Twenty urban elementary schoolyards in Cleveland, OH were included in the study. These schoolyards were selected as part of an evaluation study of schoolyard renovations [20
]. As part of this renovation, titled School Grounds as Community Parks, schoolyards received new playground equipment and safety and site improvements. Ten schools that had been renovated for at least a year were included along with ten unrenovated schools. All schools had functional playgrounds regardless of renovation status and were neighborhood schools, meaning those who attended the school lived in the surrounding area (i.e., non-magnet school).
Utilization and activity levels of the users were measured by direct observation using the System for Observing Play and Leisure Activity in Youth (SOPLAY) [21
]. Each schoolyard was observed on ten occasions outside of school time spanning evening (4:30-6:00 pm or 6:30-8:00 pm) and weekend times (10:30-12:00 am, 12:30-2:00 pm, or 2:30-4:00 pm). Features of the playground were documented using the Environmental Assessment of Public Recreation Spaces (EAPRS) [22
] one month prior to the direct observations, which took place May-July, 2005.
At the time of the study, the city's 414,500 residents were comprised of 54% African American residents, 39% White residents, 2% Asian residents and 5% residents of other races. Just under 8% of residents classified themselves as Hispanic, regardless of race. In the census tracts of the schools included in this study, an average 39% of the children under 17 lived in poverty. The participants of this study are those who came to the schoolyards during the observation periods. This protocol was reviewed by the MetroHealth Institutional Review Board and deemed exempt.
Portions of the EAPRS measurement tool that focused on the playground features and amenities were completed at each of the schoolyards including the following sections: play set and structure features, other components (separate from play set), general areas, eating/drinking features, sitting or resting features (non-trail), general aesthetics, directives and information-related features. The first author was trained on the instrument by the developer of the EAPRS. The first author in turn trained two observers. Both observers completed the EAPRS at each school site. Ninety percent of the items had good-excellent reliability and/or high percent agreement values [22
]. Any item or scale with a reliability value below 0.7 was not included in this analysis.
Using the data collected from the EAPRS tool, 14 summary exposure variables were considered. Two of the variables represented counts of the features. Total unique types of play equipment was a count of the following 10 features: presence of things to hang from, things to slide down, things to climb up/through, things to stand/walk on, swings, climbers/spin on, blacktop game, spring toy, imaginary play structure, and play panel. This variable ranged from 0, indicating none of these features were present, to 10, indicating the playground had all these features (see Table for variable distribution information). Total number of play features was a summation of the total number of each of the features mentioned above with multiple credit when a playground had more than one feature (i.e., 3 slides). This variable could range from 0-∞.
Descriptive characteristics of playground features, schools, and neighborhoods
Four of the variables represented the condition or safety of the features. Overall condition was developed by taking the mean condition of up to nine playground features (the condition of play panels is not included in the EAPRS). Condition assessed the functionality as well as the quality of the features including the presence of dents, sharp edges, rust, damage, holes, cracks, and ditches of the features. Condition ratings varied from 1 to 3 where 1 indicates poor condition and 3 indicates excellent condition. Overall cleanliness was developed by taking the mean cleanliness of up to 6 playground features (cleanliness of things to hang from, play panel, blacktop, and spring toy is not included in the EAPRS). Cleanliness took into consideration the presence of dirt, debris, trash, and paint quality of the features. Cleanliness ratings varied from 1 to 3 where 1 indicates not at all clean and 3 indicates mostly to extremely clean.
Overall quality and overall safety were created based on proposed variables in the EAPRS scoring sheet. To create these variables, certain attributes of each feature were dichotomously coded and the mean of these attributes was calculated. For quality, attributes considered for any type of equipment usually included condition, cleanliness, presence of 'special features', and appropriate draining features if applicable. For example, to rate the quality of things to stand or walk on, condition and cleanliness were given one point if they were rated as 3 and given 0 points if they were rated 1 or 2. A point was given if there was a bridge (a special feature); otherwise it was coded as zero. The score for the quality of things to stand or walk on was calculated as the mean of condition (0,1), cleanliness (0,1) and presence of a bridge (0,1). For swings, a point was given if condition and cleanliness were rated as 3, one point if there was proper ground drainage and 1 point if there was more than one swing type (in the absence of these features it was scored a 0). The quality of the swings was a mean of these four attributes. The overall quality of the playground was developed by taking mean quality score for all 9 features (play panels were not rated) and it ranged from 0-1. The safety variable was similarly constructed with various attributes of features coded dichotomously and averaged. Then the mean safety values across each of the nine features were averaged. Safety attributes generally included height off the ground and having a soft landing or railings as appropriate. This variable ranged from 0-1.
