ABSTRACT
Neighborhood Influences on Adolescent Cigarette and Alcohol Use
By Ying-Chih Chuang




            This dissertation contains three studies. The first study addresses whether neighborhood influences on adolescent substance use are mediated through parental and peer characteristics. The second study compares differences in neighborhood mechanisms on adolescent substance use between census tracts and block groups. The third study uses a typology approach for measuring neighborhoods and examines differences in the relationship between parenting and adolescent substance use across different neighborhood types.
            Study hypotheses are tested through secondary analysis of data from the Family Matters Project, a randomized experimental study designed to determine whether a family-directed intervention prevented adolescent cigarette and alcohol use. The residential addresses of adolescents were matched with 1990 Census tracts and block groups. The final data set includes 959 cases in the tract sample and 924 cases in the block-group sample.
            The findings of these studies are as follows:
            Study 1: For adolescent cigarette use, low social economic status neighborhoods increase parental monitoring, which in turn decreases adolescent smoking. For adolescent alcohol use, high SES neighborhoods increase parent drinking and low SES neighborhoods increase parental monitoring as well as peer drinking. Parent drinking, parental monitoring, and peer drinking, each in turn, has an influence on adolescent drinking.
            Study 2: The pattern of relationships in the tract model of alcohol use is similar to that in the block-group model of alcohol use. The block-group model of cigarette use shows a different pattern of relationships than the tract model of cigarette use. Both models suggest that low SES neighborhoods increase parental monitoring, but, in addition, the block-group model suggests that high SES neighborhoods decrease parent smoking and Hispanic concentration decreases adolescent smoking.
            Study 3: Six types of neighborhood are identified by this study. They are: (1) rural low SES neighborhoods; (2) urban White middle SES neighborhoods; (3) urban White high SES neighborhoods; (4) suburban White middle SES neighborhoods; (5) rural White middle SES neighborhoods, and (6) urban Black low SES neighborhoods. This study found that adolescent cigarette and alcohol use does not vary by neighborhood types, but the impact of parenting on adolescent cigarette and alcohol use differs by neighborhood types.

© 2001
Ying-Chih Chuang
All Rights Reserved
 
 

CHAPTER ONE
Paper One: Neighborhood Influences on Adolescent Cigarette and Alcohol Use: Mediating Effects through Parents and Peers

Introduction

            The purpose of this study is to understand how neighborhoods influence adolescents through parents and peers. Traditionally, parent and peer influences have been the two major focuses in research about adolescent substance use (Kandel, 1996). It has been suggested that to fully understand the nature of adolescent substance use, studies need to go beyond those interpersonal influences to consider how different social settings exert influence (Brook, Nomura, & Cohen, 1989). This study extends the interpersonal focus by including neighborhood influences in the etiology of adolescent substance use.
Only limited attention has been paid to understanding how neighborhoods influence adolescent substance use and the study findings are inconsistent (Abdelrahman et al., 1998; Allison et al., 1999; Brook, Nomura, & Cohen, 1989; Case, 1991; Crum et al., 1996; Darling & Steinberg, 1997; Dembo, 1985; Elliott et al., 1996, Karvonen & Rimpela, 1997; Simon et al., 1996). Some studies found strong neighborhood effects, which showed that residing in a disadvantaged neighborhood increased the likelihood that adolescents were offered various kinds of substances (Crum, 1996) and developed heavy drinking patterns (Karvonen, 1997). In other studies, neighborhoods were found to have small effects compared with peer substance use (Abdelrahman et al., 1998) or were found to have no influence at all on adolescent substance use (Allison, 1999).
            Mixed results concerning neighborhood influences may be due to the fact that prior studies have generally not identified mediating mechanisms. For those studies that found small or no neighborhood effects, neighborhood influences on adolescent substance use were assessed by simultaneously controlling family and peer factors in the same regression model. Because neighborhoods may exert influence on adolescents directly and indirectly through family and peer factors, treating neighborhood, family, and peer as independent domains and including them in the same regression model may obscure the total effects of neighborhoods. In order to answer the questions “Do neighborhoods matter for adolescent substance use?” and “How do neighborhoods influence adolescent substance use?,” identification of neighborhood mediating pathways is an essential task. In this study, parental and peer factors are proposed as two potential mechanisms through which neighborhoods may influence adolescents. The interconnections of neighborhood, family and peer influences are theoretically hypothesized and examined.

Theories of Neighborhood Influences

            Social disorganization theory (Shaw & McKay, 1969), Wilson’s theory (Wilson, 1987), and Coleman’s concept of social capital (Coleman, 1988) suggest potential mediating mechanisms for neighborhood influences on adolescent and child development. These three theories are not distinctly different because some of the mechanisms that they propose are similar. I first review each theory separately and then use the classification of neighborhood influences proposed by Jencks and Mayer (1990) to compare the similarities and differences across the theories.
            Shaw and McKay (1969) found that the prevalence of delinquency and other social problems are closely associated with neighborhood characteristics, such as high rates of population turnover, families on public assistance, foreign-born or Black families, and low rates of home ownership. They further found that the most disadvantaged neighborhoods usually have the lowest rental costs, which attract immigrants because they tend to be of lower economic status. Because immigrants from different places have different cultural backgrounds and speak different languages, they may have difficulties in reaching consensus on values. The most disadvantaged neighborhoods also have high residential mobility that further contributes to an inconsistency of values and norms. The result is that neighborhoods have low informal social control and low ability to socialize residents to conventional values. Lack of neighborhood consensus may further contribute to a decline of neighborhood organizations and give rise to delinquent groups. Children who live in this kind of neighborhood tend to look up to gang members because they usually have the highest economic status in the community. Local delinquent groups also give local youth the means of securing economic gain and a sense of belonging. Children are then exposed to a diversity in norms and standards of behavior, which leads to a propensity towards engaging in criminal activities.
            Wilson (1987) suggested that the origins of recent disadvantaged neighborhoods in inner-city areas come from changing economic structures and out-migration of middle-class Blacks. The out-migration of middle-class Blacks creates an environment where local basic organizations disorganize, conventional norms cannot be maintained, and high proportions of extremely poor people are isolated from the job network system. People who are excluded from the job network system then experience long-term unemployment and seek illicit activities to support their income. The large group of jobless Black males in inner-cities further creates an “unmarriageable pool,” which leads to the persistence of high rates of female headed households and out of wedlock births.
            The concept of social capital proposed by Coleman (1988) describes how social relationships can influence one’s decisions and actions, which provides an explanatory linkage between neighborhoods and individuals. Coleman defined social capital by its function. Social capital is embedded in social relationships and serves as resources for people to achieve their interests. There are three forms of social capital: (1) obligations, expectations, and trustworthiness that exist in social structures; (2) information channels imbedded in social relationships; and (3) norms and effective sanctions against deviant behaviors. Applying the concept of social capital in neighborhood research, if residents feel obligations, expectations, and trustworthiness toward their neighbors, they have more social capital to draw upon to facilitate their actions. For example, parents can ask neighbors to take care of their children while they are away. Social capital can be enhanced by certain forms of social structures: closure of social networks and presence of local organizations. The simplest example of closure of a social network is a three-person group in which member A knows member B, member B knows member C, and member C knows member A. Because all members in the group know each other, member A and B can combine forces to constrain member C if member C exhibits deviant behaviors. Applying the concept of closure of social networks to neighborhood research, if parents know other parents in the same neighborhood, they can share information and help monitor children who are not their own. Besides closure of social networks, social capital can be enhanced by the presence of local organizations. Local organizations can serve as places to gather residents’ resources and conduct collective actions to improve the quality of a neighborhood. Social capital can also be found inside a family, such as the physical presence of parents and the attention that parents pay to their children. The social capital inside a family serves as a channel through which children access their parents’ human capital (i.e., education, knowledge, and financial resources).
            These three neighborhood theories proposed by Shaw and McKay, Wilson, and Coleman overlap in terms of mechanisms of neighborhood influences on adolescent behaviors (Furstenberg & Hughes, 1997; Gephart, 1997). The mechanisms identified by these theories can be summarized by the classification of neighborhood influences proposed by Jencks and Mayer (1989; 1990). Using their classification, Wilson’s theory and social disorganization theory are interested in what Jencks and Mayer referred to as collective socialization, institutional, and epidemic models. The collective socialization models emphasize how adults outside a family influence children. Wilson’s theory proposes that an impoverished neighborhood usually lacks role models for children; therefore children are less likely to have prosocial behaviors and more likely to get involved in problem behaviors such as using illicit drugs and committing crime. Like Wilson’s theory, social disorganization theory also addresses the problem of lack of role models but further emphasizes that disappearance of informal social control is the main mechanism contributing to adolescent problem behaviors. Examples of informal social control are the willingness of an adult to intervene with teenagers hanging around the street corner or to stop people who are destroying public facilities (Sampson, Raudenbush, & Earls, 1997).
            The institution models emphasize the importance of local organizational or institutional resources. Both Wilson’s theory and social disorganization theory suggest that a disadvantaged neighborhood with low sustainable local organizations may create an environment where families and adolescents lack sufficient resources to facilitate adolescent development. For example, the quality of local schools may be affected by scarce learning equipment and lack of devoted teachers. Parks, libraries, and community centers either do not exist or are not well maintained.
            The epidemic models explain that adolescents engage in problem behaviors because other peers who live in the neighborhood also exhibit these behaviors. The epidemic models assume that problem behaviors are contagious, and mainly through peer influences (Crane, 1991). Social disorganization theory uses case studies to illustrate that youth became involved in delinquent activities by learning or pressuring from neighborhood peers who have problem behaviors.
            Jencks and Mayer also suggested the relative deprivation models. This mechanism assumes that people judge their success by the people who are around them. Once people feel that they cannot compete successfully, they may create a deviant subculture to let them psychologically adapt to their environment. For example, low SES children may have a better image about themselves and do better in their academic performance if they do not attend a school with a high proportion of high SES children. According to Jencks and Mayer, the relative deprivation models explain particularly well the behaviors that occur in an area where success is unequally distributed. For example, deviant subcultures are more likely to be found in a low SES neighborhood located next to a high SES neighborhood than in a low SES neighborhood located next to another low SES neighborhood.
            A fifth mediating path is parental relationships, which is proposed by Leventhal and Brooks-Gunn (Leventhal & Brooks-Gunn, 2000). Although Jencks and Mayer’s classification of neighborhood mechanisms informs many potential mediating paths of neighborhood influences on adolescent behaviors, the paths do not directly explain how neighborhoods influence the function of a family, which may be essential to understanding neighborhood influences on adolescent development. This mediating path, parental relationships, addresses how an adolescent’s proximate interpersonal environment, the family, serves as a linkage between the neighborhood and the adolescent. Unlike the collective socialization models, which emphasize how adults in a neighborhood influence children who are not their own, this mediating path suggests parents are like gate keepers who manage risks and opportunities for their own children (Furstenberg, 1993). Leventhal and Brooks-Gunn’s stated that neighborhoods influence a child’s or an adolescent’s well-being through parents’ characteristics, parenting behaviors, home environment, and the neighborhood’s helping social networks. By living in an impoverished neighborhood, parents are more likely to have poor mental health, inadequate coping skills, and low parenting efficacy. These characteristics, in turn, can affect parenting behaviors (Elder, et al., 1995). For example, instead of showing warmth to their children, parents in a disadvantaged neighborhood may use harsh control and verbal aggression toward their children (Earls et al., 1994). Parents in an impoverished neighborhood also show less ability to arrange a healthy home environment, such as by not making books and reading materials available at home (Klebanov et al., 1994). The mechanisms of parental relationships are also proposed by Coleman (1988). Coleman suggested that neighborhood social support may lessen parental stress associated with living in an impoverished neighborhood. For example, if parents in a neighborhood know each other and form social networks, they can share information and help each other supervise their children.
            The summary of the above theoretical mechanisms points out that neighborhoods may influence adolescents by various mediating paths including local organizations, informal social control, residents’ consensus on conventional norms, deviant peer groups, helping social networks, and parents’ characteristics. Our data are limited for examining some neighborhood mediating characteristics (e.g., informal social control and community norms), but the data provide rich family and peer characteristics such as parent-adolescent closeness, parental monitoring, parent cigarette/alcohol use, and peer cigarette/alcohol use. Our goal is to examine whether neighborhoods influence adolescent substance use through these parental and peer characteristics.

