A Virtual Audit Tool of Neighborhood Disorder using Google Street View: A Validation of LAND

• In Event: Novel Approaches to Operationalizing the Neighborhood Context

Abstract

Previous research has shown that neighborhood disorder (e.g., visual cues that may signal a deterioration or lack of social control) is a robust indicator of a host of neighborhood processes and has been linked with instances of crime, chronic stress, poor health status (Jones et al., 2011) and child behavioral problems (Jocson & McLoyd, 2015). However, the current methodological tools to reliably and efficiently quantify neighborhood disorder have lagged behind technological advances. Currently, much of the neighborhood disorder literature uses in-person audits as the primary method of assessment with lengthy checklists. However, this method, though reliable, valid, and an essential contributor to our understanding of neighborhood disorder thus far, is a time-consuming, difficult to reproduce, and cumbersome method that is limited by physical location and proximity to the neighborhoods being assessed (Furr-Holden et al., 2008; Jones et al., 2011). There have been important advances in using Google Street View (GSV) for virtual audits, including to assess neighborhood disorder (e.g., Mooney et al., 2014; Odgers et al., 2012). However, many of these methods focus on global characteristics of a given neighborhood (variably defined), therefore lacking the nuanced attention to variation that can occur on a block. In addition, there is significant variation in the ability to reliably assess signs of disordered using GSV, with studies reporting kappas ranging between .35 to .55 for most indicators. We set out to develop a valid and reliable tool for comparative analysis of neighborhood disorder via a virtual auditing method at the individual lot/parcel level to be able to identify exactly which neighborhoods experience disorder, to what extent, and the nature of the disorder.

After a literature review, existing audits were compiled and consistent elements were extracted across indices. Next, through an iterative process of coding GSV images and creating a training manual, coded items, criteria and processes were refined. We coined our neighborhood disorder coding system: Lot Level Assessment of Neighborhood Disorder (LAND). Then, a total of 710 block faces on 355 street segments were coded in Detroit, MI, an ideal place to validate the tool given high levels of vacant lots, abandoned houses, varied signs of disorder, diverse population of residents, and increase in economic decline. We used classical test and item response theories to assess the dimensionality, reliability and equivalence. We tested reliability between coders on 20% of the sample (71 segments;146 blocks faces) and reliability was adequate at both the individual lot level and block face level (see Table 1). Therefore, we can conclude that the coding scheme is reliably assessing the same constructs between coders. The next step will be to validate these codes using existing data to test the construct validity of LAND.

LAND measures neighborhood disorder via virtual audit methods across neighborhoods of differing wealth and violence levels. Its development ushers in the ability to reliably quantify within and between neighborhood conditions which affect health and can contribute to an evidence base for public health and urban planning recommendations about neighborhood conditions.

Authors

• Elizabeth Ann Shewark, Michigan State University
  Presenting Author

• Amber L Pearson, Michigan State University
  Non-Presenting Author

• C.J. Sivak, Michigan State University
  Non-Presenting Author

• Hyunseo Park, Michigan State University
  Non-Presenting Author

• S. Alexandra Burt, Michigan State University
  Non-Presenting Author