Accessibility cyberscapes and the digital divide

INTRODUCTION

Collaborative maps for rating the accessibility of specific points of interest (POIs) are growing in number. These maps collect ratings of how accessible certain POIs are for people with disabilities. For example, Wheelmap (2014) is a service focused on mapping wheelchair accessible places. These maps can be important vehicles for people to express their views on the real world accessibility they encounter on their daily travels and result in collections of accessibility data that can be leveraged as powerful assistive tools for accessible trip planning and everyday navigation. To date, little investigation of these maps has been conducted. This paper focuses on exploring if a relationship exists between the data collected in these maps and the digital divide.

The digital divide is a concept that describes a (virtual/physical) divide between those who have access to technology and those who do not have access. Generally conceived as access to communication networks such as the Internet, some scholars have identified relations between the mapping of spatial data and the digital divide along societal divisions such as race (Crutcher and Zook 2009) and religion (Shelton et al. 2014). Others (Dobransky and Hargittai 2006; Hollier 2007) posit the existence of a disability divide that relates to access to technologies/interfaces for people with low to no vision, hearing loss or limited motor control.

Further exploration of user generated accessibility datasets may shed light on characteristics of contributors of accessibility information and the coverage (in terms of both community and geographic area) of the footprints of each dataset. In this study, we begin to examine the distribution of user contributed accessibility data and try to identify factors contributing to the observed differences in the density of data across geographic areas. As an initial step, we looked at the availability of accessibility data from the angle of the digital divide. More specifically, we aim to answer the following question: to what extent can socioeconomic indicators, such as income, or location-based factors, such as urban or rural setting, explain differences in the distribution of user generated accessibility data collected in the US?

Answering this question is important because people with disabilities may benefit from data describing the environment. This kind of information may lead to an increase in independent mobility by allowing a traveler with a disability to plan a trip more thoroughly. On the other hand, if data collection is impacted by certain socioeconomic or location-based phenomena, certain areas of the map may remain empty thus limiting a traveler's options. Highlighting the existence of societal barriers to the collaborative mapping of accessibility information may extend existing studies of the disability divide which currently focus on accessible interfaces not content about accessibility.

The data we analyze are data points that are tagged with the ‘wheelchair’ key (tag) in OpenStreetMap (OSM). We count the number in each area of interest and then compare this frequency with data related to the digital divide such as median household income and geographic setting (urban or rural) collected from the US Census. We also compare the frequency with the population of people with disabilities in each area of interest. A linear regression is used to measure correlation between the constructs.

BACKGROUND

Cyberspace is a virtual “place” online. There are spatial connotations to the term cyberspace and it is used by many scholars to describe the online places in which people participate (Graham 2007). The concept of DigiPlace builds on the notion of cyberspace and describes the “place” in which cyberspace and actual place coincide (Zook and Graham 2007). An example of DigiPlace is Google's location search feature. Cyberscape is a concept used to describe the collaborative mapping of real places in cyberspace (Crutcher and Zook 2009); in other words, the collaborative creation of DigiPlaces.

The affordance to collaboratively map any location or any information about a location has impacted the professional mapping community (Elwood et al. 2012). Since the introduction of Google Earth in 2005, collaborative sites that support the creation of cyberscapes have proliferated. Goodchild (2007) calls the phenomenon volunteered geographic information (VGI) and others have termed the activity neogeography (Elwood et al. 2012). Since 2010, a handful of sites supporting the creation of cyberscapes of accessibility have emerged.

Access Together (2014) is a “tool for collecting, displaying, and acting on community accessibility data” for the United States and Canada. Access Together provides questionnaires regarding the accessibility of specific POIs for mobility impaired users, users with no or low vision, hearing impaired users, users with sensory sensitivity, and the elderly. AXSMap (2014) is a “user database” to “obtain or input information about physical accessibility of public places” in the United States. Created by an independent filmaker in 2010, AXSMap utilizes the GoogleMaps API to generate an editable map in which users can share star ratings and select the presence of certain features to describe accessibility such as “spacious” or “quiet”.

Wheelmap (2014) is a “map for wheelchair accessible locations” anywhere on the Earth. Created in 2010, Wheelmap is an accessibility project in the OpenStreetMap universe. Consequently, the OpenStreetMap API provides the basis for the interactive map and all annotations shared in Wheelmap are available in the planet.osm database. Wheelmap currently only supports wheelchair users, although they aspire to expand the system to other communities. The accessibility annotation is derived using concise statements in which all parameters should be met. For example, “yes = Entrance without steps, all rooms without steps, accessible toilet if customary in a place”.

