Article
A Geospatial Mixed Methods Approach to Assessing Campus Safety
Evaluation Review 2013, Vol. 37(5) 347-369 ª The Author(s) 2013 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0193841X13509815 erx.sagepub.com
Lisle S. Hites1,2, Matthew Fifolt2, Heidi Beck2, Wei Su2, Shatomi Kerbawy2, Jessica Wakelee2, and Ariann Nassel2
Abstract Background: While there is no panacea for alleviating campus safety concerns, safety experts agree that one of the key components to an effective campus security plan is monitoring the environment. Despite previous attempts to measure campus safety, quantifying perceptions of fear, safety, and risk remains a challenging issue. Since perceptions of safety and incidents of crime do not necessarily mirror one another, both were utilized in this investigation. Purpose: The purpose of this article is to describe an innovative, mixed methods approach for assessing campus safety at a large, urban campus in the southeast region of the United States. Method: A concurrent triangulation design was implemented to allow investigators the opportunity to collect qualitative and quantitative data simultaneously and
1
Department of Health Care Organization and Policy, School of Public Health, University of Alabama at Birmingham, Birmingham, AL, USA 2 School of Public Health, Center for the Study of Community Health, University of Alabama at Birmingham, Birmingham, AL, USA Corresponding Author: Lisle S. Hites, Department of Health Care Organization and Policy, School of Public Health, University of Alabama at Birmingham, 1665 University Boulevard, Birmingham, AL 35233, USA. Email: [email protected]
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
348
Evaluation Review 37(5)
integrate results in the interpretation phase. Data were collected from four distinct sources of information. Results: Student focus groups yielded data regarding perceptions of risk, and kernel density analysis was used to identify ‘‘hot spots’’ of campus crime incidents. Conclusion: While in many cases perceived risk and actual crime incidents were associated, incidents of hot spots of each type occurred independently with such frequency that an overall correlation of the two was not significant. Accordingly, while no significant correlation between perceived risk and crime incidents was confirmed statistically, the geospatial integration of these data suggested three types of safety conditions. Further, the combination of focus group data and spatial analyses provided a more comprehensive and, therefore, more complete understanding of the multifaceted issues related to campus safety. Keywords campus safety, geographic information system (GIS), mixed methods, perceived risk
Introduction In the United States, there are approximately 4,500 institutions of postsecondary education (Bok 2013). Higher education in the United States is comprised of research universities, comprehensive universities, liberal arts colleges, and public and private community colleges that serve approximately 21 million students annually (Hussar and Bailey 2011). In order to attract and retain these students, colleges and universities must identify and meet the expectations of students and their parents, including expectations for a safe and secure campus environment (Elliott 2002–2003; Sells 2002). Contemporary investigators have used a variety of methods to examine aspects of campus safety, including perceptions of safety, with the majority of studies employing survey research to report correlational and descriptive statistics (Blair et al. 2004; Fisher and May 2009; Janosik 2004; Jennings, Gover, and Purdrynska 2007; Reed and Ainsworth 2007; Tomsich, Gover, and Jennings 2011; Wilcox, Jordan, and Pritchard 2007; Woolnough 2009). Alternative methods for gathering data have included (a) focus groups and interviews (Kelly 2003; Kelly and Torres 2006; Ratti 2010; Starkweather 2007); (b) case studies (Danis 2007; Rader, Cossman, and Allison 2009);
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
349
and (c) historical reviews of safety and policy literature (Carter and Bath 2007; Gregory and Janosik 2007; Rasmussen, Johnson, and Midwestern Higher Education 2008). Despite previous attempts to measure campus safety, quantifying perceptions of fear, safety, and risk remains a challenging issue (Jennings, Gover, and Pudrzynska 2007). Researchers have also noted that reconciling perceptions of safety with actual incidents of crime on college campuses can also be difficult due to institutional underreporting or selectively reporting specific types of crimes (Fisher et al. 2002; Wilcox, Jordan, and Pritchard 2007). While there is no panacea for improving campus safety (either perceived or actual), safety experts agree that one of the key components to an effective campus security plan is monitoring the environment (Zdziarski et al. 2007). Fields (2009, { 10) suggests that interviews with stakeholders can ‘‘provide valuable insight into the effectiveness of current physical security measures in place, and how they align with the perceived level of vulnerability.’’ Since perceptions of safety and incidents of crime do not necessarily mirror one another (McGuire et al. 2012), both were utilized in this investigation. This article describes an innovative, mixed methods approach for assessing campus safety at a large, urban campus in the southeast region of the United States. Both methods and findings are presented.
Background and Rationale In spring 2012, university administrators at a research-intensive university in the southeastern United States began an introspective effort to examine student perceptions of safety on and around the university campus. Results from previous national and local surveys (National Collegiate Health Assessment [NCHA], Noel-Levitz Student Satisfaction Inventory [SSI], and Graduating Student Survey [GSS]) indicated that undergraduate students viewed campus safety as a topic of high priority but low student satisfaction. Due to consistently poor ratings of campus safety by students, university administrators communicated the need for a detailed assessment in order to make more informed and effective decisions regarding current and future campus safety initiatives. Therefore, university administrators contracted with internal evaluation experts to assess student perception of risk on campus. To accomplish this task, investigators proposed the use of (1) student focus groups to elicit responses from students regarding perceived risk of unsafe conditions on campus, (2) geospatial statistics to analyze patterns within these data, and (3) campus crime data in order to compare student
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
350
Evaluation Review 37(5)
perceptions of risk with actual reported incidents of crime. Researchers supplemented the original investigation with (4) a crime severity survey in order to add depth to the analysis of crime incidents on campus.
Setting This investigation was conducted at a comprehensive research university and medical complex in the southeast. With an enrollment of more than 17,000 students in liberal arts and sciences, professional schools, university-wide Graduate School, and a medical and research complex, this institution is a large, urban campus situated in the middle of a downtown business district. The physical campus encompasses 86 city blocks and has experienced rapid growth and expansion over the past four decades.
