Health & Place 24 (2013) 131–139 Contents lists available at ScienceDirect Health & Place journal homepage: www.elsevier.com/locate/healthplace Food mirages: Geographic and economic barriers to healthful food access in Portland, Oregon Betsy Breyer n, Adriana Voss-Andreae Department of Geography, PO Box 751-GEOG, Portland State University, Portland, OR 97207-0751, USA. art ic l e i nf o a b s t r a c t Article history: This paper investigated the role of grocery store prices in structuring food access for low-income Received 21 December 2012 households in Portland, Oregon. We conducted a detailed healthful foods market basket survey and Received in revised form developed an index of store cost based on the USDA Thrifty Food Plan. Using this index, we estimated 17 July 2013 the difference in street-network distance between the nearest low-cost grocery store and the nearest Accepted 28 July 2013 Available online 26 August 2013 grocery store irrespective of cost. Spatial regression of this metric in relation to income, poverty, and gentrification at the census tract scale lead to a new theory regarding food access in the urban landscape. Keywords: Food deserts are sparse in Portland, but food mirages are abundant, particularly in gentrifying areas Food access where poverty remains high. In a food mirage, grocery stores are plentiful but prices are beyond the Food prices means of low-income households, making them functionally equivalent to food deserts in that a long Spatial analysis journey to obtain affordable, nutritious food is required in either case. Results suggested that evaluation Portland Oregon of food environments should, at a minimum, consider both proximity and price in assessing healthy food access for low-income households. & 2013 Elsevier Ltd. All rights reserved. 1. Introduction ‘food mirage,’ rather than a food desert, and argue that the potential impact on health is similar—managing the challenges Much of the recent discussion around equitable food access of time, distance and cost means infrequent shopping trips and in North American cities has centered on food deserts, or areas less fresh produce. (Everett, 2011, p. 14) lacking physical access to full-service grocery stores. This paper argues that the food environment in Portland, Oregon is marked not only by food deserts but also by food mirages. In a food mirage, Taking Everett’s anecdotal evidence as a point of departure, this full service grocery stores appear plentiful but, because food prices paper identified food mirages with spatial analysis of food price data are high, healthful foods are economically inaccessible for low- from a 2011 market basket survey of healthful food in the City of income households. Food mirages are invisible using conventional Portland. We assessed store affordability for low-income households approaches to food desert identification, but affect food access for in relation to benchmark prices informed by the U.S. Department of low-income households similarly—a long journey to obtain afford- Agriculture (USDA) Thrifty Food Plan (TFP). Geographic information able, nutritious food is required either way. systems (GIS) and spatial regression were used to investigate relation- We borrow the term food mirage from Short et al. (2007), who ships among store locations, affordability, and socioeconomic vari- critique the assumption that food access arises from proximity to a ables at the census tract scale. Throughout this paper, we refer to full-service national chain grocery store. Everett (2011) also uses “food mirages” as census tracts where food access limitations stem the term in her account of a public health coalition that aimed to from a lack of affordable, healthful options rather than an absence of encourage healthier eating habits among low-income Hispanic grocery stores. Results confirmed previous findings that convention- households in North Portland. Everett reports: ally defined food deserts are rare in Portland (Sparks et al., 2010; Leete et al., 2012). Food mirages, by contrast, cover much of the city, …although North Portland has several large grocery stores, with the most extreme cases coinciding with gentrifying areas with Hispanic parents find them expensive and lacking in culturally relatively high rates of household poverty. We argue that the barriers appropriate foods, and therefore travel long distances to shop to healthful food access in Portland arise primarily from demand-side in discount supermarkets. We describe this phenomenon as a considerations – incomes, food budgets, and the spatial patterning of food prices – rather than solely the locations of full-service grocery n stores and areas of concentrated social deprivation. The paper Corresponding author. Tel.: þ 1 213 446 1650; fax: þ1 503 725 3166. E-mail addresses: betsybreyer@gmail.com (B. Breyer), concludes by connecting food mirages to inner city gentrification vossandreae@gmail.com (A. Voss-Andreae). and possible emergence of suburban food deserts. 1353-8292/$ - see front matter & 2013 Elsevier Ltd. All rights reserved. http://dx.doi.org/10.1016/j.healthplace.2013.07.008 132 B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 1.1. Food deserts risen while the price of energy-dense food (including soda) has fallen (Wendt and Todd, 2011). Basic principles of microeconomics suggest At its core, the food desert metaphor posits a linkage between that consumption of some commodity will be negatively correlated poverty, access to nutritious and affordable food, and poor diet-related with its price. An econometric meta-analysis found that a 10% increase health outcomes (Beaulac et al., 2009; Larson et al., 2009; Charreire in the price of soft drinks would reduce consumption by 8% to 10%, et al., 2010; Walker et al., 2010). Formulated generally, the concept has while a 10% reduction in the price of vegetables would increase broad explanatory power. Socioeconomic status has been identified as consumption 5.8% to 7% (Andreyeva et al., 2010). Price reductions, for a key predictor of diet across developed countries (Darmon and example through discounts, vouchers or coupons, have been shown Drewnowski, 2008; Monsivais and Drewnowski, 2009). Fruit and to increase consumption of fruits and vegetables (Glanz and Yaroch, vegetable consumption tends to vary inversely with distance to a 2004). The Special Supplemental Nutrition Program for Women, grocery store (Rose and Richards, 2004; Zenk et al., 2008), although at Infants and Children (WIC) program’s cash value vouchers, which least one study found this not to be the case (Boone-Heinonen et al., provides monthly supplementary checks to low-income women for 2011). Statistically significant correlations between socioeconomic fruit and vegetable purchases, is another type of economic subsidy status and the local food environment have been found in numerous that has been shown to increase the quantity of fresh fruit and case studies, particularly in