We conducted a field survey in August 2016 during the first month after Pokémon GO was launched in Hong Kong. Researchers approached Pokémon GO players at 5 study sites that represented 5 typical types of built environment. Participants’ data, including sex, age, income, education level, BMI (kg/m²), and the start date of Pokémon GO, were collected using a questionnaire. Participants’ physical activity data, specifically daily walking and running distances, were captured from their iPhone “health” app pages by taking pictures of their screens. Eventually, we compiled a data set covering the period from 14 days before and 21 days after the installation of Pokémon GO. Weather data were collected from Hong Kong Observatory, including total rainfall (mm), bright sunshine (hours), mean air temperature (°C), and mean wind speed (km/h).

The survey was conducted by our researchers on a face to face basis with each participant in the study areas. All participants provided written informed consent to participate prior to survey conduction. Participants received a Pokémon toy as a token of appreciation for completing the study. All data collected were anonymized, and the participants’ names and home addresses were not collected.

We approached Pokémon GO players while they were playing Pokémon GO within the study sites. To ensure the reliability and consistency of data, only iPhone 5s or 6 series users with qualified daily walking and running distances in their iPhone “health” app were included in this study. We excluded individuals who did not complete the questionnaire, were not iPhone 5s or 6 users, and those who were iPhone users but were unable to provide their daily walking and running distances data. We also excluded respondents with insufficient Pokémon GO levels to unlock the primary functions of the game. The primary function of the game was to reach level 5 and gain access to the Gyms.

We used 5 study sites, which represented 5 typical types of built environment in Hong Kong, namely, a green space, an office area, a residential area, a mixed-use area, and a retail location (Table 1). For each built environment, we selected a 200-m radial zone based on the surrounding land use [36] and set ≥60% of the surrounding land use belonging to that type of built environment as the criterion for the site selection through a GIS platform [37].

Daily walking and running distance data represented the daily physical activities of Pokémon GO players. To compare players’ physical activities before and after installation, we used the average walking and running distances each day over a period of 35 days (14 days before to 21 days after game installation). We estimated 95% CI 5.8-6.0 through a bootstrap with 500 resamples of daily walking and running distances. If there were no data on a given day, which meant the player probably did not touch his or her phone at all, or there were problems with the phone, we ignored the data recorded on that day. We excluded 15 observations from the analysis because the data values were empty.

Multilevel modeling was employed to investigate the difference of association between Pokémon GO and daily walking and running distances across players. We have multiple observations of daily walking and running distances of the same individual (from 14 days before to 21 days after the game installation); based on this cluster data structure, multilevel modeling was applied to investigate the relationship between Pokémon GO use and physical activity. At observation level, each observation has its own observation-level attributes, including whether Pokémon GO was played, the number of days played, total rainfall (mm), bright sunshine (hours), mean air temperature (degree Celsius), and mean wind speed (km/h). At player level, age, income, gender, education, BMI (kg/m²), and built environment variables were added. Further, interdependencies among different player levels were taken into account through multilevel modeling. We transferred parts of variables into dummy variables based on questions that we intended to investigate. MLwiN V.3.0 was used to conduct the multilevel modeling analysis.

This research received no specific grant from any funding agency in the public, commercial, or not-for-profit sectors. This study was approved by Survey and Behavioral Research Ethics Committee at The Chinese University of Hong Kong on August 4, 2016. The statistical code and dataset are available from the corresponding author.

Overall, 210 Pokémon GO players were included in this study; Table 2 shows the profile of the participants. We identified 2964 individuals as Pokémon GO players while they were playing Pokémon GO within the study sites. Of these, 1248 answered our questionnaire and 1028 players were excluded because they were not iPhone 5s or 6 users or were iPhone users but unable to provide their daily walking and running distance data. Additionally, we excluded those who did not complete the questionnaire (n=7) or who were unable to unlock the primary functions of the game (n=3).

Figure 1 shows the change in average daily walking and running distances of the Pokémon GO players in the period from 14 days before to 21 days after game installation. The average distance increased by 18.1% (0.96 km), from 5.30 km (SD 2.12) (before installation) to 6.26 km (SD 2.45) (after installation). A comparison of means shows that Pokémon GO was associated with increases in daily walking and running distances (F=33.825, P<.001). We observed a decrease in daily walking and running distances during the period from the 5th to 8th day after installation. During that period, Hong Kong’s weather deteriorated because of the influence of the typhoon Nida.

