Landscape characteristics affect animal control by urban residents

Ecological patterns exist within urban landscapes. Among urban patterns of biodiversity, species occurrences may coincide with interactions between humans and wildlife. However, research focused on consequences of human reaction to interactions with wildlife is limited. We evaluated landscape characteristics of rodent control behavior across two urban landscapes in California, Bakersfield, and in proximity to Santa Monica Mountain National Recreation Area (SAMO). Our data were collected prior to a recent policy ruling limiting distribution of particular rodent control products. In both locations, local biologists have observed non-target effects of rodent control products among local carnivores. Mice and rats were among the species most targeted in both locations, but squirrels and gophers also were common targets in SAMO. Carnivore species identified by biologists were among those also reported by residents as targeted for control. In both locations, those who reside in single-family structures and among lower-density development were more likely to practice rodent control. Species targeted varied by distance to open space in both locations, but by development density in SAMO only. In Bakersfield, control was distributed across the study area, but one cluster of control existed among mainly lower-density, single-family residences. In SAMO, clusters of both control (n = 2) and chemical use (n = 3) existed among single-family, lower-density areas in proximity to wash channels and relatively lush vegetation. Our results suggest possible pathways for contact between wildlife and rodent control products, but causal linkages between the two are beyond the scope of our data. Similar to other urban ecological processes, human responses to interactions with ecological phenomena may occur at both fine and landscape scales. Furthermore, our results suggest a possible feedback loop of interacting ecological and social phenomena that may provide information about human activities affecting urban wildlife populations.


INTRODUCTION
In urban landscapes, humans influence ecosystems as a result of land cover and land use conversion (Theobold et al. 1997, McKinney 2002, Faeth et al. 2005, Grimm et al. 2008, modification of biophysical and ecological processes (Arnold and Gibbons 1996, Collins et al. 2000, Paul and Meyer 2001, and alteration of species habitat and assemblages (Hope et al. 2003, Faeth et al. 2005. Research documenting relationships between urbanization and declining biodiversity, habitat degradation, policy, and human health often focuses on industry and transportation (Holdren and Ehrlich 1974, Gill and Elliott 2003, Naylor et al. 2003, Bell et al. 2004, Hadley and Wilson 2004. Although resident population is the metric used to define city size (U.S. Census Bureau data), few studies have focused specifically on residential components of urban ecosystems. With the ongoing rural-to-urban-andsuburban population shift Stoops 2002, McKinney 2002), and spatial dominance of residential development, there is a need to understand the extent that residents recognize impacts they have on their local environment. This understanding may lead to better outreach and regulation that meets the needs of both humans and other species.
Within urban systems, the presence and adaptation of wildlife and conflict with humans (Luniak 2004, DeStefano and Degraaf 2003, Gehrt et al. 2010) is a consequence of broadly shared resource needs (e.g., O'Donnell and DeNicola 2006, Hill et al. 2007, Krester et al. 2008, Hostetler and Drake 2009. For instance, residential land management (Lepczyk et al. 2004b), intentional feeding (Fuller et al. 2008), and presence of exotic species Soulé 1999, Baker et al. 2005) may enable potential for conflict between humans and wildlife. Although it is known that residential activities can impact wildlife (e.g., Lepczyk et al. 2004a, b, Faeth et al. 2005, research focusing on effects of such activities is limited (Liu et al. 2003).

Background and context
Pest control is one method by which humans may respond to conflict with wildlife. Regardless of location, control often focuses on synanthropic exotics, such as black or Norway rats (Rattus rattus and R. novegicus, respectively) and house mice (Mus musculus). However, control may also target native species, such as gophers (e.g., Thomomys spp.) and moles (e.g., Scapanus spp.). Regardless of target, the products used for control can be indiscriminant, resulting in impacts to non-target species.