Midwestern and Eastern U.S. cities (Alwitt vegetable purchases (Gleason and Pooler, 2011; Herman et al., 2006). and Donley, 1997; Moore and Diez Roux, 2006; Franco et al., 2008; Regionally, food prices co-vary with obesity rates (Todd et al., 2011). Gordon et al., 2011). Lower fruit and vegetable prices are associated with lower body As the food desert concept has become relevant to politics and weight among low-income populations (Powell and Chaloupka, 2009; policy (Ver Ploeg et al., 2009), this general formulation has been Powell et al., 2013) and larger food budgets are associated with higher operationalized for GIS analyses based on the following assump- nutrient intake (Monsivais et al., 2011; Aggarwal et al., 2012). Regions tions. First, full-service grocery stores, usually national chain with lower prices for dark green vegetables and milk tend to have stores, have been assumed to proxy for the presence of nutritious, relatively low child body-mass index (BMI); low-income households affordable food. Second, households have been assumed to buy exhibit the highest price sensitivity for these items (Wendt and Todd, food from the nearest retailer. Third, food deserts are assumed to 2011). Adolescents in the highest quintile for BMI tend to come from exist only in areas of concentrated poverty (Leete et al., 2012). lower-income households who, again, exhibit the highest sensitivity Jiao et al. (2012) also note that the mode of transportation is assumed to food prices (Auld and Powell, 2009). These findings suggest a to be the same for all residents. We refer to these assumptions as the prominent role for food prices in understanding any shifts in food conventional approach to food desert identification. consumption patterns and health outcomes, particularly among low- These assumptions allow food environments to be reduced to income households facing very tight budget constraints. quantities and scrutinized from afar. However, evidence to support their validity is mixed. On the first point, Chung and Myers (1999) and 1.3. Study area: Portland, Oregon Hubley (2011) found that chain stores offer lower prices than inde- pendently owned stores, while Cummins and Macintyre, 2002a; Conventional approaches have not found more than a few Cassady et al., 2007 and Short et al., 2007 obtained the inverse result; census tracts in Portland that satisfy the criteria for food deserts this assumption may be place-specific. The second assumption may (Margheim, 2007; Sparks et al., 2010). Many tracts along the underestimate the importance of variability among food retailers. suburban periphery are more than 1 mile from a grocery store, Store type may be an important parameter of overall food access but do not constitute food deserts because they lack sufficient (Block and Kouba, 2006; Drewnowski et al., 2012). Empirical evidence concentrations of households experiencing poverty or social has indicated that, in some U.S. cities, low-income households do not depravation (Leete et al., 2012). Redlining and disinvestment once necessarily shop at their neighborhood grocery store or even the marked closer-in neighborhoods, particularly those in North and nearest chain store because product mixes and price points vary Northeast Portland–they may have been food deserts in the late (Drewnowski et al., 2010; Hillier et al., 2011). Nationally, Supplemental 20th century (Gibson, 2007). However, these areas are rapidly Nutrition Assistance Program (SNAP) recipients live, on average, gentrifying and are largely well served by grocery stores. Citywide, 1.8 miles from a full-service grocery store but travel a mean distance lower-income areas tend to be associated with shorter distances to of 4.9 miles from home to shop for food (Ver Ploeg et al., 2009). On the the nearest grocery store (Armstrong et al., 2009). third point, barriers to healthful food access may not coincide with Despite relatively equitable spatial access to grocery stores, areas with concentrated poverty. Multiple researchers have failed to over half of Multnomah County’s adult residents are considered find food deserts in areas of social deprivation (Apparicio et al., 2007; overweight or obese, signaling the operation of some other path- Sharkey and Horel, 2008; Lee, 2012). Simultaneously, areas of con- way to poor diet-related health outcomes (Multnomah County centrated poverty have become more diffuse, even suburbanized, with Health Department, 2008). Community food assessments have recent demographic shifts (Richardson et al., 2012). These findings revealed that low-income households experience food insecurity suggest that other pathways to poor diet-related health outcomes, while living within walking distance of a full-service grocery store apart from spatial proximity to any full-service grocery store. One such (Interfaith Food and Farms Partnership, 2007). Anecdotal evidence pathway may be food prices. exists for low-income households spurning their neighborhood grocery stores and travelling long distances to shop at discount food retailers on the urban periphery (Casey, 2008; Everett, 2011). 1.2. Food prices The largest municipality within a rapidly growing metropolitan region, the City of Portland is located near the confluence of the A growing body of economics and public health research links food Willamette and Columbia Rivers. The city has a population of over prices to consumption patterns and diet-related health outcomes. 580,000 (Population Research Center, 2012) and a notable food On the basis of price per calorie, fresh produce is considerably more culture (Asimov, 2007). The Westside (west of the Willamette costly than energy-dense processed foods (Drewnowski, 2004; Rehm River) contains the downtown district and the affluent West Hills et al., 2011).1 Moreover, since 1980, the real price of fresh produce has area. The Eastside (east of the Willamette River), Interstate 205 (I-205) freeway serves as an important socioeconomic boundary, 1 However, other measures (price-per-portion) yield more comparable results with housing density and income tending to be lower directly east (Carlson and Frazão, 2012). of I-205. The study area outlined in Fig. 1 comprises 140 census  B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 133 imputed to other stores in the same chain found within the urban growth boundary (n¼74). 2.1.2. Market basket design The healthy foods market basket survey was designed in relation to the food budget embedded in the Thrifty Food Plan (TFP), which is used to determine eligibility for SNAP. The TFP reflects the prices and quantities that constitute a food budget representing ⅓ of the net income for a family at or below 130% of the poverty level (Carlson et al., 2007). Survey design excluded most highly processed foods listed in the TFP, focusing instead on produce, legumes, and whole grains. For example, whole or minimally processed vegetables and fruit items made up 55% of the healthy market basket survey, reflecting the USDA nutrition guidelines. The survey contained a total of 43 items in eight food categories: fresh produce, frozen and canned vegetables, grains, legumes, oils, meat, and dairy (Grocery CART PDX, 2012). 