There is a growing body of public health research documenting how characteristics of neighborhoods are associated with differences in the health status of residents. However, little is known about how the spatial resolution of neighborhood observational data or community audits affects the identification of neighborhood differences in health. We developed a systematic neighborhood observation instrument for collecting data at very high spatial resolution (we observe each parcel independently) and used it to collect data in a low-income minority neighborhood in Dallas, TX. In addition, we collected data on the health status of individuals residing in this neighborhood. We then assessed the inter-rater reliability of the instrument and compared the costs and benefits of using data at this high spatial resolution. Our instrument provides a reliable and cost-effect method for collecting neighborhood observational data at high spatial resolution, which then allows researchers to explore the impact of varying geographic aggregations. Furthermore, these data facilitate a demonstration of the predictive accuracy of self-reported health status. We find that ordered logit models of health status using observational data at different spatial resolution produce different results. This implies a need to analyze the variation in correlative relationships at different geographic resolutions when there is no solid theoretical rational for choosing a particular resolution. We argue that neighborhood data at high spatial resolution greatly facilitates the evaluation of alternative geographic specifications in studies of neighborhood and health.
Citation: Leonard TCM, Caughy MO, Mays JK, Murdoch JC (2011) Systematic Neighborhood Observations at High Spatial Resolution: Methodology and Assessment of Potential Benefits. PLoS ONE 6(6): e20225. https://doi.org/10.1371/journal.pone.0020225
Editor: Joseph S. Ross, Yale University School of Medicine, United States of America
Received: January 20, 2011; Accepted: April 27, 2011; Published: June 3, 2011
Copyright: © 2011 Leonard et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: This work was supported by funding from the National Science Foundation (NSF/SES-0827350), the Institute for Urban Policy Research at the University of Texas at Dallas, and the Foundation for Community Empowerment Dallas, Texas. The funders had no role in study design, data collection and analysis, decision to publish, or preparation of the manuscript.
Competing interests: The authors have declared that no competing interests exist.
In the last 15 years, public health researchers have documented disparities in health status associated with the structural and social characteristics of neighborhoods that cannot be explained by individual differences in risk profiles. A broad range of health outcomes has been considered in neighborhood research including indices of adult physical health , , , , , , , , , adult mental health , , , , , , , and child health , , , , , , , , , , , , , .
Health outcome data at smaller geographic resolution (for example spatially referenced individual level data) are becoming increasingly available, furthering the study of neighborhood effects on health. Unfortunately, secondary data sources, such as the census, may be inadequate for these studies because aggregating over census geographies (e.g. blockgroups, tracts, etc) loses much of the variation that is valuable for analysis of individual level data. To address the limitations of secondary data sources, a number of public health researchers have employed community audits to provide not only current data on neighborhoods but also direct observation of neighborhood conditions.
Two recent reviews of the use of neighborhood observation methods in public health research ,  document the wide range of approaches used in the implementation of community audit methods, and the limitations of the extant literature on these methods. Spatial resolution is one of the key factors along which built environment data differ. There are many reasons for this. First, if secondary data is used, the researcher must adjust and use the best spatial resolution available. In the absence of other available data, some information about the built environment is certainly better than none and some policy questions can be broadly assessed without high-resolution data. Second, data at higher spatial resolution is more expensive to obtain because a larger number of observations per area are needed. Costs and benefits of collecting and analyzing high-resolution vs lower resolution observational data should be considered. Additionally, the scale of the research question sometimes determines the resolution of data used. It is impractical for researchers who are studying the impact of a nation or state wide program to attempt to use the method presented here to collect high-resolution observational data. However, research studies focusing on smaller geographies (within a neighborhood or a city –sector) might find high-resolution data more beneficial. In short, the research question should determine the spatial resolution needed.
With this in mind, there is a need to examine the reliability and predictive utility of observational data collected at varying levels of aggregation. Examining the robustness of the relationship between individual outcome measures and observational data at different geographic aggregation levels is essential for investigating the presence of potential omitted variable biases that result when geographic boundaries are correlated with unobserved individual characteristics. Additionally, policy initiatives that operate at varying geographic scope would be better informed if the robustness of results to lower (and higher) level aggregation schemes were evaluated. However, due to the level of spatial aggregation at which the observational data are often collected, most of the existent data sets are not well suited for a thorough investigation of this robustness. In this paper, we present a new interdisciplinary community audit methodology for collecting data on neighborhood factors at the smallest geographic unit possible and assess the advantages of using data at such a high spatial resolution for micro-level studies in neighborhood-related health outcomes.