The results of multilevel models are shown in Table 3. A null model, a model without any input variances, which simply describes the variance at each level, was conducted first. We found that 22.5% of the variance was explained at the observation level, and this provided justification for proceeding to the subsequent analysis.

Model 1 contains observation-level variables. In model 2, the player-level variables were added and analyzed together with the observation-level variables (Table 3). In model 3, assuming that all other variables in the model were held constant, the relation between Pokémon GO behaviors and dependent variables were investigated. In model 4, on the basis of model 3, the interactions of Pokémon GO use (or no Pokémon GO use) and built environment were included (Table 4).

The results of multilevel models showed that the observation-level variables (including whether Pokémon GO was played, the number of days played, total rainfall, bright sunshine duration, mean air temperature, and mean wind speed) were significantly related to daily walking and running distances. Whether Pokémon GO was played (P=.005) and bright sunshine duration (P=.01) were significantly associated with the increased daily walking and running distances. Conversely, the number of days played (P<.001), total rainfall (P=.003), mean air temperature (P=.01), and mean wind speed (P=.04) were negatively correlated with daily walking and running distances. At the player level, our results indicated that players who were male, had high school or lower education, and exhibited normal BMI (kg/m²) were likely to have higher daily walking and running distances (Table 4). Furthermore, playing Pokémon GO in green space was associated with higher daily walking and running distances (P=.03).

Playing Pokémon GO was significantly associated with increased daily walking and running distances (P=.005), whereas the number of days played was significantly associated with decreased daily walking and running distances (P<.001). The results showed that the association between Pokémon GO and daily walking and running distances attenuated over time and may have dissappeared after players played Pokémon GO for 24 days (Table 4).

Most importantly, our results indicate that the effects of playing Pokémon GO are somewhat stronger for players with lower levels of daily physical activity and slightly stronger for those who have played the game for fewer days. These findings are obtained when the estimated effects of playing Pokémon GO and number of days played are allowed to vary across individuals in the sample. The negative estimate of the covariance of playing Pokémon GO and intercept (–0.008 in Model 4) captures the stronger health benefits of playing among respondents with shorter walking and running distances. The estimated covariance between days played and whether Pokémon GO was played of −0.00057 shows that those playing the game for a shorter period of time have, all else equal, experienced slightly less health benefits.

This study reported that after the game installation, average daily walking and running distances of Hong Kong Pokémon GO players increased by 18.1% (0.96 km，approximately 1200 steps) compared with the period before installation, which spanned for over a period of 35 days. These results were largely consistent with those of previous studies that were conducted in USA. For example, Howe et al found that the average number of daily steps for Pokémon GO players during the first week of installation increased by 955 steps [21]. Althoff and his colleagues indicated that Pokémon GO led to significant increases of more than 25% (1473 steps per day over a period of 30 days) [22]. These studies conducted in USA and the study conducted in Asia study showed that playing Pokémon GO was associated with an increase in physical activity in the first month players played the game. These results supported the idea that Pokémon GO may improve public health by promoting physical activity [32,33,38,39].

Our results showed that the increases in daily walking and running distances attenuated over time. This finding is also congruent with those of the previous studies conducted in USA [21,22]. However, the speed of attenuation was different between the study conducted by us, in Asia, and those conducted in USA. Although Howe et al [21] reported that the association was no longer observed after 6 weeks (42 days), our results indicated that the association disappeared faster, in 24 days, with Hong Kong players. This may be because Hong Kong is a relatively fast-paced society [40], which may make it easier for people to lose interest in the game. The high-density built environment, which lacks spacious places for playing location-based AR games, could also be another reason for this difference. If we wish to maximize the positive benefit from the game, initiatives should be taken to find ways to slow down attenuation.

