«Urban Problems and sPatial methods VolUme 17, nUmber 1 • 2015 U.S. Department of Housing and Urban Development | Office of Policy Development and ...»
Critical GIS and Mixed-Methods Research Several studies provide models for how to incorporate daily practices into geospatial analysis, many of them falling under the broad heading of critical GIS (O’Sullivan, 2006; Sheppard, 2005). Kwan (2008), for example, used travel logs and interviews with Muslim women soon after the terrorist attacks on September 11, 2001, to map how formerly routine daily trips to school and work became significantly shortened and filled with anxiety. Both Rogalsky (2010) and Matthews, Detwiler, and Burton (2005) used tracking and interview data to map the daily trips of welfare clients, showing how family commitments, shopping needs, and institutional demands meant regular longdistance trips, often using public transit, at a significant cost in both time and money. Knigge and Cope (2006) used grounded visualization, combining analysis of demographic data and participant observation within neighborhoods, to analyze the political battles over vacant lot space in Buffalo, New York. Critical GIS research is often more participatory in nature, prioritizing situated accounts over a supposedly objective and expansive analytical view (Pavlovskaya and St. Martin, 2007).
One primary contribution of critical GIS has been its mixed-methods approach. While some of these projects repurposed geospatial technologies for qualitative research (Cope and Elwood, 2009), others combined both quantitative and qualitative components in ways that preserve their respective strengths—breadth of view and analytical clarity in the case of quantitative work and the interpretative richness and nuance of qualitative approaches. Here, the use of mixed-methods provides a complex view of a world that is always just beyond our epistemological grasp (Elwood, 2009; Nightingale, 2003). Drawing on this work and combining quantitative and qualitative data on food-provisioning practices in complementary ways, the methodology described in the remainder of this article provides a richer understanding of the factors shaping food access at the household and neighborhood levels.
Study Context and Sampling This research project, conducted in the Twin Cities (Minneapolis/St. Paul), Minnesota, was composed of two main sections. The first section used dasymetric mapping to analyze ZIP Code-level data on the Supplemental Nutrition Assistance Program (SNAP, formerly known as food stamps).
I drew on these data to create disaggregated estimates of both client locations and benefit redemptions at SNAP-accepting retailers in neighborhoods with the highest concentration of SNAP clients.
Two main findings emerged from this analysis. First, even in neighborhoods with a supermarket, a net “outflow” of SNAP dollars was evident, meaning that clients often traveled out of these areas to use program benefits. Second, midsized grocers (for example, discounters, ethnic retailers) accounted for a much larger proportion of benefit redemptions (31 percent of the total) within these neighborhoods than they did outside them (4 percent). More information on this section of this research is available in a related publication (Shannon, 2014).
The second section, which this article details, used case studies in two low-income neighborhoods in Minneapolis, using data on daily mobility, photographs of foods and stores used by study participants, and semistructured interviews. These two neighborhoods, north and south Minneapolis (exhibit 1), shared a high density of SNAP clients but differed demographically in significant ways (exhibit 2). North Minneapolis residents are largely White or African-American, with a smaller population of immigrant Hmong families from Southeast Asia. South Minneapolis also has
large White and African-American populations, along with immigrant households from South and Central America and East Africa. At the time of the study, south Minneapolis also had six times as many midsized to large food retailers (43 versus 7) as north Minneapolis. Although these two neighborhoods provided contrasting cultural and commercial landscapes, they shared high levels of economic hardship.
I used a quota sampling method in each neighborhood, recruiting roughly equal numbers of White, African-American, and immigrant populations (Hispanic populations in south Minneapolis and Hmong populations in north Minneapolis). A summary of study participants (N = 38) is provided in exhibit 2. These participants were recruited primarily through posting flyers in public spaces (for example, libraries, sign posts, neighborhood centers) along with advertisements on the online classified system Craigslist. In a few cases, participants heard of the study through word of mouth. Participants received a gift card in return for their completion in the study.
Study Methods My study methods collected three broad forms of data from participants: (1) GPS tracks of daily mobility, (2) written and photographic diaries of food procurement, and (3) semistructured interviews about their activities during the 5-day study period. This length of time (2 weekend days and 3 weekdays) provided enough food-related trips for productive interview conversations without making the interviews overly burdensome. Many participants described their food shopping as a monthly pattern oriented around receipt of SNAP benefits, so while I considered study periods of up to 2 weeks, even these might not have been a fully representative sample. To investigate other possible food sources, I asked participants to describe any other food sources they used regularly and their reasons for doing so.
Study participants used Android smartphones to collect data on daily mobility and to take photographs of their food and food sources. The phone I chose, the LG Optimus T, had already been on the market for more than a year, reducing its price, but it had the needed hardware specifications.
I registered these phones on a daily use plan with a major U.S. provider, meaning that a small fee was charged only on days the phones were in operation and that the phones would have unlimited service on those days.
