Procedure
Data were collected in Ghent (Belgium) and San Diego, California (USA). Ghent has approximately 250,000 inhabitants, an area of 156.18 km2 and a population density of 1565 inhabitants/km2. Of all inhabitants, 18.8% belongs to ethnic-cultural minorities (mainly Turkish, Bulgarian and Moroccan); the other 81.2% is predominantly White. The distribution of inhabitants across age groups is as follows: 11.4% of the population is between 0 and 9 years, 9.9% between 10 and 19 years, 56.7% between 20 and 59, and 22.0% is older than 60 years of age (Belgian National Institute of Statistics 2011: http://www.statbel.fgov.be). San Diego counts 1,307,402 inhabitants, an area of 842.23 km2 and a population density of 1552 inhabitants/km2. San Diego’s ethnic racial distribution includes: 28.8% Hispanic/Latino, 26.2% other ethnic-racial minority (mainly Asian and African American); the other 45.0% is predominantly White. The age group distribution is as follows: 11.9% of the population is between 0 and 9 years, 12.7% between 10 and 19 years, 60.2% between 20 and 59, and 15.3% is older than 60 years of age (U.S. Bureau of the Census. American FactFinder, 2010 http://www.census.gov).
In both cities, a similar protocol was used. Parks were randomly selected from four quadrants categorized by crossing high/low walkability and high/low socioeconomic status of neighborhoods: high-walkable/high-income, high-walkable/low-income, low-walkable/high-income and low-walkable/low-income neighborhoods. These neighborhoods had been defined for previous studies examining the associations between the built environment and physical activity (Ghent; Belgian Environmental Physical Activity Study) [20] (San Diego; Neighborhood Impact on Kids study) [23]. The neighborhoods were chosen to maximize within-country variability in walkability and income. In both cities, neighborhoods consisted of clusters of administrative or population collection units (statistical sectors in Belgium, block groups in the USA), which were the smallest geographical units that had information on household income, other sociodemographic factors and objective spatial data for walkability.
In both cities, neighborhood-level walkability was determined objectively, using Geographic Information Systems (GIS)-based built environment parameters. In Ghent, neighborhood walkability included three environmental attributes previously found to be related to physical activity: net residential density, land use mix, and intersection density [24]. A detailed description of the calculation of this walkability index can be found elsewhere [24]. In San Diego, the neighborhood walkability index consisted of the same three variables plus retail floor area ratio [25]. In both cities, neighborhood-level income was determined using census-based median annual household income data (US census 2000: http://www.census.gov; Belgian National Institute of Statistics 2007: http://www.statbel.fgov.be).
In Ghent, 10 parks were randomly selected from a sampling frame of all parks in the four neighborhood quadrants: four parks were located in the low-walkable/low-income neighborhoods, while there were two parks in each of the other three quadrants. Afterwards, 10 parks were selected in San Diego after matching park size to the parks in Ghent. In San Diego, two parks were located in each of the high-walkable/low-income and the low-walkable/low-income neighborhoods, and three parks were located in each of the other quadrants. Information about park size was obtained from the City Council in Ghent and from GIS data provided in city records in San Diego, completed with data from Google Earth and information found on parks and recreation websites when needed.
In these 20 parks, park characteristics were systematically coded by two trained observers using the Environmental Assessment of Public Recreation Spaces (EAPRS) tool [14]. Characteristics of park users were observed using the System for Observing Play and Recreation in Communities (SOPARC) tool [8]. In Ghent, data were collected in August and September 2011 (summer season; mean temperature = 16.8°C, average number of days with precipitation = 10). In San Diego, SOPARC data were collected in October and November 2011 (mean temperature = 21.6°C, average number of days with precipitation = 3). EAPRS data had been collected previously by trained observers, from February to June 2008. Since 2008, no substantial park renovations were completed in the 10 selected parks in San Diego.
Two trained observers collected SOPARC and EAPRS data in Ghent and SOPARC data in San Diego. Before collecting SOPARC data, both observers completed SOPARC training, provided by Dr. McKenzie (on DVD). Before collecting EAPRS data in Ghent, the observers received a standard EAPRS training offered by the San Diego team who used the tool to collect the data in San Diego in 2008. In both cities, the first two parks were observed by both observers to ensure comparability and solve any inconsistencies between observers. The other eight parks were rated by one of the observers (each observer rated four parks per country).