In this study, we evaluated characteristics of pest control across two urban landscapes. Of particular interest were mammals as targets and use of chemical products, specifically anticoagulants. Active anticoagulant ingredients, which inhibit the clotting of blood (Amdur et al. 1991), include warfarin, brodifacoum, and bromadiolone (USFWS 1993). Such compounds intended for household use (indoor and outdoor near structures) are marketed under a variety of trade names. Exposure of non-target species may occur by direct consumption (e.g., Spurr 1995, Brakes andSmith 2005), ingestion of nonabsorbed compounds within the digestive tracts of prey (Howald et al. 1999), or indirect exposure during consumption of contaminated prey (e.g., Alterio 1996, Berny et al. 1997, Eason et al. 1999. Anticoagulants can increase risk of mortality among non-target species, but there is almost no knowledge about the mechanisms by which they travel through the environment, locations of use, and target species (Erickson and Urban 2004), and little information about general product use (e.g., US EPA 1979, Wilen 2001, Erickson and Urban 2004. Observed non-target mortality from anticoagulants is a global phenomenon (Eason and Spurr 1995). In Europe, Brakes and Smith (2005) reported small mammals consuming bait during routine rat control, which severely affected small mammal populations. Residues have been reported in avian and terrestrial predators in Europe (Shore et al. 2003, Fournier-Chambrillon et al. 2004, Walker et al. 2008 and Asia (Duckett 1984). In Canada, use of brodificoum for rat eradication presented obvious exposure to avian scavengers (Howald et al. 1999, Albert et al. 2010. Non-target mortality has been observed among raptors, other birds, and mammals in New Zealand (Dowding et al. 1999, Eason et al. 1999, Eason et al. 2002 and the United States (Littrell 1988, Stone et al. 1999, Riley et al. 2003, Riley et al. 2007, McMillin et al. 2008. Although these reports highlight non-target effects, we are unaware of any research within a landscape context, and have limited knowledge about the role of urban residents as potential contributors to the issue (Wilen 2001).
A 2008 Environmental Protection Agency (EPA) federal ruling limits the sale and distribution of 10 anticoagulant rodenticides in the United States in an effort to minimize potential for exposure by children and non-target wildlife (US EPA 2008). New requirements include minimum package size requirements, use site restrictions, sale and distribution restrictions, and use of bait stations for outdoor above-ground application (US EPA 2008). Our research is the first attempt to evaluate rodent control and anticoagulant product use across a landscape, and our data were collected prior to the 2008 ruling. Our objectives were to (1) identify species targeted for control and whether our carnivores of interest are among them, and (2) evaluate relationships between landscape characteristics and control behavior.

Study area
Our study area included two urban locations in California, USA, where local biologists have observed non-target impacts on several species (Fig. 1). One location is within the southwestern section of Bakersfield, delimited by the channelized Kern River and Highway 99. Land use included mixed-density residential development and related services (e.g., shopping plazas), golf courses, a small university campus, and some industrial development. A species of particular interest is a local urban population of San Joaquin kit fox (Vulpes macrotis mutica; e.g., Bjurlin et al. 2005), a federally endangered species that has experienced mortality from exposure to anticoagulants (McMillin et al. 2008).
The other location straddled the Ventura-Los Angeles County border in proximity to the Santa Monica Mountains National Recreation Area (hereafter SAMO). This area consisted particularly of low-and medium-density residential development interspersed among natural areas, residential services, a golf course, and limited industrial areas. Species of particular interest include bobcat (Lynx rufus), coyote (Canis latrans), and mountain lion (Puma concolor), which have been observed as having anticoagulant residues in their tissues; toxin load was directly related to use of developed habitat by two of these species (Riley et al. 2003, Riley et al. 2007. We received evidence of anticoagulant-related mortality from local biologists, as determined by necropsy analysis (Riley et al. 2007, McMillin et al. 2008. Biologists also provided locations where dead animals were found (e.g., in dens) and associated estimated home range areas and movements as gathered by radiotelemetry (B. Cypher, Endangered Species Recovery Program, California State University Stanislaus, personal communication; S. P. D. Riley, National Park Service, personal communication; sensu Riley et al. 2003). This information allowed us to identify residential areas to target for information about rodent control. We incorporated this knowledge into ArcView GIS 3.2 (Environmental Systems Research Institute, Redlands, California, USA), to establish boundaries of extents (sampling areas) based on major roads.

Data collection
Residential rodent control behavior.-A mail survey was used to collect information about residential pest control behavior. The sampling frame was the list of all residential street addresses within our study extents, and the sampling unit was the individual household. We purchased street address information from Marketing Systems Group (Fort Washington, PA), which compiles datasets from U.S. Postal Service delivery sequence files. We excluded PO boxes, seasonal homes, and mail drops (single delivery points for multiple addresses) because of our need to apply data to a spatial context. The Office of Management and Budget (Control Number 2080-0077), Office of Human Research Ethics at University of North Carolina-Chapel Hill (IRB # 08-0775), and Office of Research Integrity at Oregon State University (IRB # 4442) granted permission for use of human subjects.