Previous work in this arena, such as the extensive data collected in conjunction with the Project on Human Development in Chicago Neighborhoods (PHDCN) and many others, has focused on the face block—or the parcels facing the same street segment or “block”– as the smallest geographic level for collecting observational data, but our methodology increases the spatial resolution by analyzing the parcel. In residential neighborhoods, a parcel consists of the house and surrounding yard or all of the property that a homeowner owns and is assessed for tax purposes. Most neighborhood audit tools measure some elements of both the physical and social conditions in the neighborhood. The parcel observation methodology we present focuses exclusively on the physical dimension. Raudenbush and Sampson  analyzed PHDCN data and find that measures of physical disorder are relevant gauges of neighborhood condition at higher spatial resolution than measures of social disorder. Further, they note that reliability among face block measures of physical disorder may be improved by increasing the number of items observed for each face block. Collecting parcel observations achieves this goal. Beyond the ecometrics-based rational put forth by Raudenbush and Sampson , there are also numerous practical reasons for expanding data collection to include parcel observations: (1) it allows for maximum flexibility with regards to spatial aggregation; (2) it allows the researcher to distinguish between observations which the household has direct control (i.e. the upkeep of their yard/property) and observations which are impacted by others in the community (i.e. the upkeep of common areas such as parks or the upkeep of other properties in the neighborhood); and (3) it allows for the data and research outcomes to be related to property values which have direct policy impact through the tax base. However, while the advantages appear strong, no systematic work has been done to determine whether the advantages outweigh the costs of collecting data at this micro level.
An issue related to geographic aggregation is the problem of how to operationally define neighborhood boundaries—or put more generally, the modifiable areal unit problem (MAUP) . When addressing the question about how to measure neighborhood characteristics, Guo and Bhat  state that “we should measure what matters to people over the area that really matters to people” (p. 31). This suggests neighborhood boundaries should be selected thoughtfully and may vary depending upon the research question at hand. To date, a few investigators studying the effects of neighborhood context on health have utilized sophisticated spatial definitions of neighborhoods , , , , , but no studies have been able to comprehensively compare the utility of community audit data collected at varying levels of aggregation. This is a significant gap in the extant literature. Boone et al  find that associations between physical activity and street connectivity vary by setting and geographic scale. The same is likely true of associations between health outcomes and neighborhood observational data. Most often, however, public health researchers rely upon administrative boundaries of neighborhoods such as census tracts or census block groups, and only a single geographic scale is analyzed.
The observational instrument developed is intended to be useful for analyzing the relationship between place and individual health and well-being while avoiding biased created by the MAUP. Can data at a high spatial resolution improve studies and, in particular, public health policy implications regarding the relationship between neighborhood conditions and health? While we acknowledge that data at lower spatial resolution is helpful and sufficient in some cases, we believe that higher resolution does present advantages that can improve and fine-tune policy implications. First, data at high spatial resolution allow the researcher nearly complete flexibility in specifying neighborhood boundaries and hence a thorough investigation of MAUP-bias is possible (though we note that a thorough analysis of this issue is outside the scope of this paper). For example, studies may find a statistically significant correlation between average census tract condition and obesity. However, unless one investigates this relationship further at varying levels of geographic resolution, it is unknown if the results are biased by other omitted variables correlated with the geographic census tract definition. Second, data at high spatial resolution allow public health policy makers to identify the most appropriate geographic level for public health interventions. Continuing with the previous example, a relationship between census tract condition and health is an important observation, but policy relevance may be greatly improved if additional insight was available on the geographic scale that scarce public resources should be deployed to improve public health. Policy makers need to know the comparative implications for enacting, for example, broad-based local neighborhood clean-up initiatives throughout the census tract, versus concentrated initiatives to improve only the most blighted areas within the census tract.. Neighborhood observational data will allow the research to specify exactly which geographic definitions matter most for a particular policy implication.
The purpose of this report is three-fold. First, we describe the methodology and how the method was implemented to ensure the collection of high quality observational data. Next, we analyze the reliability of the data and the relationships found between the variables observed. Finally, we examine the utility of these data to examine neighborhood conditions at different levels of aggregation and how such data might be used in studies of neighborhoods and health.
All research involving human subjects has been approved by the Institutional Review Board of the University of Texas at Dallas (IRB Approval Number 08-33). Informed written consent for study participation was obtained from each human subject. The consent forms used were approved by the IRB committee.
The new observational method was used as part of a large cross-sectional-longitudinal research project aimed at studying the effects of publicly driven investment in a low-income, minority neighborhood. The neighborhood examined, commonly known as Fair Park, is an area of over 2000 acres with approximately 20,000 residents located less than two miles southeast of downtown Dallas, Texas. The primary reason for research interest in Fair Park is that this community received a large injection of publicly-directed investment beginning in 2009. In particular, the Dallas Area Rapid Transit (DART) authority has built a new light rail line into Fair Park and approximately $80M has been allocated for housing and infrastructure development in the neighborhood.
The neighborhood consists of thirty-two block groups falling into seven census tracts. The area is poor (median household income is $19,939) and primarily African American (70%), although the percentage of the population that is Hispanic (26%) has been growing (estimates based on US Census, 2005–2009 American Community Survey 5 Year Estimates). The study area contains primarily single-family residential housing of which approximately 52% are owner occupied.
In addition to the neighborhood observational data, a short door-to-door survey was conducted to collect basic demographic data from neighborhood residents. The 1210 households included in this survey were sampled according to a geographically weighted sampling scheme with oversampling of residents residing near the new neighborhood light rail stations. The sampling scheme was designed to facilitate a longitudinal study analyzing the impact of light rail investment in the neighborhood. Included in the survey was self-reported health status in which residents were asked to rate their physical health status as excellent, very good, good, fair or poor. Self-reported health status will be used to provide some indication of the association between neighborhood observational measures and health.