In Ghent, the study was approved by the Ethics Committee of the Ghent University Hospital. In San Diego, the study was approved by the San Diego State University Institutional Review Boards.
Measures
SOPARC
SOPARC is an objective observation tool to quantify the physical activity levels and socio-demographic characteristics of park users. SOPARC is a valid and reliable observation tool [8]. Use of SOPARC consists of defining discrete park zones, scanning a particular park zone, counting the overall number of park users in that zone, and classifying users by gender (males and females), age group (children, adolescents, adults, older adults), ethnicity (Latinos, Blacks, Non-Hispanic Whites, Other race/ethnicity) and physical activity level (sedentary, walking, vigorously active). Each park observation period was also classified by weather conditions (clear, cloudy, rainy) and darkness (dark, not dark). The number of park zones ranged from two to nine, and zones included open spaces, trails, playgrounds, swimming pools, basketball courts, sports fields, tennis courts, paths, picnic areas or shelters. In each park, observations were done during three days (two weekdays, one weekend day). On each observation day, four observation periods were conducted in each zone for about 15 minutes: in the morning (8AM), at noon, in the afternoon (3PM) and evening (7PM).
For the analyses, METs/observation period was calculated to obtain a representation of the average physical activity intensity during each observation period in a particular zone, independent of the number of visitors. To do so, a weighted MET score was given to each activity category (sedentary = 1 MET, walking = 3 METs, vigorous activity = 6 METs). These weighted MET scores were multiplied by the observed number of visitors in each physical activity category at the moment of the observation. Then, this score was divided by the total number of visitors present at the moment of the observation. This calculation produced a mean physical activity intensity score, independent of the number of visitors.
EAPRS
EAPRS is a detailed observation tool to assess park characteristics. The tool describes the physical environment of a park and focuses on the presence of park features and park amenities. Park features represent park characteristics that are essential to do physical activity. The observed park features included trails, paths, open spaces, swimming pools, playgrounds, sports fields and skating areas. Park amenities are aspects that contribute to the attractiveness of a park. The observed park amenities included places to sit, ponds/lakes, drinking fountains, picnic areas, vending amenities, restrooms, tables, bike racks and parking lots. EAPRS has good inter-rater reliability [14]. EAPRS was only used to obtain descriptive information about park characteristics.
Statistical analyses
Descriptive statistics were analyzed using IBM SPSS Statistics 19. χ2 tests were conducted to examine associations of study site, neighborhood walkability and neighborhood income with gender, age, ethnicity and activity levels of the observed park visitors. One-way ANOVA tests were conducted to examine potential differences in park size, park features and park amenities between Ghent and San Diego, between high- and low-walkable neighborhoods, and between high- and low-income neighborhoods.
Multilevel multiple regression models were conducted in MLwiN 2.25 to examine the associations of the independent variables (study site, neighborhood walkability, neighborhood income) with the outcome measures (overall number of park visitors, number of visitors being sedentary, number of visitors walking, number of visitors being vigorously active, METs/observation period), after adjusting for covariates. Multilevel modeling was applied because the null-models showed that 4.4% to 8.1% of the variance in the outcome measures was attributable to differences between parks. Two levels were included in the analyses: observations (level 1 = individual level) and parks (level 2 = group level). For the analyses with number of park visitors, number of sedentary, walking and vigorously active visitors as the outcome, the Poisson distribution of the outcome measures was taken into account. For the analyses with METs/observation as an outcome measure, the skewed outcome measure was logarithmically transformed (log10) to improve its normality [26]. Park size, day type (weekday, weekend day), time of day (morning, noon, afternoon, evening), darkness (dark, not dark) and weather (cloudy, clear rainy) were included as covariates in all analyses. In the analyses with number of visitors sedentary, walking and vigorously active as outcomes, the total number of visitors per observation was included as an additional covariate. For all analyses, significance was set at p < 0.05.