Family and Peers as Mediators between Neighborhoods and Adolescents

            Only a small number of prior studies have empirically investigated parents and peers as mediators of neighborhood influences on adolescent behaviors. The studies assessed various kinds of adolescent behaviors, such as teenage pregnancy, substance use, delinquency, and general problem behaviors (Case, 1991; Hogan & Kitagawa, 1985; Simon et al., 1996; Stern & Smith, 1995). These studies also used different ways to measure neighborhood influences. Some studies used census tracts or data from other large social surveys (Hogan & Kitagawa, 1985; Klebanov et al., 1994; Simon et al., 1996; Case, 1991), and some used the parents’ or adolescents’ reports regarding neighborhood qualities (Simons et al., 1997; Stern & Smith, 1995; Brook et al., 1989). The studies generally concluded that residing in a disadvantaged neighborhood decreases parent-adolescent closeness, parental supervision, and parental monitoring.
            Regarding parent-adolescent closeness, Klebanov et al. (1994) found that residing in a neighborhood with a high proportion of low-income people reduces a mother’s warmth toward her children. This relationship still holds after taking into account family SES, mother’s mental health, mother’s behavioral coping and social support. Brook, Nomura, and Cohen (1989) also found that neighborhoods have impacts on adolescent drug involvement through adolescents’ affection toward their parents. Specifically, when a neighborhood was rated as cohesive, safe, and a good place to live, adolescents reported higher affection toward parents and this affectionate relationship was further associated with adolescents’ lower drug involvement. Unlike the above studies, Stern and Smith (1995) found parent-child attachment was not a significant mediator, but found that parent-child involvement mediates the relationship between neighborhoods and adolescent delinquency. When a neighborhood has fewer social problems and higher residents’ satisfaction, parent-child involvement is increased. High parent-child involvement in turn reduces adolescent delinquent behaviors.
            Regarding parental supervision and monitoring, Hogan and Kitagawa (1985) found that parental control mediates negative neighborhood effects on teenage first pregnancy. Parents in a high-risk neighborhood tend to have difficulties in supervising adolescents about dating, which leads to an increase in teenage pregnancy. The significant role of supervision was also reported by Stern and Smith (1995). They found that a disadvantaged neighborhood increases adolescent delinquency indirectly through parental supervision and consistency of parents’ discipline. Simon et al. (1997) found that disadvantaged neighborhood characteristics (i.e., low quality of schools and local medical organizations) lead to ineffective parenting both directly and indirectly through social support, mother’s mental health, and mother’s perception of negative life events. When including both parent-adolescent affectionate relationship and parental monitoring to measure the quality of parenting, Simon et al. (1996) found that a low SES neighborhood negatively impacts adolescent conduct problems indirectly through reducing the quality of parenting among male adolescents; however, the same result was not found for female adolescents.
            No studies have addressed the mediating effects of parent substance use. However, prior studies showed that living in an impoverished neighborhood increases the likelihood that parents have various kinds of stresses. Mothers who live in a disadvantaged neighborhood usually do not have access to child care resources, do not have help from professionals and neighbors, and perceive greater difficulty in raising a child (Garbarion & Sherman, 1980). These enduring life strains may lead parents to use substances as a coping response to adapt to a stressful environment (Shiffman & Wills, 1985). Once parents start to use substances, adolescents may learn to use substances as well because of modeling effects (Bandura, 1986).
            Compared to the studies that focused on the role of parents as mediators, fewer studies have paid attention to the mediating effects of peer behaviors. Simon et al. (1996) reported that male adolescents who live in a low SES neighborhood are more likely to have deviant peers and having deviant peers is further associated with more male adolescent conduct problems. Simon et al. did not find that neighborhood SES is associated with having delinquent peers among female adolescents but they found deviant peers mediate the effects of the proportion of single-parent families in a neighborhood and female adolescents’ conduct problems. Case (1991) also provided evidence for the effects of neighborhood peer influence. He found that adolescent behaviors, such as crime, drug use, church attendance, alcohol use, and idleness, are influenced by the proportion of peers who have the same behaviors in the same neighborhood.

Conceptual Model

            Figure 1.1 shows the conceptual model for this study. The neighborhood characteristics include low SES, high SES, residential mobility, immigrant concentration, White and Black racial composition, and Hispanic concentration. Low SES, residential mobility, and immigrant concentration are the three main structural characteristics of a disadvantaged neighborhood suggested by social disorganization theory. I also include high SES, which follows from the series of studies by Brooks-Gunn and her colleagues (Brooks-Gunn et al., 1993; and Brooks-Gunn et al., 1997). Brooks-Gunn and her colleagues proposed that the effects of the presence of low SES neighbors are different from the effects of the lack of high SES neighbors. Each of the effects should be separately tested because they represent different neighborhood mechanisms. The presence of low SES neighbors reflects the epidemic models, which suggests the contagious influences of deviant peers on adolescents. The lack of high SES neighbors reflects the collective socialization models, which suggest that adolescents do not have role models to learn about conventional values and behaviors. Brooks-Gunn and her colleagues empirically tested high SES and low SES neighborhood effects separately and found the presence of affluent neighbors benefits children and adolescents’ cognitive development and lessens their chances of school drop out and teenage pregnancy (Brooks-Gunn et al., 1993).
            In addition to the above four neighborhood characteristics, I include White and Black racial composition as well as Hispanic concentration. Most of the previous studies did not distinguish the effects of social economic status and race or ethnicity (Massey, 1998). For example, studies usually combined the proportion of Blacks and the proportion of low SES residents to represent neighborhood poverty. Although SES and race are correlated, they represent different origins of a disadvantaged neighborhood and therefore can be considered conceptually distinct. The concentration of Blacks in a few neighborhoods reflects the prevalence of residential segregation in the U.S. today. These are the results of the persistence of white racial prejudice and discrimination in the housing markets and banking industries (Massey, 1996). Therefore, the proportion of Blacks refers to the extent of residential segregation, which cannot be shown by combining indicators of race and SES in a single neighborhood domain. The inclusion of Hispanic concentration also shows the uneven distribution of resources across neighborhoods due in part to prejudicial effects. Because the geographical concentration of Hispanics is poorly correlated with the geographical concentration of Whites and Blacks, I separate Hispanic concentration as another neighborhood domain.
            I expect that a disadvantaged neighborhood can increase adolescent substance use through reducing parent-adolescent closeness and parental monitoring and through increasing parent substance use. The three family mediating paths represent the mechanisms of parental relationships and mechanisms of collective socialization. The daily experience of living in an impoverished neighborhood creates high stress for parents (Garbarion & Sherman, 1980). Parents who are under this condition may easily become depressed, show reduced affection toward their children, reduce the time they spend with their children, have reduced energy to monitor their children, and even use substances to cope with the stressful environment (Klebanov et al., 1994; Leventhal & Brooks-Gunn, 2000; Simon et al, 1996). The other possible explanation is that the family’s norms and behaviors are easily influenced by other families living in the neighborhood (Wilson, 1991b). Residing in a neighborhood where few adults have conventional values, have regular jobs and have parenting abilities may result in parents not having role models to learn about parenting. The result is a decline in parent-adolescent closeness and parental monitoring, and an increase in parent substance use. Besides mediating effects of parents, a disadvantaged neighborhood is also expected to increase adolescent substance use through increasing the proportion of peers in the neighborhood who use substances. This path represents the mechanism of the epidemic models where adolescents’ substance using behavior is obtained by learning or pressuring from neighborhood peers (Case, 1991).
            I also expect that parents may exert their influences on adolescents through establishing a foundation for adolescents to affiliate with drug using or non-drug using peers. Studies have shown that low parent-adolescent bonding, low parental supervision and monitoring, and high parental substance use may increase adolescent substance use directly and indirectly through peer substance use (Kandel, 1996; Hoffmann, 1993; Patterson & Dishion, 1985).
            Besides indirect paths of neighborhoods on adolescents, a neighborhood may influence adolescents through other neighborhood characteristics such as the consensus of neighborhood residents, informal social control, quality of local organization, helping social networks, and the presence of delinquent gang groups (Sampson, Raudenbush, & Earls, 1997; Elliott et al., 1996; Simcha-Fagan & Schwartz, 1986). These various mediating mechanisms, which were not measured by this study, are represented by the direct path from the neighborhood characteristics to adolescents.
            The above specification about the conceptual model shows the potential causal directions from neighborhoods to parents, peers, and adolescents. However, another competing explanation should be considered, which is that the correlation between disadvantaged neighborhoods and individual behaviors is because of the non-random selection of individuals into different neighborhoods (Tienda, 1990). This selection bias is due to individual characteristics that constrain residential options and mobility for some segments of the population but not for the others. For example, low-income people have limited residential options because they can only afford cheap rents usually found in disadvantaged neighborhoods.
            In my conceptual model, because adolescents who use substances are more likely to move into or be economically forced to stay in an impoverished neighborhood, adolescent substance use is not necessarily the outcome of neighborhood influences. Therefore, controlling baseline adolescent substance use is important because it allows me to detect the net change of adolescent substance use after neighborhood contexts are measured. This may reduce the selection bias in estimating the influences of neighborhoods on adolescent substance use. The selection bias can also exist in the relationship between neighborhoods and parent substance use. However, there is not enough variation in the change of parent substance use between baseline and follow-up surveys to allow control of the selection bias by including baseline parent substance use.
            Another selection bias may exist in the relationship between peer and adolescent substance use. Previous research suggests that adolescents who use substances may choose or attract substance-using friends (Bauman & Ennett, 1996). In this case, peer selection rather than the influence of peer use accounts for the relationship between peer and adolescent substance use. In my conceptual model, the inclusion of baseline adolescent cigarette and alcohol use reduces the peer selection effects.
            The conceptual model is informed by the various theories described above. However, both social disorganization theory and Wilson’s theory were developed to explain the concentration of poverty in inner-city areas, but the data for this study are not from youth living only in inner-city areas. Although I am aware of this potential problem, I believe these theories can also adequately provide guidance in conceptualization of neighborhood problems other than inner-city areas. Social disorganization theory and Wilson’s theory were previously applied to rural samples (Simon et al., 1997; Simon et al., 1996) and national samples (Brooks-Gunn et al., 1993; Klebanov et al., 1997). These studies demonstrated that non-metropolitan neighborhood problems can be well conceptualized by using social disorganization theory and Wilson’s theory.

Study Hypotheses

            Six major hypotheses are tested in this study:
(1) Adolescents who live in a neighborhood with a higher level of disadvantaged condition, as indicated by a higher proportion of low SES residents, a lower proportion of high SES residents, higher residential mobility, a higher proportion of immigrants, a higher proportion of Blacks, or a higher proportion of Hispanics, have higher levels of cigarette and alcohol use compared with adolescents who live in a non-disadvantaged neighborhood. The relationship between each neighborhood characteristic and adolescent substance use is tested separately for cigarette and alcohol use.
(2) Adolescents who live in a neighborhood with a higher level of disadvantaged condition have lower parent-adolescent closeness, lower parental monitoring, higher parental substance use, and higher peer substance use compared with adolescents who live in a non-disadvantaged neighborhood. The relationship between each neighborhood characteristic and each parental characteristic or peer use is tested separately for cigarette and alcohol use.
(3) Adolescents who have lower parent-adolescent closeness, lower parental monitoring, higher parental substance use, and higher peer substance use have higher levels of cigarette and alcohol use than those who have higher parent-adolescent closeness, higher parental monitoring, lower parental substance use, and lower peer substance use. The relationship between each parental characteristic or peer use and each adolescent substance use is tested separately for cigarette and alcohol use.
(4) The effects of a disadvantaged neighborhood on adolescent cigarette and alcohol use are partially mediated through parent-adolescent closeness, parental monitoring, parental substance use, and peer substance use.
(5) Adolescents who have lower parent-adolescent closeness, lower parental monitoring, and higher parental substance use have more peers who use cigarette or alcohol than those who have higher parent-adolescent closeness, higher parental monitoring, and lower parental substance use. The relationship between each parental characteristic and peer use is tested separately for cigarette and alcohol use.
(6) The effects of parent-adolescent closeness, parental monitoring, and parental substance use on adolescent cigarette and alcohol use are partially mediated through peer substance use.