The instrument used for the systematic neighborhood observations was based on the integration of the Neighborhood Observation Checklist (NOC)  and existing coding methodologies used by the City of Dallas which have their history in the urban planning literature. The resulting methodology allows for observations at the parcel level and provides data relevant to the study of public health, economics, urban planning and other social science disciplines. As stated in the introduction, one of the particular ways in which parcel level observations differ from lower spatial resolution neighborhood data (e.g. faceblock, census blockgroup, cenus tract) is that individuals have great control over their own parcel condition while control over other neighborhood attributes is limited. Thus, the parcel observation has two roles: it is the building block from which lower spatial resolution observations are built and it provides individual data for the parcel's occupant. Items for the observation are described in Table 1. Some items were characterized at the parcel level, and some were characterized at the face block level. As previously noted, the primary innovation of the method is the data collected at the parcel level, which includes only measures of physical condition (as opposed to social conditions). Additionally, parcel usage data was collected for all parcels in the neighborhood while the other parcel condition codes were observed only for single-family residence parcels.
Basemaps and geodatabases were built using spatial data downloaded from the North Central Texas Council of Governments (NCTCOG, http://clearinghouse.dfwmaps.com/) to which parcel data from the Dallas Central Appraisal District (DCAD) were added. Custom data input forms were created so that data could be entered directly into laptop and tablet computers running ArcPad software (ESRI, www.esri.com). Data collection staff could select a parcel or a block by clicking on the map and the appropriate data entry form would open for coding observations.
Training of data collection staff
Data collection staff was recruited from the local community and trained in several classroom-based sessions. During these sessions, the staff was trained in the operation of the data collection software, the definitions of parcel and block attributes to be coded, and the procedures for conducting the observations. Supervisors and data collections conducted practice coding sessions in a comparable neighborhood. Data collectors were considered proficient when they had general agreement with themselves and the supervisors.
Data collection methods
Pairs of trained data collectors collaborated to record observations of the attributes for each parcel and each block within the study boundary. One member of the pair drove the vehicle, while the other member operated the computer. On some residential streets, they were able to drive slowly while making the observations, while on busier commercial streets or when discussing their observations, they would park briefly. All codes were arrived at by consensus between the two observers. The observations were conducted during January 2009 and February 2009, and all observations were conducted on a weekday between the times of 9:00 a.m. and 4:30 p.m. Over a five week period, a total of 11,552 parcel and 1,778 face block observations were completed by varying pairs of 18 data collection staff.
Quality assurance procedures of observational data
Several checks were instituted to assure data quality. First, the data collection supervisor periodically rode along with each data collection team and completed a checklist verifying the following: the observed parcel or block matched the ArcPad map, all attributes were noted, parcel use and face block condition were correctly coded, and no parcels or face blocks were missed. Second, the pairing of computer operator and driver was systematically varied to improve consistency of observations. Third, approximately 10% of each day's parcel and block observations were selected and coded a second time by a different pair of observers. This provides a set of independent observations from which inter-rater reliability can be calculated (see Results section). The two field visits for the 10% oversampling occurred at different times of day, but Raudenbush and Sampson  note that temporal variation presents little problem for observations of physical condition—the focus of the parcel data being analyzed. Finally, parcel use codes were checked against property tax codes obtained from the Dallas Central Appraisal District (DCAD) for potential discrepancies; all discrepancies were verified by a second field visit.
Data processing of observational data
Parcel and block observations were retrieved daily from each field computer and merged into the appropriate geodatabase in ESRI's ArcGIS, also ensuring that each day's work would be saved. Maps were made showing the observed parcels and blocks. Using ArcMap tools, parcels were associated with block(s) that abutted the lot, and fields added to the parcel file specifying the adjacent block IDs. This process enabled the selection of parcels fronting both sides of a block, which, together with the street, made up the face block. Each parcel was also assigned to the proper census geographies (block, block group, and tract).
Data collection costs
The primary cost of collecting observational data may be allocated to the following categories: (1) equipment, (2) GIS database creation and preparation, (3) data checking and supervision of field staff, (4) observers time in the field—including travel to and from locations and (5) the time required for data input after field observations have been made. Equipment used for this study includes ARC GIS software, 3 laptop or tablet PC's, and 1 supervisor PC. However, these are one time costs and often this equipment can be leveraged from other projects. The GIS database utilized for this work was adapted from the local tax appraiser's GIS database. While this information is not readily available in all communities, most urban areas do maintain GIS databases. GIS staff time required to assemble the GIS database and maps, prepare data entry forms, and load data and maps onto field computers was approximately 25 hours. Data checking and supervision of field staff is necessary for any observational data collection and is not necessarily any more costly for the collection of data at high spatial resolution. Working in pairs of 2, data collection staff spent an average of 10.5 minutes making observations on each face block in the neighborhood. Data input time was minimized since the data collection staff entered all observations directly into laptop and tablet computers. Four teams of data collection staff were sent out each day, and approximately 3 hours was required each day to extract and integrate the data from each laptop or tablet computer back into ArcGIS geodatabase when the staff returned from the field. This task was completed by the project supervisor and coincided with the staff supervision and data checking tasks.