2.1.3. Survey methodology In designing the survey, we attempted to model the purchasing behavior of a health-conscious and price-sensitive food shopper. Rather than comparing prices of healthful versus unhealthful foods, Fig. 1. Study area, defined by census tracts with centroids inside the City of Portland. Census tracts with limited residential land use were removed. Inset maps we assumed a preference for healthful foods and measured how indicate that Portland comprises the large central area in the Portland Metropolitan far one must travel to obtain them, depending on budget constraints. region (defined by an urban growth boundary) and show the location of Portland This methodology aimed to identify the lowest-cost, reasonably metropolitan region in the State of Oregon. healthful food available at each store. For any survey item, a store typically offered a range of options. Prices were gathered based on items with the lowest price per unit. Some variation in quantity and tracts (of 397 tracts in the Portland metropolitan region) with quality were allowed.2 All prices were converted to comparable units. centroids in the City of Portland boundary, excluding three census No data was collected regarding food brand, food quality, or other tracts dominated by industrial land use (less than 10% residential). non-price attributes that may factor into consumer decisions (local, Three socioeconomically distinct regions were defined within the organic, etc.). Thus, our methodology differed from other market study area: the Westside (west of the Willamette River), close-in basket surveys with rigid guidelines for brand or quantity (Chung and Eastside (east of Willamette River, west of the I-205 corridor), and Myers, 1999). Prices were collected in 2011 by students at Portland East Portland (east of the I-205 corridor). Community College and Portland State University as well as by community volunteers. 2. Methods 2.2. Defining and mapping food mirages 2.1. Data collection 2.2.1. Benchmark prices 2.1.1. Identifying grocery stores We defined a set of benchmark prices to compare price levels We built a spatial dataset of grocery stores by cross-referencing across stores. Benchmark prices are taken as the average price for each geocoded store addresses identified through two sources: U.S. Census survey item at Fred Meyer, a subsidiary of the Kroger national chain, North American Industry Classification System (NAICS) and SNAP and Winco, a Western U.S. regional discounter. These stores were retailers. The addresses of all ‘Supermarkets and other grocery (except selected because they had all survey items available and the average of convenience) stores’ (code¼445110) from the NAICS were obtained store prices was comparable to TFP assumptions for produce. Shopping from SimplyMap and geocoded in ArcGIS 10.0. This particular NAICS the TFP at this price level would result in a food budget comprising code tends to incompletely capture full-service grocery stores (Forsyth roughly 30% of income for households at 130% of the poverty line et al., 2010). For example, Fred Meyer, a Kroger subsidiary that (Voss-Andreae, 2011). Both retailers tended to have lower market functions as a grocery store throughout the Portland metropolitan basket costs than all other chain retailers surveyed, including other area, is coded as a department store in NAICS. To ensure completeness, national discounters such as Wal-Mart and Target. we cross-referenced the resulting dataset with SNAP retailers, then Benchmark prices served as the basis for the affordability index manually edited based on local knowledge, Google StreetView, inter- (AI), a calculated value that reflected store affordability for low- net research, and site visits. income households (at or below 130% of the poverty line). Since this paper concerns access to healthful foods, particularly Specifically, AI functions as a multiplier indicating the increased fresh produce, we defined a grocery store as any food retailer with a cost of a basket of goods relative to benchmark prices. It is defined minimum of 10 fresh produce items available. This definition included in (1) as the ratio of observed market basket cost to benchmark small and independent grocers, who can serve as key food access basket cost: points in areas not served by chain stores (Bodor et al., 2008; Short ∑nj¼ 1 X ij et al., 2007) but excluded discount retailers that may sell non- AIi ¼ ð1Þ ∑nj¼ 1 Bi perishable food items but not dairy or produce. Following refinement by these criteria, the final dataset contained a total of 79 grocery stores lying inside of the study area boundary or within one kilometer of it. 2 For example, the survey calls for a 16 oz. bottle of vegetable oil, but if a 32 oz. All of these stores were surveyed. To avoid edge effects, another 51 bottle was less expensive per-unit, the price of a 32 oz. bottle was collected and the chain stores outside the study area were surveyed and price level was price halved. 134 B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 where Xij is the sum of costs for item j at store i and Bj is the 2.3.2. Spatial autocorrelation benchmark cost for item j. Any store with AI ¼1 has prices Spatial regression techniques were used to identify relationships comparable to benchmark. AI ¼2 indicates that the same basket with socioeconomic variables while accounting for spatial autocorre- of goods would cost twice as much as the same basket at bench- lation to avoid spurious correlations that arise from spatial depen- mark prices. We assumed that only stores with AI r1 are afford- dencies among adjacent geographic units (Rogerson, 2010). Robust able for low-income households, in the sense that SNAP recipients Lagrange Multiplier diagnostics from ordinary least squares regres- on a TFP budget would be able to keep costs of a healthful diet sions, conducted in OpenGeoDa v.1.0.1 (Anselin et al., 2006), indicated within 30% of income. Stores with AI r 1 were considered low-cost that a spatial lag model was appropriate. Spatial lag regression takes stores. Similar to Jiao et al. (2012), we determined that grocery the form of: stores with AI 4 1.4 are high-cost stores. Y i ¼ β0 þ β1 X i þ ρWY j þ εi ð3Þ where Yi the food environment variable for tract i, Xi is the socio- 2.2.2. Deriving ‘potential’ food mirages economic predictor for tract i, βi is the regression coefficient, ρ is a Apparicio et al. (2007) have argued that food access cannot be spatial autoregressive coefficient, WYj is the spatially lagged food described in a single measure because distance, density, and variety all environment variable averaged over five nearest neighbors, and εi is factor into a comprehensive view of the food environment. However, a