Multiple mailings and a toll-free number for participant questions were used in an effort to increase response rate (Dillman 2000). Sample size was based on desired sampling error (65%) and statistical power (80%), and we assumed lower-than-average response rate because of survey administration by a government agency (Dillman 2000). In September 2007, both English and Spanish versions of questionnaires were sent to randomly selected households in both locations (n ¼ 4,000 per site, N ¼ 8,000). The overall response rate for the survey was 25% (n ¼ 2,001; Bakersfield ¼ 780; SAMO ¼ 1,221). Morzillo and Mertig (2011a) provide further details about surveys returned incomplete, non-respondents, and the non-response follow-up. Ultimately, responses from Bakersfield and SAMO were v www.esajournals.org evaluated separately because of demographic differences between the two locations (Morzillo and Mertig 2011a).
We used a two-step process to identify chemical product users among our respondents. First, we identified control behavior participants by selecting those who answered ''yes'' to the question, ''Have you or anyone else tried to control rodents or other animals on your property'' (within the past five years; Bakersfield n ¼ 320, SAMO n ¼ 720). Second, we identified chemical product users by selecting those who answered ''yes'' when asked if chemical rodenticides have been used on their property (Bakersfield n ¼ 141, SAMO n ¼ 338). Chemical product users then were asked to select all products used from among: (1) anticoagulants, (2) dehydration, (3) fumigants, (4) nerve agent, (5) zinc phosphide, (6) unsure, and (7) other. Examples of each chemical were provided to assist subjects with product identity. Responses were coded such that 1 ¼ use of each chemical (1-7), and 0 ¼ non-use, and allowed us to identify anticoagulant users specifically.
Target species.-To evaluate target species, we asked, ''Which of the following types of animals have you or someone else been trying to control on your property?'' Respondents were asked to select applicable species from a list provided on the survey (Appendix A). We also provided space to specify ''other'' wildlife not included in the list. All non-mammal species (e.g., cockroaches, rattlesnakes) not pertinent to our objectives were removed from further analysis.
Background variables.-We used eight demographic and seven behavior variables to evaluate respondent characteristics (variable names in parentheses; Appendix A). Demographic variables included household size (Hhsize), children in household (Children), residential tenure (Tenure), home ownership (Own), sex (Sex), age (Age), education (Education), and household income (Income).
Of the seven behavior variables (Appendix A), three focused on pets: presence of pets (Pets), whether pets are allowed outdoors unsupervised (PetsUnsupOut), and whether pets are fed outdoors (PetsFeedOut). The other four activity variables were based on knowledge that individual behaviors often are directly associated with personal emotional relationship with the environment (Hinds and Sparks 2008), and included v www.esajournals.org participation in environmental service activities (EnvService), participation in wildlife-related activities (WldfAct), importance of environmental amenities in location of residence (NatReside), and concern about non-target effects (Concern). Morzillo and Mertig (2011a) provide details about data reduction techniques, scale construction, and creation of behavior variables.
Landscape variables.-We evaluated relationships between residential control behavior and four urban landscape variables: building age, building structure, development density, and distance from open space (Appendix B). Based on correlations between building age and pest presence (Turner andBishop 1998, Berkowitz et al. 2002), we hypothesized that residents in older buildings would be no more likely to practice control than those in newer buildings. Berkowitz et al. (2002) also reported greater pest control among apartment buildings than houses, and so we hypothesized a greater likeliness of pest control among multiple-than single-family structures. More broadly, Adgate et al. (2000) suggested no differences in pesticide use patterns between urban and rural census tracts. However, ecological processes within urban systems are spatially heterogeneous , McKinney 2008. Therefore, we suspected that human responses to ecological processes may follow patterns similar to the processes themselves, and hypothesized that residents among lower density development and closest to open space are more likely to practice control.