Results and Discussion
In analyzing the observational data, we were interested first in the inter-rater reliability of the data collection effort. Second, we analyze the relationships within the data via factor analysis and a comparison of the differences between face blocks that were rated as being of overall high and low quality. Last, we explored multiple geographic aggregation schemes and compare the information available from the data depending on the aggregation scheme used.
Inter-rater reliability was calculated using percent agreement based on the 10% sample of parcels for which 2 independent observations were recorded. For parcel-level items, agreement ranged from .66 to .998, with an average level of agreement of .91. For face block-level items, agreement ranged from .65 to .99, with an average level of agreement of .84. The results compare well with community audit measures as reported by Brownson et al. . The higher repeatability for parcel level observations is likely due to the smaller observational unit and suggests that aggregates built up from parcel observations may produce more reliable results.
Relationships within the data
Factor analysis using a principal factor method with a promax rotation was implemented to distill the parcel-level data into summary variables. Two factors with eigenvalues greater than 1 were identified, and factor loadings for the parcel items are displayed in Table 2. Items that failed to load on either factor with a loading of at least .30 were dropped. The first factor, Aesthetics was comprised of housing condition, peeling paint, boarded windows, unkempt lawn, fence in poor shape, and trash in yard. The second factor, Security was comprised of presence of barred windows and doors. The internal reliability of the factors was .61 and .83, respectively; correlation between the factors was .14, p<.001. Summary scores for each factor were calculated by coding each parcel-level item as 1 to represent undesirable attributes and summing the component items. This method was chosen rather than using factor loadings to preserve the intuition behind the Aesthetic and Security variables—i.e. Aesthetic is the number of undesirable attributes of a parcel. Higher scores on the Aesthetic factor, the primary parcel-level indicator of neighborhood condition, indicated more “eye-sore” conditions on a parcel.
The items recorded at the face-block level did not lend themselves to a data reduction method such as factor analysis. To examine the interrelations of these items, therefore, we compared the characteristics of face blocks receiving an overall rating of “desirable” with those rated “undesirable”. The results and chi square statistics (Fisher exact) are displayed in Tables 3 and 4. Desirable blocks were wider, less congested, had better maintained streets and sidewalks, more street lamps, and fewer unrestrained dogs. Additionally, desirable face blocks are more likely to contain parcels with fewer Aesthetic concerns (the Spearman rank correlation coefficient is −.20, p<.01).
Geographic aggregation results
Of key interest is how the parcel data compare to more aggregated data used in other studies. More geographically aggregated data are easier and potentially less costly to obtain; however, the aggregation may wash out key variations that occur within neighborhoods. For studies of neighborhood influences, we would argue that such variations that occur (for instance) between parcels on the same face block or consecutive streets in the same census block might be important.
To examine the range and variability of observed measures at different levels of aggregation, we aggregated the parcel-level scores for aesthetics, security and the overall desirability of each face block to geographic groupings commonly used in the literature: face block, census block, census block group, and census tract. Aggregation was achieved by taking the average of each variable over the specified geography. Geographic boundaries that are meaningful in the sense that they represent true boundaries for different neighborhood types should delineate homogeneous data groupings. One would expect the range of the mean values for each grouping to not decrease at higher levels of aggregation and the average mean absolute deviation (MAD) within each grouping to not increase considerably. However, as we see in Table 5, as the level of aggregation increased, the range of scores for each observed measure decreased and the average MAD increased.
To further examine the impact of aggregation on the data, we compared the parcel values for Aesthetic to the block group average values of Aesthetic; results are displayed in Figure 1. Parcels shaded black have a block group average value for Aesthetic that differ by one or more standard deviations (one SD = 1.14) from the actual parcel value. Over 25% of the parcels are mischaracterized by the block group average value of Aesthetic by at least one standard deviation. This is important when considering causal relationships between individual outcomes and neighborhood. An individual may influence the level of Aesthetic of his/her own parcel, but may have little control over Aesthetic at the face block (or higher) level.
The spatial analysis may be taken a step further by computing local Moran's I statistics for each geographic aggregation level using ESRI ArcMap Cluster and Outlier Analysis tool in the Spatial Statistics toolbox. The local Moran's I is a local indicator of spatial association (LISA) and takes on statistically significant positive (negative) values when there is spatial clustering of similar (dissimilar) values . The spatial clustering of data is at the heart of analyzing the influence of place on observed outcome variables such as physical activity, child development, crime or economic development. High or low degrees of spatial clustering can result in very different policy implications. For example, a block group of average neighborhood condition might in fact be composed of two distinct clusters—one with high neighborhood quality and one of low neighborhood quality. Further, spatial clustering that occurs along the boundaries between neighborhoods might result in significant edge effects that result in misleading correlative analysis.