Method A mixed methods approach was used by researchers to examine perceived risk from multiple, complementary perspectives (Brewer and Hunter 1989). By definition, a mixed methods technique involves the collection, analysis, and synthesis of both quantitative and qualitative data to address a specific problem or concern. A concurrent triangulation design was implemented to allow investigators the opportunity to collect qualitative and quantitative data simultaneously and integrate results in the interpretation phase (Creswell et al. 2003). Quantitative and qualitative data were collected at the same point in time and used to enhance one another (Onwuegbuzie and Collins 2007). Investigators identified both quantitative and qualitative data as critical to this study; therefore, the concurrent triangulation design was best suited for this study as it placed equal value on both types of data. To augment the rich, descriptive information from focus groups, investigators employed innovative but often underutilized evaluation tools available in spatial statistics and geographic information systems (GIS; Azzam and Robinson 2013). In addition to its spatial statistical capabilities and ability to combine multiple data sets into a single map, ‘‘GIS offers a visual way of detecting patterns in data that may have remained unnoticed through other traditional methods of analysis’’ (Azzam and Robinson 2013, 208).
Phase I: Perception of Risk on Campus Qualitative data were collected from 10 student focus groups. Student participants were recruited via e-mail and personal invitations by members of
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
351
the Offices of Student Life and Student Housing and Residential Life. Focus group sessions were scheduled and conducted near the end of the spring 2012 semester.
Sample A total of 61 students participated in the campus safety focus groups (n ¼ 61) with an average of six students per session (Krueger and Casey 2000). Investigators used stratified convenience sampling to ensure that the focus groups reflected the diverse student population of the overall university in terms of gender, ethnicity, academic status, and academic major. Further, investigators made efforts to recruit students who represented the various patterns of student living and student traffic at the institution (e.g., students living on and off campus). There were several, self-selected homogenous focus groups including gender (all female), international, and off-campus students.
Interview Protocol Investigators developed a protocol to ensure questions were asked consistently across focus groups. The protocol included seven questions related to student perceptions of general safety risk on campus and in order to identify specific locations on campus perceived by students to be unsafe. Students identified these locations using a mapping exercise technique (described below). Focus group questions included the following: 1. 2. 3. 4.
5. 6. 7.
How safe do you feel walking around [institution]? How does this perception differ at various times throughout the day (e.g., morning, afternoon, and evening)? In general, what would help you feel safer? Take your colored dots and indicate ‘‘hotspots’’ on the map related to your concerns about campus safety (Day and Night spots; see Figure 1). Looking at the identified spots, what can we do to help you feel more secure at each location? Do you have other observations or comments? What is the best way to communicate with you regarding campus safety?
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
352
Evaluation Review 37(5)
Figure 1. Campus safety codes.
Facilitators Six facilitators conducted the campus safety focus groups, including two undergraduate students, two graduate students, one faculty member, and one staff member. Experienced facilitators conducted a frame-ofreference style training session/practice focus group to mitigate idiosyncrasies among facilitators and thereby establish internal consistency among the group (Woehr 1994). In addition to establishing a collective understanding of the process, this training session provided individuals with an opportunity to ask questions and offer recommendations for improving the focus group process.
Data Collection Two facilitators were assigned to each focus group, with one assuming the role of lead facilitator and the other as note taker. Whenever possible, facilitators were matched according to demographic factors in an attempt to improve candidness of responses and prevent participants from feeling uncomfortable among dissimilar interviewers. For example, one focus group consisted of undergraduate, female-only participants. This group was paired with female-only focus group facilitators. Focus group sessions were scheduled for 1 hr each and all notes were recorded by hand. To promote discussion and eliminate barriers to participation, the faculty and staff members agreed to serve as note takers, so that student facilitators could assume the lead role. All of the notes were compiled by investigators and stored electronically on a password-protected server for the purposes of thematic analysis. In addition to discussing the questions as outlined in the interview protocol, a primary data collection tool for student focus groups was a campus mapping exercise. Each student was given a set of color-coded dots and stars. Shapes corresponded to day and night, while colors corresponded
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
353
to varying levels of safety concern based on the basic traffic signal pattern of red, yellow, and green (see Figure 1). For daytime safety concerns, students were instructed to place color-coded dots on campus map locations where they felt unsafe. There were no limits to the number of dots that a student could use throughout this exercise. Once students completed this task, they were given a set of color-coded stars to mark risk concerns at night. In the focus groups, students differentiated day and night by the transition between daytime and evening courses and/or the change from daylight to dusk.
Data Analysis Once all focus group notes were received from the note takers, investigators assigned codes to the qualitative responses based on similar content. Codes were grouped together into categories and then aggregated into broader themes, consistent with Hatch’s (2002, 152) model of inductive analysis in which ‘‘categories emerge from the analysis of the data set as a whole.’’ Emergent codes and themes were managed using a text-to-table application in Microsoft Word. The following five themes were consistent across the 10 focus groups: 1. 2. 3. 4.
5.
Poor communication/lack of communication regarding campus safety; Little or no student knowledge regarding the use of campus safety call boxes; Expressed need/desire for greater campus police presence; The need for improved visual differentiation between locations considered to be ‘‘on campus’’ and those that were considered to be ‘‘off-campus’’; Expressed need/desire for improved lighting on campus.
Verification Techniques Investigators employed a number of techniques to ensure the trustworthiness of the qualitative findings. Specifically, three distinct verification procedures were used, including (a) intercoder agreement; (b) content coding via triangulation of data sources; and (c) descriptive analysis. Intercoder Agreement. Investigators conducted a preliminary analysis of the data to record key ideas and concepts. Codes and themes were verified
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
354
Evaluation Review 37(5)
using an iterative process of examination in order to reach consensus among the three reviewers. Reviewers reached consensus for 95% of codes and themes, far surpassing the 80% standard recommended by Miles and Huberman (1994) for establishing trustworthiness of data. Content Coding via Triangulation of Data Sources. According to Creswell and Miller (2000, 4), ‘‘triangulation . . . is a systematic process of sorting through the data to find common themes or categories by eliminating overlapping areas.’’ From the initial national and campus survey data, investigators synthesized information from focus group notes and the campus map exercise to determine overarching themes as related to students’ perceptions of risk on campus. By using multiple collection methods, reviewers were able to cross-check findings and reach consensus. Descriptive Analysis. Facilitators and note takers gathered detailed notes of students’ voiced perceptions of risk and experiences on campus. Through these stories, students provided details and described various contexts that were absent from previous student assessments (NCHA, SSI, and GSS). Additionally, focus group participants provided a sense of authenticity to the project as key informants who navigate the campus on a daily basis (Creswell and Miller 2000). By analyzing and reporting these descriptive data from focus group participants, researchers were able to represent a comprehensive view of student perception of risk on campus to university administrators.