random error term. Spatial regressions were conducted for two Sparks et al. (2010) and Leete et al. (2012) found reasonable agreement dependent variables: distance to the nearest grocery store and between these food access measures in the case of Portland, so we potential food mirage distance. To determine whether the influence looked at access only in terms of distance. We measured the street- of predictor variables differed across the study area, we conducted network (rather than Euclidean) walking distance from census block stratified analysis in three regions: Westside, close-in Eastside, and centroids (n¼7376) to the nearest store in ArcGIS 10.0. We repeated East Portland (see Fig. 1 and Section 1.3) East Portland. the process for only those stores with AIr1. We took the difference between these store distances as the food mirage value for each block. Blocks were chosen because they are the finest scale available from 3. Results the U.S. Census, with an average of 52 blocks nested within each census tract. Eq. (2) was then used to obtain Di, a census tract-level 3.1. Store distance and food affordability estimate of potential food mirage distance from the population- weighted mean block-level distances: Grocery stores were fairly evenly distributed throughout the study ∑ p ðmin jdsb Þ area (Fig. 2a). The average street-network distance to the nearest Di ¼ b A i b ð2Þ ∑ b A i pb grocery store was 0.7 mile (1.1 km). Without considering prices, food access would appear best near the city center, where store distances where pb is the block-level population and dsb is the distance to are less than 0.25 mile (0.4 km), and worst along the peripheries of the nearest store for block b in tract i. We considered Di to be ‘potential’ study area, where store distances exceed 2 mile (3.2 km). When food mirages because the presence of an actual economic barrier grocery stores are classified by price level, however, a different spatial arises only where large food mirage distances coincide with low distribution is evident. Most low-cost stores were clustered in the incomes. vicinity of I-205. As a result, the average street-network distance to the nearest low-cost grocery store is 2.5 mile (4 km). Based on tract-level 2.2.3. Identifying actual food mirages population and poverty rates, we estimated that 81% of all people in To identify actual food mirages, we eliminated census tracts poverty in Portland reside in census tracts that are more than 1 mile with a low-cost store within 1 mile (1.6 km), since no economic from a low-cost store, representing 13% of the total population. barrier was present. We also removed tracts with annual incomes In contrast, 20% of the low-income population resides in census tracts above the 75th percentile ($60,170), as these residents can that are more than 1 mile from any type of grocery store, representing presumably afford to drive to any grocery store. 95% of households 3% of the population. A negative correlation between AI and proximity in poverty reside in the remaining 85 census tracts, from 140 total to the nearest affordable store was found, such that store distance tracts. Within this subset, we distinguished between three types of increases nonlinearly as AI falls to 1 (Fig. 2b). food access situations that differ based on the distance to the nearest store: o0.5 mile (extreme food mirage), between 0.5 and 3.2. A typology of food mirages 1 mile (moderate food mirage), and 4 1 mile (akin to a low- density food desert or, more specifically, a food ‘hinterland’ as The techniques in Section 2.2.2 were used to derive the described by Leete et al. (2012)). population-weighted distance to nearest grocery, nearest low- cost store, and food mirage distance by census tract (Fig. 3a–c). 2.3. Spatial regression Although few areas of the city were more than 1 mile from a grocery store, much of the City of Portland is not served by a low- 2.3.1. Model development cost store. As a result, much of the city is a potential food mirage We developed a series of regression models to explore correlations for households facing tight budget constraints. between the food environment and socioeconomic context at the Fig. 4 used the techniques described in Section 2.2.3 to identify census tract scale. Using stepwise linear regression, we selected the and map actual food mirages. Extreme food mirages, symbolized following socioeconomic predictor variables: median family income, by the smallest circles, tended to appear in the close-in Eastside. Food poverty rate, and % change in white population from 2000 to 2010. hinterlands, symbolized by the largest circles, occured mainly in Note that the poverty rate in Portland does not always co-vary with census tracts east of I-205. Summary statistics in Table 1 indicated median income (correlation¼0.37) due to affordable housing require- that extreme food mirages tend to have the lowest incomes and ments. These data were derived from the 2006–2010 American highest rates of poverty, as well as the largest increases in white Community Survey. We take % change in white population between population (mean increase 5.08%). By contrast, food hinterlands were 2000 and 2010 U.S. Census as a proxy for gentrification, which has associated with higher mean incomes and diversifying populations marked Portland’s cultural landscape, particularly in areas of the close- (mean decrease in white population 3.3%), although poverty rates in Eastside (Gibson, 2007). were comparable to extreme food mirages. Of the 81% of people in  B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 135 Fig. 2. (a) Spatial distribution of grocery stores by cost. The majority of low-cost stores, defined by affordability index (AI r1), are spatially clustered in or near East Portland. (b) Relationship between AI value and block-level distance to the nearest grocery store. Although Portland residents live an average of 0.7 mile (1.1 km) from the nearest grocery store, they must travel an average of 1.9 mile (3.1 km) farther to reach the nearest low-cost grocery store. poverty residing more than 1 mile from a low-cost store, the majority (65%) live in either moderate or extreme food mirages and must travel, on average, 1.8 miles (2.9 km) past the nearest grocery store to arrive at the nearest low-cost store. For the remainder, the nearest grocery store is low-cost. 3.3. Results of regression analysis Spatial regression results for store distance indicated that, after accounting for the effect of spatial autocorrelation, higher income tracts were associated with longer store distances, although the same was true for higher-poverty census tracts (Table 2). This result arose Fig. 3. Population-weighted street-network distance by census tract (n¼ 140). from the fact that some downtown census tracts were characterized (a) Distance to nearest grocery store. 