Table 6 presents the percentage of geographic units with statistically significant spatial clustering at each aggregation level. The degree to which spatial clustering occurs—measured as a percentage of total geographic units–decreases considerably at higher levels of aggregation; an obvious consequence of aggregation. Figure 2 illustrates this observation by focusing on a small quadrant in a residential section of the neighborhood. This quadrant being analyzed is indicated in Figure 1. Spatial clustering of like values is indicated in black, while spatial clustering of dissimilar values is shown in grey. Un-shaded areas do not exhibit statistically significant spatial clustering. Observing only the block group level data, one would conclude that there is little spatial clustering in this section of the neighborhood, when in fact many of the parcels are spatially clustered. A similar pattern emerges throughout the neighborhood. The role of spatial clustering of observational data is very important from a policy perspective because it can provide guidance as to whether policy should tackle small concentrated areas of high concern or should be applied on a larger, less concentrated scale.
Predictive Utility for Health Outcomes
Table 7 presents the results of a simple demonstration of the utility of the Aesthetics score obtained from our neighborhood observational data at predicting self-reported health status. A univariate ordered logit model was applied to determine the relationship between self-reported health status (M = 2.78, SD = 1.07) and Aesthetics at varying levels of geographic aggregation. The odds ratios and corresponding 95% confidence intervals are reported in Table 7.
Aesthetics at the census tract level has the highest odds ratio and statistical significance. However, we find that no statistically significant relationship can be found at the census block group aggregation. The relationship reappears at the faceblock and parcel level. These results illuminate the need to improve policy research by exploring different aggregation schemes and the necessity of a method for neighborhood observations that has high spatial resolution so that the next step can be taken: understanding why the relationships change with spatial aggregation. This perhaps overly simplistic demonstration does not necessarily indicate that one aggregation level is best for all cases. Instead, this simple example indicates that the estimated associations vary depending upon geographic aggregation, and the research question should dictate the aggregation level used—or put differently changing the aggregation changes the research question being answered. For example, if we want to know which low income census tracts to target for a public health information campaign, then tract level is appropriate. However, if we want to know how to best invest public funds to improve public health through neighborhoods, and policy makers must choose between specific small-scale projects, the aggregation level used should match the scope of the projects in question. Further, when theory does not clearly define a superior aggregation level, results should be evaluated with consideration for their robustness to alternative geographic specifications.
This paper presents and applies a new methodology for obtaining community audit data of neighborhood condition at the smallest geographic unit, the parcel. Our instrument for systematic neighborhood observations presents a high degree of inter-rater reliability, factor analysis revealed logical relationships within the data, and the factors derived from the data were predictive of health status. Of key interest to our work was assessing the costs and benefits of obtaining data at such a high spatial resolution.
Obtaining observation data is not without costs. There are some significant up front costs that are required for observational data collection at any geographic aggregation level—GIS expertise, software and hardware for data entry and organization, and supervision for project staff. The primary additional cost of obtaining observational data at high spatial resolution is time. Our systematic neighborhood observations averaged 10.5 minutes per face block. This compares favorably with other instruments designed for observational data collection at lower geographic resolution. The average time for observations for these lower-resolution instruments ranged from 10.6 to 20 minutes/segment . This suggests that collecting parcel level data is not overly expensive or time-consuming if face block or street segment observations are already being conducted once the requisite GIS is in place.
However, exploratory analysis of the data reveals that much detail is lost when the data are averaged to higher levels of aggregation. Therefore, parcel level data may provide many benefits. Higher levels of aggregation result in less variability among observations and lack of resolution when identifying statistically significant spatial clustering. This lack of resolution may be a key hindrance in uncovering correlative relationships between neighborhood condition and observed outcome measures when the parcel level attributes are not observed.
Another key advantage to obtaining small area geo-referenced data is the ability to explore different aggregations, both a variety of census geographies and also many other more flexible aggregation schemes such as the neighborhood condition within 250, 500 or 1000 feet of a parcel. This allows an array of research questions to be answered without the limitation of the aggregation scheme available in the data and the answers to the research question can provide more specific policy recommendations about the geographic scale of policy interventions. For example, child physical activity is likely influenced by neighborhood conditions based upon the routes a child might take through a neighborhood. The route through the neighborhood from the child's home to school dictates the relevant section of the neighborhood that matters for the decision of whether to walk or drive to school. The quality of this route may be distorted if proper geographic aggregation is not employed. Further, our results indicate variation in associative relationships between health status and Aesthetics due to geographic resolution. These differences allow the results to inform policy to a much greater extent because the relationship between individual parcel upkeep, near neighborhood condition and more distant neighborhood condition can be explicitly explored.
However, we should note that the utility and necessity of data at high spatial resolution will vary depending upon the research question being analyzed. Broad policy questions may effectively be answered with data at lower resolution. Further, analysis that requires observational data over large areas (e.g. counties, states, etc) are likely not practical applications for parcel-level data because of the time required for a sufficient amount of sampling for coverage in these larger areas. Nevertheless, the methodology and associated analysis presented here clearly indicate that collecting affordable high quality neighborhood data at the micro level is possible. Further, there are significant differences in how a particular location is classified in terms of neighborhood condition depending upon the geographic aggregation scheme used. In the absence of a sound, tested theoretical basis for defining a particular geographic aggregation, it is important to analyze the effect that geographic aggregation has on the correlative relationships being studied. Otherwise one cannot understand how the results are being affected by the particular geographic definitions used for the study. The systematic neighborhood observation methodology presented here demonstrates how parcel level data can be collected at minimal additional cost to answer these important questions.