Three additional variables described the amenities including presence of benches and presence of trash cans, which were both rated as yes or no as well as coverage/shade for resting features, which ranged from 1-3 with 1 being poor coverage and 3 being excellent coverage. Finally, five additional characteristics were coded but were not deemed sufficiently reliable. These included: presence of any lighting at play structures or blacktops; shade/coverage over the general play structure; presence of trash; a summary graffiti variable; and the presence of rules/regulatory signs.
Direct observation of utilization of the schoolyard and physical activity levels of those at the playground was completed using SOPLAY. This system involves scanning specific areas from left to right at a rate of about 1 person per second and noting characteristics of the users within the area. Characteristics documented in our study included gender, age group (adult versus child), and activity level. Participant's activity levels were classified into one of three groups: sedentary, walking (moderately active) or vigorously active. Separate scans were completed for women, men, boys and girls, so that activity levels within each of these categories could be considered. Ultimately all adults were combined regardless of gender because of low utilization by adults.
Schoolyards were divided into target areas that supported uniform activities and could be viewed entirely from one vantage point. The maximum number of target areas at a school was 6 and the average was 3.75. Each schoolyard was visited 10 times for one hour and thirty minutes. At each of the 10 observation sessions at a school, each target area was observed 9 times. Using momentary-time sampling, each target area was scanned for women by observer one and for men by observer two, followed by a second scan for girls by observer one and boys by observer two. Over the course of a ten minute period, each target area would be visited once. This process was completed a total of nine times per target area per observation session. The nine scans of each group (adults, girls and boys) were averaged to provide an average number of attendees during a scan for that group in a target area for that observation session. In total, our study considers 200 observations for adults, girls and boys across the 20 schools (10 observations per school). For the utilization analyses, the average number of persons across all target areas was examined. Hence, the 200 observations are an average of all 9 scans within an observation period and are averaged across all target areas. For the physical activity level analyses, the average of the 9 scans in the playground target area only were examined. Hence, the 200 observations are an average of the 9 scans of the playground target area only within an observation period.
Training for the assessors was provided and all assessors were required to meet reliability criteria [20
]. Training sessions included both classroom lectures taught by the creator of SOPLAY and practice field observations. During the classroom lectures, assessors were shown how to utilize each section of the code sheet and how to discriminate between physical activity levels. Prerecorded video tapes of adolescents being physically active were also used for coding practice during the classroom lectures. In addition, practice field observations were completed over a week. During the study, additional reliability assessments were completed at 15% of the observation periods by having a third assessor complete an independent assessment. Reliabilities were found to be acceptably high (ICC = 0.71-0.97) [20
]. Eleven trained assessors conducted all observations. There were two raters at each of the 200 observations (not including reliability assessments where there were three raters). The mean number of observation sessions per assessor was 36; the range was 20-47 observation sessions.
The outcome for the utilization analyses was utilization of the schoolyard, regardless of target area. Utilization describes the average number of persons per scan in any target area at the schoolyard at each observation session. Utilization by adults, girls and boys are considered separately. The outcome for the physical activity level analyses was the proportion active. For any participant at the schoolyards in the target area that contained the play set, the proportion of those who were either moderately or vigorously active was calculated by summing those in the walking or vigorous category and dividing by the total number of persons in the play set target area regardless of activity level. The proportion active for adults, girls and boys are examined separately. The rationale for limiting the target area in the physical activity level analyses to the playground target area only is that the features of the playground seemed unlikely to affect physical activity levels elsewhere on the schoolyard though the features of the playground were theorized to potentially have an effect on one's decision to come to the schoolyard, which is why all target areas were included for the utilization analyses.
