Last updated: January 20, 2025
Article
New Research Shows Mobile Devices Are a Powerful Way to Learn about Visitors
U.S. national parks have long relied on question-based surveys to capture snapshots of who visitors are and what they like to do. Data from mobile devices provide a complementary, park-wide perspective not previously available.
By David Pettebone, Jake Jorgenson, and Pamela Ziesler
Image credit: NPS
Located in central Virginia, approximately 100 miles south of Washington, DC, Richmond National Battlefield Park and Maggie L. Walker National Historic Site preserve the memory of significant events in American Civil War history.
They’re a large, disconnected group of historical sites in the City of Richmond—formerly a Confederate stronghold—and in Henrico, Hanover, and Chesterfield counties. Over the last decade, like many other national parks, these parks have had to operate with fewer staff. This meant they had to seasonally or permanently close some visitor centers, which resulted in fewer contacts with visitors.
Yet major cultural shifts have drastically altered the commemorative landscape of the City of Richmond, raising public awareness about the continued significance of the Civil War and the Reconstruction Era. These shifts have potentially changed how visitors use parks like these. In light of these changes, both parks recognized they must operate more strategically to better serve visitors and the larger Richmond area community. To do that, they first had to understand how visitors use the park as well as their needs and interests.
As described in our 2023 visitor study report, we used commercially available mobile device location data (called “mobility” data) to estimate visitor use, movement, and demographics in the two parks. We found significant discrepancies between what the 2023 mobility data showed and the results of the most recent (2010) conventional, questionnaire-based survey of park visitors. Our study demonstrates that mobility data can provide important visitor use insights hard to obtain by other means.
Evolving Outreach
The parks’ outreach to visitors has evolved over the last decade. Previously, the parks used their websites to direct visitors to begin their trip at the Tredegar Visitor Center, a time-honored way for agency staff to interact with visitors. But a few significant events have changed how people use the parks and obtain information. After the agency’s centennial celebration in 2016, anecdotally, park staff saw more visitors using the battlefield green spaces to recreate. The COVID-19 pandemic appeared to exacerbate this, with fewer people visiting visitor centers. Then in 2021, the National Park Service released its mobile application, enabling visitors to download maps and information about the parks and avoid visitor centers altogether.
The parks wanted to know how visitors used park facilities like these and to determine if current programs were serving their intended purpose.
Nevertheless, park staff wanted to interact more with local and residential users. Recognizing that visitor centers were no longer the hubs where people began their trip, Interpretation chief Stephanie Pooler established the “Roving Ranger” program. This program sent staff to strategic park locations like Malvern Hill, Drewry’s Bluff, and Shelton House in Totopotomoy Creek to connect with visitors. The future of the Tredegar Visitor Center was also under discussion. The parks wanted to know how visitors used park facilities like these and to determine if current programs like Roving Ranger were serving their intended purpose. They realized they needed to know more about park visitors to effectively serve them.
Image credit: NPS
The National Park Service has long used questionnaire-based park visitor surveys to find out which locations people visit and how long they stay there. Researchers have also asked visitors to carry GPS units to document where they travel inside parks. Yet in some park settings, both of these approaches have limitations related to sample size and representation. Urban parks made up of buildings scattered throughout a city, for example, may have no formal entrance stations or gates. So some visitors may not encounter park staff handing out questionnaires or researchers giving out GPS units.
Urban parks may have no formal entrance stations or gates. So some visitors may not encounter park staff handing out questionnaires or researchers giving out GPS units.
Mobility data can help parks overcome these limitations, improving our understanding of visitation in hard-to-survey locations. Unlike questionnaire-based visitor surveys, collecting mobility data doesn’t require approval from the Office of Management and Budget for compliance with the Paperwork Reduction Act. This reduces the administrative burden on parks of conducting visitor use research. We used commercially available mobility data to obtain the visitor use data the parks needed.