Geospatial Analysis With assistance from the Office of Campus Planning and a local architectural firm, investigators used ArcGIS 10.0 (ESRI) digital mapping software to create a comprehensive map of the university campus with major streets, buildings, and landmarks within a superimposed reference grid. Student perception of safety risk, expressed as color-coded dots and stars (see Figure 1), were plotted on electronic copies of the campus map for each of the 10 focus groups and then combined into a composite view of the campus based on input from all 10 groups. Hot Spot Analysis. Once all of the points were plotted on the composite map, disparate groupings of perceived risk began to appear. However, it was difficult to discern distinct clusters or patterns solely through visual inspection. Since there are many different spatial statistical methods for identifying areas of concentration or ‘‘hot spots’’ (Everitt 1974; Getis et al. 2000),
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
355
investigators tested multiple approaches to reveal locations identified by students as areas of perceived risk on campus. Ultimately, kernel density analysis (KDA) was selected for its ability to interpolate a smoothed surface from the student-identified points for areas of low to high perceived risk. KDA is a commonly used spatial statistical method that creates a smoothed value which is calculated from a weighted average derived from all of the values. The weighted average is calculated from the probability distribution centered on the kernel (Bailey 1994; Lampe and Hauser 2011). Compared to other methods for spatial analysis (e.g., Gettis-Ord Gi* and point density analysis), KDA minimizes the effect of unbalanced sample distribution by fitting a weight function at each data point. Additionally, the smoothing effect of KDA allows for better data visualization (Maciejewski et al. 2010; Rana and Dykes 2003). Recognizing that perceived risk might vary significantly by the time of day, analyses were stratified by time. Day and night were differentiated by the transition between daytime and evening courses and/or the change from daylight to dusk. Investigators used the Kernel Density tool in ArcGIS 10.0 (ESRI) to calculate the magnitude per unit area of data points (count/square foot) representing students’ perceptions of risk on campus. This tool divided the study extent into a predefined matrix of square cells (10 ft. 10 ft.), and a circular neighborhood, or bandwidth (radius ¼ 200 ft.), around each cell was considered in calculating a density value for that cell. If no points fell within the neighborhood of a particular cell, that cell was assigned a density value of 0. Before creating a kernel density layer, each feature class (dot or star) representing students’ perceptions of risk was weighted based on perceived strength or level of severity (see Figure 1). This weighting system was accounted for within the parameter settings of the Kernel Density tool (ESRI) outlined in Table 1. Analysis of Campus Map. Guided by the focus group mapping activity, areas of magnitude were graphically displayed on the campus map so that focus group comments could be used for side-by-side comparisons (comments vs. dots). Pockets of perceived risk appeared on both daytime and nighttime visualizations of the KDA, although frequency and severity of perceived risk were greater on the nighttime map (see Figure 2). During the daytime hours, hot spots were situated in areas of high vehicular traffic (i.e., busy intersections), remote locations with little pedestrian traffic, and areas in which campus boundaries were unclear. As demonstrated by KDA and corroborated by student comments, perceptions of risk were often associated with locations on the outskirts of campus and areas where individuals perceived by students to pose a threat (e.g., homeless people, panhandlers, vagrants) were typically present.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
356
Evaluation Review 37(5)
Table 1. Kernel Density Tool Parameters. Tool Parameter
Perceived Risk
Population field
Perception level of severity (weighted 3, 2, or 1 based on campus safety codes. See Figure 1.) Output cell size 10 ft. 10 ft. Neighborhood (circle) r ¼ 200 ft. Area units Acres
Crime Incidents Crime severity weighted based on survey results (Recentered z score. See Table 5.) 10 ft. 10 ft. r ¼ 200 ft. Acres
Figure 2. Kernel density of perceived risk on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
Phase II: Crime Incidents on Campus To compare student perception of risk on campus with actual incidents of crime, investigators reviewed crime incident data provided by campus police over a 3-year period 2009–2011 (see Table 2).
Data Analysis Crime incident data were geocoded using the Geocode Addresses tool in ArcGIS 10.0 (ESRI). Street addresses were matched with longitude and latitude coordinates at a 99% success rate with matching scores ranging
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
357
Table 2. Number of Crime Incidents Recorded Per Year. Year
Number of Reported Incidents
2009 (June–December) 2010 (January–December) 2011 (January–December) Total
264 489 468 1,221
between 80 and 100 on a scale of 0–100. Crime incident data included attributes, such as the type of crime; location of crime; and time of day, month, and year. When appropriate, the victim’s gender and affiliation with the institution were also included (i.e., student, faculty/staff, or visitor). Using the Uniform Crime Reports (UCR; U.S. Department of Justice, Federal Bureau of Investigations 2004), incidents of crime were organized by type under the following UCR categories: Part I: Violent Crimes, Part I: Property Crimes, and Part II: All Other Crimes. Crime incidents were also stratified and analyzed based on the time of day the incident occurred. To be consistent with Phase I of this study, day and night were once again differentiated by the transition between daytime and evening courses and/or the change from daylight to dusk. To systematically stratify the crime data, investigators defined daytime hours as 6:00 a.m. to 5:00 p.m. and nighttime hours as 5:01 p.m. to 5:59 a.m. Contrary to perceptions of risk in which students identified a greater number of hot spots at night, actual crime data revealed a higher number of crimes occurred during the day. Larceny, both of misdemeanor and felony offenses, represented the largest percentage of crimes reported within all data (see Figure 3).
Crime-Type Severity Survey The same students who participated in the focus groups were contacted one year later and asked to rate how various types of crimes would make them feel about going to a location on campus in which this crime occurred. Definitions of crimes can be found in Table 3. Response options were spread across an 8-point Likert-type type scale and ranged from 0 (It would not concern me to go to this location) to 8 (I would try to avoid this location). Of the original 61 student participants, 55 students were still in the university e-mail system and 24 (44%) responded to the survey.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
358
Evaluation Review 37(5)
Figure 3. Crime types reported in the 2009–2011 campus police data.
Survey responses were used to assign weights to each crime type in accordance with a z score, which was calculated using the mean and standard deviation of the crime severity survey rating through a linear transformation (see Tables 4 and 5). Since negative weights would cause interpretative problems with hot spot analysis results, z scores were recentered at 3 to ensure a positive weight value for all crime types. This weighting system was a necessary step in adding more depth to the following hot spot analysis of crime incidents occurring on campus.
