Accurately capturing exposure and access to different food and physical activity environments is of great interest to researchers engaged in identifying environmental correlates of dietary patterns, physical activity and obesity. In the past two decades GIS techniques have been increasingly used for these purposes . In order to understand environmental contexts of populations, researchers have often used “buffers” of a certain distance around key environments such as homes, schools, work sites, and parks .
Straight line buffers (also called circular, crow flies, or airline buffers) simply go out a certain distance in a straight line from the facility or place, creating a circle (if the item being buffered is a point). In contrast, network or street distance buffers try to more closely approximate the experience of moving around an area by measuring out a certain distance on the street network and then using some method of joining the dots. Network buffers have been made easy to use in the past decade and a half by algorithms in GIS programs; previous to the development of these algorithms, creating them manually was a tedious task. However, in the mid-2000s, Esri ArcGIS software abruptly changed its method for calculating such buffers making older analyses not directly comparable to recent assessments. Esri (originally Environmental Systems Research Institute) is the dominant developer of GIS software with a 30 % market share internationally in 2009, almost double that of its closest competitor, Integraph . This change, made by a popular GIS developer, highlights the proprietary nature of much GIS software and its design for professional rather than research contexts.
This paper explores an alternative to using proprietary algorithms—constructing a network buffer from scratch. It answers two questions. First, is it possible to create an approach that can be consistently and easily replicated across software versions and platforms (i.e. hardware and operating systems)? Second, does such an approach accurately measure where people can get to from a starting point and the environments they experience along the way? That is, does it measure relevant parts of the local environment?
In this paper, we describe how the sausage buffering approach was developed to be repeatable across platforms and places. The proposed sausage-shaped buffer, buffers all roads by a consistent dimension (the “radius”) out from the center line. We examine how this approach compares with existing alternatives in terms of (a) size and shape of buffers, (b) measurements of the food and physical activity environments, and (c) correlations between environmental features and health-related behaviors among adolescents.
The sausage buffering approach has three main strengths. First, it is theoretically defensible as it directly measures the environment near the streets or paths along which people travel. Second, it has much in common with other proprietary techniques meaning that it provides comparable data to measures already in existence. Third, it is easily reproducible across GIS platforms and program versions, meaning that it provides a stable and reliable measure that can be used in the future and by those using different GIS programs . In the obesity field, one other study has published such an approach, in a preliminary methods paper without results, called the “street buffer” . However, to the best of our knowledge, no previous studies have systematically compared the sausage-shaped buffer approach with other buffer types, particularly in a health context.
The analysis described here highlights three issues with using GIS in health research that frame the current paper: 1) reliability, 2) validity, and 3) utility across places and datasets. In regards to reliability, the more sophisticated, multifunctional, and widely-used GIS programs have been designed primarily for use in professional practice where the ability to replicate work is not as highly valued as it is in health research. Available documentation for many such GIS programs is not clear about their internal algorithms or even about the definition of terms (as will be noted below, ArcGIS uses some non-standard terminology that it does not fully define in its documentation).
Thus, research users of GIS need to figure out how to create reliable methods that can be used across platforms, both commercial and open source . In terms of food and physical activity environments, most discussions of reliability have focused on survey and audit tools where inter-rater reliability, test-retest reliability, and item consistency are most important . Reliability of GIS buffers instead relates to the repeatability of the measure across programs and platforms. Such repeatability is particularly important in international comparative work. For example, while a few software companies dominate the global market, there is regional variation in market share. Repeatability across platforms is a fundamental prerequisite for comparison. It is this issue of repeatability that led us to develop the sausage buffer, i.e., for the purpose of conducting studies of food and physical activity environments and allowing comparison across time and place.
The second issue is validity. Buffers approximate the environments experienced by populations. They are a great improvement over measures using pre-existing geographies, such as census tracts, to define “neighborhoods” or local areas in that they are centered on the participant. Network buffers go one step further than straight line buffers and approximate the extent of the environment that is experienced by people moving along streets. But it is important to create buffers that closely match the geography of the local food or physical activity environment as it is experienced or perceived by study participants. What do people see, smell, and hear as they walk, cycle, or drive along streets? How far from the street centerline does this experienced environment extend? Does the measurement geography (e.g. the buffer) approximate experience in a way that is relevant for health behaviors? Research on how people perceive their neighborhoods reveals that such perceived neighborhoods vary greatly in size and shape . However, the sausage buffer has face validity in terms of assessing local environments relevant for health behaviors. Like other buffer types, sausage buffers can be centered on such important places as homes, schools, or work sites and capture nearby environments that provide settings or contexts for health-related choices (e.g. local access to green space or healthy food options).
Finally, the third issue relates to what Lytle , in reviewing survey and audit tools, calls “other related measurement qualities.” Two examples given in the review completed by Lytle are utility of a tool across (a) populations and (b) health concerns. In the case of GIS and buffering, other relevant issues would be utility across (c) places and (d) datasets. It is important to examine whether a measure makes sense in different kinds of physical environments and with different kinds of street and path network data, such as might be available in different parts of a country or across countries. Again, the sausage buffer should perform as well as other buffering methods, and because of its simplicity it may do better than some. That is, a sausage buffer approach appears to have no disadvantages as compared to other methods and it has strengths in repeatability.
To date, little research in the area of food, physical activity, and obesity has looked explicitly at buffering. Only a few papers have reported results for both straight line and network buffers (e.g. [8, 9]). Oliver et al.  explicitly compared straight line and network buffers in terms of associations with walking for leisure and for errands, finding the network buffer to be better than the circular buffer for this purpose. Burton et al.  published a paper on the methods of the HABITAT longitudinal study of change in physical activity in middle-aged adults in Brisbane and included a diagram of three buffer types: circular, network, and street (the last akin to the sausage buffer). They used MapInfo, another major GIS program, to create the buffers. The current paper builds on this previous research by providing a more substantial discussion of buffering in regards to studies of population health.