A Complement to Conventional Surveys
Mobility data are a type of “big data,” meaning extremely large quantities of information, that record location. Cell service providers passively collect these data, like GPS navigation data, from people’s cellular devices when users turn on app location permissions. Data warehouses are data management systems that store these passively collected data from over 250,000 individual apps. Social science researchers can purchase anonymous mobility data—scrubbed of all content that identifies individuals—for specific study areas and periods of time. This information gives us an efficient, consistent way to analyze visitation at multiple sites. That’s very difficult to do with more conventional data collection methods like questionnaire-based surveys.
Mobility data provide a less detailed, park-wide perspective over a longer period of time.
Mobility data enable us to understand the movement patterns of individuals. From them, we can gain insights like the time of day and days of the week people visit and where visitors are coming from. But mobility data are not a substitute for conventionally acquired survey data; the two types of data complement each other. Conventional survey data give us detailed snapshots of people’s activities in certain places at certain times, and mobility data provide a less detailed, park-wide perspective over a longer period of time.
The National Park Service’s Social Science Program hired a contractor to acquire, organize, clean, and analyze the mobility data for this project. We consulted park staff to identify the geographic areas for which the parks wanted to acquire visitation data. In total, we acquired over 133 million individual records for January 1, 2022, to August 31, 2023. Cleaning these data was a critical step before analyzing them. This is because uncleaned mobility data have duplicate observations, observations from aircraft, and GPS errors, which need to be removed to ensure accurate results. We analyzed the cleaned data for visitor origin, demographics, and cross-visitation patterns (how visitors move across park sites), among other factors.
Understanding Visitor Origins
We wanted to understand how visitors who live close to the parks use them. Because device data were collected continuously, we could infer approximate home origins from frequent evening locations. If we saw a device at a location during evening hours for three or more months per year, we assumed it to be a home location.
U.S. privacy regulations require mobility data providers to record home and work locations somewhat inaccurately, forcing error into the data. But although estimates of visitor locations weren’t precise, we could still make assumptions about home origins on a large geographic scale, like city, county, and state. We defined “local” users as visitors with a home location in Richmond City, Henrico County, Hanover County, or Chesterfield County. We defined “residents” as local users who lived in the Greater Richmond Metro Area.
Visitors were predominantly from Virginia for all battlefields and sites.
The data indicated that the majority (80 percent) of park visitors were local. Visitors were predominantly from Virginia (94 percent) for all battlefields and sites, followed by the neighboring states of North Carolina (1 percent) and Maryland (1 percent). Sixty seven percent of the devices in Parker’s Battery and North Anna had local origins. In contrast, 94 percent of the devices in Chickahominy Bluff had local origins. This is likely because North Anna is open to researchers but not the public, and Parker’s Battery is a small site south of Richmond in an industrial area, but Chickahominy Bluff is in a residential area in the city of Richmond. Chimborazo, a popular site located in the center of Richmond, received the highest proportion of residents (19 percent), followed by Chickahominy Bluff (18 percent) and North Anna (15 percent).
Detecting Visitation Patterns
Data from each device in the mobility dataset had a unique identifier, so we could tell when the device traveled across multiple park locations. We could do this for daily visitation data, which told us when a device was detected at park locations throughout a day. Or we could do this for annual visitation, analyzing when a device was observed at locations throughout a calendar year.
Image credit: NPS
Jackson Ward had the highest levels of cross-visitation compared to all other park sites. This isn’t surprising, as Jackson Ward is located in downtown Richmond, near Maggie L. Walker National Historic Site and the Tredegar Iron Works. Forty percent of people who visited the Garthright House also visited Cold Harbor on the same day, nine percent were seen at Gaines’ Mill, and three percent also visited Fort Harrison. In contrast, visitors to Beaverdam Creek, Chickahominy Bluff, North Anna, or Tredegar were more likely to visit only one of these sites on a given day.