Geospatial Analysis Hot Spot Analysis. The Kernel Density tool in ArcGIS 10.0 (ESRI) was used to calculate the magnitude per unit area (count/square foot) of crime incidents occurring from June 2009 until December 2011, during daytime and nighttime hours. With the exception of the weighting systems applied to the data, the parameter settings for calculating the kernel density of crime incidents were similar to those used in Phase I of this study (see Table 1). For this calculation, perception level of severity was replaced with crime type severity survey results. Analysis of Campus Map. Once the hot spot areas of crime incidents were graphically displayed on the campus map, pockets of crime intensity appeared on both daytime and nighttime visualizations (see Figure 4). Similar to the findings from Phase I, daytime and nighttime crime incident hot spots were situated in areas of high vehicular traffic (i.e., busy intersections). The hot spots that displayed most intensely occurred near a major hospital located
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
359
Table 3. Crime Definitions. Type of Crime
Definition
Arson
Willful or malicious burning of another’s property (please assume that no individuals were injured) Burglary Unlawful entry of a structure to commit theft or a felony Felony larceny Nonviolent theft of property valued less than US$500 Misdemeanor larceny Nonviolent theft of property valued greater than US$500 Robbery Theft using force/violence or threat of force/ violence Nonforcible sexual offenses Engaging in sexual intercourse with another person who is unable to consent Forcible sexual offenses Engaging in sexual intercourse with another person by force/against their will Murder/nonnegligent murder Willful killing of one human being by another Motor vehicle thefts Theft or attempted theft of a motor vehicle Unlawful breaking and entering Without the consent of the owner, breaking into of a vehicle (UBEV) and entering a vehicle with the intent to commit any felony or theft Pedestrian/vehicle accidents Accidents involving a pedestrian and a vehicle
Table 4. Uniform Crime Reports Categories. Uniform Crime Reports (UCR) Categories Part I: violent crimes
Part I: property crimes
Part II: all other crimes
Type of Crime Murder/nonnegligent murder Forcible sexual offenses Robbery Arson Burglary Misdemeanor larceny Felony larceny Motor vehicle thefts Nonforcible sexual offenses Pedestrian/vehicle accidents UBEV
on campus. Minor differences were observed between day and night crime incident hot spot locations. However, night crime hot spots appeared to be more dispersed across campus than the day crime hot spots.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
360
Evaluation Review 37(5)
Table 5. Crime Weighting. Type of Crime Murder/nonnegligent murder Forcible sexual offenses Nonforcible sexual offenses Robbery Motor vehicle thefts UBEV Burglary Misdemeanor larceny Felony larceny Pedestrian/vehicle accidents Arson
M
SD
Z score
Z score (recentered at 3)
8.42 8.13 7.63 7.46 7.29 7.08 6.63 6.21 6.21 5.63 4.50
1.53 1.65 1.88 2.17 1.94 2.15 2 2.5 2.43 2.2 2.41
1.38 1.13 0.69 0.54 0.4 0.21 0.18 0.54 0.54 1.05 2.03
4.38 4.13 3.69 3.54 3.4 3.21 2.82 2.46 2.46 1.95 0.97
Note. UBEV ¼ unlawful breaking and entering of a vehicle.
Phase III: Integrative Analysis The final phase of this study applied an integrative analysis to (1) look at correlation between students’ perceptions of risk and actual crime incidents on campus and (2) to visually assess each specific incidence of perceived and actual crime incidents to look for incidences of perceived insecurity not supported by crime statistics and crime areas not recognized by perceived insecurity.
Correlation A buffer with a 100-foot radius was applied to each data point representing a crime incident. Within each buffer, the mean kernel density of perceived risk was calculated using the ZonalStatsWOverlaps tool (Clark n.d.) in ArcGIS 10.0 (ESRI). The same technique was used to calculate the mean kernel density value of actual crime. Investigators performed a Pearson’s correlation analysis to find any potential correlations between students’ perceptions of risk and actual crime incidents on campus. In general, perceived risk and actual crime incidents were not associated, r(263) ¼ .09, ns; although the correlation was slightly stronger for night incidents, r(109) ¼ .14, ns, than for day incidents, r(152) ¼ .02, ns.
Integration of Hot Spots To further explore campus safety, hot spots representing perceived risk and actual crime incidents were integrated into one map for analysis (see
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
361
Figure 4. Kernel density of crime incidents on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
Figure 5). Although the overall correlation between perceived risk and crime incidents was not statistically significant, the geospatial integration of these data identified the following three conditions: 1.
2.
3.
Locations of high perceived risk, but low crime incidents. a. Students are unnecessarily concerned as no safety risk appears to be present. b. Priority area for education/media–public relations intervention to alleviate unnecessary concerns. Locations of high crime incidents, but low perceived risk. c. Students are unaware of a potential real threat. d. Highest priority for safety intervention. Locations of high crime incidents and high perceived risk. e. Students are justifiably concerned about a location. f. High priority for safety intervention.
Locations were identified using a three-step procedure: Step 1. Based on calculated z scores, the Reclassify tool in ArcGIS 10.0 (ESRI) was used to recode kernel density raster values of perceived risk and crime incidents into three, equal interval classes indicating low, medium, and high density. All values below 1 were assigned a value of 0. Perceived risk categories were recoded into
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
362
Evaluation Review 37(5)
Figure 5. Integration kernel density hot spots of perceived risk and crime incidents on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
values of 0, 1, 2, and 3, and crime categories were recoded into values of 00, 10, 20, and 30 (see Table 6). Step 2. Using the Raster Calculator tool in ArcGIS 10.0 (ESRI), reclassified perceived risk values were added to the reclassified crime incident values based on their locations on the map. The tool created an integrated layer with both crime incidents and perceived risk on campus. Resulting values of the integration were represented as two-digit numbers in which the tens digit corresponded to the actual crime level and the units digit corresponded to the perceived risk level. Step 3. Finally, the integrated layer was used to identify locations of interest. Locations with values of 30 or 31 were identified as areas with high crime incidents, but low perceived risk. Conversely, locations with values of 03 or 13 represented areas of high perceived risk, but low crime incidents. A value of 33 would have indicated a match of high crime and high perception, although no such matchups were identified in this study.