Method

Data
            The data for this study were obtained from the Family Matters Project, a randomized experimental study designed to determine whether a family-directed intervention prevented adolescent cigarette and alcohol use (Bauman et al., 2001a; Bauman, et al., 2001b). To identify adolescents age 12 to 14 and their parents living throughout the contiguous United States, 63,811 telephone numbers were generated by random digit dialing. Of those numbers, 2,395 were estimated to be the household having an eligible parent-adolescent pair. Fifty-five percent of those eligible pairs finished baseline interviews, which generated 1316 parent-adolescent pairs. Parents and adolescents were then randomly assigned to either the experimental or the control group. Family Matters was implemented from July, 1996 to September, 1997. The experimental group received four booklets, which were mailed in sequence. These booklets served as triggers to encourage the interaction between parents and adolescents to prevent cigarette and alcohol use. Following each booklet, health educators called the parents by phone to encourage participation and clarify questions. Parents and adolescents were interviewed by phone calls at baseline and at three and twelve months after completing the program (follow-up one and follow-up two). Seventy-nine percent of the baseline adolescents finished the follow-up two interviews. This study uses the responses of adolescents who completed all three interviews (1014 cases) and whose addresses could be matched to census tracts (1280 cases), which generates 959 cases in the final sample. In the final sample, Whites comprised 78.5%; Blacks comprised 9.6%; Hispanics comprised 7.5%; and other race/ethnicity comprised 4.5%. Age ranged from 12 to 14. Half of the sample was male (49.2%). The majority of adolescents’ mothers graduated from high school or had some college education (64.4%); 30% graduated from college; 5.4% did not graduate from high school.
            To understand the influence of attrition on the sample, we compared adolescents who finished only the baseline interview and those who finished all three interviews (Bauman et al., 2001b). Adolescents who only finished the baseline interview were more likely to be non-White, have a mother with lower education, live in a single-parent home, and to be baseline alcohol and cigarette users.

Sample Assessment

            This study uses random digit dialing to sample parent-adolescent pairs, which can potentially generate coverage errors. Because people who are low-income, minority, and live in rural areas are less likely to be in a household with a telephone, the study sample may be under-representative of these areas and populations (Lavrakas, 1998; Groves, 1989). To understand how the strategy of random digit dialing affects my sample, I compare the sample of adolescents in this study with the sample in census data and with the samples in other social surveys.
            Table 1.1 compares the adolescents in this study with the adolescents 12 to 14 years of age in census data on demographic characteristics. There are more adolescents who are White and whose mothers are college graduates in my sample than in census data. I also compared adolescents in this study with adolescents participating in two other national studies: Monitoring the Future (Johnston, O’Malley, & Bachman, 1998) and The National Longitudinal Study of Health (Add Health) (Bearman, Jones, & Udry, 1997). To enhance comparability, I limit the comparison with Monitoring the Future to 8th graders and Add Health to those 12 to 14 years of age. More 8th graders in my study use alcohol (69.2%) than in Monitoring the Future (54.5%), but fewer 8th graders in my study use cigarette (36.96%) than in Monitoring the Future (46.4%). My sample also has fewer adolescents who ever use cigarette (26.9%) than adolescents in Add Health (46%). Because incompatible questions were asked in Add Health regarding alcohol use, I am unable to compare the rate of adolescent alcohol use in my sample to the rate of adolescent alcohol use in Add Health.
            The data for this study have several strengths. The data come from a randomly selected sample throughout the nation with only one adolescent per neighborhood. This study design is good for estimating the effect size of neighborhoods on individual behavior via regression coefficients because there is no inefficient dependence across observations within each neighborhood (Duncan & Raudenbush, 1999). The data comprise longitudinal outcome measurements, which can reduce the selection bias and strengthen the causal inferences in estimating the influences of neighborhoods on individual behaviors. The data also have multi-level information including neighborhood, family, peer, and adolescent characteristics, which is needed for examining the interconnections of different domains of influences on adolescent substance use.

Measurement
Neighborhood characteristics
            Neighborhood characteristics are developed using the 1990 Dicennial Census for census tracts. Previous literature suggested that a neighborhood is defined as a social-spatial unit. The construct of a neighborhood is reinforced by the name and the symbolic meaning that residents share (Chaskin, 1998; Rapoport, 1997). Originally, census tracts were created as the geographic unit to study neighborhoods. Census tracts were defined by visible boundaries and intended to be as homogeneous as possible regarding population characteristics, economic status, and living conditions (U.S. Department of Commerce and Bureau of the Census, 1994). Using census tracts as the unit of neighborhoods, the spatial influences of a neighborhood can be addressed.
            In this study, I include six neighborhood characteristics. Low SES is measured by three indicators: the proportion of residents whose family income is less than $12,500, the proportion of males who are jobless, and the proportion of residents who are under the poverty line (Std. Cronbach ?=0.78). High SES is measured by three indicators: the proportion of residents whose family income is more than $75,000, the proportion of residents who have professional or managerial occupations, and the proportion of residents who have more than 12 years of education (Std. Cronbach ?=0.87). The cut points of family income to represent high SES and low SES follows Coulton and her colleagues’ suggestion (1996). They adapted the federally defined poverty threshold as the cut point of low SES. This threshold was set at $12, 674 for a family of four in 1989. They also suggested that the cut point of high SES should represent the top 12% of the family income distribution, which is about $75,000 in 1990 census data. Residential mobility is measured by the proportion of residents who lived in the same house in 1985 and the proportion of households which have been occupied by the owners for more than 10 years. Immigrant concentration is measured by the proportion of residents who are foreign born and the proportion of households which are language isolated. White and Black racial composition is measured by the proportion of residents who are White and the proportion of residents who are Black. Hispanic concentration is measured by the proportion of residents who are Hispanic. The means and standard deviations of neighborhood variables are presented in Table 1.2.

Parental and peer characteristics
            Parental and peer characteristics are developed from follow up-one data. Parental characteristics are created by taking the average of reports about fathers and mothers. Parent-adolescent closeness measures attachment, affection, and child-centerness of a parent-adolescent relationship. This concept is measured by four indicators: (1) “How often does your mother (father) kiss or hug you?” (2) “How close do you feel toward her (him)?” (3) “ Does your mother (father) spend time just talking with you?” and (4) “Does your mother (father) do fun things with you together?” with responses ranging from “very much/very often” to “not at all” along a 4-point scale (Std. Cronbach ?=0.82). Higher values indicate higher closeness. Parental monitoring is defined as parental knowledge and awareness about a child’s location and activities. This concept is measured by four indicators: (1) “Does your mother (father) try to know what you do with your free time?” (2) “Does your mother (father) try to know where you are most afternoons after school?” (3) “Does your mother (father) really know what you do with your free time?”, (4) “Does your mother (father) really know where you are most afternoons after school?” with responses ranging from “always” to “not at all” along a 4-point scale (Std. Cronbach ?=0.83). Higher values represent higher monitoring. Parent smoking is measured by asking adolescents: “About how many cigarettes do you think your mother (father) now smokes in a day” with responses ranging from “more than a pack a day” to “no cigarettes” along a 5-point scale. Parent drinking is measured by asking adolescents: “On the average, about how much alcohol do you think your mother (father) now drinks in a day?” with responses ranging from “4 or more drinks a day” to “none at all” along a 5-point scale. For both smoking and drinking measures, higher values indicate heavier substance use. Because parent smoking and parent drinking are highly skewed, they were recoded as dichotomized variables, with 0 = does not smoke/drink and 1 = does smoke/drink in a day. Peer smoking and peer drinking are measured by asking adolescents to separately indicate how many of their 3 best friends smoke or drink. Peer smoking and peer drinking were recoded as 0 = none of friends try smoking/drinking and 1 = one or more friends try smoking/drinking. Here we assume that the best friends of adolescents with age range from 12 to 14 have a high possibility of living in the same neighborhood because adolescents at this age range usually spend more time with other adolescents who live near by and meet at school than with adolescents whom they only can meet at school.

Adolescent cigarette and alcohol use
            Adolescent substance use is developed from baseline and follow-up two interviews. Adolescent cigarette use is measured by the question: “How much have you ever smoked cigarettes in your life?” Adolescents’ responses range from “more than 20 whole cigarettes” to “none at all, not even a puff” along a 5-point scale. Adolescent alcohol use is measured by the question: “How much alcohol have you ever had in your life?” Adolescents’ responses range from “more than 20 whole drinks” to “none at all, not even a sip” along a 6-point scale. For adolescent cigarette and alcohol use, the higher values represent heavier use. Because adolescent cigarette and alcohol use are highly skewed, they were recoded as 0 = no use of cigarette/alcohol and 1 = use of cigarette/alcohol.

Control variables
            Five control variables developed from the baseline data are included in the analysis. These variables are adolescents’ age, sex, race, mother’s education, and treatment condition. Sex was coded as 0 = female and 1 = male. Race is measured by four categories: White, Black, Hispanic and other race/ethnicity. I created three dummy coded variables and used White as the reference group. Mother’s education is measured by a 3-point scale, including graduated from high school or less, some college education, and college graduates. I created two dummy coded variables and used “less or graduated from high school or less” as the reference group. Treatment condition is measured by identifying whether the adolescent belongs to the experimental group or control group. The experimental group was coded as 1 and the control group was coded as 0.

Analysis Plan
            The conceptual model of this study is represented by a set of structural equations. Because the data come from a randomly selected sample throughout the nation with only one adolescent per neighborhood, there is no dependence across observations within each neighborhood. In this sampling scheme, the conceptual model of this study can be represented by regression models or structural equations (Duncan & Raudenbush, 1999; Duncan et al., 1997). I use Mplus as the statistical modeling program, because it has special modeling capabilities for both continuous and categorical data (Muthén & Muthén, 1998). The weighted least squares method with robust standard errors and mean-adjusted Chi-square test statistic (WLSM) is used as the estimator in the analysis. Because the model contains both categorical and continuous variables, the correlation matrix estimated from these variables is unlikely to behave like ordinary sample moments (Jöreskog & Sörbom, 1996). The weighted least squares method must be used instead of the maximum likelihood method or generalized least squares method. I use factor score of indicators for each neighborhood concept and treat the factor scores as exogenous variables in each model because neighborhood variables are highly skewed.
            Because of limited sample size, analysis at the first stage was conducted separately for each characteristic and each substance for the purpose of identifying which neighborhood characteristics were candidates for further analysis. At the second stage, neighborhood characteristics that were significantly associated with any parental characteristic or adolescent substance use were integrated into a final model of cigarette use or a final model of alcohol use. At the third stage, certain paths were added or trimmed for the purpose of achieving a better model fit. All findings are evaluated at a significance level of .05.