Our analysis of cross-visitation patterns showed that visitors who weren’t from the area were more likely to visit multiple locations in a single day. External visitors to Garthright House were nine percent more likely than locals to also visit Gaines’ Mill, eight percent more likely to also visit Cold Harbor, and six percent more likely to also visit Fort Harrison on the same day.
Estimating Demographics
We estimated demographics like age and ethnicity profiles of park visitors from these mobility data. But demographics are the least accurate type of information gleaned from mobility data, because estimates are based on U.S. census block data. A census block consists of 250 to 550 households, which equates to about 4,000 people. We estimated the demographic characteristics of visitors by calculating the percentage of different groups in each census block rather than on information about individuals. For example, if 1,000 people from a census block visited the park and 40 percent of that census block identified themselves as African American, we would classify 400 of those park visitors as African American. This was a rather rough estimate, but it provided insight on a park-wide level—a scale not previously available.
Visitors across the parks were fairly evenly distributed by age but skewed slightly toward younger people.
In our study, visitors across the parks were fairly evenly distributed by age but skewed slightly toward younger people. We estimated that 64 percent of visitors were 44 years of age or under. Certain sites had a larger share of children under the age of 18, such as Totopotomoy Creek (26 percent), Beaverdam Creek (24 percent), and Chickahominy Bluff (23 percent). We found the highest percentage of visitors over 65 years of age at North Anna (23 percent), Fort Harrison (23 percent), and Malvern Hill (20 percent).
Overall, we estimated that 53 percent of devices were from people who would likely identify themselves as White. Thirty-seven percent were from people likely identifying themselves as Black or African American. Smaller percentages of the devices were from visitors likely to identify themselves as Asian (four percent) or “other” (two percent). Totopotomoy Creek and Cold Harbor had a higher percentage of devices from people who would likely identify as White (87 percent and 86 percent, respectively) compared to other locations. Compared to most locations, Chickahominy Bluff had more (76 percent) device days (the daily proportion of devices detected near a location) from visitors who live in an area where more people identify as African American.
Divergent Results
There are substantial differences between the 2023 mobility data study results and the results of the 2010 park survey. The mobility results found that 80 percent of park visitors were local—from the city of Richmond and its surrounding counties. But the 2010 survey found only 33 percent of visitors—survey respondents—were local. The 2010 survey results also indicated more visitors coming from other states than the mobility data did.
Image credit: NPS / D. Pettebone
The 2023 mobility data showed a relatively uniform distribution of visitor ages, leaning somewhat toward younger visitors. In contrast, the 2010 survey results showed an older population of visitors. The 2023 mobility data estimated that 64 percent of visitors were 44 years and younger. But the 2010 survey found only 47 percent of visitors were 45 years and younger. In the 2010 study, a non-response bias analysis of the 2010 data found statistically significant differences in age between respondents (older) and non-respondents (younger). In other words, younger people didn’t complete the survey to the degree that older people did. This may have introduced a bias toward older visitors in the 2010 data.
Image credit: NPS / D. Pettebone
The mobility data suggested a more diverse visitor population than the 2010 survey data.
The mobility data suggested a more diverse visitor population than the 2010 survey data. The mobility data estimated that 37 percent of park visitors would likely identify themselves as African American. But the 2010 survey data reported only five percent of visitors identifying as African American. Both studies report much lower percentages of visitors identifying themselves as other races.
Image credit: NPS / D. Pettebone
Different Study Designs
The 2010 survey’s sampling design may have contributed to the differences between the 2010 results and the mobility study results for local versus external visitors and White versus African American or other visitors. In 2010, the park was telling the public to start their trips at the park’s visitor centers. This led park staff to direct the 2010 researchers to focus their sampling efforts there. They assumed that this would enable them to recruit a representative segment of visitors to participate in the survey. But this may have resulted in fewer local visitors represented, even with the high volume of visitors who pass through the visitor center. This is because the survey likely captured visitors who followed the park website’s recommendation to begin their trip at the visitor center, presumably something locals wouldn’t do.
Less use of the visitor centers due to the pandemic and other factors could also partly account for why the 2023 results were so different from the 2010 results.
