Ecological models postulate that health behavior changes are a function of psychological, social, policy, and physical environmental factors [1, 2]. Numerous authors and agencies have identified environmental and policy intervention as promising strategies for creating population-wide changes in physical activity and obesity [3–6]. Current evidence of a relationship between the built environment and physical activity is generally supportive but there are limitations . An important limitation is that virtually all studies have been conducted in settings with restricted environmental variability. Restricted variability yields attenuated estimates of associations , so the magnitude of associations between environmental characteristics and physical activity is likely underestimated [9, 10]. An accurate assessment of such associations requires greater environmental variability than any one country or region can offer. The International Physical Activity and the Environment Network (IPEN; http://www.ipenproject.org) has set out to support coordinated data collection in countries with diverse environments and populations. IPEN uses common study design and measurement to produce more reliable and valid effect size estimates.
The Neighborhood Environment Walkability Scale, (NEWS)  and its abbreviated form (NEWS-A) , are the measures of perceived neighborhood environment selected for the IPEN initiative. The NEWS and NEWS-A assess perceived environmental characteristics, stemming in part from the urban planning literature , believed to influence walking and other forms physical activity. Initial evidence for their criterion validity and reliability has been documented in four countries across four continents [11, 14–17]. Except for the neighborhood characteristic of not many/any cul-de-sacs, test-retest reliability of the individual items of the NEWS and NEWS-A was moderate to high . Significant associations were observed between the NEWS subscales and objective  as well as self-report measures of physical activity and walking [12, 16, 17]. Additionally, scores on the NEWS were strongly associated with corresponding objectively-measured constructs of neighborhood environments [11, 16, 19].
Two recent studies examined the measurement models of the original and Australian versions of the NEWS [12, 16]. The measurement models described the relationships of the items to the theoretical constructs measured by the scales . In other words, they identified groupings of items measuring distinct perceived neighborhood-environment constructs (e.g. environmental aesthetics, traffic hazards, and access to services).
To maximize the variability in environmental attributes, both studies adopted a stratified two-stage cluster sampling strategy whereby participants were recruited from specific areas (here, census blockgroups, the smallest geographical units for which census bureaus publish demographic data) selected according to their objectively-measured walkability and socio-economic status (SES; i.e. median income). Stratification by SES likely enhanced the representativeness of the sample because, otherwise, low SES respondents might have been underrepresented . Two distinct measurement models of the NEWS and NEWS-A, one for each level of variation in the data, were examined to address violations of the statistical assumption of independence of observations resulting from the adopted sampling strategy . Thus, measurement models were defined at the individual (based on within-census blockgroup variations in the responses to the items) and blockgroup levels (based on the between-census blockgroup variations) [12, 16].
The individual-level measurement models were based on differences in responses between study participants living in the same blockgroups and described the way perceived environmental attributes (represented by the NEWS items) covaried within census blockgroups. The differences in responses may have resulted from actual environmental differences within a blockgroup (e.g. differences in traffic load or aesthetics across locations within a blockgroup), response biases (e.g. tendency to provide extreme ratings), and/or perceptual biases (e.g. anxious respondents' tendency to overestimate the risk of crime in their neighborhood) . In contrast, the blockgroup-level measurement models were based on the blockgroup average ratings of the items and indicated how perceived environmental attributes covaried between blockgroups. These models likely reflected the way environmental attributes clustered objectively across blockgroups. In fact, the average rating of a blockgroup characteristic can be considered a relatively reliable and valid indicator of the objective environment. This is because response and perceptual biases are likely to be random effects that, by definition, cancel out when summed across respondents and, hence, have no impact on the average rating for a blockgroup. Importantly, blockgroup-level factors were found to be strongly correlated with corresponding objective measures .
The authors recommended scoring the NEWS according to the individual-level measurement model for three main reasons: (1) the individual-level factors more accurately represented constructs commonly used in the urban planning and transportation fields ; (2) they likely indicate how perceptions of environmental attributes group together into factors, while blockgroup-level factors likely represent patterns of associations between objective environmental attributes; (3) they are likely to be more generalizable across locations and populations than are blockgroup-level factors .
The measurement model of the Australian NEWS mostly resembled that of the original version, but differed in significant ways . For instance, while traffic-related items formed a unique individual-level latent factor in the original NEWS tested in some USA cities, in the Australian version they split into two weakly correlated factors – namely, traffic safety and traffic hazards. Although dissimilarities in factorial structures were partly attributed to substantive item-content differences between the two versions of the NEWS , they also raise concerns about the reliability and generalizability of the original measurement model to different geographical and cultural settings. Hence, it was necessary to cross-validate the original NEWS and NEWS-A in a geographical location and population different from those used in the original validation study (i.e. Seattle, Washington region). Such information is important for establishing common, valid scoring protocols, which in turn can provide a more accurate estimation of a dose-response relationship of the perceived built environment with physical activity and obesity. Thus, the current paper reports the individual- and blockgroup-level factor structures of the NEWS and NEWS-A tested in the Baltimore, Maryland – Washington, DC region, which is a demographically- and environmentally-dissimilar city to Seattle, WA.
Specifically, according to the 2003 American Community Survey, Seattle was the most educated larger city in the USA, with 52% of residents aged 25 and over having attained at least a bachelor's degree . In contrast, the percentage of highly educated residents in Baltimore was 35.6. The 2000 median household income in Baltimore was approximately $32,500, while in Seattle it was $49,500. Seattle had 70.5% of White and only 7.8% of African American residents, while the percentage of African Americans in Baltimore was 63.8, and that of Whites 31.4. Baltimore is located on the East coast of the United States, while Seattle is located on the West coast. Baltimore and Seattle are similar in size, terrain, urban layout (grid pattern), and are both considered "cities of neighborhoods". However, with its climate and geographical location, Seattle provides more ample access to a variety of outdoor activities. Also, Seattle has higher population density, more traffic congestion problems, but lower crime rates than Baltimore [22, 23].
We hypothesized that the individual-level factor structures of the NEWS and NEWS-A, derived from the original validation sample (Seattle, Washington, USA), would show a sufficient level of fit to the data from the cross-validation sample (Baltimore, MD – Washington, DC, USA) due to them being in part a function of psychological principles that apply across diverse subgroups. We also hypothesized that the original blockgroup-level factor structures of the NEWS and NEWS-A would show poorer fit to the data from the cross-validation sample than their individual-level counterparts due to them reflecting patterns of associations between objective environmental factors, which likely vary across geographical locations.