For Bakersfield, building age and building type were derived from the Kern County Land Assessor's Office database (http://www.recorder. co.kern.ca.us/index.php; Appendix B). All land use information for Bakersfield was obtained from the Kern County Development Services Agency (http://www.co.kern.ca.us/gis/) GIS database. Residential development density was defined by the five-class Kern County general land use designations. Overlapping classes signifies that some mixed densities existed within the same block. All open space maintained relatively urban characteristics (i.e., ''altered open space'') and included vacant lots along Kern River, parks and recreation areas, and schoolyards.
For SAMO, building age and building type were derived from the Los Angeles (http:// assessor.lacounty.gov) and Ventura (http:// assessor.countyofventura.org/) County Assessor's Office databases. Building age and type were calculated similarly as for Bakersfield (Appendix B). Land use data for SAMO were obtained from the National Park Service (D. Kamradt, National Park Service, personal communication) and the National Land Cover Dataset (Fry et al. 2009). Residential density was determined based on existing designations. Open space included two categories: natural areas (e.g., county parks, National Recreation Area units) and ''altered open'' (golf courses, developed parks, schoolyards). To account for management differences, we completed calculations for natural areas alone (''natural''), and all open space collectively (''natural plus altered'').
Statistical analysis.-SPSS 16.0.1 (SPSS, Chicago, Illinois, USA) was used for to evaluate relationships between survey responses and landscape variables. Pearson's r and chi-square were used to test bivariate relationships, as appropriate (Sokal and Rohlf 1995). Alpha values were defined at the 95% confidence interval (a ¼ 0.05).
To define scale of and evaluate spatial distribution of control behavior, ArcGIS 9.3.1 (Environmental Systems Research Institute, Redlands, California, USA) and Moran's I test (Moran 1950, Fortin andDale 2005) were used to quantify a global spatial autocorrelation, with the null hypothesis of a random spatial pattern. Essentially, Moran's I allowed us to identify whether the pattern of values across the study area tend to be clustered, random, or dispersed. The inherent clustering of humans and our use of human activity as the variables of interest necessitated identification of a neighborhood-sized scale. Because neighborhoods were adjacent to each other, we conducted the Moran's I analysis at 100 m intervals, starting at 100 m and ending at 2,000 m from each respondent practicing control. The resulting Moran's I Z-scores from each distance were graphed to define the distance that locations of respondents maintained the greatest amount of autocorrelation.
Because global statistics such as Moran's I are designed only to evaluate general autocorrelation of a particular landscape pattern, an additional test was needed to evaluate ''membership'' of individual respondents among those with similar v www.esajournals.org control behavior (i.e., at the appropriate scale of the behavior). Therefore, we used the Getis-Ord test (G(i )*; Getis and Ord 1992) to evaluate details of clustering of control behavior at the scale of control behavior. Respondents who practiced control and used chemicals served as two feature classes of interest. For large n sizes (i.e., .30), Getis and Ord (1996) suggest a distance band that includes at least 30 neighbors. A Zone of Indifference conceptualization of spatial relationships was selected because it allows the minimum number of neighbors to be met, and does not establish a finite boundary on neighbors that fall near the edge of the distance band. Respondents near the edge of the distance threshold influence group membership within the distance threshold, which is important for high concentration areas. The threshold distance value (greatest amount of autocorrelation) was defined by Moran's I results (above). G(i )* values with Z . 1.96 were considered significant at the 95% confidence level, and suggested rejection of a null hypothesis (i.e., distribution containing no clusters).

Target species
For both locations, rats and mice were the species controlled for most frequently (Fig. 2). Squirrels, gophers, and rabbits were controlled for by at least 15% of respondents in SAMO. A variety of ''other'' target species were identified, including kit fox in Bakersfield and the three carnivore species of interest (bobcat, coyote, mountain lion) in SAMO (Table 1).

Sample characteristics across the landscape
For Bakersfield, average (6SD) residential structure age was 23 (614) years. A majority were single-family residences (80%), and among low (70%) and high-medium (18%) density categories. Average distance to open space was 290 (6197) m. Respondents with larger households, children in the household, younger respondents, those with less formal education, and those with lower incomes were more likely to reside further from open space (Table 2). Smaller households, households without children, those with shorter tenure, renters, younger residents, as well as those with lower incomes, without pets (and who do not let pets outside unsupervised or feed pets outside), participate less in wildliferelated activities, and have greater concern about non-target effects were more likely to reside among greater development density.