27 tracts were found to be located more by both relatively high poverty rates and high median family incomes. than 1 mile from any grocery store. These census tracts are either conventionally- defined food deserts (areas of concentrated poverty) or food hinterlands (see Leete Gentrifying census tracts were associated with shorter store distances. et al., 2012). (b) Distance to nearest low-cost store. 116 census tracts are more than Note that coefficients are higher for East Portland, since grocery stores 1 mile away from a low-cost store. (c) Food mirage distance, calculated by are generally more dispersed in this region (Table 1). These relation- subtracting (b) from (a). 96 census tracts are associated with a food mirage distance ships were inverted with respect to potential food mirage distance of more than 1 mile. 136 B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 Distance to Table 2 nearest store Spatial regression results for store distance by census tract (n ¼140). Coefficients < 1/2 mile are smaller than one would obtain using ordinary least square regression because variation in the response variable has been partitioned between the socio- 1/2 - 1 mile economic variables and the spatial lag. Units for coefficients are feet per unit of > 1 mile the predictor variable. Households Spatial Median family income Household Gentrification Spatial in poverty extent (US$1000) poverty rate Lag 0 - 50th 51st - 75th Study area 44.48849nnn 60.57993nn  54.33194nn 0.5246nnn Close-in 25.49712n 54.8377nn  13.1580 0.3658nn 76th - 100th eastside Far-out 174.4405nnn 181.5422nn 80.9157 0.4540nn No access eastside barrier Westside 49.99646nn 43.2656  231.5638nn 0.2868 Fig. 4. Map of ‘actual’ food mirages, with household poverty overlay. Tracts where no n p ¼ 0.05. access barrier is present, either because low-cost stores are within 1 mile or because nn p ¼ 0.01, 0.001. tract-level median family income was in the top quartile, were removed. In the urban nnn po 0.0001. core, grocery stores tend to be plentiful (small circles) but prices are high. Along the periphery, stores are in general more dispersed (larger circles) but low-cost stores are available. Households in poverty reside in both the urban core and suburban periphery. Table 3 Spatial regression results for mirage distance by census tract (n¼140). Signs of all Table 1 relationships are the inverse of results for store distance. Units for coefficients are Mean values for tract-level variables for study area partitioned into food access feet per unit of the predictor variable. situations: extreme food mirage (nearest store o 0.5 mile), moderate food mirage (0.5 mileo nearest storeo 1 mile), food hinterland (nearest store 4 1 mile), no food Spatial Median family income Household Gentrification Lag mirage (low cost storeo 1 mile), and high incomes (median income 475th per- extent (US$1000) poverty rate centile for study area). Extreme food mirages are associated with the lowest incomes, highest rates of gentrification, and highest poverty rates. Study area  16.2482  9.9357 32.3834 0.9529nnn Close-in  39.9766nn  50.6037n 48.74478n 0.9630nnn Income Households Gentrification Nearest Food eastside in poverty (% change store mirage Far-out  66.2258  29.5479  99.8234 0.9230nnn (%) white) (miles) distance eastside (miles) Westside  12.6220  42.3495 142.7623 0.7620nnn n Extreme p ¼ 0.05. nn food $38,690 22.40 5.08% 0.37 2.55 p ¼ 0.01, 0.001. nnn mirage po 0.0001. Moderate food $44,821 17.91 1.18% 0.70 2.01 These barriers are particularly relevant for low-income households mirage in gentrifying areas of Portland, where stores are plentiful but Food $46,363 20.17  3.30 1.30 1.53 prices are uniformly high. hinterland No food It is no coincidence that food mirages are at their most extreme in $54,178 13.17  2.18% 0.53 0.17 mirage gentrifying census tracts of North and Northeast Portland. In these High areas, a wide variety of urban amenities, including higher-cost $86,029 7.67  0.63% 0.91 2.41 incomes grocery stores, have recently clustered to service increasingly affluent Study area $54,067 16.36 0.62% 0.72 1.92 (and mostly white) populations—similar shifts are underway in numerous American cities (Quastel, 2009; Hyra, 2012). In the case (Table 3). Potential food mirage distance was negatively associated of Portland, this shift is associated with increasing costs of living, in with median family income and the household poverty rate, while terms of both housing and food, which has lead to displacement of positively associated with gentrification. Stratified regression revealed low-income households into suburban areas of the East Portland, that coefficients were statistically significant in the close-in Eastside where the cost of living is lower (Gibson, 2007). In effect, low-income area of the city, indicating that the link between socioeconomic households are migrating out of extreme food mirages and into food context and food mirages was strongest in this region. hinterlands. To the extent that this process is concentrating poverty in certain East Portland census tracts, the same processes that have produced food mirages could be producing the beginnings of 4. Discussion suburban food deserts. 4.1. Gentrification in food mirages 4.2. Food justice The conventional food desert approach presumes that grocery Although food mirages can be measured quantitatively and store prices are reasonably equivalent, such that any full-service mapped in a GIS, identifying potential remedies requires a more grocery store proxies for access to affordable food. Our survey critical, discursive approach. Issues of class and food justice are demonstrated that grocery stores in the same city offered drasti- sensitive to their framing (Cummins and Macintyre, 2002b). If poor cally different price points, such that many stores are not afford- diet-related health outcomes are framed as problems of proximity able for low-income households. Food access depends on store to a grocery store – any grocery store – then simply by adding grocery affordability, which must be understood as a function of income. stores to under-served areas would seem to resolve the problem. As income increases, so do the number of affordable stores Guthman (2011) contends that the discourse around food justice has and thus spatial proximity to affordable food. By foregrounding placed too much emphasis on supply-side, geographic measures of interactions among income, price, and proximity, a food mirage access at the expense of demand-side, economic barriers related to approach captures otherwise-invisible barriers to healthful diets. household income and budget constraints. In doing so, the discourse  B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 137 neatly sidesteps complex issues of affordability, need, and class. hinterlands. The logical outcome of this process is the production of Supporting this assertion, Short et al. (2007) argue that efforts to areas with both lower population density and lower income—in short, increase food availability in low-income neighborhoods “should