The main function of these phones was collecting GPS data on daily mobility. It was difficult to find a suitable GPS-tracking application for Android. I used three applications during the course of the study, because my first choice was discontinued 1 month into the trial and the second option was unstable on the project phones. The final solution was FollowMee (https://www.followmee.com), a third party application that proved to be the most reliable and that provided data in a spreadsheet format easily transferred into GIS software. I set the application to record locations every 5 minutes, which allowed the phones to last a day on a single charge. This approach provided sufficient data to identify the general neighborhoods where individuals spent their time. Although GPS trackers provide greater temporal and spatial accuracy than phones, they add expense and require study participants to carry an additional device. The additional accuracy was also not necessary in this case. To protect privacy, I added noise to these GPS data, random numbers for both latitude and longitude that fell within a range of +/-167 meters (0.0015 decimal degrees). Visual tests showed that this approach made determining the location of home or workplaces significantly more difficult. I shared maps that demonstrate this additional uncertainty with participants in discussions of study risk before enrollment. For my final interviews for participants, I created and discussed a map of each participant’s daily trips, using these data (see discussion in the following section).
Participants also used these smartphones to take pictures of food sources they used and the foods they procured during the study period. These images provided a ground-level view of these sites and created a visual link between food sources and the varieties of foods people purchased and gathered. Participants’ food sources included workplace kitchens and friends’ homes, along with supermarkets and restaurants (exhibit 3). Each phone automatically synced photos to an online storage service, enabling me to check the quality of images during the study period. Because smartphones automatically georeference photos, these photos could be placed on a map using software such as Google Earth. Although most participants had no trouble using the phone’s camera, images did sometimes suffer from poor quality, primarily blurriness or low light for outside shots.
Exhibit 3 Participants’ Photos of Food Sources and Purchased Foods In addition to taking photographs, participants kept a written record of any food source used during the study period. This record included a shopping diary that listed store names and locations;
information about when, how, and with whom they visited the stores; and how much money they spent. Although participants also had the option of using their smartphone to complete these diary records via an online form, all but one participant preferred the paper version. Participants also saved receipts from any food purchase made during the study period. The receipts verified store locations. Except in the case of sit-down and fast-food restaurants, the receipts were coded based on type of food purchased (for example, dairy, meats, dry goods, and produce). This approach provided detailed data for analyzing food-shopping patterns; I could summarize trip characteristics based on which products were purchased to determine food-shopping patterns.
Two interviews framed participation in the study. An initial interview, which lasted 20 to 30 minutes, included collecting participants’ background information and providing them instructions for their role in the study. A second semistructured interview after the study period lasted 40 to 75 minutes. This interview had three sections. First, we examined the map of GPS data, and participants identified the main locations where they had spent time, talking about their daily routines.
Second, participants described each food source they used, talking about their reasons for using the place and their impressions of it. Third, we spoke more generally about other food sources they did or did not use on an ongoing basis and the ways they thought their food options could be improved.2 These interviews provided essential insights into factors shaping how and where these participants got food. For example, one woman traveled far across town to go grocery shopping rather than use the supermarket in her neighborhood, largely because of her limited walking ability and the direct bus route to this preferred store. Another woman gathered a group of friends for an early morning trip to a suburban Wal-Mart store to take advantage of their once-a-week meat specials. The details of these trips would have been difficult to discern using only GPS data.
A full version of this interview protocol is available at https://www.scribd.com/doc/257040526/Closing-Interview-Schedule.
In keeping with previous research using active-interviewing strategies, the goal of both interviews was to develop a shared understanding of the factors shaping participants’ food-provisioning strategies (Holstein and Gubrium, 1995). Indeed, when I asked participants for their thoughts about the study, the most common response by far was how much they had learned about their own foodprovisioning habits through their participation. These interviews were coded inductively using the qualitative software NVivo. Codes were based on my research questions, including themes such as perceptions of distance, store quality, and quality of foods within different store types.
After an initial analysis of these data, I invited participants to a followup focus group in which they could respond to my initial conclusions. One focus group was held for each study area, and participants were offered a free meal for their participation. My initial results largely focused on the notable variety of stores participants used and the high number of trips they made to stores outside their neighborhoods. To me, these results demonstrated a significant degree of individual mobility, a finding that ran counter to existing research that focused only on the neighborhood environment.
Participants pushed back against this interpretation, however, particularly in the north Minneapolis group. To them, their shopping patterns were direct effects of the high prices and low quality of foods in their local stores. Comparing their neighborhood stores to suburban locations, where prices and quality both were more favorable, several wanted to know how my research would improve what they saw as a clear injustice. These focus groups shaped my subsequent reporting on this research, which ultimately focused on the need to situate research and policy on food access within a broader framework of neighborhood disorder and segregation and on the need to strengthen transit systems.
Lessons Learned and Recommendations GIS-based approaches to studying food accessibility document significant disparities in the food sources available to urban residents. Relying on measures of spatial proximity, however, fails to incorporate other important factors that shape food provisioning, including perceptions of the neighborhood environment and local stores and the resident’s daily mobility. The alternative approach used in this study addresses this issue by using GIS as part of a mixed-methods approach. GIS and qualitative techniques were complementary in this case. GPS data on daily mobility and shopping diaries provided a full picture of participants’ activities during the study period, providing ways to assess their daily activity space and visualize their food provisioning. Interviews and photographs contextualized this mobility, providing an “on the ground” perspective of the food sources participants used and their reasons for doing so.