Results

            The neighborhood characteristics included in the final model of cigarette use are Low SES and White and Black racial composition. The neighborhood characteristics included in the final model of alcohol use are Low SES, High SES, and White and Black racial composition.
            Table 1.3 presents the goodness-of-fit measures that reflect how close the data are to the model. The Chi-square measure tests the hypothesis that the covariance matrix predicted by the model is equal to the population covariance matrix of the observed variables. The Chi-square test is sensitive to the number of cases so that a large sample increases the chances that the Chi-square test is significant (Kline, 1998). Other fit indices can help to evaluate the fit of the model to the data. For example, the comparative fit index (CFI) shows the proportional improvement of the overall fit of the model relative to a null model in which all the variables are uncorrelated. The comparative fit index has a maximum value of 1.0. A higher value of this measure indicates a better fit. The other fit index, the root mean square error of approximation (RMSEA), is the residual between the predicted and observed covariance matrix. A smaller value of this measure indicates a better fit. If the fit is perfect, then RMSEA equals zero (Kline, 1998).
            The fit of the initial models (Model one) is poor (Cigarette: Chi(110)= 3727, CFI= 0.66, RMSEA= 0.18; Alcohol: Chi(116)= 3897, CFI= 0.66, RMSEA= 0.18). To improve model fit, I added two paths to each of the models. I allowed correlated residuals of parental closeness and parental monitoring because they both represent the quality of a parent-adolescent relationship. In addition, I added a path from baseline adolescent substance use to peer substance use because adolescents who use substances are more likely to affiliate with friends who also use substances (Bauman & Ennett, 1996).
            Table 1.3 shows that the fit of model two, which has two additional paths, as largely improved (Cigarette: Chi(108)=764, CFI=0.93, RMSEA=0.08; Alcohol: Chi(114)=561, CFI=0.96, RMSEA=0.06). Originally, closeness and monitoring each was significantly associated with adolescent or peer substance use in model one. However, after adding the correlated residuals between closeness and monitoring, closeness is not associated with adolescent or peer substance use. Monitoring stands out as a more important predictor of adolescent substance use than does closeness.
            In order to reduce the complexity of the model, I trimmed the model when the standardized structural coefficients were less than 0.05. This criterion is based on retaining significant paths but also keeping the integrity of the models. The fit of the trimmed model (model three) is improved slightly (Cigarette: Chi(127)=641, CFI=0.95, RMSEA=0.06; Alcohol: Chi(134)=467, CFI=0.97, RMSEA=0.05). Although the Chi-square is significant, the sample size of this study (959) usually guarantees a significant Chi-square for a model with this number of variables. In addition, because other fit indicators suggest an acceptable level of fit, I decided to accept model three as the final model.
            Table 1.4 presents the measurement model of cigarette use. For the two latent concepts, parental closeness and parental monitoring, the standardized regression coefficients show a moderately strong relationship between each latent concept and their indicators. The pattern of the measurement model of alcohol use is similar to the measurement model of cigarette use.
            Figure 1.2 presents the standardized structural coefficients for the relations among neighborhood characteristics, parental characteristics, the peer characteristic, and adolescent cigarette use. Figure 1.3 is for adolescent alcohol use. Standardized coefficients for all variables are presented in Table 1.5 and Table 1.6, respectively, for adolescent cigarette use and adolescent alcohol use. Unstandardized coefficients and standard errors are presented in Appendix 1.1 and Appendix 1.2, respectively, for adolescent cigarette use and adolescent alcohol use. Figure 1.2 shows that Low SES has indirect effects on adolescent cigarette use through parental monitoring and peer smoking. Contrary to predictions, parents living in a low SES neighborhood increase parental monitoring on their children, which in turns reduces peer smoking and adolescent cigarette use. While Low SES is significantly related to parental monitoring, White and Black racial composition is not related to any mediating or outcome factor.
            In addition to parental monitoring, adolescent cigarette use is directly influenced by baseline smoking, parent smoking, and peer smoking. Each of the predictors increases the likelihood that the adolescent smokes. Consistent with previous literature about adolescent substance use, peer smoking in this study is a strong predictor of adolescent smoking (Flay et al., 1994). However, Figure 1.2 demonstrates that peer smoking is influenced by baseline adolescent smoking, parental monitoring, and parent smoking, therefore, the total effects of peer smoking are not as substantial as what previous literature suggested (Bauman & Ennett, 1996; Kandel, 1996). Specifically, adolescent baseline smoking behaviors, parental monitoring, and parent smoking can influence whether adolescents affiliate with smoking friends.
            The model of adolescent alcohol use (Figure 1.3) shows that Low SES has indirect effects on adolescent alcohol use through parental monitoring and peer drinking. A low SES neighborhood increases the level of parental monitoring and the possibility that adolescents have drinking friends. Parental monitoring reduces adolescent alcohol use and peer drinking increases adolescent alcohol use. Another neighborhood factor, High SES, has indirect effects on adolescent alcohol use through parent drinking. Contrary to predictions, parents in a high SES neighborhood are more likely to use alcohol, which in turns increases adolescent alcohol use. White and Black racial composition is not associated with any mediating or outcome factor.
            Figure 1.3 also shows that peer drinking mediates the relationship between baseline drinking and adolescent alcohol use and the relationship between parental monitoring and adolescent alcohol use. However, peer drinking is not a mediator between parent drinking and adolescent alcohol use.
            In order to compare the relative influences of neighborhood, parent, and peer characteristics on adolescent cigarette or alcohol use, Table 1.7 presents the decomposition of the effects of these characteristics on adolescent cigarette or alcohol use. Neighborhood characteristics have small total effects on adolescent cigarette and alcohol use. The effects that Low SES decreases adolescent cigarette and alcohol use through parental monitoring are cancelled out by the effects that Low SES increases adolescent cigarette and alcohol use through peer substance use (Figure 1.2 and Figure 1.3).
            Comparing the relative influences of parental and peer characteristics on adolescent cigarette use, parental monitoring and peer smoking show stronger effect on adolescent cigarette use, while closeness has the weakest effects on adolescent cigarette use. In terms of adolescent alcohol use, parental monitoring, parent drinking, and peer drinking all have strong effects on adolescent alcohol use but parental closeness does not have effects on adolescent alcohol use.

Discussion

            This study found neighborhoods influence parental and peer characteristics, which perform as mediators between neighborhoods and adolescent cigarette or alcohol use. This study found parents living in a low SES neighborhood increase parental monitoring. This result is not consistent with the findings of a small number of prior studies, which suggested that a poor neighborhood decreases parental supervision and monitoring (Hogan & Kitagawa, 1985; Stern & Smith, 1995; Simon et al., 1997; Simon et al., 1996). The differences in the findings between this study and prior studies may come from characteristics of the study sample, which results in the different reaction of individuals toward neighborhood environments.
            This study uses a randomly selected sample of adolescents throughout the nation and matched the addresses of adolescents with census tracts, which is different from the high-risk population samples or neighborhoods in prior studies (Simon et al., 1997; Simon et al., 1996; Stern & Smith, 1995; Hogan & Kitagawa, 1985). Previous literature has suggested that neighborhood effects on social problems are nonlinear. When the density of disadvantaged neighborhood characteristics rises beyond a critical level, there should be a sharp increase in social problems in the neighborhood (Crane, 1991). This nonlinear effect results from both the level of disadvantaged neighborhood characteristics and the susceptibility of individuals and families to the environment. Using the critical value suggested by Wilson (1991b), only 1.77% of the neighborhoods in this study have more than 40% of people below poverty line. In addition, parents who completed this study are more likely to be White, highly educated, and are less likely to live in single-parent households (Bauman et al., 2001b). These facts suggest that the neighborhood sample in this study is not at the bottom of the distribution of neighborhood quality and parents in this study have more resources to conduct effective parenting because they have higher education. Under these conditions, a neighborhood with a higher proportion of low SES residents may not decrease parental monitoring but rather increases parents’ attempts to protect their children from a risky neighborhood environment by conducting parental monitoring.
            The other explanation may come from the unique parenting strategies employed by parents who live in a low SES neighborhood. Some studies suggested that parents in a low SES neighborhood have to use stringent parenting strategies, such as confining adolescents to the household, chaperoning adolescents on their daily rounds in the neighborhood, and even using corporal punishment to enforce parental orders (Burton & Jarrett, 2000; Furstenberg, 1993; Jarrett, 1997; Jarrett, 1995). Although these stringent parenting strategies have been argued to be harmful to adolescent development, studies recently have started to show that these stringent parenting strategies are beneficial to adolescents or at least not harmful if they occur in an impoverished neighborhood (Burton & Jarrett, 2000; Simons et al., forthcoming). These stringent parenting strategies are regarded as necessary to keep adolescents away from dangerous neighborhood activities, such as violence or illicit drug vending. This phenomenon may partially explain why this study shows that low SES neighborhoods increase parental monitoring, which decreases adolescent substance use. However, future research needs to be conducted to understand the extent to which a low SES neighborhood increases stringent parenting strategies and whether these parenting strategies are beneficial for adolescent development.
            This study also found that Low SES increases the likelihood that adolescents have drinking friends. This result demonstrates that neighborhoods may influence adolescent drinking through the epidemic models (Jencks & Mayer, 1990). The epidemic models assume that problem behaviors are contagious and are propagated mainly through peer influences in a neighborhood. Although this study does not directly measure concentration of peer substance use in a neighborhood, this study shows that the presence of low SES neighbors is significantly associated with drinking friends. The result is consistent with the study of Simon et al. (1996). They found that adolescents who live in low SES neighborhoods are more likely to report having deviant peers, and having deviant peers is further associated with more male adolescent conduct problems including substance use.
            This study also found that a high SES neighborhood increases the chances that parents use alcohol. This study measured parental drinking by asking adolescents whether their parents used alcohol. Although this measure can detect whether adolescents model their parents’ drinking behaviors or whether adolescents can access alcohol at home, this measure cannot describe whether these parents have problem drinking behaviors. In addition, because wine drinking has become a popular culture in professional classes in recent decades, White-collar households are known to be the main consumers of wine (Fuller, 1991). It is possible that a high SES neighborhood may have a higher rate of people who use alcohol but have a lower rate of people who have problem drinking behaviors, such as drunk driving or heavy drinking.
            This study provides good evidence for arguing that the importance of peer substance use on adolescent substance use has been over emphasized in research on adolescent substance use. Previous literature suggested that peer substance use is the strongest and most consistent predictor of adolescent substance use (Bauman & Ennett, 1996). This study shows that adolescent baseline substance use predicts peer substance use, which in turn predicts adolescent follow-up substance use (Bauman & Ennett, 1996; Kandel, 1996). The result of this study suggests that peer substance use and adolescent substance use are mutually reinforcing processes. While peers can influence adolescent substance use through role modeling and norm changing, adolescents who use substances can also influence peers through the same processes (Kandel, 1996).
            This study also found that peer substance use is predicted by parental monitoring and parent substance use. Traditionally, parental influence has been regarded as much less influential than peer influences on adolescent substance use (Kandel, 1996). Recently, studies have started to investigate the linkages between parental and peer influences. They suggested that a close adolescent-parent relationship or strong parental monitoring can change adolescents’ orientation toward friends who do not use drugs (Bogenschnider et al., 1998). This study found that parental monitoring reduces the likelihood that adolescents have smoking or drinking friends and parental smoking increases the likelihood that adolescents have smoking friends. These parent-peer linkages suggest that parental characteristics are important factors in understanding how peers exert influences on adolescent substance use.
            A limitation of this study is the lack of longitudinal neighborhood variables. It is possible that the relationship between neighborhoods and individual behaviors is because of a non-random selection of individuals into different neighborhoods and is not because of neighborhood influences (Tienda, 1990). Therefore, a longitudinal assessment of neighborhood variables is needed for estimating neighborhood effects on parents, peers, and adolescents. This study only uses 1990 Census data for the estimation of neighborhood effects so that the casual direction of some of the relationships between neighborhood variables and parental or peer characteristics may not be able to be specified. For example, the correlation between low SES neighborhoods and peer drinking may be because deviant peers are more likely to move into low SES neighborhoods.
            In summary, this study demonstrates that neighborhoods influence parental and peer factors, which can further influence adolescent cigarette and alcohol use. This study also demonstrates that the interrelationship among neighborhoods, parents, peers, and adolescent substance use needs to be investigated by using various theories as a guide. This study applies theories about neighborhood influences, family relationships, and adolescent substance use in explaining the mechanism in each path of the interrelationship among neighborhoods, parents, peers, and adolescent substance use.
 