not suburban food deserts. Although food mirages may arise from distract from ensuring that residents can pay for it” (Short et al., 2007, different processes than food deserts, they result in the same problem p. 362). These critiques are particularly relevant in Portland, where for low-income households—limited access to healthful foods and food access depends on not only grocery store locations but also food long travel times to get to an affordable grocery store or supermarket. prices and the patterning of affluence and poverty. Food affordability is undoubtedly a difficult concept. No single price By the same token, framing food access as merely a problem of level captures the term’s meaning for all demographics since afford- price neglects its systemic context. For example, a proliferation of ability must be tied to incomes, budget constraints, and, to some Wal-marts throughout low-income neighborhoods has been proposed extent, preferences. Characterizing food price variation requires careful as a solution to inequitable food access (The White House, Office of the measurement and considerable effort. Our measure of food mirages First Lady, 2011). However, our price survey found that Walmart prices hinges on a set of assumptions, embedded in the affordability index were slightly above our benchmark for affordability, mainly because of (AI), that are specific to low-households: incomes at or below 130% above-benchmark prices for fresh produce, which comprised a large of the poverty line, where 30% of income is spent on food. Although share of our market basket. Apart from price, the nutritional value capturing food price variation and defining food affordability are of Wal-mart produce has been called into question (Clifford, 2013). empirically challenging, these issues must be at least considered in Furthermore, in considering an intervention in the food environment, any complete evaluation of food access, particularly access for low- one must balance any improvements in food access against possible income households. negative externalities imposed on households, communities, and to Food access is primarily an issue of income and class. As such, the food system at large (Goetz and Swaminathan, 2006; Neumark food prices matter. They cannot be overlooked in a food environment et al., 2007; Dube et al., 2007; Courtemanche and Carden, 2011; Davis assessment because members of low-income households are likely et al., 2012; Hauter, 2012). Another possible remedy to food mirages price-sensitive shoppers. Spatial patterning of store prices is a critical, would be intervention in agricultural sector to slow the growth in yet under-articulated, dimension of the food environment. By draw- fresh produce prices, but this would require consideration of broader ing attention to food mirages in Portland, this analysis invoked food production priorities. At the minimum, food access depends on broader questions related to gentrification, income distribution, and both price and proximity; however, any intervention must be eval- food production priorities. The survey results presented here suggest uated in its broader context. that research efforts aiming to describe and interpret the food environment should consider implementing healthy market basket 4.3. Study limitations surveys to complement data obtained remotely. Results also suggest that policy-makers seeking to improve health outcomes by interven- We have focused solely on food price variation over urban space in ing in the food environment on behalf of low-income households relation to socioeconomic context at the census tract scale. We believe within a rubric of food justice should consider the possible effects of this focus is justified, given that economic barriers to accessing their actions on price variation over urban space. affordable, nutritious food within cities largely remain unexamined. Our assessment of the food environment was, however, limited in that it ignored other attributes that may influence food consumption Acknowledgements decisions, including quality, convenience, cultural appropriateness, and brand loyalty. Our assessment implicitly assumed households do We are grateful to Christina Friedle, Hunter Shobe, Allison all their shopping at one store, which may not be the case. Further, we Jones, Alexa Todd, Daron McCaulley, Carly Vendegna, and Fiona only considered street-network walking distance, so our results do Gladstone for their project support, assistance with food price data not pertain to other modes of transportation, which can significantly collection, contributions to the initial analysis, and review of the impact effective access for households (Jiao et al., 2012). See initial manuscript. We also thank Suzanne Briggs and David Banis Armstrong et al. (2009) for a careful analysis of the dimensions of for contributing their insights, and the Northwest Health Founda- cultural appropriateness and modes of transportation in the case of tion for funding the community-based project that served as the Portland’s food environment. inspiration for this study. 4.4. Conclusion References This study has investigated how store locations and prices interact with socioeconomic variation to structure the food environment in Aggarwal, A., Monsivais, P., Drewnowski, A., 2012. Nutrient intakes linked to better Portland, Oregon. It has provided evidence that conventionally defined health outcomes are associated with higher diet costs in the US. Public Library of Science 7 (5), 237533. food deserts do not sufficiently describe the barriers to healthful food Alwitt, L.F., Donley, T.D., 1997. Retail stores in poor urban neighborhoods. Journal of access faced by Portland’s low-income households. Price-based bar- Consumer Affairs 31 (1), 139–164. riers were shown to exist in areas that would not appear problematic Andreyeva, T., Long, M.W., Brownell, K.D., 2010. The impact of food prices on consumption: a systematic review of research on the price elasticity of demand from a conventional food desert standpoint. On average, the nearest for food. American Journal of Public Health 100 (2), 216–222. grocery store is 0.7 mile (about a 30-min round-trip walk for the Anselin, L., Syabri, I., Kho, Y., 2006. GeoDa: an introduction to spatial data analysis. average person), but the nearest low-cost grocery store is 1.9 miles Geographical Analysis 38 (1), 5–22. Apparicio, P., Cloutier, M.S., Shearmur, R., 2007. The case of Montréal’s missing food farther away (nearly a 2-h round-trip walk). deserts: evaluation of accessibility to food supermarkets. International Journal This study has posited a typology of food mirages. The most of Health Geographics 6 (4), 1–13. extreme food mirages occur where grocery stores are abundant but Armstrong, K., Chapin, E., Chastain, A., Person, J., VanRheen, S., White, S., 2009. Foodability: Visioning for Healthful Food Access in Portland. (Last accessed 11 costly, as is the case in denser, often gentrifying, neighborhoods. August 2012 from). By contrast, food hinterlands occur in lower-density neighborhoods, Asimov, E., 2007. In Portland, a Golden Age of Dining and Drinking. 