For each school year, observations were weighted to ensure that data were representative of the enrollment population for that year. Weights were calculated using a raking process, with race/ethnicity, a combination of borough and district public health office (DPHO) neighborhoods (neighborhoods defined by low income and disproportionate rates of morbidity and mortality), free lunch status (free versus not free), grade, sex, age, and school type (elementary versus middle) as population marginal control totals.‡ To test for obesity prevalence trends from 2006-07 to 2010-11, a multivariate model was built that included a linear term for time, along with sex, age, race/ethnicity, school borough, free lunch status, DPHO, place of birth, language spoken at home, and an interaction of age, sex, and race/ethnicity, as covariates. School and student codes were used as cluster variables, and statistical procedures that account for intercluster correlation were used to ensure that variance estimates were calculated correctly. Separate multivariate models were built to test trends for age group, race/ethnicity, and socioeconomic status. The significance level for all analyses was set at p<0.05. For presentation of prevalence estimates by school neighborhood poverty, school postal codes were characterized by the percentage of residents living below the federal poverty level (as defined by the 2000 U.S. Census). The percentage of residents living below the poverty level in the school postal code area was categorized as low (<10% of residents), medium (10% to <20%), high (20% to <30%), and very high (≥30%).