Study Limitations The current study has several potential limitations that may have implications for generalizability of findings. Students were recruited for focus
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
363
Table 6. Reclassification of Kernel Density Raster Values. Perceived Risk Kernel Density Value
A Geospatial Mixed Methods Approach to Assessing Campus Safety
Evaluation Review 2013, Vol. 37(5) 347-369 ª The Author(s) 2013 Reprints and permission: sagepub.com/journalsPermissions.nav DOI: 10.1177/0193841X13509815 erx.sagepub.com
Lisle S. Hites1,2, Matthew Fifolt2, Heidi Beck2, Wei Su2, Shatomi Kerbawy2, Jessica Wakelee2, and Ariann Nassel2
Abstract Background: While there is no panacea for alleviating campus safety concerns, safety experts agree that one of the key components to an effective campus security plan is monitoring the environment. Despite previous attempts to measure campus safety, quantifying perceptions of fear, safety, and risk remains a challenging issue. Since perceptions of safety and incidents of crime do not necessarily mirror one another, both were utilized in this investigation. Purpose: The purpose of this article is to describe an innovative, mixed methods approach for assessing campus safety at a large, urban campus in the southeast region of the United States. Method: A concurrent triangulation design was implemented to allow investigators the opportunity to collect qualitative and quantitative data simultaneously and
1
Department of Health Care Organization and Policy, School of Public Health, University of Alabama at Birmingham, Birmingham, AL, USA 2 School of Public Health, Center for the Study of Community Health, University of Alabama at Birmingham, Birmingham, AL, USA Corresponding Author: Lisle S. Hites, Department of Health Care Organization and Policy, School of Public Health, University of Alabama at Birmingham, 1665 University Boulevard, Birmingham, AL 35233, USA. Email: [email protected]
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
348
Evaluation Review 37(5)
integrate results in the interpretation phase. Data were collected from four distinct sources of information. Results: Student focus groups yielded data regarding perceptions of risk, and kernel density analysis was used to identify ‘‘hot spots’’ of campus crime incidents. Conclusion: While in many cases perceived risk and actual crime incidents were associated, incidents of hot spots of each type occurred independently with such frequency that an overall correlation of the two was not significant. Accordingly, while no significant correlation between perceived risk and crime incidents was confirmed statistically, the geospatial integration of these data suggested three types of safety conditions. Further, the combination of focus group data and spatial analyses provided a more comprehensive and, therefore, more complete understanding of the multifaceted issues related to campus safety. Keywords campus safety, geographic information system (GIS), mixed methods, perceived risk
Introduction In the United States, there are approximately 4,500 institutions of postsecondary education (Bok 2013). Higher education in the United States is comprised of research universities, comprehensive universities, liberal arts colleges, and public and private community colleges that serve approximately 21 million students annually (Hussar and Bailey 2011). In order to attract and retain these students, colleges and universities must identify and meet the expectations of students and their parents, including expectations for a safe and secure campus environment (Elliott 2002–2003; Sells 2002). Contemporary investigators have used a variety of methods to examine aspects of campus safety, including perceptions of safety, with the majority of studies employing survey research to report correlational and descriptive statistics (Blair et al. 2004; Fisher and May 2009; Janosik 2004; Jennings, Gover, and Purdrynska 2007; Reed and Ainsworth 2007; Tomsich, Gover, and Jennings 2011; Wilcox, Jordan, and Pritchard 2007; Woolnough 2009). Alternative methods for gathering data have included (a) focus groups and interviews (Kelly 2003; Kelly and Torres 2006; Ratti 2010; Starkweather 2007); (b) case studies (Danis 2007; Rader, Cossman, and Allison 2009);
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
349
and (c) historical reviews of safety and policy literature (Carter and Bath 2007; Gregory and Janosik 2007; Rasmussen, Johnson, and Midwestern Higher Education 2008). Despite previous attempts to measure campus safety, quantifying perceptions of fear, safety, and risk remains a challenging issue (Jennings, Gover, and Pudrzynska 2007). Researchers have also noted that reconciling perceptions of safety with actual incidents of crime on college campuses can also be difficult due to institutional underreporting or selectively reporting specific types of crimes (Fisher et al. 2002; Wilcox, Jordan, and Pritchard 2007). While there is no panacea for improving campus safety (either perceived or actual), safety experts agree that one of the key components to an effective campus security plan is monitoring the environment (Zdziarski et al. 2007). Fields (2009, { 10) suggests that interviews with stakeholders can ‘‘provide valuable insight into the effectiveness of current physical security measures in place, and how they align with the perceived level of vulnerability.’’ Since perceptions of safety and incidents of crime do not necessarily mirror one another (McGuire et al. 2012), both were utilized in this investigation. This article describes an innovative, mixed methods approach for assessing campus safety at a large, urban campus in the southeast region of the United States. Both methods and findings are presented.
Background and Rationale In spring 2012, university administrators at a research-intensive university in the southeastern United States began an introspective effort to examine student perceptions of safety on and around the university campus. Results from previous national and local surveys (National Collegiate Health Assessment [NCHA], Noel-Levitz Student Satisfaction Inventory [SSI], and Graduating Student Survey [GSS]) indicated that undergraduate students viewed campus safety as a topic of high priority but low student satisfaction. Due to consistently poor ratings of campus safety by students, university administrators communicated the need for a detailed assessment in order to make more informed and effective decisions regarding current and future campus safety initiatives. Therefore, university administrators contracted with internal evaluation experts to assess student perception of risk on campus. To accomplish this task, investigators proposed the use of (1) student focus groups to elicit responses from students regarding perceived risk of unsafe conditions on campus, (2) geospatial statistics to analyze patterns within these data, and (3) campus crime data in order to compare student
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
350
Evaluation Review 37(5)
perceptions of risk with actual reported incidents of crime. Researchers supplemented the original investigation with (4) a crime severity survey in order to add depth to the analysis of crime incidents on campus.
Setting This investigation was conducted at a comprehensive research university and medical complex in the southeast. With an enrollment of more than 17,000 students in liberal arts and sciences, professional schools, university-wide Graduate School, and a medical and research complex, this institution is a large, urban campus situated in the middle of a downtown business district. The physical campus encompasses 86 city blocks and has experienced rapid growth and expansion over the past four decades.