26. New York where grocery stores tend to be more affordable for low-income Times. (September 2007. Last accessed 11 August 2012 from). households but distance to nearest store exceeds 1 mile. We sug- Auld, M.C., Powell, L.M., 2009. Economics of food energy density and adolescent body weight. Economica 76 (304), 719–740. gested that the processes of gentrification have lead to the displace- Beaulac, J., Kristjansson, E., Cummins, S., 2009. A systematic review of food deserts. ment of low-income households from extreme food mirages to food Preventing Chronic Disease 6 (3), A105. 138 B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 Block, D., Kouba, J., 2006. A comparison of the availability and affordability of a Grocery CART PDX, 2012. Healthful Food Market Basket Survey. Last accessed 11 August market basket in two communities in the Chicago area. Public Health Nutrition 2012 from /http://grocerycartpdx/the-food-basketS. 9 (7), 837–845. Guthman, J., 2011. Weighing In. University of California Press, Berkeley. Bodor, J.N., Rose, D., Farley, T.A., Swalm, C., Scott, S.K., 2008. Neighbourhood fruit Hauter, W., 2012. Why Walmart can’t Fix the Food System. Food & Water Watch and vegetable availability and consumption: the role of small food stores in an Report. Last accessed 20 November from 〈http://documents.foodandwater urban environment. Public Health Nutrition 11 (4), 413–420. watch.org/doc/FoodandWaterWatchReportWalmart022112.pdf〉. Boone-Heinonen, J., Gordon-Larsen, P., Kiefe, C.I., Shikany, J.M., Lewis, C.E., Popkin, Herman, D.R., Harrison, G.G., Jenks, E., 2006. Choices made by low-income women B.M., 2011. Fast food restaurants and food stores: longitudinal associations with provided with an economic supplement for fresh fruit and vegetable purchase. diet in young to middle-aged adults: the CARDIA study. Archives of Internal Journal of the American Diet Association 106 (5), 740–744. Medicine 171 (13), 1162–1170. Hillier, A., Cannuscio, C.C., Karpyn, A., McLaughlin, J., Chilton, M., Glanz, K., 2011. Carlson, A., Lino, M., Juan, W.Y., Hanson, K., Basiotis, P.P., 2007. Thrifty Food Plan, How far do low-income parents travel to shop for food? Empirical evidence 2006. (CNPP-19). U.S. Department of Agriculture, Center for Nutrition Policy from two urban neighborhoods. Urban Geography 32 (5), 712–729. and Promotion. Hubley, T.A., 2011. Assessing the proximity of healthy food options and food deserts Carlson, A., Frazão, E., 2012. Are Healthy Foods Really more Expensive ? It Depends in a rural area in Maine. Applied Geography 31 (4), 1224–1231. on How you Measure the Price. Economic Information Bulletin No. (EIB-96) 50, Hyra, D., 2012. Conceptualizing the new urban renewal: comparing the past to the Economic Research Service, U.S. Department of Agriculture. Last accessed present. Urban Affairs Review 48 (4), 498–527. 11 August 2012 from 〈http://www.ers.usda.gov/publications/eib-economic-in Interfaith Food and Farms Partnership, 2007. Everyone Eats! A Community Food formation-bulletin/eib96.aspx〉. Assessment Report for areas of North and Northeast Portland. Ecumenical Casey, J., 2008. Portland’s low-income neighborhoods are city’s ‘food deserts.’. The Ministries of Oregon’s Interfaith Network for Earth Concerns. Last accessed 11 Oregonian 15 November 2008. Last accessed 11 August 2012 from /http:// August 2012 from 〈http://www.emoregon.org/pdfs/IFFP/IFFP_NNE_Portland_ www.oregonlive.com/health/index.ssf/2008/11/living_in_a_food_desert.htmlS. Food_Assessment_full_report.pdf〉. Cassady, D., Jetter, K.M., Culp, J., 2007. Is price a barrier to eating more fruits and Jiao, J., Moudon, A.V., Ulmer, J., Hurvitz, P.M., Drewnowski, A., 2012. How to identify vegetables for low-income families? Journal of the American Dietetic Associa- food deserts: measuring physical and economic access to supermarkets in King tion 107 (11), 1909–1915. County, Washington. American Journal of Public Health 102 (10), e32–e39. Charreire, H., Casey, R., Salze, P., Simon, C., Chaix, B., Banos, A., Badariotti, D., Weber, C., Larson, N.I., Story, M.T., Nelson, M.C., 2009. Neighborhood environments: dispa- Oppert, J., 2010. Measuring the food environment using geographical information rities in access to healthy foods in the U.S. American Journal of Preventive systems: a methodological review. Public Health Nutrition 13 (11), 1773–1785. Medicine 36 (1), 74–81. Chung, C., Myers, S.L., 1999. Do the poor pay more for food? An analysis of grocery store Lee, H., 2012. The role of local food availability in explaining obesity risk among availability and food price disparities. The Journal of Consumer Affairs 33 (2), young school-aged children. Social Science & Medicine 74 (8), 1193–1203. 276–296. Leete, L., Bania, N., Sparks-Ibanga, A., 2012. Congruence and coverage: alternative Clifford, S. 2013. Walmart Strains to Keep Aisles Stocked Fresh. The New York approaches to identifying urban food deserts and food hinterlands. Journal of Times, 3 April 2013. Last accessed 1 July 2013 from 〈http://www.nytimes. Planning Education and Research 32 (2), 204–218. Margheim, J., 2007. The Geography of Eating Well: Food Access in the Metroscape. com/2013/04/04/business/walmart-strains-to-keep-grocery-aisles-stocked. Periodic Atlas of the Metroscape (2007), 13-19. Last accessed 11 August 2012 html?pagewanted ¼ all&_r ¼ 0〉. from 〈http://pdx.edu/sites/www.pdx.edu.ims/files/ims_mscape07atlas.pdf〉. Courtemanche, C., Carden, A., 2011. Supersizing supercenters? The impact of Monsivais, P., Aggarwal, A., Drewnowski, A., 2011. Following federal guidelines to Walmart supercenters on body mass index and obesity. Journal of Urban increase nutrient consumption may lead to higher food costs for consumers. Economics 69 (2), 165–181. Health Affairs (Millwood) 30 (8), 1471–1477. Cummins, S., Macintyre, S., 2002a. A systematic study of an urban foodscape : the Monsivais, P., Drewnowski, A., 2009. Lower-energy-density diets are associated price and availability of food in greater Glasgow. Urban Studies 39 (11), with higher monetary costs per kilocalorie and are consumed by women of 2115–2130. higher socioeconomic status. Journal of the American Dietetic Association 109 Cummins, S., Macintyre, S., 2002b. “Food deserts”—evidence and assumption in (5), 814–822. health policy making. British Medical Journal 325 (7361), 436–438. Moore, L.V., Diez Roux, A.V., 2006. Associations of neighborhood characteristics Darmon, N., Drewnowski, A., 2008. Does social class predict diet quality? American with the location and type of food stores. American Journal of Public Health Journal of Clinical Nutrition 87 (5), 1107–1117. 