All analyses considered 11 covariates: 4 measures representing school demographics and the surrounding area; the square footage of the schoolyard; 4 neighborhood availability/accessibility measures; the temperature at the time of the first scan for an observation session; and renovation status of schoolyard (yes/no). The 4 variables representing the demographics of the school and the surrounding area included: 1) the enrollment in the school; 2) the percentage of school students that are African American; 3) the population of children 17 and under in the census tract of the school; and 4) the percentage of children 17 and under living in poverty in the census tract of the school. Schools that were renovated as part of the "Schoolyards as Community Parks" program were designated as renovated [20
]. Ten individually matched schoolyards were designated as unrenovated schoolyards. The 4 neighborhood availability/accessibility measures included: 1) a count of parks and 2) a count of the number of city run recreation centers and pools within a .75 mile network distance of schoolyards (both based on data from the city's Planning Department); 3) a count of the number of child-friendly commercial venues for activity within a .75 mile network distance of schoolyards (based on data from InfoUSA and Internet yellow pages); and 4) a measure of street connectivity (i.e., alpha index, which measures the number of circuits or loops relative to number of possible circuits or loops using a .75 mile distance ) [23
]. Variable distribution information for the school and neighborhood covariates are presented in table .
All analyses were performed using Stata/IC 11.0 (StataCorp, College Station, TX). All regression models accounted for the clustering by school. Descriptive analyses were performed to examine the means and standard deviations of the different playground, school, and the surrounding neighborhood attributes as well as the average number of persons and their levels of activity.
For the utilization analyses, four different count models were considered including the negative binomial model, the poisson model, the zero-inflated poisson model, and the zero-inflated negative binomial model. Model residuals suggested negative binomial regression models be used. Given the small number of observations (200), smaller number of schools (20) and the large number of covariates (11), unadjusted negative binomial regressions were performed to determine which of the 11 covariates were significantly associated with any of the 3 utilization outcomes. For parsimonious models, only those covariates with significant associations for any group were controlled for in the primary analyses examining the associations between playground attributes and utilization.
Therefore in the utilization analyses, the primary analyses were multivariable negative binomial regressions, which adjusted for the school clustering, that predicted the average number of persons at the schoolyard during a scan as a function of each of the 9 playground attributes separately while controlling for the following significant covariates: a count of parks within a .75 mile network distance of schoolyards, the population of children 17 and under in the census tract of the school, and the enrollment in the school. Formal tests of interaction between each playground attribute and renovation status were performed for all 3 utilization outcomes (adults, girls, and boys) to examine effect modification by renovation status. Stratified analyses by renovation status were performed because the majority of the interaction tests were significant. Lastly, one negative binomial regression with all the significant playground attributes found from the primary multivariable negative binomial regression analyses was completed for both the renovated schoolyards and the unrenovated schoolyards while still controlling for the significant school/neighborhood attributes.
For the physical activity level analyses, three different outcome transformations were considered including log transformation, reciprocal transformation and untransformed. Analyses suggested that for boys and girls an untransformed proportion active outcome was the best fit. In adults, the log transformation of the proportion active outcome was the better fit. Separate unadjusted linear regressions were performed to determine which of the 11 covariates were significantly associated with each of the 3 proportion active outcomes (log transformed in adults only). For parsimonious models, only those covariates with significant associations were controlled for in the primary analyses examining the associations between playground attributes and proportion active.
Therefore in the physical activity level analyses, the primary analyses were multivariable linear regressions, which adjusted for the school clustering, that predicted the proportion moderately or vigorously active in the target area that contained the play set as a function of each of the 9 playground attributes separately while controlling for the following significant covariates: the population of children 17 and under in the census tract of the school; total number of free recreational centers and pools within a 0.75 network mile of the school; a count of parks within a .75 mile network distance of schoolyards; and renovation status. Formal tests of interaction between each playground attribute and renovation status were performed for all 3 proportion active outcomes (adults, girls, and boys) to examine effect modification by renovation status. No significant interactions were found therefore results were not stratified by renovation status but controlled for renovation status. Lastly, one multivariable linear regression with all the significant playground attributes found from the primary multivariable linear regression analyses was completed while still controlling for the significant school/neighborhood attributes. Since at least one person has to be in the playground target area to calculate the proportion active, the sample sizes for this outcome varies by group. For adults, girls and boys, the sample sizes were 61, 87, 99, respectively, meaning for 139, 113 and 101 observations there were no adults, boys or girls respectively in the playground target area.
The incident rate ratios (IRRs) and 95% confidence intervals (CIs) were reported from the negative binomial regression models and the beta coefficients and 95% CIs were reported from the linear regression models. Statistically significant results, defined as P ≤ 0.05, are discussed.