Neighborhoods, Schools, and Adolescent Violence: Ecological Relative Deprivation, Disadvantage Saturation, or Cumulative Disadvantage?

While school FRPL has been shown to be an incomplete measure of school socioeconomic disadvantage, FRPL status nevertheless captures dimensions of educational disadvantage that other income-based indicators do not (Domina et al., 2018). Beyond income, school-level rates of parental education have also been shown to shape educational achievement and attainment (Crosnoe, 2009; Owens, 2010). Because the neighborhoods and violence literature tends to emphasize consequences of exposure to low-socioeconomic resources (Chang et al., 2016), aggregated reports of low-parent education and parental welfare receipt are used in combination with school FRPL reports to more fully capture school socioeconomic disadvantage.

To understand the representativeness of the analytic sample with respect to school FRPL rates, Add Health school FRPL rates were compared with those reported by the NCES for the earliest available time period (1999-2000) (McFarland et al., 2019). The NCES reports the proportion of public school students attending “low poverty,” “mid-low poverty,” “mid-high poverty,” and “high poverty” categories of schools. Low poverty schools are those where < =25% of the student body is FRPL eligible. The cutoffs for mid-low poverty, mid-high poverty, and high poverty schools are 26-50%, 51-75%, and > =76%, respectively. In 1999-2000, 45% of public school students attended low poverty schools, 25% attended mid-low poverty schools, 16% attended mid-high poverty schools, and 12% attended high poverty schools. These categories were recreated using the Add Health FRPL data. In the present analytic sample weighted according to Add Health guidelines, 48.9% of students attend low poverty schools, 33.2% attend mid-low poverty schools, 12.6% attend mid-high poverty schools, and 5.3% attended high poverty schools. Thus, students attending “high poverty” schools are likely underrepresented in the current sample. This underrepresentation may bias the present results, but it is important to note that effects of nonresponse bias in multivariable models have been demonstrated be to minimal when controlling for design variables (Amaya & Presser, 2017; Rindfuss et al., 2015).

In these analyses, the measures of racial composition and student population size are based on aggregations of available in-school and in-home Add Health survey responses when NCES data are missing.