variable (due
to the number of people that visited the survey locations for
PokéStops). Conversely, the findings of the geostatistical
analysis provided critical context to our survey results. For
instance, without the geostatistical analysis, our results about
people spending money and visiting new places while
playing Pokémon GO are a uniformly positive story. With
the geostatistical analyses, it becomes a story at least
partially about the reinforcing of existing advantages.
Moreover, as expected, since the time of our survey the
global interest in Pokémon GO has waned  (although by
no means dissolved entirely [1,53,86]). Many of the survey
locations, for instance, now have many fewer players than
they did during the period of the survey.
Limitations and Future Work
While our field survey considered five countries, our
geostatistical analyses focused only on specific regions of the
United States. Future work should seek to expand the
geographic reach of these analyses to more areas of the U.S.
and, critically, to different countries. While many countries
have challenges associated with race, ethnicity, and equality
similar to those in the U.S., they tend to have different
geographic structures and histories than their American
versions. The same is true with regard to the relationships
between urban and rural areas. While we expect that the
phenomena we observed generalize internationally at least in
part, it is would be interesting to see the extent of the validity
of this generalization. Additionally, expanding our field
survey to more countries would have similar benefits.
Selecting interview locations using interviewers’ local
knowledge of active Pokémon GO player locations was the
likely only feasible approach to gain insights from a large
number of players quickly. However, the use of more formal
geographic sampling strategies to survey a representative
group of Pokémon GO players would have been preferred if
more time were available. We note, however, that we
observed relatively little geographic variation in our survey
results: besides local differences in geographical structures
(e.g. suburbs are predominantly a US concept), we found
little difference in reported playing and geographical
movement behaviors across the 5 countries surveyed.
Another important limitation and direction of future work
relates to the breadth of the concepts of movement and
places. This paper asked two specific questions: how has
movement changed and which types of places are
advantaged or disadvantaged. However, there are a number
of other questions one could ask about these concepts in
relationship to Pokémon GO and location-based gaming. For
instance: How does Pokémon GO alter the senses of place of
players in places both previously known and unknown to
them? How does the visible extent available to players affect
movement? What types of people are involved in Pokémon
GO-related “migration”? Do players revisit locations they
first discover in Pokémon GO? More broadly, while this
paper began the study of the geography of Pokémon GO, we
have a lot more to learn about the geographic dimension of
Pokémon GO and location-based gaming more generally.
We are releasing our complete survey results so that other
researchers may conduct additional analyses using this data
The paper provided the first detailed snapshot of the
geography of a widely democratized location-based game.
While we expect that some of our findings will not generalize
beyond Pokémon GO and very similar games, others provide
early insight into the geography a world in which location-
based gaming and related technologies are more widespread.
In several important cases, these insights are “canaries in the
coal mine”, providing warnings that can inform the design of
safer and less racially- and ethnically-biased technologies.
We wish to thank GroupLens Research for their feedback,
especially Daniel Kluver and Loren Terveen. This research
was supported in part by Tekes –the Finnish Funding Agency
for Innovation as part of ‘The Naked Approach’ program and
the Volkswagen Foundation through a Lichtenberg
1. Jennifer Allen. The past and future of location based AR
games like Pokemon Go. Retrieved January 4, 2017
2. Nikolaos M. Avouris and Nikoleta Yiannoutsou. 2012. A
review of mobile location-based games for learning
across physical and virtual spaces. J. UCS 18, 15: 2120–
3. Marek Bell, Matthew Chalmers, Louise Barkhuus,
Malcolm Hall, Scott Sherwood, Paul Tennent, Barry
Brown, Duncan Rowland, Steve Benford, Mauricio
Capra, and Alistair Hampshire. 2006. Interweaving
mobile games with everyday life. In Proceedings of the
SIGCHI conference on Human Factors in computing
systems (CHI ’06), 417–426.
4. Staffan Björk, Jennica Falk, Rebecca Hansson, and Peter
Ljungstrand. 2001. Pirates! Using the physical world as
a game board. In In Proceedings of Interact 2001, 9–13.