We wish to thank Timothy Bray and Moyo Kimathi for assistance in instrument design and data collection. We also wish to thank anonymous reviewers for their valuable input.
Conceived and designed the experiments: JCM JM MC TL. Performed the experiments: JM. Analyzed the data: TL MC. Wrote the paper: TL MC JM JCM.
- 1. Bosma H, van de Mheen HD, Borsboom GJ, Mackenbach JP (2001) Neighborhood socioeconomic status and all-cause mortality. Am J Epidemiol 153: 363–371.H. BosmaHD van de MheenGJ BorsboomJP Mackenbach2001Neighborhood socioeconomic status and all-cause mortality.Am J Epidemiol153363371
- 2. Yen IH, Kaplan GA (1999) Neighborhood social environment and risk of death: Multilevel evidence from the Alameda County Study. Am J Epidemiol 149: 898–907.IH YenGA Kaplan1999Neighborhood social environment and risk of death: Multilevel evidence from the Alameda County Study.Am J Epidemiol149898907
- 3. Diez-Roux AV, Nieto FJ, Muntaner C, Tyroler HA, Comstock GW, et al. (1997) Neighborhood environments and coronary heart disease: A multilevel analysis. Am J Epidemiol 146: 48–63.AV Diez-RouxFJ NietoC. MuntanerHA TyrolerGW Comstock1997Neighborhood environments and coronary heart disease: A multilevel analysis.Am J Epidemiol1464863
- 4. Franzini L, Caughy MO, Spears W, Fernandez Esquer ME (2005) Neighborhood economic conditions, social processes, and self-rated health: A multilevel latent variables model. Soc Sci Med 61: 1135–1150.L. FranziniMO CaughyW. SpearsME Fernandez Esquer2005Neighborhood economic conditions, social processes, and self-rated health: A multilevel latent variables model.Soc Sci Med6111351150
- 5. Malmstrom M, Sundquist J, Johansson SE (1999) Neighborhood environment and self-reported health status: A multilevel analysis. Am J Public Health 89: 1181–1186.M. MalmstromJ. SundquistSE Johansson1999Neighborhood environment and self-reported health status: A multilevel analysis.Am J Public Health8911811186
- 6. Gordon-Larsen P, Nelson MC, Page P, Popkin BM (2006) Inequality in the built environment underlies key health disparities in physical activity and obesity. Pediatrics 117: 417–424.P. Gordon-LarsenMC NelsonP. PageBM Popkin2006Inequality in the built environment underlies key health disparities in physical activity and obesity.Pediatrics117417424
- 7. Kipke MD, Iverson E, Moore D, Booker C, Ruelas V, et al. (2007) Food and park environments: Neighborhood-level risks of childhood obesity in East Los Angeles. J Adolesc Health 40: 325–333.MD KipkeE. IversonD. MooreC. BookerV. Ruelas2007Food and park environments: Neighborhood-level risks of childhood obesity in East Los Angeles.J Adolesc Health40325333
- 8. Nelson MC, Gordon-Larsen P, Song Y, Popkin BM (2006) Built and social environments: Associations with adolescent overweight and activity. Am J Prev Med 31: 109–117.MC NelsonP. Gordon-LarsenY. SongBM Popkin2006Built and social environments: Associations with adolescent overweight and activity.Am J Prev Med31109117
- 9. Wickrama KAT, Wickrama KAS, Bryant CM (2006) Community influence on adolescent obesity: Race/ethnicity differences. Journal of Youth and Adolescence 35: 647–657.KAT WickramaKAS WickramaCM Bryant2006Community influence on adolescent obesity: Race/ethnicity differences.Journal of Youth and Adolescence35647657
- 10. Gee GC (2002) A multilevel analysis of the relationship between institutional and individual racial discrimination and health status. Am J Public Health 92: 615–623.GC Gee2002A multilevel analysis of the relationship between institutional and individual racial discrimination and health status.Am J Public Health92615623
- 11. Gutman LM, Sameroff SJ (2004) Continuities in depression from adolescent to young adulthood: Contrasting ecological influences. Dev Psychopathol 16: 967–984.LM GutmanSJ Sameroff2004Continuities in depression from adolescent to young adulthood: Contrasting ecological influences.Dev Psychopathol16967984
- 12. Leventhal T, Brooks-Gunn J (2003) Moving to Opportunity: An experimental study of neighborhood effects on mental health. Am J Public Health 93: 1576–1582.T. LeventhalJ. Brooks-Gunn2003Moving to Opportunity: An experimental study of neighborhood effects on mental health.Am J Public Health9315761582
- 13. Martinez P, Richters JE (1993) The NIMH community violence project: II. Children's distress symptoms associated with violence exposure. Psychiatry 56: 22–35.P. MartinezJE Richters1993The NIMH community violence project: II. Children's distress symptoms associated with violence exposure.Psychiatry562235
- 14. Ross CE (2000) Neighborhood disadvantage and adult depression. J Health Soc Behav 41: 177–187.CE Ross2000Neighborhood disadvantage and adult depression.J Health Soc Behav41177187
