1) The document describes a study that measured levels of social cohesion and collective efficacy across 343 Chicago neighborhoods using survey data and census information.
2) Two measures of social cohesion and informal social control - willingness to help neighbors and sense of shared values - were highly correlated, so the researchers combined them into a single measure of collective efficacy.
3) The researchers then used hierarchical statistical modeling to analyze how collective efficacy varied within and between individuals and neighborhoods, while accounting for missing data. They modeled individual responses as varying around a latent neighborhood average.
1) The document describes a study that measured levels of social cohesion and collective efficacy across 343 Chicago neighborhoods using survey data and census information.
2) Two measures of social cohesion and informal social control - willingness to help neighbors and sense of shared values - were highly correlated, so the researchers combined them into a single measure of collective efficacy.
3) The researchers then used hierarchical statistical modeling to analyze how collective efficacy varied within and between individuals and neighborhoods, while accounting for missing data. They modeled individual responses as varying around a latent neighborhood average.
1) The document describes a study that measured levels of social cohesion and collective efficacy across 343 Chicago neighborhoods using survey data and census information.
2) Two measures of social cohesion and informal social control - willingness to help neighbors and sense of shared values - were highly correlated, so the researchers combined them into a single measure of collective efficacy.
3) The researchers then used hierarchical statistical modeling to analyze how collective efficacy varied within and between individuals and neighborhoods, while accounting for missing data. They modeled individual responses as varying around a latent neighborhood average.
fight broke out in front of their house, and tested both survey measures against inde- gence of a residential
a residential stability factor is con-
(v) the fire station closest to their home was pendently recorded incidents of homicide sistent with much past research (13). threatened with budget cuts. “Social cohe- aggregated to the NC level (23). Homi- sion and trust” were also represented by five cide is one of the most reliably measured Analytic Models conceptually related items. Respondents crimes by the police and does not suffer were asked how strongly they agreed (on a the reporting limitations associated with The internal consistency of a person measure five-point scale) that “people around here other violent crimes, such as assault and will depend on the intercorrelation among are willing to help their neighbors,” “this is rape. items and the number of items in a scale. a close-knit neighborhood,” “people in this Ten variables were constructed from The internal consistency of a neighborhood neighborhood can be trusted,” “people in the 1990 decennial census of the popula- measure will depend in part on these factors, this neighborhood generally don’t get along tion to reflect neighborhood differences in but it will hinge more on the degree of with each other,” and “people in this neigh- poverty, race and ethnicity, immigration, intersubjective agreement among informants borhood do not share the same values” (the the labor market, age composition, family in their ratings of the neighborhood in last two statements were reverse coded). structure, homeownership, and residential which they share membership and on the Responses to the five-point Likert scales stability (see Table 2). The census was sample size of informants per neighborhood were aggregated to the neighborhood level independent of the PHDCN CS; more- (26). To study reliability, we therefore for- as initial measures. Social cohesion and in- over, the census data were collected 5 mulated a hierarchical statistical model rep- formal social control were closely associated years earlier, which permitted temporal resenting item variation within persons, per- across neighborhoods (r 5 0.80, P , sequencing. To assess whether a smaller son variation within neighborhoods, and 0.001), which suggests that the two mea- number of linear combinations of census variation between neighborhoods. Compli- sures were tapping aspects of the same la- characteristics describe the structure of cating the analysis is the problem of missing tent construct. Because we also expected the 343 Chicago neighborhoods, we con- data: inevitably, some persons will fail to that the willingness and intention to inter- ducted a factor analysis (24). respond to some questions in an interview. vene on behalf of the neighborhood would Consistent with theories and research We present our hierarchical model as a series be enhanced under conditions of mutual on U.S. cities, the poverty-related vari- of nested models, one for each level in the trust and cohesion, we combined the two ables given in Table 2 are highly associat- hierarchy (27). scales into a summary measure labeled col- ed and load on the same factor. With an Level 1 model. Within each person, Yijk, lective efficacy (21). eigenvalue greater than 5, the first factor the ith response of person j in neighborhood The measurement of violence was is dominated by high loadings (.0.85) for k, depends on the person’s latent perception achieved in three ways. First, respondents poverty, receipt of public assistance, un- of collective efficacy plus error: were asked how often each of the follow- employment, female headed-families, and OaD 9 ing had occurred in the neighborhood dur- density of children, followed by, to a lesser ing the past 6 months: (i) a fight in which extent, percentage of black residents. Yijk 5 pjk 1 p pijk 1 eijk (1) p51 a weapon was used, (ii) a violent argument Hence, the predominant interpretation re- between neighbors, (iii) a gang fight, (iv) volves around concentrated disadvan- Here Dpijk is an indicator variable taking on a sexual assault or rape, and (v) a robbery tage—African Americans, children, and a value of unity if response i is to item p in or mugging. The scale construction for single-parent families are differentially the 10-item scale intended to measure col- perceived neighborhood violence mir- found in neighborhoods with high con- lective efficacy and zero if response i is to rored that for social control and cohesion. centrations of poverty (25). To represent some other item. Thus, ap represents the Second, to assess personal victimization, this dimension parsimoniously, we calcu- “difficulty” of item p, and pjk is the “true each respondent was asked “While you lated a factor regression score that weight- score” for person jk and is adjusted for the have lived in this neighborhood, has any- ed each variable by its factor loading. difficulty level of the items to which that one ever used violence, such as in a mug- The second dimension captures areas of person responded (28). The errors of mea- ging, fight, or sexual assault, against you or the city undergoing immigration, especial- surement, eijk, are assumed to be indepen- any member of your household anywhere ly from Mexico. The two variables that dent and homoscedastic (that is, to have in your neighborhood?” (22). Third, we define this dimension are the percentage equal standard deviations). of Latinos (approximately 70% of Latinos Level 2 model. Across informants within in Chicago are of Mexican descent) and neighborhoods, the latent true scores vary Table 2. Oblique rotated factor pattern (Loadings the percentage of foreign-born persons. randomly around the neighborhood mean: $ 0.60) in 343 Chicago neighborhoods. (Data are Similar to the procedures for concentrated from the 1990 census.) disadvantage, a weighted factor score was pjk 5 hk 1 rjk, rjk ; N~0,tp! (2) created to reflect immigrant concentra- Here hk is the neighborhood mean collec- Variable Factor loading tion. Because it describes neighborhoods tive efficacy, and random effects rjk associ- Concentrated disadvantage of ethnic and linguistic heterogeneity, ated with each person are independently, Below poverty line 0.93 there is reason to believe that immigrant normally distributed with variance tp, that On public assistance 0.94 concentration may impede the capacity of is, the “within-neighborhood variance.” Female-headed families 0.93 residents to realize common values and to Level 3 model. Across neighborhoods, Unemployed 0.86 Less than age 18 0.94 achieve informal social controls, which in each neighborhood’s mean collective effi- Black 0.60 turn explains an increased risk of violence cacy hk varies randomly about a grand Immigrant concentration (1–5, 7). mean: Latino 0.88 The third factor score is dominated by Foreign-born 0.70 two variables with high (.0.75) loadings: h k 5 g 1 u k, uk ; N~0,th! (3) Residential stability the percentage of persons living in the same where g is the grand mean collective effica- Same house as in 1985 0.77 Owner-occupied house 0.86 house as 5 years earlier and the percentage cy, uk is a normally distributed random effect of owner-occupied homes. The clear emer- associated with neighborhood k, and th is
920 SCIENCE z VOL. 277 z 15 AUGUST 1997 z www.sciencemag.org