5. Katherine W. Byington and David C. Schwebel. 2013.
Effects of mobile Internet use on college student
pedestrian injury risk. Accident Analysis & Prevention
6. Shira Chess. 2014. Augmented regionalism: Ingress as
geomediated gaming narrative. Information,
Communication & Society 17, 9: 1105–1117.
7. J. Paul Elhorst. 2010. Applied Spatial Econometrics:
Raising the Bar. Spatial Economic Analysis 5, 1: 9–28.
8. Martin Flintham, Steve Benford, Rob Anastasi, Terry
Hemmings, Andy Crabtree, Chris Greenhalgh, Nick
Tandavanitj, Matt Adams, and Ju Row-Farr. 2003.
Where on-line meets on the streets: experiences with
mobile mixed reality games. In Proceedings of the
SIGCHI conference on Human factors in computing
systems (CHI ’03), 569–576.
9. Robin Flowerdew and Murray Aitkin. 1982. A method of
fitting the gravity model based on the Poisson
distribution. Journal of regional science 22, 2: 191–202.
10. Ruth García-Gavilanes, Yelena Mejova, and Daniele
Quercia. 2014. Twitter ain’t without frontiers:
economic, social, and cultural boundaries in
international communication. In Proceedings of the
17th ACM conference on Computer supported
cooperative work & social computing (CSCW ’14),
11. Annie Gentes, Aude Guyot-Mbodji, and Isabelle
Demeure. 2010. Gaming on the move: urban experience
as a new paradigm for mobile pervasive game design.
Multimedia systems 16, 1: 43–55.
12. Eric Gilbert, Karrie Karahalios, and Christian Sandvig.
2008. The network in the garden: an empirical analysis
of social media in rural life. In Proceedings of the
SIGCHI Conference on Human Factors in Computing
Systems (CHI ’08), 1603–1612.
13. Eric Gilbert, Karrie Karahalios, and Christian Sandvig.
2010. The Network in the Garden: Designing Social
Media for Rural Life. American Behavioral Scientist 53,
14. Mark Graham, Stefano De Sabbata, and Matthew A.
Zook. 2015. Towards a study of information
geographies: (im)mutable augmentations and a
mapping of the geographies of information. Geo:
Geography and Environment 2, 1: 88–105.
15. Mark Graham, Bernie Hogan, Ralph K. Straumann, and
Ahmed Medhat. 2014. Uneven Geographies of User-
Generated Information: Patterns of Increasing
Informational Poverty. Annals of the Association of
American Geographers 104, 4: 746–764.
16. Mark Graham and Matthew Zook. 2013. Augmented
realities and uneven geographies: exploring the
geolinguistic contours of the web. Environment and
Planning A 45, 1: 77 – 99.
17. Mark Graham, Matthew Zook, and Andrew Boulton.
2013. Augmented reality in urban places: contested
content and the duplicity of code. Transactions of the
Institute of British Geographers 38, 3: 464–479.
18. Mordechai Haklay. 2010. How good is volunteered
geographical information? A comparative study of
OpenStreetMap and Ordnance Survey datasets.
Environment and planning B: Planning and design 37,
19. Scott A. Hale. 2015. Cross-language Wikipedia Editing
of Okinawa, Japan. In Proceedings of the 33rd Annual
ACM Conference on Human Factors in Computing
Systems (CHI ’15), 183–192.
20. Darren Hardy, James Frew, and Michael F. Goodchild.
2012. Volunteered geographic information production
as a spatial process. International Journal of
Geographical Information Science 26, 7: 1191–1212.
21. Eszter Hargittai. 2007. Whose Space? Differences
Among Users and Non-Users of Social Network Sites.
Journal of Computer-Mediated Communication 13, 1:
22. Rami Hashish, PhD, D. P. T. Principal, and National
Biomechanics Institute. 400AD. Pokémon GO... And
Get Injured? | Huffington Post. The Huffington Post.
Retrieved September 18, 2016 from
23. Brent Hecht and Darren Gergle. 2010. The Tower of
Babel Meets Web 2.0: User-Generated Content and Its
Applications in a Multilingual Context. In CHI ’10:
28th International Conference on Human Factors in
Computing Systems (CHI ’10), 291–300.