 

CHAPTER TWO
Paper Two: A Comparison of Neighborhood Influences on Adolescent Cigarette and Alcohol Use by Using Census Tracts and Census Block Groups

Introduction

            The purpose of this paper is to examine the extent to which the spatial size of a neighborhood makes a difference in explaining neighborhood mechanisms on adolescent substance use. Specifically, this paper compares neighborhood mechanisms on adolescent substance use using census tracts and block groups. Census tracts and census block groups are small geographical units created by the U.S. Census Bureau for studying neighborhoods. Census tracts and census block groups provide spatial information about a neighborhood such as population characteristics. A census tract comprises on average 4,000 people, which is larger than a census block group. A census tract is on average 3.7 times larger than a census block group.
            The importance of neighborhoods to adolescent development has gained considerable research attention in the past decade. Previous studies generally concluded that effects of neighborhoods on adolescent behaviors exist but are not substantial (Gephart, 1997; Ellen & Turner, 1997; Jencks & Mayer, 1990; Leventhal & Brooks-Gunn, 2000). However, these studies suffer many methodological limitations. One of the main criticisms concerns the conceptualization and measurement of “neighborhoods.” Most previous neighborhood research used census tracts to represent neighborhoods because of cost efficiency and convenience (Furstenberg & Hughes, 1997). More recently, scholars have questioned whether census tracts can adequately represent neighborhoods.
            The most frequently cited limitation of using census tracts as neighborhoods is that census tracts only comprise neighborhood population characteristics and do not necessarily reflect social processing variables (e.g., neighborhood norms or informal social control) (Furstenberg & Hughes, 1997). The second limitation is that census tracts are usually regarded as geographic units that are much larger than the developmental neighborhood niches of children and adolescents (Burton et al., 1997). The third limitation is that the size of a neighborhood perceived by its residents is usually smaller than a census tract and closer to a census block group (Rapoport, 1997; Elliott et al. forthcoming).
            Recently, studies have used census block groups instead of census tracts, thereby keeping the convenience of using census data but overcoming some of the limitations associated with census tracts (Coulton et al., 1996). Although census block groups still do not define neighborhood social processing variables, researchers argue that block groups better represent an area in which residents can get around by walking and develop close interactions (Coulton et al., 1996). Nevertheless, few studies have evaluated the appropriateness of using census block groups as an alternative to census tracts. The difference between using census tracts and block groups in neighborhood research remains unknown.
            Using adolescent substance use as an example, this study answers the questions “Do neighborhood effects on adolescent substance use differ when neighborhoods are defined by census block groups than census tracts?” and “How do neighborhood mechanisms differ when neighborhoods are defined by these two different geographical scales?” This study first discusses the definition of a neighborhood and then discusses how neighborhood definition changes according to the neighborhood geographical scale. Finally, this study separately tests a conceptual model of neighborhood influences on adolescent substance use by census tracts and block groups. This study proposes that block-group neighborhood influences are more likely than tract neighborhood influences to be correlated with family characteristics, which serve as mediators between neighborhoods and adolescent substance use.

Definition of Neighborhoods

            The term “neighborhood” seems to be commonly understood by the public, but there is little consensus on what it really means (Galster, 1986). In the social science literature, researchers have defined a neighborhood in various ways (Chaskin, 1997). Some defined neighborhoods as a social unit, which is generated by natural economic competition and selection (Frisbie & Kasarda, 1988). Some regarded neighborhoods as a group of social networks (Wellman, 1979). Others defined neighborhoods as a small physical place around the dwellings of the residents (Rapoport, 1997). Although researchers have different definitions of “neighborhoods,” some core elements can be found in most of the literature, which can be summed up as the following:
            A neighborhood is a small and spatially circumscribed area surrounding homes, which has a name and other symbolic identities (Keller, 1968; Hunter, 1974). In the neighborhood, residents have homogenous population characteristics (i.e., social economic status, ethnicity, values, attitudes, life style, and culture) (Rapoport, 1997). Residents socially interact and form various social networks (Wellman, 1979). A neighborhood is not an isolated social unit. It is an immediate social context that provides residents the organization of daily social activities and connections to other opportunity structures and resources in the larger society (Gephart, 1997; Warren, 1987).
            One of the reasons that neighborhoods are difficult to define is because people perceive neighborhoods differently. Studies have found that the definitions of neighborhoods vary by residents who are in different stages of life course and social positions (Guest & Lee, 1984; Lee & Campbell, 1997). For example, African-Americans, elderly, women, unemployed adults, long-term residents, and the people who are actively involved in neighborhood activities define a neighborhood by its social relationships. On the other hand, Whites, younger people, well-educated people, and employed people define a neighborhood by its physical characteristics and spatial size.
            To summarize from the above review, a neighborhood can be defined as a spatial, social, and perceptional unit (Chaskin, 1997). Theoretically, a neighborhood is most clearly defined when its spatial, social, and perceptional characteristics are congruent (Rapoport, 1997). For example, when natural boundaries, social networks, local organizations, and the symbolic meanings to residents all come together, neighborhoods are most likely to be defined. Nevertheless, in most cases, a universal definition of a neighborhood is not possible to reach. Chaskin (1997; 1998) suggested that neighborhood definition is usually a political product, which is decided by the negotiating processes between residents’ and outsiders’ conceptualizations and between geographical and social factors. He further suggested that a neighborhood should be defined according to the program aims and the research purposes.

Using Census Tracts and Block Groups to Represent Spatial Neighborhoods

            In previous research, the most common approach to defining a neighborhood was to use census tracts. Census tract data provide spatial information about a neighborhood, including geographical and population characteristics. Census tracts usually comprise 4,000 people and 1,500 housing units on average with a range of 2,500 to 8,000 residents and 1,000 to 3,000 housing units (U.S. Department of Commerce and Bureau of the Census, 1994). Census tracts were created by visible boundaries such as highways, streets, rivers, and railroads and were created to be as homogeneous as possible with population characteristics. The approach of using census tracts to define a neighborhood represents the traditional urban planning viewpoint, which concerns whether the size of the population can sufficiently sustain local organizations. For example, Perry suggested the ideal size of a neighborhood is around 5,000 people (Perry, 1929; Chaskin, 1998).
            Using census tracts to define neighborhoods has been questioned in recent years. Many studies found that the size of a census tract is usually larger than what residents perceive to be their neighborhood (Elliott et al., forthcoming; Rapoport, 1997). In addition, an area with 2,500 to 8,000 people is likely to include more people than residents can possibly know and interact with daily. Therefore, some studies proposed to use census block groups instead of census tracts. A census block group is a subdivision of a census tract. On average, a census tract comprises 3.7 block groups. A block group has around 250 to 550 housing units. Because a block group is much smaller than a census tract, residents are more likely to interact closely with the other residents in the same block group.
            The difference in neighborhood effects from using census tracts and census block groups is unclear. The only study that has systematically compared census block groups with tracts was conducted by Elliott and colleagues (Elliott et al., forthcoming). They compared residents’ perceptions regarding neighborhood sizes with census data. They found that more residents identified the neighborhood size as closer to block groups than tracts. Two procedures were further applied to evaluate the validity of census tracts and block groups to represent neighborhoods. Elliott and colleagues found the correspondence between the aggregated residents’ responses and census data regarding the same question (i.e., how many families in their neighborhood are poor) is higher when block groups are applied. In addition, the homogeneity of residents’ responses is also higher in block groups than tracts. The study concluded that census block groups are a better unit for defining neighborhoods than census tracts are.
            Another interesting finding of that study was that although block groups were demonstrated to be a better unit of neighborhoods, residents changed their perceptions regarding the neighborhood size when different topics were introduced into the interviews. For example, when residents were asked about social services in the neighborhood, half of the residents who originally identified blocks or block groups as the neighborhood size changed to identify tracts or tract groups as the size of their neighborhoods. In contrast, when they were asked about adolescents’ chances of realizing their educational and occupational goals, residents tended to choose a smaller neighborhood size. They also found that race and class influenced residents’ perceptions of the neighborhood size. The neighborhood sizes perceived by Black and lower SES residents were smaller than the neighborhood sizes perceived by White and higher SES residents.
            The findings of Elliott and his colleagues empirically support the suggestions made by Chaskin (1997; 1998) and Gaslter and Killen (1995). Chaskin and Gaslter and Killen recommended that studies should consider changing the neighborhood scale for different outcomes or processes of interest. For example, for adolescent school education, studies should consider matching neighborhood boundaries to school districts because difference in school resources creates distinct school environments for adolescents. For adolescent job opportunities, studies should consider the spatial variation in information about job vacancies and the spatial variation in the racial or gender discrimination. The appropriate scale to study adolescent job opportunities is recommended to be larger than a census tract but smaller than a metropolitan area (Galster & Killen, 1995). Regarding the appropriate neighborhood scale in this study, which focuses on the mechanisms of neighborhood influences on adolescent substance use through parental characteristics, I expect that the smaller scale of census block groups is a better neighborhood unit than census tracts. I will describe the reasons more fully and provide the hypotheses later in the paper.

Neighborhood Mechanisms on Adolescent Behaviors

            Having addressed the meaning of neighborhood size to neighborhood research, the theoretical mechanisms of neighborhood influences on adolescent behaviors are discussed in this section before introducing the conceptual model for this study. For a more detailed review of neighborhood mechanisms on adolescent behaviors, see paper one “Neighborhood Influences on Adolescent Cigarette and Alcohol Use: Mediating Effects through Parents and Peers.”
            This study follows the classification of neighborhood mechanisms proposed by Jencks and Mayer (1990). In their review, they proposed three models to explain how disadvantaged neighborhoods negatively influence adolescents. The first model is the epidemic models, which suggest that neighborhoods affect adolescents through the contagious influences of peers who live in the same neighborhood. Because a disadvantaged neighborhood usually comprises a large group of teenagers who have problem behaviors, adolescents who grow up in the neighborhood may easily accept deviant norms and learn problem behaviors as well. The second model is the collective socialization models, which focus on how adults in the neighborhood influence children who are not their own. A disadvantaged neighborhood usually has a high proportion of adults who have lost their jobs and become involved in criminal activities. Adolescents then do not have role models in the neighborhood for conventional behaviors (Wilson, 1987). The third model is the institutional models, which suggest that neighborhoods influence adolescents through the quality of local organizations (Wilson, 1987; Shaw & McKay, 1969). For example, in a disadvantaged neighborhood, the quality of local schools may be affected by scarce learning equipment and lack of devoted teachers. Parks, libraries, and community centers either do not exist or are not well maintained.
            Jencks and Mayer used these three models to explain how neighborhoods influence adolescents directly, but Jencks and Mayer’s models do not explain how neighborhoods influence adolescents indirectly through families. Leventhal and Brooks-Gunn (2000) proposed a fourth model, parental relationships, which regards parents as gate keepers, who manage risk and opportunities for their children (Furstenberg, 1993; Furstenberg et al., 1999). Specifically, Leventhal and Brooks-Gunn suggested that neighborhoods influence adolescents’ well-being through parents’ characteristics, parenting behaviors, home environments, and helping social networks. For example, parents who live in a disadvantaged neighborhood are more likely to have poor mental health, inadequate coping skills, and low efficacy. These characteristics, in turn, can affect parenting behaviors such as using harsh control and verbal aggression toward their children (Earls et al., 1994). Parents in a disadvantaged neighborhood also show less ability to arrange a healthy home environment that can facilitate adolescents’ learning. For example, few books and reading materials are available at home (Klebanov et al., 1994). Parents in this kind of neighborhood cannot obtain social support from their neighbors. The lack of helping social networks reduces the possibility that parents can lessen the stress associated with living in a dangerous and impoverished neighborhood.
            These four models are not mutually exclusive. For example, adolescents’ school performances may be influenced by peers’ school performances, other adults’ orientation toward education, schools’ resources, and their own parents’ attitude toward education. Each factor represents a mediating path that transfers neighborhood influences to adolescents. This study focuses primarily on interpersonal mediating mechanisms in the neighborhood, which include collective socialization models and parental relationships. Using census tracts and block groups to examine these mechanisms, this study intends to understand whether the neighborhood mechanisms affect adolescents differently when neighborhood size changes from a tract to a block group.

