The WISEWOMAN trial, conducted from May 2003 through December 2004, was based in one community health center in Wilmington, NC, a moderately sized town surrounded by smaller coastal and agricultural areas. At the time of the study, the census bureau population estimate for Wilmington was 93,292, and estimates for the least to most populated surrounding incorporated places where participants lived ranged from 483 to 5192 people (21). 

For the first measure of proximity, three items were used to assess perceived distance to the closest PA resource: "How close to your home, in miles, is the closest [school; gym or recreation center; park]?" Three continuous perceived distance variables were created.

For the second measure of proximity, neighborhood was defined as a 10-minute drive from home, and three items measured perceived existence of PA resources in the participant's neighborhood: "Is there a [school; gym or recreation center; park or trail] where you could exercise in your neighborhood?" Response options were yes, no, and don't know, and three dichotomous variables were created to indicate presence (yes) or absence (no, don't know) of a resource in the neighborhood. Although survey measures were not validated, previous work has demonstrated that similar items have moderate to large reliability (22,23).

The following locations were geocoded in ArcGIS Version 9.1 (Environmental Systems Research Institute, Redlands, Calif): participants' home addresses, parks, gyms and recreation centers, and public schools that allowed the public to use PA facilities. Participants' home addresses were mapped as points, and 10% of all participants' addresses with the lowest match scores were checked against the appropriate road network file. (The match score indicates how closely elements of the address being mapped match the location where it was mapped.) Only one address was identified as being incorrectly mapped. The point was deleted and the correct address was manually geocoded.

Park addresses from the three counties were obtained using the Internet and by contacting each county's parks and recreation department. Internet search terms included "parks in New Hanover County North Carolina," and returns included sites such as New Hanover County Parks (24) and the Insider's Guide to North Carolina's Southern Coast and Wilmington (25). Although larger park size is associated with increased use (26), all parks, regardless of size, were geocoded because small parks may be part of an aesthetically pleasing neighborhood environment. State-, county-, or city-maintained miniparks, neighborhood parks, community parks, large urban parks, sports complexes, and natural resource areas were included (27). Before the WISEWOMAN intervention trial began, qualitative interviews were conducted with women from the study area to learn about community factors related to PA and nutrition (28). In the qualitative interviews, women mentioned driving to use PA resources. Thus, the street addresses for all parks were geocoded, as well as additional parking areas for large parks. A total of 107 park points were geocoded, representing 102 parks.

Gyms and recreation centers in each county were identified using the Internet and by requesting information from parks and recreation departments and the New Hanover County Department on Aging. Fifty-seven gyms and recreation centers were mapped, which included 19 city- or county-funded centers (city gyms, community centers, senior centers) and 38 commercial, fee-for-use gyms.

A list of all public schools (n = 65) in the three counties was obtained from the Internet (29). Each of the 65 schools was contacted to determine whether the general public was allowed to use its PA facilities. Facilities available for public use at schools included open fields and tracks. Because only the schools that allowed public use were expected to influence PA, these were the only ones geocoded. One university, six high schools, four middle schools, and 11 elementary schools were geocoded, for a total of 22 schools.

Triangulation of data sources has been suggested to further explore quantitative associations between environmental variables and PA (30). Therefore, qualitative interviews were conducted by the first author and reviewed to identify PA resources women reported currently or previously using and PA resources they reported that a friend or relative used. Participants in the qualitative study (n = 28) were not enrolled in the WISEWOMAN Project but lived in southeastern North Carolina and were similar to WISEWOMAN participants in age and socioeconomic status (31). Because of the similarities among the women, we assumed that resources identified in the qualitative interview were resources that WISEWOMAN participants would be most likely to use. Study participants were recruited by contacting key informants from a local senior center, two community health centers, and a community recreation center. Key informants then asked women who worked at or used the centers if they would be interested in participating in the interviews. Resources reported by women were geocoded and included two school tracks, one mall, four city- or county-funded recreation centers, and eight parks. (The tracks, recreation centers, and parks included in this layer were a subset of those geocoded as described above.)

The distance from each participant's home address to the closest PA resource along the road network was calculated using the ArcGIS Network Analyst extension. The number of each type of PA resource in 1- and 2-mile Euclidean ("as the crow flies") buffers was calculated using the Network Analyst intersect tool. Attribute tables generated in ArcGIS were exported into Excel and then imported into SAS Version 8.2 (SAS Institute, Cary, NC) for further analyses.