24. Brent Hecht, Johannes Schöning, Licia Capra, Afra
Mashhadi, Loren Terveen, and Mei-Po Kwan. 2013.
2013 Workshop on Geographic Human-Computer
Interaction. In CHI ’13 EA: Extended Abstracts of the
31st CHI Conference on Human Factors in Computing
25. Brent Hecht, Johannes Schöning, Thomas Erickson, and
Reid Priedhorsky. 2011. Geographic human-computer
interaction. In CHI ’11 EA: 29th ACM Conference on
Human Factors in Computing Systems (Extended
Abstracts) (CHI EA ’11), 447–450.
26. Brent Hecht and Monica Stephens. 2014. A Tale of
Cities: Urban Biases in Volunteered Geographic
Information. In ICWSM ’14: Eight International AAAI
Conference on Weblogs and Social Media.
27. Christopher Huffaker. 2016. There are fewer Pokemon
Go locations in black neighborhoods, but why? bnd.
Retrieved September 21, 2016 from
28. Nathan Hulsey and Joshua Reeves. 2014. The gift that
keeps on giving: Google, Ingress, and the gift of
surveillance. Surveillance & Society 12, 3: 389.
29. Isaac Johnson and Brent Hecht. 2015. Structural Causes
of Bias in Crowd-derived Geographic Information:
Towards a Holistic Understanding. Retrieved
September 17, 2016 from http://www-
30. Isaac Johnson, Yilun Lin, Toby Li, Andrew Hall, Aaron
Halfaker, Johannes Schöning, and Brent Hecht. 2016.
Not at Home on the Range: Peer Production and the
Urban/Rural Divide. In CHI ’16: 34th International
Conference on Human Factors in Computing Systems.
31. Ryuichi Kitamura and Toon Van Der Hoorn. 1987.
Regularity and irreversibility of weekly travel behavior.
Transportation 14, 3: 227–251.
32. Sheila G. Klauer, Feng Guo, Bruce G. Simons-Morton,
Marie Claude Ouimet, Suzanne E. Lee, and Thomas A.
Dingus. 2014. Distracted driving and risk of road
crashes among novice and experienced drivers. New
England journal of medicine 370, 1: 54–59.
33. Linna Li, Michael F. Goodchild, and Bo Xu. 2013.
Spatial, temporal, and socioeconomic patterns in the use
of Twitter and Flickr. Cartography and Geographic
Information Science 40, 2: 61–77.
34. Tony Liao and Lee Humphreys. 2015. Layar-ed places:
Using mobile augmented reality to tactically reengage,
reproduce, and reappropriate public space. New Media
& Society 17, 9: 1418–1435.
35. Carsten Magerkurth, Adrian David Cheok, Regan L.
Mandryk, and Trond Nilsen. 2005. Pervasive games:
bringing computer entertainment back to the real world.
Computers in Entertainment (CIE) 3, 3: 4–4.
36. Momin M. Malik, Hemank Lamba, Constantine Nakos,
and Jürgen Pfeffer. 2015. Population Bias in Geotagged
Tweets. In ICWSM 2015 Working on Standards and
Practices in Large-Scale Social Media Research.
37. Afra Mashhadi, Giovanni Quattrone, and Licia Capra.
2013. Putting ubiquitous crowd-sourcing into context.
In CSCW ’13: 2013 ACM Conference on Computer
Supported Cooperative Work, 611–622.
38. Sorin Matei, Sandra J. Ball-Rokeach, and Jack
Linchuan Qiu. 2001. Fear and misperception of Los
Angeles urban space a spatial-statistical study of
communication-shaped mental maps. Communication
Research 28, 4: 429–463.
39. Sebastian Matyas, Christian Matyas, Christoph
Schlieder, Peter Kiefer, Hiroko Mitarai, and Maiko
Kamata. 2008. Designing location-based mobile games
with a purpose: collecting geospatial data with
CityExplorer. In Proceedings of the 2008 international
conference on advances in computer entertainment
technology (ACE ’08), 244–247.
40. Frans Mäyrä. Note on Pokémon Go: fantasy of
collecting and exploring “urban nature.” Retrieved
September 17, 2016 from
41. Michael G. McNally. 2007. The four step model.
Handbook of transport modelling 1: 35–41.