For SAMO, average residential structure age was 28 (619) years. A majority were singlefamily residences (73%), and among medium-(62%) and high-(28%) density categories. Average distance to natural plus altered open space (112 6 106 m) was less than natural open space (127 6 106 m). Respondents with longer tenure, less formal education, lower incomes, those who leave pets out unsupervised, and those who considered natural amenities to be less important were more likely to reside further from open space (natural or natural plus altered; Table 2). Older residents were more likely to be further from natural areas. Those less likely to own their own homes and participate in environmental service activities were more likely to be further from open space. Smaller households, those with children, newer residents, renters, females, younger residents, those without pets (and who do not let pets outside unsupervised or feed pets outside), those less likely to be involved in environmental service activities, and those most concerned about non-target impacts on wildlife were more likely to reside among greater development density.

Landscape characteristics affecting control behavior
For Bakersfield, neither control (r ¼ 0.04, p ¼ 0.29), nor chemical use (r ¼À0.04, p ¼ 0.32) varied with building age. However, those in singlefamily structures were more likely to control (v 2 ¼ 11.42, df ¼ 1, p ¼ 0.001) and use chemicals (v 2 ¼ 8.33, df ¼ 1, p ¼ 0.004) than those in multiplefamily structures. Distance from open space did not vary between those who do and do not control (r ¼ 0.01, p ¼ 0.72), or use chemicals (r ¼ À0.12, p ¼ 0.10). Across all respondents who control, distance to open space did not differ for those who target mice, rats, or gophers; those targeting squirrels were more likely closer to open space (r ¼ 0.14, p ¼ 0.009). Respondents practicing control were more likely to reside among lower density development (r ¼À0.11, p ¼ 0.003); this relationship did not hold true for chemical users (r ¼À0.04, p ¼ 0.61). Compared to all respondents who practice control, species controlled for did not vary across development density. Density and distance were not correlated (r ¼ À0.04, p ¼ 0.25).
Also for Bakersfield, we rejected our null hypothesis that control was randomly distributed (Z ¼ 1.95, p ¼ 0.05); maximum autocorrelation existed at a distance of 1,800 m (Fig. 3). Although no prominent clusters emerged for chemical use, one prominent cluster emerged for control (n ¼ 95). This cluster corresponded to both the western boundary of the extent, and was in proximity to both a golf course and multiple areas of ongoing residential construction. Among this cluster, average structure age, and average distance to (altered) open space were less than that for the overall sample (Table 3). All but one structure were single family residences, and most were among the low and high-medium density development categories. Mice and rats were the main species targeted, but at a lower proportion than across all respondents who control. Neither control nor use of chemicals exhibited global autocorrelation based on Moran's I scores. One potential ''hotspot'' of control existed among an area with relatively lower density and newer buildings near the extent boundary.
For SAMO, whether or not a respondent practiced control did not vary by building age (r ¼ À0.01, p ¼ 0.84); this was consistent for chemical (r ¼ À0.01, p ¼ 0.69) users. However, those who control (v 2 ¼ 65.10, df ¼ 2, p , 0.001) and use chemicals (v 2 ¼ 30.16, df ¼ 1, p , 0.001) were more likely to reside in single-family rather than multiple-family structures. Respondents closer to open space were more likely to control for rodents than those further away, regardless of whether open space was natural (r ¼ À0.11, p , 0.001) or natural plus altered (r ¼ À0.06, p ¼ 0.044). This relationship held true for chemical users (natural r ¼ À0.12, p , 0.001; natural plus altered r ¼ À0.07, p ¼ 0.017). Across all Fig. 2. Frequency (percent) of respondents who control for each target species on their property in two locations in California. Bakersfield is illustrated using white bars (n ¼ 317); SAMO is illustrated using black bars (n ¼ 718). See Table 1 for a detailed breakdown of the ''other'' category.  Fig. 3. Z-score values of distance bands from Moran's I test were used to determine level of spatial autocorrelation of respondents that practiced control in Bakersfield (gray) and SAMO (black). Greater Z-scores indicate greater spatial autocorrelation (i.e., clustering of similar behavior). A star denotes maximum value of spatial autocorrelation for each location.