We looked at the racial makeup of visitors, as indicated by the mobility study, in the park sites where the 2010 survey was distributed. The Cold Harbor and Tredegar visitor centers, where park staff distributed most of the 2010 survey questionnaires, had relatively low numbers of likely African American visitors according to our study. Less use of the visitor centers due to the pandemic and other, previously mentioned factors could also partly account for why the 2023 results were so different from the 2010 results, which concentrated on visitor center contacts.
Image credit: NPS / D. Pettebone
It’s also possible that the questionnaire-based survey sampling methods failed to capture a representative sample of visitors at these two geographically open and scattered parks, because they don’t have conventional entrance stations where staff can hand out questionnaires. This may have skewed the 2010 survey results toward visitors at visitor centers and other attractions where staff distributed the survey.
Image credit: NPS
Recent work from the Pew Research Center found that over 97 percent of Americans owned cellphones and 90 percent of American adults owned smartphones. There were few differences in cellphone and smartphone ownership among different racial groups. Black adults owned smartphones at a slightly lower proportion (84 percent) than other groups but owned cell phones at extremely high rates (96 percent), comparable to other racial groups. Different groups could have location information turned on/off at different rates, which could introduce bias into mobility data. Cell phone usage behavior is an important topic that merits further research to understand the representativeness of mobility data.
Nevertheless, we cannot conclude that the 2010 visitor survey’s sampling was biased toward certain visitors relative to the 2023 study. This is because the data from these two studies were collected 13 years apart under different circumstances using different study designs. The survey researchers and the National Park Service assumed the 2010 survey was based on a representative probability sample and representative of all the people who visited the pre-selected survey locations. But we cannot assume the 2023 mobility data are a representative sample of all the people visiting all the park areas. This is because the data were not randomly collected. For this reason, statistical comparisons between these two datasets would not be meaningful.
Scientific research using big mobility data is becoming more common, and researchers are identifying strengths and limitations to these data.
That being said, the 2023 mobility dataset is robust because of its geographic and temporal scale along with its large sample size. It provides data from over 400,000 individual devices over a 20-month period for all the park areas. The town of Richmond has good cell phone service infrastructure, ensuring clear cell coverage signals. Also, the park has obvious geographic boundaries within which to detect visitor use. Scientific research using big mobility data is becoming more common, and researchers are identifying strengths and limitations to these data, for example in monitoring compliance with social distancing recommendations during the pandemic. For these reasons, we believe the results of our mobile device study accurately represent visitor use and characteristics.
Capitalizing on Clever Cars
Richmond National Battlefield Park and Maggie L. Walker National Historic Site aren’t the first national parks to use mobility data. Social scientists have also done mobility data studies of visitors at Grand Teton National Park, Saguaro National Park, and Blue Ridge Parkway, among others. Jenn Newton, a social scientist at Grand Teton, said, “There are clear patterns of where people go, but also where they stop in many places.”
These data were really accurate because they showed exactly where the cars were.
The Grand Teton study used data from cars with special systems that connect to the internet. These data were really accurate because they showed exactly where the cars were and how the drivers behaved, like when they braked hard or speeded up. But Newton noted that these mobility data had limitations. First, not many cars have these systems, so the data came from a smaller group of vehicles. Second, only certain types of cars (ones with the systems) were included in the dataset, which could have led to biased results. Third, park staff needed to adjust or "calibrate" the data by comparing mobility data to count data that had been verified by direct observation. This was necessary to help them understand how many people were actually using the park and make sure the data were useful.
Image credit: NPS
A map showing Yellowstone National Park and Grand Teton National Park, as well as adjacent roads and towns. The map features a dark green arrow starting in Yellowstone and ending in Grand Teton at a label that reads “GRTE 56%.” This indicates that 56 percent of Wejo Trips entering from Yellowstone went into Grand Teton.