MVPA was measured using the ActiGraph accelerometer (ActiGraph, LLC, Fort Walton Beach, Fla), a valid and reliable measure of PA (32,33). Participants were instructed to wear the accelerometer for 7 consecutive days (with wear time starting the day after the clinic visit) during all waking hours. The ActiGraph measures uniaxial acceleration over investigator-specified time intervals, or epochs (1 minute in this study). The ActiProcess data reduction program (Catellier, 2004) was used to determine valid wearing time and to generate variables for use in subsequent analyses. Epochs contained within strings of 20 or more minutes of consecutive zeros were eliminated because we assumed the accelerometer was not worn. The minimum criterion for days worn was 4 days. A minimum of 6 valid hours defined a valid day: 78% (184/236) of the cohort had valid PA data. The average wearing time for the 184 participants was 11.2 hours per day and 5.7 days worn.

Average minutes of MVPA were calculated by dividing the sum of total MVPA minutes by the total number of days the accelerometer was worn. Minutes of MVPA were generated by imposing count cutpoints from Swartz et al (34), where moderate PA was 574 to 4944 accelerometer counts per minute and vigorous PA was greater than or equal to 4945 counts per minute. These cutpoints were used because participants in the Swartz et al (34) study were the most similar to WISEWOMAN participants in body mass index (BMI) and age, and the activities performed by study participants were similar to those likely to be performed by WISEWOMAN participants at a park, gym, or school.

Urbanicity was a potential effect modifier in the current analyses (35,36). If a participant reported a Wilmington address, she was categorized as urban; if she lived in a surrounding smaller town, she was categorized as rural. Randomization group status was examined as a covariate because the WISEWOMAN intervention aimed to increase participants' PA. The following additional covariates were examined because of their previously established independent association with PA or because they were potential confounders of the relationship between proximity (objectively measured and perceived) to resources and PA: age in years, calculated from self-reported birth date (continuous); BMI, measured at baseline as weight in kilograms divided by height in meters squared (continuous); education (dichotomized by high school graduation); baseline self-reported annual household income (greater than or equal to $10,000 and less than $10,000); self-reported race (white and nonwhite); and smoking status at baseline (dichotomous). A dichotomous variable that indicated whether participants had medical problems that limited their ability to walk briskly was also examined as a potential covariate.

Of the 236 WISEWOMAN participants, geocoded data were collected on 199. Women whose addresses were not geocoded were generally similar to women whose addresses were geocoded, yet they were more likely to live in a rural area and lived significantly farther from the health center. As mentioned above, 184 had valid accelerometer data. Women without valid accelerometer data were significantly younger and lived farther from the health center. Of the 180 women who completed the survey containing perceived distance questions, 100 answered all three perceived distance questions. Participants with incomplete perceived distance data were significantly older and more likely to have an annual household income of less than $10,000. Therefore, the final sample was 123 women for the analyses examining 1) the correlation between perceived and objectively measured distance and 2) the association between MVPA minutes and perceived and objectively measured distance to resources. The final sample was 155 women for the analyses examining 1) agreement between perceived and objectively measured existence of resources in the neighborhood and 2) associations between MVPA minutes and resources in the neighborhood.

For the continuous variables of distance in miles, the intraclass correlation coefficient (ICC) and Pearson correlation coefficient (r) were used to assess agreement between GIS network distance to closest PA resource and perceived distance to the closest resource. Correlation coefficients were classified as small (0.10–0.29), moderate (0.30–0.50), and large (>0.50) (37). A SAS macro written by Robert Hamer (38) was used to calculate the ICC for perceived and objectively measured distance. There were two raters of interest: 1) women’s perceptions of distance and 2) GIS network distance. These two raters were not chosen from a larger group of possible raters and were the only raters of interest; thus, they were considered fixed effects. The paired observations of perceived and objectively measured distance were obtained from a larger population of potential WISEWOMAN participants; thus, they were considered random effects. Therefore, the two-way mixed model (ICC 3,1 in the output) was used because the raters (women’s perceptions and GIS distance) were fixed effects, and the ratings of distance were random effects.

For the dichotomous variables indicating presence or absence of resources, the kappa (κ) statistic and percentage agreement were used to assess agreement between perceived and objectively measured existence of PA resources in the neighborhood. Differences in agreement were separated into six categories: 1) less than chance agreement (κ ≤0); 2) slight agreement (κ = 0.01–0.20); 3) fair agreement (κ = 0.21–0.40); 4) moderate agreement (κ = 0.41–0.60); 5) substantial agreement (κ = 0.61–0.80); and 6) almost perfect agreement (κ = 0.81–0.99) (39). All correlations and measures of agreement were calculated for the total sample. They were also calculated separately for urban and rural participants and for participants with MVPA minutes above and below the median. Previous work suggests that perceptions of the neighborhood environment may differ based on urbanicity (40) or PA levels (22). Correlations and measures of agreement were also stratified by intervention group because the intervention included tools designed to increase awareness of PA and nutrition resources