Also for SAMO, we also rejected our null hypothesis that survey respondents practicing control were randomly distributed (Z ¼ 2.98, p ¼ 0.003); maximum autocorrelation existed at a distance of 1,400 m (Fig. 3). Among respondents exhibiting control behavior, two prominent ''hotspot'' clusters (n ¼ 108) emerged in the northeastern and southern parts of the study area. Both were among low density development and in proximity to wash channels that contained more lush vegetation than surrounding hills. Compared to the overall sample, landscape characteristics generally were similar among those who control (Table 3). However, notable distinctions among the clusters include closer proximity to open space, and less frequency in targeting rats. Three prominent ''hotspot'' clusters (n ¼ 156) emerged among respondents using chemicals, two of which overlapped with the two control clusters and a third at the western boundary of the extent. The third cluster was located among a medium density and singlefamily structure area that abutted natural open space. Compared to the overall sample, landscape characteristics for clustered chemical users generally illustrated similar trends as for those who control (Table 3). However, compared to those who control, chemical users were more frequently among single-family structures, medium development density, and slightly further from open space. The southern cluster that emerged for control and chemical use suggested a high concentration of anticoagulant users among it, and coincided with proximity to land being converted to new residential development. Therefore, control and use of chemicals exhibited global autocorrelation, with ''hotspots'' among single family structures in relatively lower density development categories near open space.

DISCUSSION
Urbanization impacts ecosystem pattern (Luck and Wu 2002) and processes (McDonnell and Pickett 1990, McDonnell et al. 1997, Neil and Wu 2006. Few studies have evaluated impacts of human activities at the household scale (Liu et al. 2003), or human reaction to environmental effects. While evaluating urban residential pest control behavior, we discovered that patterns of pest control may mimic underlying ecological Note: Values in parentheses are standard deviations.
The ubiquitous targeting of mice and rats may be a function of the adaptability of these species to urban areas and inside buildings, as well as a psychological phenomenon. Rats generally are not well-liked (Kellert and Berry 1980) and, in another part of our study, a direct relationship existed between negative attitudes toward rodents and control behavior (Morzillo and Mertig 2011b). Others have linked reaction to rats to perception of human welfare (Bratt 2009). Beyond the scope of our analysis, we speculate that exotic rats and mice are more likely observed indoors and less likely to be perceived as part of the local environment than other species, which contributes to widespread control behavior of them.
The carnivores of interest (coyote, bobcat, mountain lion, kit fox) were among the species targeted for control. Three (of 15) respondents who reported targeting those species also reported use of anticoagulants; all were from SAMO and all targeted coyotes (Morzillo unpublished data). Collectively, those respondents reside in 15-to 22-year-old single-family structures among varying development densities and distances from open space and place control products both outdoors and indoors. Indoor and outdoor observation of pests and property damage initiated control for all three respondents, who were minimally or not concerned about nontarget effects on wildlife. Our integration of survey and ecological data allow us to observe potential pathways for possible contact between carnivores and anticoagulants, but conclusions about resident misuse or intentional use of anticoagulants to harm carnivores are beyond the scope of our data. Also, although anticoagulant use and targeting of carnivores corresponds with areas of observed impacts (Riley et al. 2007, McMillin et al. 2008, we cannot make causal inferences linking specific locations of use, target, and non-target effects. Addressing our second objective, particular infrastructure characteristics are linked to a greater likelihood of control behavior. Similar to our first hypothesis, building age did not affect likeliness to control. Similar to our second hypothesis, those among single-family structures were more apt to control than those among multiple-family structures. Response bias toward single-family homes likely exists because in both locations the proportion of both homeowners and residents of single-family structures was greater than actual (U.S. Census Bureau data). Underrepresentation of multiple-family units may be attributed to property managers and landlords assuming pest control (Morzillo and Mertig 2011b). Regardless, factors beyond building characteristics affect control behavior.