Method A mixed methods approach was used by researchers to examine perceived risk from multiple, complementary perspectives (Brewer and Hunter 1989). By definition, a mixed methods technique involves the collection, analysis, and synthesis of both quantitative and qualitative data to address a specific problem or concern. A concurrent triangulation design was implemented to allow investigators the opportunity to collect qualitative and quantitative data simultaneously and integrate results in the interpretation phase (Creswell et al. 2003). Quantitative and qualitative data were collected at the same point in time and used to enhance one another (Onwuegbuzie and Collins 2007). Investigators identified both quantitative and qualitative data as critical to this study; therefore, the concurrent triangulation design was best suited for this study as it placed equal value on both types of data. To augment the rich, descriptive information from focus groups, investigators employed innovative but often underutilized evaluation tools available in spatial statistics and geographic information systems (GIS; Azzam and Robinson 2013). In addition to its spatial statistical capabilities and ability to combine multiple data sets into a single map, ‘‘GIS offers a visual way of detecting patterns in data that may have remained unnoticed through other traditional methods of analysis’’ (Azzam and Robinson 2013, 208).
Phase I: Perception of Risk on Campus Qualitative data were collected from 10 student focus groups. Student participants were recruited via e-mail and personal invitations by members of
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
351
the Offices of Student Life and Student Housing and Residential Life. Focus group sessions were scheduled and conducted near the end of the spring 2012 semester.
Sample A total of 61 students participated in the campus safety focus groups (n ¼ 61) with an average of six students per session (Krueger and Casey 2000). Investigators used stratified convenience sampling to ensure that the focus groups reflected the diverse student population of the overall university in terms of gender, ethnicity, academic status, and academic major. Further, investigators made efforts to recruit students who represented the various patterns of student living and student traffic at the institution (e.g., students living on and off campus). There were several, self-selected homogenous focus groups including gender (all female), international, and off-campus students.
Interview Protocol Investigators developed a protocol to ensure questions were asked consistently across focus groups. The protocol included seven questions related to student perceptions of general safety risk on campus and in order to identify specific locations on campus perceived by students to be unsafe. Students identified these locations using a mapping exercise technique (described below). Focus group questions included the following: 1. 2. 3. 4.
5. 6. 7.
How safe do you feel walking around [institution]? How does this perception differ at various times throughout the day (e.g., morning, afternoon, and evening)? In general, what would help you feel safer? Take your colored dots and indicate ‘‘hotspots’’ on the map related to your concerns about campus safety (Day and Night spots; see Figure 1). Looking at the identified spots, what can we do to help you feel more secure at each location? Do you have other observations or comments? What is the best way to communicate with you regarding campus safety?
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
352
Evaluation Review 37(5)
Figure 1. Campus safety codes.
Facilitators Six facilitators conducted the campus safety focus groups, including two undergraduate students, two graduate students, one faculty member, and one staff member. Experienced facilitators conducted a frame-ofreference style training session/practice focus group to mitigate idiosyncrasies among facilitators and thereby establish internal consistency among the group (Woehr 1994). In addition to establishing a collective understanding of the process, this training session provided individuals with an opportunity to ask questions and offer recommendations for improving the focus group process.
Data Collection Two facilitators were assigned to each focus group, with one assuming the role of lead facilitator and the other as note taker. Whenever possible, facilitators were matched according to demographic factors in an attempt to improve candidness of responses and prevent participants from feeling uncomfortable among dissimilar interviewers. For example, one focus group consisted of undergraduate, female-only participants. This group was paired with female-only focus group facilitators. Focus group sessions were scheduled for 1 hr each and all notes were recorded by hand. To promote discussion and eliminate barriers to participation, the faculty and staff members agreed to serve as note takers, so that student facilitators could assume the lead role. All of the notes were compiled by investigators and stored electronically on a password-protected server for the purposes of thematic analysis. In addition to discussing the questions as outlined in the interview protocol, a primary data collection tool for student focus groups was a campus mapping exercise. Each student was given a set of color-coded dots and stars. Shapes corresponded to day and night, while colors corresponded
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
353
to varying levels of safety concern based on the basic traffic signal pattern of red, yellow, and green (see Figure 1). For daytime safety concerns, students were instructed to place color-coded dots on campus map locations where they felt unsafe. There were no limits to the number of dots that a student could use throughout this exercise. Once students completed this task, they were given a set of color-coded stars to mark risk concerns at night. In the focus groups, students differentiated day and night by the transition between daytime and evening courses and/or the change from daylight to dusk.
Data Analysis Once all focus group notes were received from the note takers, investigators assigned codes to the qualitative responses based on similar content. Codes were grouped together into categories and then aggregated into broader themes, consistent with Hatch’s (2002, 152) model of inductive analysis in which ‘‘categories emerge from the analysis of the data set as a whole.’’ Emergent codes and themes were managed using a text-to-table application in Microsoft Word. The following five themes were consistent across the 10 focus groups: 1. 2. 3. 4.
5.
Poor communication/lack of communication regarding campus safety; Little or no student knowledge regarding the use of campus safety call boxes; Expressed need/desire for greater campus police presence; The need for improved visual differentiation between locations considered to be ‘‘on campus’’ and those that were considered to be ‘‘off-campus’’; Expressed need/desire for improved lighting on campus.
Verification Techniques Investigators employed a number of techniques to ensure the trustworthiness of the qualitative findings. Specifically, three distinct verification procedures were used, including (a) intercoder agreement; (b) content coding via triangulation of data sources; and (c) descriptive analysis. Intercoder Agreement. Investigators conducted a preliminary analysis of the data to record key ideas and concepts. Codes and themes were verified
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
354
Evaluation Review 37(5)
using an iterative process of examination in order to reach consensus among the three reviewers. Reviewers reached consensus for 95% of codes and themes, far surpassing the 80% standard recommended by Miles and Huberman (1994) for establishing trustworthiness of data. Content Coding via Triangulation of Data Sources. According to Creswell and Miller (2000, 4), ‘‘triangulation . . . is a systematic process of sorting through the data to find common themes or categories by eliminating overlapping areas.’’ From the initial national and campus survey data, investigators synthesized information from focus group notes and the campus map exercise to determine overarching themes as related to students’ perceptions of risk on campus. By using multiple collection methods, reviewers were able to cross-check findings and reach consensus. Descriptive Analysis. Facilitators and note takers gathered detailed notes of students’ voiced perceptions of risk and experiences on campus. Through these stories, students provided details and described various contexts that were absent from previous student assessments (NCHA, SSI, and GSS). Additionally, focus group participants provided a sense of authenticity to the project as key informants who navigate the campus on a daily basis (Creswell and Miller 2000). By analyzing and reporting these descriptive data from focus group participants, researchers were able to represent a comprehensive view of student perception of risk on campus to university administrators.