42. Keith Mitchell, Duncan McCaffery, George Metaxas,
Joe Finney, Stefan Schmid, and Andrew Scott. 2003.
Six in the city: introducing Real Tournament-a mobile
IPv6 based context-aware multiplayer game. In
Proceedings of the 2nd workshop on Network and
system support for games, 91–100.
43. Markus Montola, Jaakko Stenros, and Annika Waern.
2009. Pervasive games: theory and design. Morgan
Kaufmann Publishers Inc.
44. Jack L. Nasar and Derek Troyer. 2013. Pedestrian
injuries due to mobile phone use in public places.
Accident Analysis & Prevention 57: 91–95.
45. Anastasios Noulas, Salvatore Scellato, Renaud
Lambiotte, Massimiliano Pontil, and Cecilia Mascolo.
2012. A tale of many cities: universal patterns in human
urban mobility. PloS one 7, 5: e37027.
46. Catia Prandi, Valentina Nisi, Paola Salomoni, and Nuno
Jardim Nunes. 2015. From gamification to pervasive
game in mapping urban accessibility. In Proceedings of
the 11th Biannual Conference on Italian SIGCHI
47. Giovanni Quattrone, Licia Capra, and Pasquale De
Meo. 2015. There’s No Such Thing as the Perfect Map:
Quantifying Bias in Spatial Crowd-sourcing Datasets.
In Proceedings of the 18th ACM Conference on
Computer Supported Cooperative Work & Social
48. Giovanni Quattrone, Afra Mashhadi, and Licia Capra.
2014. Mind the Map: The Impact of Culture and
Economic Affluence on Crowd-mapping Behaviours. In
Proceedings of the 17th ACM Conference on Computer
Supported Cooperative Work & Social Computing
(CSCW ’14), 934–944.
49. Giovanni Quattrone, Davide Proserpio, Daniele
Quercia, Licia Capra, and Mirco Musolesi. 2016. Who
Benefits from the “Sharing” Economy of Airbnb? In
Proceedings of the 25th International Conference on
World Wide Web (WWW ’16), 1385–1394.
50. Shilad W. Sen, Heather Ford, David R. Musicant, Mark
Graham, Oliver S.B. Keyes, and Brent Hecht. 2015.
Barriers to the Localness of Volunteered Geographic
Information. In Proceedings of the 33rd Annual ACM
Conference on Human Factors in Computing Systems
(CHI ’15), 197–206.
51. SIGCHI Ethics Working Group. 2016. ACM SIGCHI
Research Ethics. Retrieved September 18, 2016 from
52. Chris Smith, Daniele Quercia, and Licia Capra. 2013.
Finger on the pulse: identifying deprivation using transit
flow analysis. In Proceedings of the 2013 conference on
Computer supported cooperative work (CSCW ’13),
53. Mike Sonders. 2016. Pokémon GO daily revenue: On
the decline, but there’s still good news. SurveyMonkey
Intelligence. Retrieved January 4, 2017 from
54. Chaoming Song, Zehui Qu, Nicholas Blumm, and
Albert-László Barabási. 2010. Limits of Predictability
in Human Mobility. Science 327, 5968: 1018–1021.
55. Fred Stutzman and Jacob Kramer-Duffield. 2010.
Friends Only: Examining a Privacy-enhancing
Behavior in Facebook. In Proceedings of the SIGCHI
Conference on Human Factors in Computing Systems
(CHI ’10), 1553–1562.
56. Jacob Thebault-Spieker, Loren G. Terveen, and Brent
Hecht. 2015. Avoiding the South Side and the Suburbs:
The Geography of Mobile Crowdsourcing Markets. In
Proceedings of the 18th ACM Conference on Computer
Supported Cooperative Work & Social Computing
(CSCW ’15), 265–275.