96 (2), 325–331. Davis, J., Merriman, D., Samayoa, L., Flanagan, B., Baiman, R., Persky, J., 2012. The Multnomah County Health Department, 2008. Community Health Assessment Impact of an Urban Wal-Mart Store on Area Businesses: An Evaluation of One Quarterly 3 (3), 1–4. Chicago Neighborhood’s Experience. Center for Urban Research and Learning, Neumark, D., Zhang, J., Ciccarella, S., 2007. The Effects of Wal-Mart on Local Labor Loyola University Chicago. Last accessed 11 August 2012 from 〈http://ecom Markets. Institute for the Study of Labor IZA DP No.2545. Last accessed 20 mons.luc.edu/curl_pubs/3〉. November 2012 from 〈http://papers.ssrn.com/sol3/papers.cfm?abstract_id=958704〉. Drewnowski, A., 2004. Obesity and the food environment: dietary energy density Population Research Center, 2012. Oregon Annual Population Report. Portland State and diet costs. American Journal of Preventive Medicine 27 (Suppl. 3), 154–162. University. Last accessed 11 October 2012 from 〈http://www.pdx.edu/prc/ Drewnowski, A., Aggarwal, A., Vernez Moudon, A., 2010. The Supermarket Gap: annual-oregon-population-report〉. How to Ensure Equitable Access to Affordable, Healthy Foods. Center for Public Powell, L.M., Chaloupka, F.J., 2009. Food prices and obesity: evidence and policy Health and Nutrition Research Brief (May), University of Washington, p. 1–4. implications for taxes and subsidies. Milbank Quarterly 87 (1), 229–257. Last accessed 11 August 2012 from /http://depts.washington.edu/uwcphn/ Powell, L.M., Chirqui, J.F., Kahn, T., Wada, R., Chaloupka, F.J., 2013. Assessing the reports/cphnbrf051910.pdfS. potential effectiveness of food and beverage taxes and subsidies for improving Drewnowski, A., Aggarwal, A., Hurvitz, P.M., Monsivais, P., Moudon, A.V., 2012. Obesity public health: a systematic review of prices, demand and body weight out- and supermarket access: proximity or price? American Journal of Public Health 102 comes. Obesity Review 14 (2), 110–128. (8), e74–e80. Quastel, N., 2009. Political ecologies of gentrification. Urban Geography 30 (7), 694–725. Dube, A., Lester, T.W., Eidlin, B., 2007. Firm Entry and Wages: Impact of Wal-Mart Rehm, C.D., Monsivais, P., Drewnowski, A., 2011. The quality and monetary value of Growth on Earnings Throughout the Retail Sector. UC Berkeley: Institute for diets consumed by adults in the United States. American Journal of Clinical Research on Labor and Employment. Last accessed 20 November 2012 from Nutrition 94 (5), 1333–1339. 〈http://escholarship.org/uc/item/22s5k4pv〉. Richardson, A.S., Boone-Heinonen, J., Popkin, B.M., Gordon-Larsen, P., 2012. Are Everett, M., 2011. Practicing anthropology on a community-based public health neighbourhood food resources distributed inequitably by income and race in coalition: lessons from HEAL. Annals of Anthropological Practice 35 (2), 10–26. the USA? Epidemiological findings across the urban spectrum.. BMJ Open 2 (2), Forsyth, A., Lytle, L., Van Riper, D., 2010. Finding food: issues and challenges in e000698. using geographic information systems to measure food access. Journal of Rogerson, P., 2010. Statistical Methods for Geography. Sage Publications Ltd., London Transport and Land Use 3 (1), 43–65. p. 283. Franco, M., Diez Roux, A.V., Glass, T.A., Caballero, B., Brancati, F.L., 2008. Neighbor- Rose, D., Richards, R., 2004. Food store access and household fruit and vegetable use hood characteristics and availability of healthy foods in Baltimore. American among participants in the U.S. Food Stamp Program. Public Health Nutrition 7 (8), Journal of Preventive Medicine 35 (6), 561–567. 1081–1088. Gibson, K.J., 2007. Bleeding Albina: a history of community disinvestment. Sharkey, J.R., Horel, S., 2008. Neighborhood socioeconomic deprivation and minor- Transforming Anthropology 15 (1), 3–25. ity composition are associated with better potential spatial access to the Glanz, K., Yaroch, A.L., 2004. Strategies for increasing fruit and vegetable intake in ground-truthed food environment in a large rural area. The Journal of Nutrition grocery stores and communities: policy, pricing, and environmental change. 138 (3), 620–627. Preventative Medicine 39 (Suppl. 2), S75–S80. Short, A., Guthman, J., Raskin, S., 2007. Food deserts, oases, or mirages? Small Gleason S., Pooler J., 2011. The Effects of Changes in WIC Food Packages on markets and community food security in the San Francisco Bay area. Journal of Redemptions. Altarum Institute. Last accessed January 21, 2013 from 〈http:// Planning Education and Research 26 (3), 352–364. www.altarum.org/files/pub_resources/Effects%20of%20Changes%20to%20the% Sparks, A.L., Bania, N., Leete, L., 2010. Comparative approaches to measuring food 20WIC%20Food%20Package_December%202011final.pdf〉. access in urban areas: the case of Portland, Oregon. Urban Studies 48 (8), Goetz, S., Swaminathan, H., 2006. Wal-Mart and county-wide poverty. Social 1715–1737. Science Quarterly 87 (2), 211–226. Todd, J.E., Leibtag, E.S., Penberthy, C., 2011. Geographic Differences in the Relative Gordon, C., Purciel-Hill, M., Ghai, N.R., Kaufman, L., Graham, R., Van Wye, G., 2011. Price of Healthy Foods. Economic Information Bulletin No. (EIB-78) 40, Measuring food deserts in New York City’s low-income neighborhoods. Health Economic Research Service, U.S. Department of Agriculture. Last accessed & Place 17 (2), 696–700. 9 April 2013 from 〈http://www.ers.usda.gov/media/128007/eib78.pdf〉.  B. Breyer, A. Voss-Andreae / Health & Place 24 (2013) 131–139 139 Ver Ploeg, M., Breneman, V., Farrigan, T., Hamrick, K., Hopkins, D., Kaufman, P., et al., Service, U.S. Department of Agriculture. Last accessed 11 August 2012 from 2009. Access to Affordable and Nutritious Food—Measuring and Understanding 〈http://ers.usda.gov/publications/err-economic-research-report/err118.aspx〉. Food Deserts and Their Consequences: Report to Congress; U.S. Department of The White House, Office of the First Lady, 2011. First Lady Michelle Obama Announces Agriculture Economic Research Service. Last accessed 11 August 11, 2012 from Nationwide Commitments to Provide Millions of People Access to Healthy, Afford- 〈http://www.ers.usda.gov/Publications/AP/AP036/〉. able Food in Underserved Communities. Last accessed 11 August 2012 from 〈http:// Voss-Andreae, A., 2011. The Food Landscape for Affordable Housing Residents in www.whitehouse.gov/the-press-office/2011/07/20/first-lady-michelle-obama- Portland: A Healthy Foods Access Initiative of the Housing Organizations of announces-nationwide-commitments-provide-milli〉. Color Coalition (unpublished report). Zenk, S.N., Lachance, L.L., Schulz, A.J., Mentz, G., Kannan, S., Ridella, W., 2008. Walker, R.E., Keane, C.R., Burke, J.G., 2010. Disparities and access to healthy food in the Neighborhood retail food environment and fruit and vegetable intake in a United States: a review of food deserts literature. Health & Place 16 (5), 876–884. multiethnic urban population. American Journal of Health Promotion 23 (4), Wendt, M., Todd, J.E., 2011. The Effect of Food and Beverage Prices on Children’s 255–264. Weights. Economic Information Bulletin No. (EIB-118) 29, Economic Research