Geospatial Analysis With assistance from the Office of Campus Planning and a local architectural firm, investigators used ArcGIS 10.0 (ESRI) digital mapping software to create a comprehensive map of the university campus with major streets, buildings, and landmarks within a superimposed reference grid. Student perception of safety risk, expressed as color-coded dots and stars (see Figure 1), were plotted on electronic copies of the campus map for each of the 10 focus groups and then combined into a composite view of the campus based on input from all 10 groups. Hot Spot Analysis. Once all of the points were plotted on the composite map, disparate groupings of perceived risk began to appear. However, it was difficult to discern distinct clusters or patterns solely through visual inspection. Since there are many different spatial statistical methods for identifying areas of concentration or ‘‘hot spots’’ (Everitt 1974; Getis et al. 2000),
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
355
investigators tested multiple approaches to reveal locations identified by students as areas of perceived risk on campus. Ultimately, kernel density analysis (KDA) was selected for its ability to interpolate a smoothed surface from the student-identified points for areas of low to high perceived risk. KDA is a commonly used spatial statistical method that creates a smoothed value which is calculated from a weighted average derived from all of the values. The weighted average is calculated from the probability distribution centered on the kernel (Bailey 1994; Lampe and Hauser 2011). Compared to other methods for spatial analysis (e.g., Gettis-Ord Gi* and point density analysis), KDA minimizes the effect of unbalanced sample distribution by fitting a weight function at each data point. Additionally, the smoothing effect of KDA allows for better data visualization (Maciejewski et al. 2010; Rana and Dykes 2003). Recognizing that perceived risk might vary significantly by the time of day, analyses were stratified by time. Day and night were differentiated by the transition between daytime and evening courses and/or the change from daylight to dusk. Investigators used the Kernel Density tool in ArcGIS 10.0 (ESRI) to calculate the magnitude per unit area of data points (count/square foot) representing students’ perceptions of risk on campus. This tool divided the study extent into a predefined matrix of square cells (10 ft. 10 ft.), and a circular neighborhood, or bandwidth (radius ¼ 200 ft.), around each cell was considered in calculating a density value for that cell. If no points fell within the neighborhood of a particular cell, that cell was assigned a density value of 0. Before creating a kernel density layer, each feature class (dot or star) representing students’ perceptions of risk was weighted based on perceived strength or level of severity (see Figure 1). This weighting system was accounted for within the parameter settings of the Kernel Density tool (ESRI) outlined in Table 1. Analysis of Campus Map. Guided by the focus group mapping activity, areas of magnitude were graphically displayed on the campus map so that focus group comments could be used for side-by-side comparisons (comments vs. dots). Pockets of perceived risk appeared on both daytime and nighttime visualizations of the KDA, although frequency and severity of perceived risk were greater on the nighttime map (see Figure 2). During the daytime hours, hot spots were situated in areas of high vehicular traffic (i.e., busy intersections), remote locations with little pedestrian traffic, and areas in which campus boundaries were unclear. As demonstrated by KDA and corroborated by student comments, perceptions of risk were often associated with locations on the outskirts of campus and areas where individuals perceived by students to pose a threat (e.g., homeless people, panhandlers, vagrants) were typically present.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
356
Evaluation Review 37(5)
Table 1. Kernel Density Tool Parameters. Tool Parameter
Perceived Risk
Population field
Perception level of severity (weighted 3, 2, or 1 based on campus safety codes. See Figure 1.) Output cell size 10 ft. 10 ft. Neighborhood (circle) r ¼ 200 ft. Area units Acres
Crime Incidents Crime severity weighted based on survey results (Recentered z score. See Table 5.) 10 ft. 10 ft. r ¼ 200 ft. Acres
Figure 2. Kernel density of perceived risk on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
Phase II: Crime Incidents on Campus To compare student perception of risk on campus with actual incidents of crime, investigators reviewed crime incident data provided by campus police over a 3-year period 2009–2011 (see Table 2).
Data Analysis Crime incident data were geocoded using the Geocode Addresses tool in ArcGIS 10.0 (ESRI). Street addresses were matched with longitude and latitude coordinates at a 99% success rate with matching scores ranging
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
357
Table 2. Number of Crime Incidents Recorded Per Year. Year
Number of Reported Incidents
2009 (June–December) 2010 (January–December) 2011 (January–December) Total
264 489 468 1,221
between 80 and 100 on a scale of 0–100. Crime incident data included attributes, such as the type of crime; location of crime; and time of day, month, and year. When appropriate, the victim’s gender and affiliation with the institution were also included (i.e., student, faculty/staff, or visitor). Using the Uniform Crime Reports (UCR; U.S. Department of Justice, Federal Bureau of Investigations 2004), incidents of crime were organized by type under the following UCR categories: Part I: Violent Crimes, Part I: Property Crimes, and Part II: All Other Crimes. Crime incidents were also stratified and analyzed based on the time of day the incident occurred. To be consistent with Phase I of this study, day and night were once again differentiated by the transition between daytime and evening courses and/or the change from daylight to dusk. To systematically stratify the crime data, investigators defined daytime hours as 6:00 a.m. to 5:00 p.m. and nighttime hours as 5:01 p.m. to 5:59 a.m. Contrary to perceptions of risk in which students identified a greater number of hot spots at night, actual crime data revealed a higher number of crimes occurred during the day. Larceny, both of misdemeanor and felony offenses, represented the largest percentage of crimes reported within all data (see Figure 3).
Crime-Type Severity Survey The same students who participated in the focus groups were contacted one year later and asked to rate how various types of crimes would make them feel about going to a location on campus in which this crime occurred. Definitions of crimes can be found in Table 3. Response options were spread across an 8-point Likert-type type scale and ranged from 0 (It would not concern me to go to this location) to 8 (I would try to avoid this location). Of the original 61 student participants, 55 students were still in the university e-mail system and 24 (44%) responded to the survey.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
358
Evaluation Review 37(5)
Figure 3. Crime types reported in the 2009–2011 campus police data.
Survey responses were used to assign weights to each crime type in accordance with a z score, which was calculated using the mean and standard deviation of the crime severity survey rating through a linear transformation (see Tables 4 and 5). Since negative weights would cause interpretative problems with hot spot analysis results, z scores were recentered at 3 to ensure a positive weight value for all crime types. This weighting system was a necessary step in adding more depth to the following hot spot analysis of crime incidents occurring on campus.