The map also has a light brown arrow that goes from Yellowstone to a label that reads “Town/County 28%,” outside of Grand Teton National Park. This indicates that 28 percent of Wejo Trips went to towns and counties outside the park.
A yellow arrow on the map goes from Yellowstone and turns east, pointing to a label that reads “US-287 11%.” This shows that 11 percent of Wejo trips went to this highway.
Four small, light yellow arrows point from Yellowstone to the following four labels (with percentages of Wejo trips in parentheses), in an anticlockwise direction: Airport (<1%), Teton Pass (3%), US-89 (2%), and US-191 (<1%).
The map legend is on the lower right of the map and has color-coded rectangles that correspond to percentage categories for trips entering from Yellowstone. The legend also has symbols for park boundary, state boundary, and highways. It indicates which direction is north and shows the scale of the map in miles.
Newton sees the potential for mobility data to improve the National Park Service’s understanding of how visitors’ trips fit together regionally. She thinks parks can use this knowledge to improve how they interact or talk with visitors. For example, she’d like to coordinate consistent messaging among parks to promote awareness and encourage safe visitor behavior around issues such as food storage and wildlife management. When asked how she would use mobility data in future studies, Newton said she hoped to answer questions about cross-boundary visitation between Grand Teton, Bridger Teton National Forest, and other locations in the area.
Meeting People Where They Are
The Richmond National Battlefield Park and Maggie L. Walker National Historic Site mobility study’s cross-visitation and visitor origin analyses gave park staff a sound, data-driven basis for making decisions about new programs, staffing, and facility improvements. The study results confirmed what staff already thought: people were using cultural spaces like battlefield sites as green spaces for recreation.
Our data can help determine where park rangers are needed; for example, in highly visited areas with no visitor centers. The data can also help parks plan programing on the importance of preserving battlefield landscapes to deter people from going off-trail. Interpretation chief Pooler said Richmond National Battlefield Park now highlights trails and approved, non-battlefield, recreational green spaces. It does this on its website and through messaging. It also encourages visitors to use the National Park Service mobile app.
“We can’t make decisions about visitor use without these types of data.”
Based partly on our mobility study results, Richmond National Battlefield Park closed the visitor contact station at the Tredagar Iron Works in 2024, deciding to concentrate outreach activities on the riverfront and other park sites. “This will allow park staff to connect better with visitors and focus on programming with the American Civil War Museum,” said Pooler. She also said relocating rangers to locations where visitors go will improve staff morale. She sees these changes as an opportunity to increase awareness about the park and the agency. “I would do [another mobility study] again in five years,” she added, “to see if we are meeting people in correct spots and times.”
Image credit: NPS
“These data allow us to see not just where people are going but what they are doing,” emphasized park superintendent Scott Teodorski. “It's important for us to meet people where they are. We can’t make decisions about visitor use without these types of data.” Our data helped staff at Richmond National Battlefield Park and Maggie L. Walker National Historic Site recognize blind spots and mistaken assumptions they had about visitor use. These affected how they operated and staffed the parks. Now that they know visitors are likely visiting field locations more often than visitor centers, the parks are taking steps toward their goal of better connecting with the local community.
Finding Answers to Difficult Questions
The National Park Service has worked to increase public awareness of public lands, especially among youth and historically under-represented groups of people. But is the agency reaching new and diverse audiences? Without research like the 2023 mobility study, that’s difficult to know. The agency relies on questionnaire-based visitor surveys to answer questions about visitors and their experiences. But this kind of survey data collection is difficult, expensive, and has some limitations when collecting representative data at park units with porous public access.
This is especially true for parks that are spread out across different buildings or locations, like Richmond National Battlefield Park and Maggie L. Walker National Historic Site. Our study shows the value of using data from mobile devices to analyze visitation in places like these, where it’s challenging to carry out conventional surveys. This new data source is not a substitute for visitor surveys but can provide important insights about visitor populations not previously available. More research will help us better understand the benefits and limitations—and exciting potential—of mobility data.
About the authors