- 15. Xue Y, Leventhal T, Brooks-Gunn J, Earls F (2005) Neighborhood of residence and mental health problems of 5- to 11-year-olds. Arch Gen Psychiatry 62: 554–563.Y. XueT. LeventhalJ. Brooks-GunnF. Earls2005Neighborhood of residence and mental health problems of 5- to 11-year-olds.Arch Gen Psychiatry62554563
- 16. Muntaner C, Easton WW, Miech R, O'Campo P (2004) Socioeconomic position and major mental disorders. Epidemiol Rev 26: 53–62.C. MuntanerWW EastonR. MiechP. O'Campo2004Socioeconomic position and major mental disorders.Epidemiol Rev265362
- 17. O'Campo P, Rao RP, Gielen AC, Royalty W, Wilson M (2000) Injury-producing events among children in low-income communities: the role of community characteristics. Journal of Urban Health 77: 34–49.P. O'CampoRP RaoAC GielenW. RoyaltyM. Wilson2000Injury-producing events among children in low-income communities: the role of community characteristics.Journal of Urban Health773449
- 18. O'Campo PJ, Xue X, Wang M, Caughy MO (1997) Neighborhood risk factors for low birthweight in Baltimore: A multilevel analysis. Am J Public Health 87: 1113–1118.PJ O'CampoX. XueM. WangMO Caughy1997Neighborhood risk factors for low birthweight in Baltimore: A multilevel analysis.Am J Public Health8711131118
- 19. Papas MA, Alberg AJ, Ewing R, Helzlsouer KJ, Gary TL, et al. (2007) The built environment and obesity. Epidemiol Rev 29: 129–143.MA PapasAJ AlbergR. EwingKJ HelzlsouerTL Gary2007The built environment and obesity.Epidemiol Rev29129143
- 20. Rauh VA, Landrigan PJ, Claudio L (2008) Housing and health: Intersection of poverty and environmental exposures. Ann N Y Acad Sci1136: 276–288.VA RauhPJ LandriganL. Claudio2008Housing and health: Intersection of poverty and environmental exposures.Ann N Y AcadSci1136276288
- 21. Sellstrom E, Bremberg S (2006) The significance of neighbourhood context to child and adolescent health and well-being: A systematic review of multilevel studies. Scandinavian Journal of Public Health 34: 544–554.E. SellstromS. Bremberg2006The significance of neighbourhood context to child and adolescent health and well-being: A systematic review of multilevel studies.Scandinavian Journal of Public Health34544554
- 22. Wright RJ, Fischer EB (2003) Putting asthma in context: Community influences on risk, behavior and intervention. In: Kawachi I, Berkman LF, editors. Neighborhoods and health. New York, NY: Oxford University Press Inc. pp. 233–262.RJ WrightEB Fischer2003Putting asthma in context: Community influences on risk, behavior and intervention.I. KawachiLF BerkmanNeighborhoods and healthNew York, NYOxford University Press Inc233262
- 23. Burdette HL, Whitaker RC (2005) A national study of neighborhood safety, outdoor play, television viewing and obesity in preschool children. Pediatrics 116: 657–662.HL BurdetteRC Whitaker2005A national study of neighborhood safety, outdoor play, television viewing and obesity in preschool children.Pediatrics116657662
- 24. Franzini L, Elliott MN, Cuccaro P, Schuster M, Gilliland J, et al. (2007) Influences of physical and social neighborhood environments on children's physical activity and obesity. Am J Public Health 99: 271–278.L. FranziniMN ElliottP. CuccaroM. SchusterJ. Gilliland2007Influences of physical and social neighborhood environments on children's physical activity and obesity.Am J Public Health99271278
- 25. Molnar BE, Gortmaker SL, Bull FC, Buka SL (2004) Unsafe to play? Neighborhood disorder and lack of safety predict reduced physical activity among urban children and adolescents. American Journal of Health Promotion 18: 378–386.BE MolnarSL GortmakerFC BullSL Buka2004Unsafe to play? Neighborhood disorder and lack of safety predict reduced physical activity among urban children and adolescents.American Journal of Health Promotion18378386
- 26. Beyers JM, Bates JE, Pettit GS, Dodge KA (2003) Neighborhood structure, parenting processes, and the development of youths' externalizing behaviors: A multilevel analysis. Am J Community Psychol 31: 35–53.JM BeyersJE BatesGS PettitKA Dodge2003Neighborhood structure, parenting processes, and the development of youths' externalizing behaviors: A multilevel analysis.Am J Community Psychol313553
- 27. Caughy MO, O'Campo PJ, Nettles SM (2008) The effect of residential neighborhood on child behavior problems in first grade. Am J Community Psychol 42: 39–50.MO CaughyPJ O'CampoSM Nettles2008The effect of residential neighborhood on child behavior problems in first grade.Am J Community Psychol423950