These examples hint at a growing desire to utilize collaborative tools to benefit local communities. One issue that is important to explore is – what determines which areas of the map accumulate accessibility information and which do not? Researchers have used the concept of cyberscape to investigate the impact of racialized landscapes on the creation of a cyberscape of hurricane alerts during Hurricane Katrina (Crutcher and Zook 2009) and the distributions of various religions across the world (Shelton et al. 2012). Other studies have examined the influence of gender (Stephens 2013) and urban status of a mapped location (Baginski et al. 2014) on map content. In this work, we are interested in cyberscapes of accessibility.

STUDY DESIGN

This study focuses on collaboratively mapped accessibility information in the OSM project. This dataset is the largest available collaboratively created dataset of accessibility information available today. The dataset includes data collected via Wheelmap and data collected through OSM directly via iD or another editor. An extract for North America (for March 2014) was collected using the osmosis tool (OSM osmosis 2014) and filtered for nodes that included the ‘wheelchair’ key (e.g., wheelchair = yes, wheelchair = no, wheelchair:description = ‘free text', etc).

Hilbert (2011) acknowledges the digital divide as a “multidimensional and complex” challenge and provides evidence, from several Latin American countries, that the most explanatory attributes for understanding individuals and the digital divide are income and education. Crutcher and Zook (2009) incorporate the African American population, total population, median household income, population of renters and those over 65 years of age into a series of models in their study of racialized cyberscapes. Baginski et al. (2014) and Hilbert (2011) note the division between urban and rural areas as being a key indicator for the digital divide.

For this preliminary study, we evaluate the OSM wheelchair points against measures of median household income, geographic setting and the population of people with disabilities. Median household income is chosen as a stronger socioeconomic indicator than level of education. Geographic setting is chosen as the location-based factor. The population of people with disabilities is included because these communities are the primary constituents of accessibility datasets. Data describing income, urban area designation and population of people with disabilities was collected from the American Community Survey datasets available from the U.S. Census. The OSM wheelchair points were then examined at the census tract level and a linear regression was performed (Table 2).

PRELIMINARY ANALYSIS

11301 unique POIs in North America included at least one wheelchair key/value pair. Of these POIs, 676 have more than one wheelchair key/value pair. The two areas with the highest number of data points are Ontario, Canada and California, U.S. (Figure 1). We chose to focus on the U.S. in this study and focus on California. California includes 1683 wheelchair points concentrated in urban areas (Figure 2). The most densely packed census tract is near Santa Cruz, CA (Figure 3) and represents 32.7% of the wheelchair points in California (n=550).

Census tracts in California that included locations marked with one or more wheelchair keys in OSM have, on average, a lower overall population and a lower population of people with disabilities, however, the median household income is higher than the general trend for census tracts across California (Table 1). When compared to trends for all census tracts in California and those tracks with data points, the Santa Cruz tract has less people with disabilities and a lower median household income than the average but a much higher population overall. Closer inspection of the Santa Cruz census tract reveals that the area includes the University of California Santa Cruz campus. This may help explain the lower average income and higher than average population found in the Santa Cruz tract.

The variables shown in Table 1 are used in a regression analysis for census tracts in California that have wheelchair data points (Table 2). The final model, using all variables, showed significant reliability (F) in predictor variables, and several significant individual variables (t); however, the explanatory power of the model was considerably low. Income, geographic setting and the population of people with disabilities individually explain less than 1% each, and collectively, with the total population, they explain 2% of the frequency in the data. The results did not improve when all census tracts in California were used.

CONCLUSION & FUTURE DIRECTIONS

This study examined a set of data describing wheelchair accessibility in OpenStreetMap and its relation to the digital divide. The study sought to understand – to what extent can socioeconomic indicators, such as income, or location-based factors, such as urban or rural setting, explain differences in the distribution of user generated accessibility data collected in the US? We compared the distribution of user-generated accessibility data in California to data related to the digital divide (median household income and geographic setting) and the population of people with disabilities in each area of interest. Based on a linear regression, our chosen constructs explained a very small amount of the data distribution in our current dataset. A potential cause for this result is that we separate the data with accessibility tags from the OSM dataset as a whole in our analysis. By doing this, we may have missed patterns in the overall dataset. Additionally, the excessive data found on the University of Santa Cruz campus may indicate a stronger relationship between other constructs, such as the level of education, and amount of data on the map.

Our next step includes (1) extending the dataset to include all OSM data within a small area and (2) adding additional measures of the digital divide, such as level of education and availability of access points. Analysis of all the OSM data within a single map tile may aid our understanding of the digital divide and OSM in general, which will inform our specific interest in accessibility data within OSM and the digital divide. The additional measures of the digital divide – such as level of education, age, number of access points – may lead to an improved model with more explanatory power.