For both locations, our data supported our third hypothesis (more control among lowerdensity development), but our fourth hypothesis (more control closer to open space) was supported only in SAMO. We speculate that landscape configuration likely affects patterns of control. Geographic (Redman 1999) and gradient characteristics (McDonnell et al. 1997, Dow 2000 are phenomena of urban systems. Generally, inverse relationships exist between species richness and distance from urban core (Blair and Launer 1997, Clergeau et al. 1998, Germaine and Wakeling 2001, Williams et al. 2005, McKinney 2008. Although a gradient-related biodiversity analysis was beyond the scope of our objectives, human responses to interactions with ecological phenomena may occur at both fine and landscape scales. Although targeting mice and rats was widespread, spatial variation existed in control of other species. Community-level diversity often is greatest at intermediate levels of residential development (Blair and Launer 1997, Blair 2004, Buchans and Thompson 2006. In Bakersfield, control behavior for squirrels may operate as a function of land cover, whereas both land cover and land use are factors in SAMO (Theobold 2004). Bakersfield is topographically flat with development in gridded patterns and limited natural open space. Scrubland and intensive agriculture surround Bakersfield, whereas urban residential areas contain squirrel habitat features including canopy cover (Williamson 1983, McPherson and Nilon 1987, Bowers and Breland 1996, Gurnell et al. 2002, Nelson et al. 2005) and water (Peterson et al. 1999) not as available elsewhere. Therefore, in Bakersfield, residential areas may function as habitat islands (Gehrt and Chelsvig 2004), from which squirrels have difficulty dispersing (Angold et al. 2006). Comparatively, for SAMO, landscape influences distribution based on both topography and v www.esajournals.org ownership. Natural areas of SAMO contains both chaparral and wooded riparian zones, and the interspersion of both may aid in squirrel movement out of, yet aid gophers and rabbits movement into, residential areas.
Thus, we suspect that fine-scale features affect spatial variation in control for native wildlife, as well. Fraterrigo and Wiens (2005) suggested that building density as well as fine-scale habitat features affect species persistence. Fine-scale features may include vegetation cover (Hennings and Edge 2003, Chamberlain et al. 2004, Atwood et al. 2004) and complexity (Germaine andWakeling 2001, Crooks et al. 2004), as well as berry-producing shrubs (Melles et al. 2003), birdfeeders (Fuller et al. 2008, and den sites and available foraging area (Randa and Yunger 2006). In our study areas, the propensity of nonnative lush vegetation and fruit trees in yards likely attract wildlife; resulting damage prompts control. In fact, several survey respondents noted use of control products when yard fruit is ripe, or commented that they have observed pests ''chewing on tree fruit,'' and ''eating fruit off of the trees. Others suggested causes of pest presence such as ''neighbor needs to trim grapefruit trees'' or ''property owners neglect[ing] of existing fruit trees.'' Thus, we suspect that both landscape and fine-scale characteristics both influence wildlife processes that, in turn, result in patterns of human reaction to them.
We also highlight possible influence of socioeconomics on observed trends. For instance, income was related consistently to both distance and density variables. Income has been positively associated with leaf area index in Indiana (Jensen et al. 2004), as well as plant diversity across Phoenix, another desert metropolis containing expanses of non-native vegetation (Hope et al. 2003). We also speculate potential for cognitive dissonance (Festinger 1957), such that those closer to open space in SAMO were more likely to indicate importance of natural features near their residence, yet more likely to participate in control behavior that consequentially may affect the nearby natural features such as wildlife.
Our results suggest a possible feedback loop of interacting ecological and social phenomena (McIntyre et al. 2000) that calls for further exploration. Humans create appealing wildlife habitat, wildlife use this habitat, presence of and actions by wildlife causes human reaction to wildlife, and humans repair and reinforce appealing habitat thereby attracting wildlife. We speculate whether further exploration of the social pattern and processes of pest control may reveal similar patterns of ecological processes at both local and landscape scales (Melles et al. 2003, Chamberlain et al. 2004, Sparks et al. 2005, Herrmann et al. 2010), but intra-site control patterns reflect landscape heterogeneity. From a policy and management perspective, such human-wildlife conflict may be resolved by encouraging residents in lower density areas and near open space to adhere to manufacturer guidelines for pest control product use and minimizing presence of attractants on their property.

ACKNOWLEDGMENTS
We thank B. Cypher, C. Davis, A. Fairbrother, J. Hagar, D. Kamradt, A.G. Mertig, E. Minor, D. Olszyk, S.P.D. Riley, R. Sauvajot, M. Santelmann, one anonymous reviewer, and all of the California residents who completed the survey. The information in this document has been funded in part by the U.S. Environmental Protection Agency. It has been subjected to review by the National Health and Environmental Effects Research Laboratory and approved for publication. Approval does not signify that the contents reflect the views of the Agency, nor does mention of trade names or commercial products constitute endorsement or recommendation for use. This is contribution number WED-11-091 of the Western Ecology Division.