Conceptual Model

Model Specification
            The conceptual model of this study shows that a disadvantaged neighborhood increases adolescent cigarette and alcohol use directly and indirectly through reducing parental closeness, reducing parental monitoring, and increasing parent cigarette and alcohol use, which is presented in Figure 2.1. Six neighborhood characteristics are included in the model: low SES, high SES, residential mobility, immigrant concentration, White and Black racial composition, and Hispanic concentration. The selection of the first four characteristics is based on social disorganization theory (Shaw & McKay, 1969), Wilson’s theory (1987) and Brooks-Gunn and her colleagues’ series of studies (1993; 1997). These characteristics together define a neighborhood where residents are hypothesized to have low normative consensus, low organizational support, weak social ties, and low informal social control. Local delinquent groups are formed because the neighborhood does not have the ability to control them. The factor of high SES is separated from low SES because each factor represents different potential neighborhood mechanisms. A neighborhood that has a low proportion of high SES residents means that adolescents do not have role models to learn about conventional behaviors. On the other hand, a neighborhood that has a high proportion of low SES neighbors suggests that adolescents are more likely to learn problem behaviors from deviant friends. The inclusion of White and Black racial composition follows the suggestion that studies should conceptually separate the effects of SES and race because they represent different origins of a disadvantaged neighborhood (Massey, 1998). The concentration of Blacks in a few neighborhoods reflects the prevalence of residential segregation in U.S. today. These are the results of the persistence of white racial prejudice and discrimination in the housing markets and banking industries (Massey, 1996). Therefore, the proportion of Blacks refers to the extent of residential segregation, which cannot be shown by combining indicators of race and SES in a single neighborhood domain. The inclusion of Hispanic concentration also shows the uneven distribution of resources across neighborhoods due to the racial effects. Because the geographical concentration of Hispanics is poorly correlated with the geographical concentration of Whites and Blacks, I separate Hispanic concentration as another neighborhood domain.
            A disadvantaged neighborhood is expected to decrease parent-adolescent closeness, decrease parental monitoring, and increase parent cigarette and alcohol use. Changes in these three domains of family life are expected to increase adolescent cigarette and alcohol use. These three mediating paths represent the mechanism of parental relationships and the mechanism of collective socialization. The experience of living in an impoverished neighborhood creates long-term stress for parents (Garbarion & Sherman, 1980). Parents who live in a dangerous and impoverished neighborhood may become depressed, show reduced affection toward their children, reduce the time they spend with their children, have reduced energy to monitor their child, and even use substances to cope with the stressful environment (Klebanov et al., 1994; Leventhal & Brooks-Gunn, 2000; Simon et al, 1996). The other possible explanation is that parents’ norms and behaviors are influenced by other parents who live in the same neighborhood (Wilson, 1991). A disadvantaged neighborhood may have many adults who do not have a regular job, get involved in delinquent activities, and have poor skills in family management. Parents then easily come to tolerate deviant behaviors and relax the standards in family management. The result is a decline in parent-adolescent closeness, and parental monitoring, and an increase of parent substance use. In my conceptual model, I allowed correlated residuals for the latent concepts of parental closeness and parental monitoring because they both represent the well being in parent-adolescent relationships.
            A neighborhood is also expected to have direct influences on adolescent cigarette and alcohol use. The data are limited for examining some other neighborhood mediating characteristics such as residents’ consensus on conventional norms, informal social control, quality of local organizations, helping social networks, and culture of delinquency (Sampson, Raudenbush, & Earls, 1997; Elliott et al., 1996; Simcha-Fagan & Schwartz, 1986). These mediating neighborhood characteristics are included in the direct path in the model.
            In my conceptual model, because adolescents who use substances are more likely to move into or be economically forced to stay in an impoverished neighborhood, adolescent substance use is not necessarily the outcome of neighborhood influences. Therefore, including baseline adolescent substance use is important because it allows me to detect the net change of adolescent substance use after their neighborhood contexts are measured. This may reduce the selection bias in estimating the influences of neighborhoods on adolescent substance use. The selection bias can also exist in the relationship between neighborhoods and parent substance use. However, there is not enough variation in the change of parent substance use between baseline and follow-up surveys so that I cannot control the selection bias by including baseline parent substance use in the model.

A Comparison of Census Tracts with Block groups
            Comparing census tracts with block groups, I argue that neighborhood characteristics are more likely to be significantly associated with parental characteristics when neighborhood characteristics are measured by census block groups than census tracts. Because the main interest of this model is in whether parental characteristics act as mediators between neighborhood characteristics and adolescent substance use, the spatial variation of the parental characteristics should be considered. Unlike for studying neighborhood organizations (e.g., schools) or economic opportunity structures (e.g., job market), I argue that a smaller geographical scale is better for determining interpersonal relationships in the neighborhood. According to the review of neighborhood mechanisms by Jencks and Mayer (1990) as well as Levethel and Brooks-Gunn (2000), parental closeness, parental monitoring, and parent substance use are influenced by neighborhoods because parents are socialized by deviant norms or have long-term stress and lack of social support. The socialization of deviant norms usually involves a large amount of social interactions, such as conversations or observations, which can be most efficiently achieved within a small geographic area where residents can get around by walking. Parents are also more likely to obtain support if helpers live near by. Therefore, a small and proximate geographical area surrounding homes is more likely to capture neighborhood interpersonal influences. From previous studies, census tracts are claimed to be too large to study residents’ daily social interactions, so I argue that block-group neighborhood measures are more likely to be significantly associated with parental closeness, parental monitoring, and parent substance use.
            In contrast to supporting census block groups for representing neighborhoods when the mediating paths are considered, I think the direct path linking neighborhoods and adolescents is more controversial. The direct path includes neighborhood mechanisms other than parental characteristics, such as the locations where adolescents can buy cigarette and alcohol, informal social control to stop adolescents from using substances, substance use prevention programs provided by local agencies, and neighborhood peers who use substances. The spatial variations of these mechanisms are different so that these mechanisms need to be studied in different neighborhood scales. For example, for the mechanism of neighborhood availability of cigarette and alcohol, studies should consider the locations of stores or outlets where adolescents can buy cigarette or alcohol. For the mechanisms of neighborhood informal social control, studies should consider the sufficient number of residents, who can combine local forces to intervene problem behaviors. The appropriate scale to study about neighborhood informal social control is recommended to be larger than a census tract but smaller than a community (Sampson, Raudenbush, & Earls, 1997). Because different spatial scales need to be specified for different neighborhood mechanisms, I am not able to hypothesize whether block groups are a better unit than tracts in the direct path because this path includes various mechanisms.

Study Hypotheses

            Five major hypotheses are tested in this study. The first four hypotheses are tested separately by census tracts and census block groups. The hypotheses are as follows:
(1) Adolescents who live in a neighborhood with a higher level of disadvantaged condition, as indicated by a higher proportion of low SES residents, a lower proportion of high SES residents, higher residential mobility, a higher proportion of immigrants, a higher proportion of Blacks, and a higher proportion of Hispanics, have higher levels of cigarette and alcohol use compared with adolescents who live in a non-disadvantaged neighborhood. The relationship between each neighborhood characteristic and adolescent substance use is tested separately for cigarette and alcohol use.
(2) Adolescents who live in a neighborhood with a higher level of disadvantaged condition have lower parent-adolescent closeness, lower parental monitoring, and higher parental substance use compared with adolescents who live in a non-disadvantaged neighborhood. The relationship between each neighborhood characteristic and each parental characteristic is tested separately for cigarette and alcohol use.
(3) Adolescents who have lower parent-adolescent closeness, lower parental monitoring, and higher parental substance use have higher levels of cigarette and alcohol use than those who have higher parent-adolescent closeness, higher parental monitoring, and lower parental substance use. The relationship between each parental characteristic and each adolescent substance use is tested separately for cigarette and alcohol use.
(4) The effects of a disadvantaged neighborhood on adolescent cigarette and alcohol use are partially mediated through parent-adolescent closeness, parental monitoring, and parental substance use.
(5) Census block groups are a better unit for defining neighborhoods than census tracts, which is shown in the following hypotheses:
The relationship between each disadvantaged neighborhood latent construct (i.e., a high proportion of low SES residents, a low proportion of high SES residents, high residential mobility, a high proportion of immigrants, a high proportion of Blacks, and a high proportion of Hispanics) and parent-adolescent closeness, and parental monitoring, and parental substance use are more likely to be significant when neighborhood characteristics are measured by census block groups than census tracts. These hypotheses can also be shown in the magnitude of the relationships between neighborhoods and parental behaviors and in the proportion of variance of parental behaviors explained by neighborhood constructs.

Methods

Data
            The data come the Family Matters Project, a randomized experimental study designed to determine whether a family-directed intervention prevented adolescent cigarette and alcohol use (Bauman et al., 2001a; Bauman, et al., 2001b). Adolescents aged 10 to 14 and their parents were identified by random digit dialing throughout the United States. After finishing baseline interviews (55.4% completed baseline interviews), 1316 pairs of adolescents and parents were randomized assigned to either experimental or control groups. Family Matters was implemented from July, 1996 to September, 1997. The experimental group received four booklets by mail in sequence. The booklets served as triggers for parents to conduct activities with adolescents to prevent tobacco and alcohol use. Following each booklet were phone calls from health educators for answering questions and encouraging participation. Three months and twelve months after completing the four booklets, both experimental and control groups received follow-up one and follow-up two interviews. Seventy-nine percent of the baseline adolescents finished the follow-up two interviews. This study uses the responses of adolescents who completed all three interviews (1014 cases) and whose addresses could be matched to census tracts (1280 cases) and census block groups (1237 cases), which generates 924 cases in the final sample. Regarding the demographic characteristics of the sample, Whites comprise 78.5%; Blacks comprise 9.6%; Hispanics comprise 7.5%; and others comprise 4.5%. Age ranges from 12 to 14. Half of the sample is male (49.2%). The majority of adolescents’ mothers graduated from high school or had some college education (64.4%); 30% graduated from college; 5.4% did not graduate from high school.
            To understand the influence of attrition on the sample, adolescents who only finished the baseline interview were compared with those who finished all three-wave of interviews (Bauman et al., 2001b). Adolescents who only finished the baseline interview were more likely to be non-White, have mothers with lower education, live in single-parent homes, and to be baseline alcohol and cigarette users.

Measurement
Neighborhood Characteristics
            Neighborhood characteristics, including low SES, high SES, residential mobility, immigrant concentration, White and Black racial composition, and Hispanic concentration, are developed from 1990 Census tracts and block groups. Low SES is assessed by three items: the proportion of residents who have family income less than 12,500, the proportion of males jobless, and the proportion of residents who are under poverty line (Std. Cronbach ?=0.78 for census tracts and ?=0.64 for census block groups). The correlation coefficients between the items measured by tracts and by block groups are 0.85, 0.75, and 0.75 respectively. High SES is assessed by three items: the proportion of residents whose family income is more than 75,000, the proportion of residents who have professional or managerial occupations, and the proportion of residents whose education is more than 12 years (Std. Cronbach ?=0.87 for census tracts and ?=0.80 for census block groups). The correlation coefficients between the items measured by tracts and by block groups are 0.90, 0.88, and 0.91 respectively. The cut points of family income to represent high SES and low SES follows Coulton and colleagues’ suggestion (1996). They adapted the federally defined poverty threshold as the cut point of low SES. This threshold was set at $12,674 for a family of four in 1989. They also suggested that the cut point of high SES should represent the top 12% of the family income distribution, which is about $75,000 in 1990 census data. Residential mobility is assessed by the proportion of residents who lived in the same house in 1985 and the proportion of households that have been occupied by the owners for more than 10 years. The correlation coefficients are 0.81 and 0.81 between tracts and block groups, respectively. Immigrant concentration is assessed by the proportion of residents who are foreign born and the proportion of households which are language isolated. The correlation coefficients are 0.92 and 0.89 between tracts and block groups, respectively. White and Black racial composition is measured by the two items, the proportion of residents who are White and the proportion of residents who are Black. The correlation coefficients are 0.92 and 0.93 between tracts and block groups respectively. Hispanic concentration is measured by the proportion of residents who are Hispanic. The correlation coefficient is 0.87 between tracts and block groups.

Parental characteristics
            Parental characteristics are developed from the Family Matters follow up-one data. Parental characteristics are created by taking the average of reports about fathers and mothers. Parent-adolescent closeness measures attachment, affection, and child-centerness of a parent-adolescent relationship. This concept is measured by four indicators: (1) “How often does your mother (father) kiss or hug you?” (2) “How close do you feel toward her (him)?” (3) “ Does your mother (father) spend time just talking with you?” and (4) “Does your mother (father) do fun things with you together?” with responses ranging from “very much/very often” to “not at all” along a 4-point scale (Std. Cronbach ?=0.82). Higher values indicate higher closeness. Parental monitoring is defined as parental knowledge and awareness about a child’s location and activities. This concept is measured by four indicators: (1) “Does your mother (father) try to know what you do with your free time?” (2) “Does your mother (father) try to know where you are most afternoons after school?” (3) “Does your mother (father) really know what you do with your free time?”, (4) “Does your mother (father) really know where you are most afternoons after school?” with responses ranging from “always” to “not at all” along a 4-point scale (Std. Cronbach ?=0.83). Higher values represent higher monitoring. Parent smoking is measured by asking adolescents: “About how many cigarettes do you think your mother (father) now smokes in a day” with responses ranging from “more than a pack a day” to “no cigarettes” along a 5-point scale. Parent drinking is measured by asking adolescents: “On the average, about how much alcohol do you think your mother (father) now drinks in a day?” with responses ranging from “4 or more drinks a day” to “none at all” along a 5-point scale. For both smoking and drinking measures, higher values indicate heavier substance use. Because parent smoking and parent drinking are highly skewed, they were recoded as dichotomized variables, with 0 = does not smoke/drink and 1 = does smoke/drink in a day.