To examine the association between MVPA minutes per day and perceived and objectively measured proximity to PA resources, the following independent variables were used: 1) GIS network distance to the closest resource, 2) perceived distance to the closest resource, 3) number of resources in GIS buffers, and 4) perceived existence of resources in the neighborhood. An additional variable was generated from the data gathered during the qualitative interviews (New Hanover County participants only, n = 132): GIS network distance to the closest PA resource. Because MVPA minutes were skewed, the natural logarithm of MVPA minutes was used as the continuous dependent variable. The association between urbanicity and GIS network distance was examined by comparing mean distance to each type of resource for urban participants compared with rural participants. Because urbanicity was associated with GIS network distance, it was not examined as a moderator (41). Multiple linear regression was used to estimate the associations between independent variables and the natural logarithm of MVPA minutes, adjusting for individual-level covariates. Covariates were retained in the final model if they had either an independent relationship with MVPA (P value for the parameter estimate <.05) or were confounders (changed the parameter estimate for the independent variable ≥10%). Model parsimony was a goal.

Table 3 reports agreement between perceived and objectively measured distance to closest PA resources. Pearson correlations between perceived and objectively measured distance were moderate to large for overall categories, ranging from 0.40 to 0.54. ICCs ranged from moderate for parks (0.33) to small for gyms (0.23) and schools (0.20). Correlations were generally higher for rural participants and for women in the intervention group.

Table 4 shows κ coefficients and percentage agreement for perceived and objectively measured existence of PA resources in 1- and 2-mile buffers. There was slight to fair agreement between the perceived and objective measures. κ coefficients for all participants in the 1-mile buffer category ranged from 0.14 for gyms and 0.15 for schools to 0.39 for parks. Most disagreement for parks and gyms resulted from women who reported no park or gym in the neighborhood when the GIS measure indicated one in the buffer representing the neighborhood. However, for rural women, most disagreement occurred when the woman perceived existence of a park or gym in her neighborhood when the GIS measure indicated none. Most disagreement for schools arose from women who reported a school in the neighborhood when the GIS measure indicated none. In general, rural participants had higher κ coefficients and percentage agreement compared with urban participants. For parks and gyms, the GIS measures using the 1-mile buffer yielded greater agreement compared with measures using the 2-mile buffer. For schools, there was greater agreement using the 2-mile buffer compared with the 1-mile buffer.

Table 5 shows associations between MVPA minutes per day and two measures of proximity: 1) distance to the closest PA resources, models 1 through 12; and 2) density of PA resources within 1-mile buffers, models 13 through 24. For each type of resource, the first model in the set includes the objective measure of proximity, the second model includes the perceived measure, and the third model includes both perceived and objective measures of proximity.

Models 1 through 12 in Table 5 report crude and adjusted standardized parameter estimates for the cross-sectional associations between natural logarithm-transformed MVPA minutes and perceived and objectively measured distance to the closest resources. The only statistically significant association was for perceived distance to gyms when both perceived and objectively measured distance were included in the model. Perceived distance to gyms and schools had greater parameter estimates compared with objectively measured distance, so that greater perceived distance was associated with less MVPA. Final models adjusted for age and BMI accounted for 8% to 15% variation in MVPA.

Models 13 through 24 in Table 5 show crude and adjusted standardized parameter estimates of cross-sectional associations between natural logarithm-transformed MVPA minutes and perceived existence of and objectively measured number of resources within 1-mile buffers. In all models, the number of resources in the buffer was inversely related to MVPA, against the expectation that a greater number of facilities would be associated with more activity. There was a statistically significant association between the number of schools within the 1-mile buffer and minutes of MVPA (model 19, P = .04; model 21, P = .03). Of two women of the same age (53 years) and BMI (31 kg/m²), one woman with no school within her 1-mile buffer averaged 105.3 minutes of MVPA per day while the other woman with two schools within her 1-mile buffer averaged 83.2 minutes of MVPA per day (P = .04). Variance explained by the models adjusted for age and BMI ranged from 8% to 11%.

The amount of time that women wore the accelerometer influenced the quality of PA data obtained. Thus, we examined differences in wear time by individual characteristics. We found that women who wore the accelerometer all 7 days had a lower average BMI than women who wore it 4 to 6 days (P = .006, data not shown). We stratified the two models in Table 5 that demonstrated statistically significant associations (i.e., models 6 and 19) by days the accelerometer was worn. Stratifying by days worn had no effect on the association between perceived distance to gyms and MVPA minutes. The association between number of schools within the 1-mile buffer and MVPA minutes was stronger and statistically significant for women who wore the accelerometer for 7 days (standardized parameter estimate = −0.38, P = .01, n = 44) compared with women who wore it 4 to 6 days (standardized parameter estimate = −0.08, P = .36, n = 111).

There was no association between distance to resources identified through qualitative interviews and MVPA minutes, adjusting for age and BMI (standardized parameter estimate for GIS network distance = 0.06, P = .45).