- 28. Kohen DE, Brooks-Gunn J, Leventhal T, Hertzmann C (2002) Neighborhood income and physical and social disorder in Canada: Associations with young children's competencies. Child Dev 73: 1844–1860.DE KohenJ. Brooks-GunnT. LeventhalC. Hertzmann2002Neighborhood income and physical and social disorder in Canada: Associations with young children's competencies.Child Dev7318441860
- 29. Brownson RC, Hoehner CM, Day K, Forsyth A, Sallis JF (2009) Measuring the built environment for physical activity: State of the science. Am J Prev Med 36: S99–S123.RC BrownsonCM HoehnerK. DayA. ForsythJF Sallis2009Measuring the built environment for physical activity: State of the science.Am J Prev Med36S99S123
- 30. Schaefer-McDaniel N, Caughy MO, O'Campo PJ, Gearey W (2010) Examining methodological details of neighborhood observations and the relationship to health: A literature review. Soc Sci Med 70: 277–292.N. Schaefer-McDanielMO CaughyPJ O'CampoW. Gearey2010Examining methodological details of neighborhood observations and the relationship to health: A literature review.Soc Sci Med70277292
- 31. Raudenbush SW, Sampson RJ (1999) Ecometrics: Toward a science of assessing ecological settings, with application to the systematic social observation of neighborhoods. Sociological Methodology 29: 1–41.SW RaudenbushRJ Sampson1999Ecometrics: Toward a science of assessing ecological settings, with application to the systematic social observation of neighborhoods.Sociological Methodology29141
- 32. Openshaw S, Taylor PJ (1981) The modifiable areal unit problem. In: Wigley N, Bennett RJ, editors. Quantitative Geography. London: Routledge. pp. 60–69.S. OpenshawPJ Taylor1981The modifiable areal unit problem.N. WigleyRJ BennettQuantitative GeographyLondonRoutledge6069
- 33. Guo JY, Bhat CR (2007) Operationalizing the concept of neighborhood: Application to residential location choice analysis. Journal of Transport Geography 15: 31–45.JY GuoCR Bhat2007Operationalizing the concept of neighborhood: Application to residential location choice analysis.Journal of Transport Geography153145
- 34. Caughy MO, Hayslett-McCall KL, O'Campo PJ (2007) No neighborhood is an island: Incorporating distal neighborhood effects into multilevel studies of Child Deval competence. Health & Place 13: 788–798.MO CaughyKL Hayslett-McCallPJ O'Campo2007No neighborhood is an island: Incorporating distal neighborhood effects into multilevel studies of Child Deval competence.Health & Place13788798
- 35. Chaix B, Merlo J, Subramanian SV, Lynch J, Chauvin (2005) Comparison of a spatial perspective with the multilevel analytic approach in neighborhood studies: The case of mental and behavioral disorders due to psychoactive substance use in Malmo, Sweden, 2001. Am J Epidemiol 162: 171–182.B. ChaixJ. MerloSV SubramanianJ. LynchChauvin2005Comparison of a spatial perspective with the multilevel analytic approach in neighborhood studies: The case of mental and behavioral disorders due to psychoactive substance use in Malmo, Sweden, 2001.Am J Epidemiol162171182
- 36. Freisthler B, Bruce E, Needell B (2007) Understanding the geospatial relationship of neighborhood characteristics and rates of maltreatment for black, Hispanic, and white children. Social Work 52: 7–16.B. FreisthlerE. BruceB. Needell2007Understanding the geospatial relationship of neighborhood characteristics and rates of maltreatment for black, Hispanic, and white children.Social Work52716
- 37. Moudon AV, Lee C, Cheadle AD, Garvin C, Johnson DJ, et al. (2006) Operational definitions of walkable neighborhood: Theoretical and empirical insights. Journal of Physical Activity and Health 3: S99–S117.AV MoudonC. LeeAD CheadleC. GarvinDJ Johnson2006Operational definitions of walkable neighborhood: Theoretical and empirical insights.Journal of Physical Activity and Health3S99S117
- 38. Ross NA, Tremblay S, Graham K (2004) Neighborhood influences on health in Montreal, Canada. Soc Sci Med 59: 1485–1494.NA RossS. TremblayK. Graham2004Neighborhood influences on health in Montreal, Canada.Soc Sci Med5914851494
- 39. Boone-Heinonen J, Popkin BM, Song Y, Gordon-Larsen P (2010) What neighborhood area catpures built environment features related to adolescent physical activity? Health and Place 16: J. Boone-HeinonenBM PopkinY. SongP. Gordon-Larsen2010What neighborhood area catpures built environment features related to adolescent physical activity?Health and Place16
- 40. Caughy MO, O'Campo PJ, Patterson J (2001) A brief observational measure for urban neighborhoods. Health & Place 7: 225–236.MO CaughyPJ O'CampoJ. Patterson2001A brief observational measure for urban neighborhoods.Health & Place7225236
- 41. Anselin L (1995) Local indicators of spatial association: LISA. Geographical Analysis 27: 93–115.L. Anselin1995Local indicators of spatial association: LISA.Geographical Analysis2793115