57. Leah L. Thompson, Frederick P. Rivara, Rajiv C.
Ayyagari, and Beth E. Ebel. 2013. Impact of social and
technological distraction on pedestrian crossing
behaviour: an observational study. Injury prevention 19,
58. firstname.lastname@example.org. pgoapi - a pokemon go api lib in
python. Retrieved from
59. U.S. Census Bureau. 1999. Explanation of Race and
Hispanic Origin Categories. Retrieved September 15,
60. Alice G. Walton. Pokémon GO Is Actually A Health
Hazard, Study Finds. Forbes. Retrieved September 21,
61. Yingzi Wang, Nicholas Jing Yuan, Defu Lian, Linli Xu,
Xing Xie, Enhong Chen, and Yong Rui. 2015.
Regularity and conformity: location prediction using
heterogeneous mobility data. In Proceedings of the 21th
ACM SIGKDD International Conference on Knowledge
Discovery and Data Mining, 1275–1284.
62. Nick Wingfield and Mike Isaac. 2016. Pokémon Go
Brings Augmented Reality to a Mass Audience. The
New York Times. Retrieved September 20, 2016 from
63. D. Yvette Wohn. Why Pokemon Go is Popular: Maybe
Not The Reasons You Think. Retrieved September 17,
2016 from https://yvettewohn.com/2016/07/17/why-
64. Tse-Chuan Yang, Aggie J. Noah, and Carla Shoff. 2013.
Exploring Geographic Variation in US Mortality Rates
Using a Spatial Durbin Approach: Mortality and Spatial
Durbin. Population, Space and Place 21, 1: 18–37.
65. Lana Yarosh. Notes from the Field: Pokémon GO!
Retrieved September 16, 2016 from
66. Dennis Zielstra, Hartwig H. Hochmair, and Pascal Neis.
2013. Assessing the Effect of Data Imports on the
Completeness of OpenStreetMap – A United States
Case Study. Transactions in GIS 17, 3: 315–334.
67. Dennis Zielstra and Alexander Zipf. 2010. A
comparative study of proprietary geodata and
volunteered geographic information for Germany. In
68. 2017. Pokémon Go. Wikipedia. Retrieved from
69. Pokémon Go tops Twitter’s daily users, sees more
engagement than Facebook. TechCrunch. Retrieved
September 15, 2016 from
70. Five themes of geography. Wikipedia, the free
encyclopedia. Retrieved September 20, 2016 from
71. The Five Themes of Geography. Retrieved September
20, 2016 from
72. The Demographics of Ingress. Retrieved September 15,
2016 from http://simulacrum.cc/2013/01/23/the-
73. How the gurus behind Google Earth created “Pokémon
Go.” Mashable. Retrieved September 16, 2016 from
74. The results are in: Your favourite Ingress feature of
2015. Edinburgh Enlightened. Retrieved September 20,
2016 from http://blog.enl-ed.uk/2016/01/04/the-best-
75. Pokemon Go - the world’s most dangerous game? Mail
Online. Retrieved September 21, 2016 from
76. How “pokemon Go” Is Taking Tech Into Dangerous,
Uncharted Waters. FoxNews.com. Retrieved September
17, 2016 from
77. Girl hit by car while playing “Pokemon Go,” blames the
game. WESH. Retrieved September 17, 2016 from
78. These Charts Show That Pokemon Go Is Already in
Decline. Bloomberg.com. Retrieved September 17,
79. PokemonGoMap/PokemonGo-Map. GitHub. Retrieved
September 21, 2016 from
80. National Center for Health Statistics Urban Rural
Classification Scheme for Counties. Retrieved
September 11, 2013 from
81. Population estimates, July 1, 2015, (V2015). Retrieved
September 17, 2016 from
82. How “Pokémon GO” Can Lure More Customers To
Your Local Business. Forbes. Retrieved September 17,
83. The monetization promise and pitfalls of Pokémon Go.
TechCrunch. Retrieved September 17, 2016 from
84. An Alternative Way To Monetize Pokémon Go. Wall
Street Journal. Retrieved September 18, 2016 from
85. Men fall from cliff playing Pokémon Go. CNN.
Retrieved September 17, 2016 from
86. How Niantic Labs Plans to Follow Pokemon Go’s
Crushing Success | Inc.com. Retrieved January 4, 2017
Note: This version of the paper contains a fix for a reference
issue that appeared in the original version.