From multilevel models, we found that total rainfall, bright sunshine duration, mean air temperature, and mean wind speed were significantly related to Pokémon GO players’ physical activity levels. The link between the weather and physical activity is well established. An increased duration of daily bright sunshine could increase daily walking activity [41] and participation in outdoor activities [42]; higher rainfall and wind speed are likely to be associated with lower physical activity levels [43-46]; the temperature could have different impacts on physical activities across areas and seasons [47]. For the season during which we conducted our study in Hong Kong, we found that the duration of bright sunshine was significantly associated with increased daily walking and running distances, whereas total rainfall, mean air temperature, and mean wind speed were negatively correlated with the distances. To the best of our knowledge, this is the first study that controlled the weather variables to provide a more precise relationship between Pokémon GO behaviors and physical activity levels. On December 7, 2017, the new version of Pokémon GO was introduced and included weather features, in addition to an in-game weather visual map, weather near Pokémon players will impact Pokémon in a variety of ways. This update made weather an even more important factor for the players [48].

The relationship between physical activity and built environment has been well investigated in recent decades [17]. As a location-based AR game, Pokémon GO is highly connected with built environment and both of these variables could be associated with players’ physical activity levels. In previous studies, however, built environment was not taken into consideration or they showed little impact on physical activity levels [21,22]. We examined Pokémon GO together with built environment and found that green space had a significant positive relationship with daily walking and running distances (Table 4). This result indicated that Pokémon GO may have encouraged players to use the green space around them to engage in physical activity. Nature-based recreation has decreased by 25% in the last 40 years [49], endangering the health benefits associated with nature and green space [50]. To solve this problem, Pokémon GO shows great potential to build connections between green space and opportunities for physical activity.

By controlling for the weather variable at the observation level and for demographics and built environment at the player level, our results showed that players with shorter daily walking and running distances exhibited stronger associations with Pokémon GO. These associations also lasted longer than players with relatively longer distances before installation. This result confirmed the previous findings from the studies conducted in USA that Pokémon GO is able to reach low activity populations [22,26]. Governments and public health agencies may consider the possibility of using location-based AR technology to improve physical activity and public health.

To the best of our knowledge, this study was one of the earliest and the only studies conducted during the Pokémon GO craze (from July 2016 to August 2016). This work offers 3 main contributions. First, we combined field surveys with automatically recorded physical activity data from mobile devices. The field study gave us an idea of who was playing the game outside and engaging in physical activity. Second, this paper was the first paper that investigated the relationship between Pokémon GO and physical activity with the weather and built environment added as covariates. Third, the difference in associations between Pokémon GO use and physical activity among different players was investigated. Such information is important to enhance the positive impact brought by future games or interventions. Further, as the first objective physical activity study of Pokémon GO in Asia, this study provided evidence together with other previous studies, which indicated that Pokémon GO was associated with short-term increases in physical activity levels.

Our study had some limitations. First, our study population was a representative of active Pokémon GO players; it was not a random sample of the Hong Kong population. It could not represent all the Pokémon GO players in Hong Kong. It is possible that some players may have only played at midnight (or during other irregular times), which was not within our survey time, or some people may have only played Pokémon GO in a place that was dissimilar to our study areas.

Second, this was a one group pretest-posttest design study. We acknowledge that we lack a control group; however, our data indicate that Pokémon GO alone was associated with physical activity changes in participants. During the study time, there was no campaign, parade, or other large-scale physical activity-related event that happened in Hong Kong. There were also no dramatic weather changes that could have led to increases in physical activity levels.

Further, we used daily walking and running distances as a proxy measurement of physical activity. We do acknowledge the possibility that other physical activities (eg, swimming and basketball) were likely to be not recorded if the players did not carry their phones during those activities.

We studied 210 Pokémon GO players in Hong Kong during the Pokémon GO craze of 2016. Our results indicated that after the installation of the game, the average daily walking and running distances increased by 18.1% (0.96 km, approximately 1200 steps) compared with the levels measured before the installation. However, this association attenuated over time and was estimated to disappear after 24 days. The results of multilevel models indicated that the weather should be considered in this kind of research and the association between Pokémon GO use and physical activity was stronger among less physically active people compared with physically active people. We found that a game like Pokémon GO has the potential to build new links between humans and green space and to encourage people to engage in physical activity outdoor. Having a better understanding of the relationship between Pokémon GO use and physical activity may cast light on the future efforts needed to promote public health on a global scale.