Adolescent cigarette and alcohol use
            Adolescent substance use is developed from the baseline and follow-up two interviews. Adolescent cigarette use is measured by the question: “How much have you ever smoked cigarettes in your life?” Adolescents’ responses range from “more than 20 whole cigarettes” to “none at all, not even a puff” along a 5-point scale. Adolescent alcohol use is measured by the question: “How much alcohol have you ever had in your life?” Adolescents’ responses range from “more than 20 whole drinks” to “none at all, not even a sip” along a 6-point scale. For both adolescent cigarette and alcohol use, the higher values represent heavier use. Because adolescent cigarette and alcohol use are highly skewed, they were recoded as 0 = no use of cigarette/alcohol and 1 = use of cigarette/alcohol.

Control variables
            Five control variables developed from baseline data are included in the analysis. These variables are adolescents’ age, sex, race, and mother’s education, and treatment condition. Sex was coded as 0 = female and 1 = male. Race is measured by four categories: White, Black, Hispanic and other race/ethnicity. I created three dummy coded variables and used White as the reference group. Mother’s education is measured by a 3-point scale, including graduated from high school or less, some college education, and college graduates. I created two dummy coded variables and used “less or graduated from high school” as the reference group. Treatment condition is measured by identifying whether the adolescent belongs to the experimental group or the control group. The experimental group was coded as 1 and the control group was coded as 0.

Analysis Plan
            The conceptual model is represented by a set of structural equations. Because the data come from a randomly selected sample throughout the nation with only one adolescent per neighborhood, there is no dependence across observations within each neighborhood. In this sampling scheme, the conceptual model of this study can be represented by regression models or structural equations (Duncan & Raudenbush, 1999; Duncan et al., 1997). Mplus is used as the statistical modeling program, because it has special modeling capabilities for both continuous and categorical data (Muthén & Muthén, 1998). The weighted least squares method with robust standard errors and mean-adjusted Chi-square test statistic (WLSM) is used as the estimator in the analysis. Because the model of this study contains both categorical and continuous variables, the correlation matrix estimated from these variables are unlikely to behave like ordinary sample moments (Jöreskog & Sörbom, 1996). The weighted least squares method must be used instead of maximum likelihood method or generalized least squares method. I use the factor score for each neighborhood concept and treat the factor scores as exogenous variables in each model because neighborhood variables are highly skewed.
            All analyses are processed separately by either census tracts or census block groups. Because of limited sample size, the conceptual model was first examined separately for each neighborhood characteristic and each substance for the purpose of identifying which neighborhood characteristics were candidates for further analysis. Neighborhood characteristics that were significantly associated with any parental characteristic or adolescent substance use were integrated into a final model of cigarette use or a final model of alcohol use. For the comparison between tract models and block-group models, the same neighborhood characteristics were retained in the tract models and the block-group models based on each substance. The tract models and the block-group models were compared in the pattern of significant relationships in the model, magnitude of the path coefficients, and the proportion of variance that neighborhood latent constructs explain for the parental characteristics and adolescent substance use.

Results

            I first focus on tract neighborhood influences on adolescent cigarette use, which is presented in Figure 2.2. Standardized coefficients are presented in Table 2.1. Unstandardized coefficients and standard errors are presented in Appendix 2.1. The model fit indices suggest a reasonable fit between the data and the model (Chi(108)=713.513; CFI=0.94; RMSEA=0.07). Although the Chi-square statistics is significant, the sample size of this study (924) usually guarantees a significant Chi-square for a model with this number of variables. The standardized regression coefficients between parental closeness and its indicators range from 0.62 to 0.83. The standardized regression coefficients between parental monitoring and its indicators range from 0.77 to 0.87. The same pattern of relationships between parental closeness, parental monitoring, and their indicators is also found in the model of block-group influences on cigarette use, tract influences on alcohol use, and block-group influences on alcohol use.
            According to Figure 2.2, Low SES has indirect effects on adolescent cigarette use through parental monitoring. Contrary to the predictions, Low SES is positively associated with parental monitoring (0.121). Parents who live in a low SES neighborhood tend to report higher parental monitoring, which in turns reduces adolescent cigarette use. Please refer to paper one “Neighborhood Influences on Adolescent Cigarette and Alcohol Use: Mediating Effects through Parents and Peers” for a detailed explanation.
            Comparing the relative influences of parental closeness, parental monitoring, and parent smoking on adolescent cigarette use, parental monitoring is the strongest predictor. Parental monitoring has negative effects on adolescent cigarette use (-0.255), suggesting that adolescents who have higher parental monitoring are less likely to use cigarette. Parent smoking has positive effects on adolescent cigarette use (0.133). Adolescents who have smoking parents are more likely to use cigarette. While parental monitoring and parent smoking are significantly associated with adolescent cigarette use, parental closeness is not significantly associated with adolescent cigarette use. Part of the reason is that the correlation between the residuals of parental closeness and parental monitoring affects the significance of parental closeness.
            The pattern of the significant relationships in the model of block-group neighborhood influences on adolescent cigarette use is different from that in the model of tract neighborhood influences on adolescent cigarette use. The results of block-group neighborhood influences on adolescent cigarette use are presented in Figure 2.3. Standardized coefficients are presented in Table 2.1. Unstandardized coefficients and standard errors are presented in Appendix 2.1. The model fit indices show a similar level of fit as that of the tract model (Chi(108)=717.66; CFI=0.94; RMSEA=0.07). Like the model of tract neighborhood influences on adolescent cigarette use, Low SES measured by block groups has indirect effects on adolescent cigarette use through parental monitoring (0.117). However, the model of block-group influences shows two additional significant paths in which High SES is associated with parental smoking and Hispanic concentration is associated with adolescent cigarette use. High SES measured by block groups has indirect effects on adolescent cigarette use through parent smoking (-0.145), indicating that high SES neighborhoods reduce parent smoking, which in turn reduces adolescent cigarette use. Hispanic concentration in a block-group neighborhood has negative direct effects on adolescent cigarette use (-0.086). Adolescents who live in a neighborhood with a high proportion of Hispanics are less likely to report cigarette use.
            After comparing the tract influences on adolescent cigarette use with the block-group influences on adolescent cigarette, I next focus on the comparison between tract and block-group influences on adolescent alcohol use. Figure 2.4 presents the results of tract influences on adolescent alcohol use. Standardized coefficients are presented in Table 2.2. Unstandardized coefficients and standard errors are presented in Appendix 2.2. The model fits the data reasonably well (Chi(102)=488.41; CFI=0.96; RMSEA=0.06). Contrary to the predictions, Low SES has negative effects on adolescent alcohol use through increasing parental monitoring (0.093). Parents who live in a low SES neighborhood are more likely to closely monitor their children, which in turn reduces adolescent alcohol use. High SES has positive effects on adolescent alcohol use through increasing parent drinking (0.186), which also contradicts the hypothesis that High SES would reduce parent drinking. Parents who live in a high SES neighborhood are more likely to use alcohol, which in turns increases adolescent alcohol use. Please refer to paper one “Neighborhood Influences on Adolescent Cigarette and Alcohol Use: Mediating Effects through Parents and Peers” for a detailed explanation.
            The block-group model of alcohol use shows a similar pattern of relationships as that of the tract model of alcohol use (Chi(102)=483.53; CFI=0.96; RMSEA=0.06). Figure 2.5 presents the results of block-group influences on adolescent alcohol use. Standardized coefficients are presented in Table 2.2. Unstandardized coefficients and standard errors are presented in Appendix 2.2. Low SES influences adolescent alcohol use through increasing parental monitoring (0.101) and High SES influences adolescent alcohol use through increasing parental drinking (0.213). Parental closeness is not predicted by any neighborhood characteristic and is not associated with adolescent alcohol use.
            After presenting the pattern of relationships in the models, I compared the block-group models with the tract models in the magnitude of path coefficients and the proportion of variance that the model explains for parental characteristics and adolescent substance use. According to Table 2.1 and Table 2.2, there is not a consistent pattern showing whether block-group models have bigger or smaller coefficients than tract models. Besides, the differences in the magnitude of the coefficients between block-group models and tract models are small. For example, in Table 2.1, the coefficients between Low SES and parent smoking is larger in the block-group model (-0.145) than in the tract model (-0.072), while the coefficients between Low SES and parental monitoring is smaller in the block-group model (0.117) than in the tract model (0.121).
            Table 2.3 presents the proportion of variance that the model explains for parental characteristics, adolescent cigarette use, and adolescent alcohol use. For both adolescent cigarette use and adolescent alcohol use, the block-group model explains essentially the same proportion of variance in parental closeness, parental monitoring, parent drinking, and adolescent alcohol use as the tract model does.

Discussion

            This study shows limited support for the hypothesis that census block groups are a better unit for defining neighborhoods than census tracts when the study purpose is to determine whether parental characteristics act as mediators between neighborhoods and adolescent substance use. With respect to adolescent alcohol use, the pattern of relationships in the block-group model is similar to that in the tract model. Both models suggest that Low SES increases parental monitoring and High SES increases parent drinking. However, with respect to adolescent cigarette use, the pattern of relationships shown in the block-group model is different from that in the tract model. Specifically, both models show a significant relationship between Low SES and parental monitoring, but the block-group model also suggests that High SES decreases parent smoking and Hispanic concentration decreases adolescent cigarette use. Regarding the magnitude of the relationships and the proportion of variance in parental and adolescent characteristics explained by the model, the tract models are similar to the block-group models.
            This study suggests that neighborhoods measured by census tracts and census block groups do not substantially differ in the prediction of parental characteristics and adolescent substance use. These results are not consistent with the findings of prior studies. Elliott and colleagues (forthcoming) found that the size of a neighborhood perceived by its residents is closer to a block group than a tract. They validated this result by estimating the homogeneity of residents’ responses and the correspondence between the aggregated residents’ responses and census data. They concluded that census block groups are a better unit for defining neighborhoods then census tracts. Unlike the results found by Elliott and his colleagues, some other researchers suggested that the best neighborhood size is larger than a tract. Sampson and colleagues (1997) identified neighborhoods in Chicago by using multiple criteria such as physical boundaries, local knowledge of residents, and demographic homogeneity. The average neighborhood size recommended by Sampson is about 25 blocks (Sampson, Raudenbush & Earls, 1997).
            The contradictory evidence in regard to the best neighborhood size may be because researchers use different methods to evaluate the appropriateness of a neighborhood size. For example, this study evaluates the neighborhood size based on the ability of neighborhoods to predict parental behaviors, while Elliott et al. evaluated the neighborhood size based on the residents’ perception. Another explanation for the contradictory evidence may be because researchers use different sizes to infer different neighborhood mechanisms (Gaslter & Killen, 1995). The approach of using census data is to use aggregated residents’ responses to be the proximate measures of neighborhood mechanisms so that neighborhoods measured by different sizes may actually represent different neighborhood mechanisms. For example, while neighborhoods measured by block-groups can represent the immediate surrounding environment of a household, neighborhoods measured by tracts or larger can represent local institutional resources, which are usually located outside a block group. In this study, neighborhoods measured by block groups may influence parenting behaviors through parents’ interaction with neighbors, while neighborhoods measured by tracts or larger may influence parenting behaviors through the quality of local institution. Therefore, the similar tract and block-group effects found in this study does not necessarily suggest that neighborhood size does not matter, but may suggest that different neighborhood sizes capture different aspects of a neighborhood that contribute to the prediction of parental behaviors.
            This study also shows that High SES is positively associated with parent drinking and Low SES is negatively associated with parent smoking, which suggests that smoking but not drinking may be a sensitive indicator of low SES neighborhoods. This result may reflect the societal norm toward different substances. There is general agreement that smoking is perceived as a more deviant behavior than drinking. Smoking may be more acceptable in a low SES neighborhood because low SES neighborhoods have low ability to socialize residents in conventional values and have low standards toward deviant behaviors. On the other hand, wine drinking has become part of popular culture in professional classes in recent decades. High SES people are known to be the major consumers of alcohol (Substance Abuse and Mental Health Service Administration, 1999). These observations may explain why Low SES is positively associated with parent smoking, while High SES is positively associated with parent drinking. Note that “alcohol use” in this study is not equivalent to “problem drinking.” Contrary to the high rates of alcohol use in high SES neighborhoods, problem drinking such as binge drinking and heavy drinking are more prevalent among low SES people and therefore problem drinking are more likely to exist in low SES neighborhoods (Substance Abuse and Mental Health Service Administration, 1999).
            This study found that block-group neighborhoods with Hispanic concentration decrease adolescent cigarette use. This result suggests that neighborhoods exert their influences on adolescents through the collective socialization models, which emphasize how adults outside a family influence children (Jencks & Mayer, 1990). Because Hispanics are known to have the lowest rate of cigarette use compared to other ethnic groups except Asians, adolescents in a neighborhood with a high proportion of Hispanics may be less likely to learn to use cigarette from the adults living in the same neighborhood (Substance Abuse and Mental Health Service Administration, 1999).
            The limitation of this study is lack of longitudinal neighborhood variables. It is possible that the relationship between neighborhoods and individual behaviors is because of non-random selection of individuals into different neighborhoods and not because of neighborhood influences (Tienda, 1990). Therefore, a longitudinal assessment of neighborhood variables is needed for estimating neighborhood effects on parents and adolescents. This study only uses 1990 Census for the estimation of neighborhood effects so that the casual direction of the relationships between neighborhood variables and parental or adolescent variables may not be able to be specified. For example, the correlation between Low SES and parent smoking may be because smoking parents are more likely to move into low SES neighborhoods.
            In summary, this study demonstrates the importance and the process of adjusting neighborhood size according the outcome of the study. Although this study found that census tracts and census block-groups are similar in the ability to represent a neighborhood, this study did find that the pattern of relationships among neighborhoods, parents, and adolescent cigarette use differs when neighborhoods are measured by different neighborhood sizes.
 