Geospatial Analysis Hot Spot Analysis. The Kernel Density tool in ArcGIS 10.0 (ESRI) was used to calculate the magnitude per unit area (count/square foot) of crime incidents occurring from June 2009 until December 2011, during daytime and nighttime hours. With the exception of the weighting systems applied to the data, the parameter settings for calculating the kernel density of crime incidents were similar to those used in Phase I of this study (see Table 1). For this calculation, perception level of severity was replaced with crime type severity survey results. Analysis of Campus Map. Once the hot spot areas of crime incidents were graphically displayed on the campus map, pockets of crime intensity appeared on both daytime and nighttime visualizations (see Figure 4). Similar to the findings from Phase I, daytime and nighttime crime incident hot spots were situated in areas of high vehicular traffic (i.e., busy intersections). The hot spots that displayed most intensely occurred near a major hospital located
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
359
Table 3. Crime Definitions. Type of Crime
Definition
Arson
Willful or malicious burning of another’s property (please assume that no individuals were injured) Burglary Unlawful entry of a structure to commit theft or a felony Felony larceny Nonviolent theft of property valued less than US$500 Misdemeanor larceny Nonviolent theft of property valued greater than US$500 Robbery Theft using force/violence or threat of force/ violence Nonforcible sexual offenses Engaging in sexual intercourse with another person who is unable to consent Forcible sexual offenses Engaging in sexual intercourse with another person by force/against their will Murder/nonnegligent murder Willful killing of one human being by another Motor vehicle thefts Theft or attempted theft of a motor vehicle Unlawful breaking and entering Without the consent of the owner, breaking into of a vehicle (UBEV) and entering a vehicle with the intent to commit any felony or theft Pedestrian/vehicle accidents Accidents involving a pedestrian and a vehicle
Table 4. Uniform Crime Reports Categories. Uniform Crime Reports (UCR) Categories Part I: violent crimes
Part I: property crimes
Part II: all other crimes
Type of Crime Murder/nonnegligent murder Forcible sexual offenses Robbery Arson Burglary Misdemeanor larceny Felony larceny Motor vehicle thefts Nonforcible sexual offenses Pedestrian/vehicle accidents UBEV
on campus. Minor differences were observed between day and night crime incident hot spot locations. However, night crime hot spots appeared to be more dispersed across campus than the day crime hot spots.
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
360
Evaluation Review 37(5)
Table 5. Crime Weighting. Type of Crime Murder/nonnegligent murder Forcible sexual offenses Nonforcible sexual offenses Robbery Motor vehicle thefts UBEV Burglary Misdemeanor larceny Felony larceny Pedestrian/vehicle accidents Arson
M
SD
Z score
Z score (recentered at 3)
8.42 8.13 7.63 7.46 7.29 7.08 6.63 6.21 6.21 5.63 4.50
1.53 1.65 1.88 2.17 1.94 2.15 2 2.5 2.43 2.2 2.41
1.38 1.13 0.69 0.54 0.4 0.21 0.18 0.54 0.54 1.05 2.03
4.38 4.13 3.69 3.54 3.4 3.21 2.82 2.46 2.46 1.95 0.97
Note. UBEV ¼ unlawful breaking and entering of a vehicle.
Phase III: Integrative Analysis The final phase of this study applied an integrative analysis to (1) look at correlation between students’ perceptions of risk and actual crime incidents on campus and (2) to visually assess each specific incidence of perceived and actual crime incidents to look for incidences of perceived insecurity not supported by crime statistics and crime areas not recognized by perceived insecurity.
Correlation A buffer with a 100-foot radius was applied to each data point representing a crime incident. Within each buffer, the mean kernel density of perceived risk was calculated using the ZonalStatsWOverlaps tool (Clark n.d.) in ArcGIS 10.0 (ESRI). The same technique was used to calculate the mean kernel density value of actual crime. Investigators performed a Pearson’s correlation analysis to find any potential correlations between students’ perceptions of risk and actual crime incidents on campus. In general, perceived risk and actual crime incidents were not associated, r(263) ¼ .09, ns; although the correlation was slightly stronger for night incidents, r(109) ¼ .14, ns, than for day incidents, r(152) ¼ .02, ns.
Integration of Hot Spots To further explore campus safety, hot spots representing perceived risk and actual crime incidents were integrated into one map for analysis (see
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
361
Figure 4. Kernel density of crime incidents on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
Figure 5). Although the overall correlation between perceived risk and crime incidents was not statistically significant, the geospatial integration of these data identified the following three conditions: 1.
2.
3.
Locations of high perceived risk, but low crime incidents. a. Students are unnecessarily concerned as no safety risk appears to be present. b. Priority area for education/media–public relations intervention to alleviate unnecessary concerns. Locations of high crime incidents, but low perceived risk. c. Students are unaware of a potential real threat. d. Highest priority for safety intervention. Locations of high crime incidents and high perceived risk. e. Students are justifiably concerned about a location. f. High priority for safety intervention.
Locations were identified using a three-step procedure: Step 1. Based on calculated z scores, the Reclassify tool in ArcGIS 10.0 (ESRI) was used to recode kernel density raster values of perceived risk and crime incidents into three, equal interval classes indicating low, medium, and high density. All values below 1 were assigned a value of 0. Perceived risk categories were recoded into
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
362
Evaluation Review 37(5)
Figure 5. Integration kernel density hot spots of perceived risk and crime incidents on campus. (A) Day (6:00 a.m. to 5:00 p.m.). (B) Night (5:01 p.m. to 5:59 a.m.).
values of 0, 1, 2, and 3, and crime categories were recoded into values of 00, 10, 20, and 30 (see Table 6). Step 2. Using the Raster Calculator tool in ArcGIS 10.0 (ESRI), reclassified perceived risk values were added to the reclassified crime incident values based on their locations on the map. The tool created an integrated layer with both crime incidents and perceived risk on campus. Resulting values of the integration were represented as two-digit numbers in which the tens digit corresponded to the actual crime level and the units digit corresponded to the perceived risk level. Step 3. Finally, the integrated layer was used to identify locations of interest. Locations with values of 30 or 31 were identified as areas with high crime incidents, but low perceived risk. Conversely, locations with values of 03 or 13 represented areas of high perceived risk, but low crime incidents. A value of 33 would have indicated a match of high crime and high perception, although no such matchups were identified in this study.
Study Limitations The current study has several potential limitations that may have implications for generalizability of findings. Students were recruited for focus
Downloaded from erx.sagepub.com at UNIV OF PITTSBURGH on March 30, 2015
Hites et al.
363
Table 6. Reclassification of Kernel Density Raster Values. Perceived Risk Kernel Density Value