 

CHAPTER THREE
Paper Three: A Typology of Disadvantaged Neighborhoods: Neighborhood Influences on the Relationship between Parenting and Adolescent Cigarette and Alcohol Use

Introduction

            This study has three purposes. The first purpose is to identify a typology of neighborhoods according to disadvantaged conditions and geographical locations that are characterized by urban, suburban, or rural areas. The second purpose is to identify the effects of neighborhood types on adolescent substance use and parenting. The third purpose is to understand whether the relationship between parenting and adolescent substance use varies by neighborhood type.
            The influences of neighborhoods on adolescent development have recently become a popular area of research. Neighborhood influences have been examined on various adolescent behaviors, such as academic performance, teenage pregnancy, delinquency, and substance use. Recent reviews of neighborhood research on child and adolescent development have pointed out that previous empirical studies suffer many methodological limitations (Jencks & Mayer, 1990; Gephart, 1997; Leventhal & Brooks-Gunn, 2000). Thus, the findings from extant literature remain inconclusive about how neighborhoods influence adolescent behaviors. One of the major limitations emphasized by these reviews is the measurement of neighborhood context.
            The most typical approach to measuring neighborhood context has been to use factor analysis to identify dimensions of a neighborhood among census variables (Leventhal & Brooks-Gunn, 2000). For example, the most common dimensions found in previous studies are low SES, residential mobility, and racial heterogeneity. Although this strategy can identify the underlying dimensions of a neighborhood, it assumes that all neighborhoods are homogeneous in possessing these dimensions (Esbensen & Huizinga, 1990). This strategy ignores the possibility that a diversity of neighborhoods with different combinations of these dimensions might exist. In addition, because previous studies nearly all focused on inner-city areas (Gephart, 1997; Leventhal & Brooks-Gunn, 2000), a disadvantaged neighborhood thus becomes a synonym for an inner-city neighborhood with low SES, high residential mobility, and high racial heterogeneity. However, disadvantaged neighborhoods are not homogeneous with the same combination of neighborhood dimensions. For example, rural Blacks, Hispanics, and Native Americans also have disadvantaged neighborhoods, which are characterized by racial homogeneity and high rates of poverty (Snipp, 1996).
            Recognizing that neighborhoods are not homogeneous in common indicators of disadvantage is important for examining the relationship between neighborhoods and adolescent behaviors. Disregarding neighborhood diversity may mask the true relationships that exist between neighborhood characteristics and adolescent behaviors. For example, if the relationship between neighborhoods and adolescent behaviors is found to be small, it may be an indication of greater neighborhood diversity, with different strength or directions of association between neighborhood context and adolescent behaviors across different kinds of neighborhoods. Therefore, it is necessary to articulate the neighborhood type under which relationships are observed (Rapkin & Luke, 1993).
            Using a typology approach can help in discerning underlying forces and processes in each type of neighborhood. For example, if two neighborhoods have high poverty rates but one has high and the other has low residential mobility, the underling mechanisms that drive neighborhood influences on adolescents can be quite different. The first kind of neighborhood may influence adolescents mainly through the low neighborhood cohesion because residents do not know each other. The second kind may influence adolescents mainly through lack of role models for conventional behaviors. Adolescents in the second kind of neighborhood have lesser chances to connect with adults who have conventional norms and behaviors because of the concentration effects of deviant adult neighbors and the low turnover of residents in and out of the neighborhood.
            This study uses cluster analysis to develop a neighborhood typology using the characteristics identified by neighborhood theories. Cluster analysis is a statistical procedure that attempts to classify samples into relatively homogeneous groups based on some entities (Aldenderfer & Blashfield, 1984). Cluster analysis is used because this method is especially appropriate for identifying natural groupings of neighborhoods that may not be apparent from previous studies. After identifying the neighborhood typology, adolescent substance use, parenting and other parental behaviors (i.e., closeness, monitoring, and parental substance use), and the relationships between the above parental characteristics and adolescent substance use are examined within each neighborhood type.

Neighborhood Typologies in Previous Studies

            Social science researchers have a long history of proposing different theoretical models of neighborhood types (Fellin, 1995). For example, Warren and Warren (1977) categorized neighborhoods into six types in terms of social identity, social interaction, and linkages to the wider communities. In recent years, Figueira-McDonough (1991) developed four neighborhood types using two dimensions: population factors (i.e., poverty and mobility) and organizational factors (i.e., social network and material resources). Although different neighborhood typologies have been theoretically suggested, few empirical studies have been conducted so that the application of these neighborhood typologies in different settings is still unknown (Chow, 1998).
            Two recent empirical studies that identified neighborhood typologies were conducted by Esbensen and Huizinga (1990) and Chow (1998). Esbensen and Huizinga (1990) conducted a cluster analysis among many demographic variables and found that three types of disadvantaged neighborhoods existed in their data. The first type was traditional disadvantaged neighborhoods with high rates of poverty and racial mix. The second type was also low in economical status and comprised high rates of unmarried persons and high rates of residential mobility. The third type had a majority of Blacks and many single-parent households.
            Chow (1998) conducted a cluster analysis along three factors: poverty-related conditions, crime, and number of infant deaths. She found four types of neighborhood in her data. The first type was stable neighborhoods where the rates of poverty, crime and infant deaths were low. The second type was transitory neighborhoods that were characterized by high rates of infant death but low rates of poverty and crime. The third type was distressed neighborhoods with high rates of poverty but crime and infant death were not so serious. The fourth type was extreme neighborhoods with high rates of poverty and crime but low rates of infant death.
            Esbensen and Huizingas’ (1990) and Chow’s (1998) studies demonstrated that neighborhoods are not homogeneous. Even among neighborhoods that are all economically disadvantaged, the combinations or directions of neighborhood characteristics are quite different, suggesting the need for a typology approach to studying about neighborhood context.

Rural and Suburban Neighborhoods

            Although Esbensen and Huizinga (1990) and Chow (1998) empirically developed neighborhood typologies, both studies were based on the data collected in metropolitan areas. If a nation-wide data set is used to develop a neighborhood typology, studies must consider whether neighborhoods are located in urban, suburban, or rural areas.
            Compared with the intense research attention paid to urban poverty, rural neighborhoods have received much less attention. Researchers who are interested in rural neighborhoods claim that the severity of neighborhood disadvantage is much higher in rural than in urban areas such that a larger number of residents are under poverty line in rural areas (Rural Sociological Society, Task Force on Persistent Rural Poverty, 1993). Although not directly focusing on rural neighborhood disadvantages, one study reported that neighborhood population density is negatively associated with adolescent school rates of cigarette and alcohol use, which shows that residing in a less crowded neighborhood context may contribute to adolescent substance use at the school level (Ennett, et al., 1997).
            Rural neighborhoods have characteristics that are distinctly different from urban neighborhoods. Rural neighborhoods are known to have a higher proportion of children and elderly, have a higher proportion of two-parent households, and have more homogeneous social networks that involve many kinship ties (Beggs, et.al., 1996; Hofferth & Iceland, 1998). These characteristics may create a local environment that is structurally and culturally different from urban neighborhoods.
            Located geographically between urban neighborhoods and rural neighborhoods are suburban neighborhoods. No studies have specifically examined the influences of suburban neighborhoods on adolescent behaviors, so it is unknown whether or how suburban neighborhoods exert influences on adolescents. A suburban neighborhood is usually located adjacent to a central city and often includes self-governing municipalities or townships (Fellin, 1995). Traditionally, suburban neighborhoods have a high proportion of residential housing but now are increasingly occupied by business and industrial units (Fellin, 1995). For addressing the limitation that few studies have investigated how neighborhoods with different geographical locations may influence parents and adolescents, this study includes the dimension of urban, suburban, and rural in the process of developing a neighborhood typology.

Neighborhood Influences on Parents and Adolescents: Theoretical Perspectives

            In addition to developing a neighborhood typology, this study intends to understand whether adolescent behaviors and parenting vary across different types of neighborhoods. Jencks and Mayer (1990) provided a comprehensive review of the mechanisms of neighborhood influences on adolescent behaviors, which provides some guidance about how different types of neighborhoods transfer their influences to adolescents. They proposed that neighborhoods influence adolescents through epidemic models, collective socialization models, and institutional models. For a detailed review of neighborhood mechanisms on adolescents, please refer to paper one “Neighborhood Influences on Adolescent Cigarette and Alcohol Use: Mediating Effects through Parents and Peers.”
            Regarding the mechanisms of neighborhood influences on parenting, no theories are specifically developed for explaining how neighborhoods influence parenting. Most neighborhood studies have adapted the theoretical frameworks of social disorganization theory (Shaw & McKay, 1969), Wilson’s theory (Wilson, 1987), and Coleman’s concept of social capital (Coleman, 1988). Although these three theories address neighborhood influences in general, they all explain to some extent how neighborhoods influence parenting. Summarizing from the three theories, neighborhoods may influence parenting through the following mechanisms: (1) organizational and interpersonal supports, and (2) neighborhood norms about family management and deviant behaviors.
            The first mechanism of neighborhood influences on parenting refers to the effects of low organizational and interpersonal support in a disadvantaged neighborhood. A disadvantaged neighborhood is usually characterized by low sustainable local organizations, which create an environment where parents do not have sufficient resources to facilitate adolescent development. For example, the quality of local schools may be affected by scarce learning equipment and lack of devoted teachers. Parks, libraries, and community centers either do not exist or are not well maintained. A disadvantaged neighborhood also lacks supportive social networks because residents do not know each other, which influence the possibility that parents can obtain help from local social networks.
            The second mechanism of neighborhood influences on parenting is through neighborhood norms about family management and deviant behaviors. Wilson (1991a; 1991b) discussed that the concentration of extremely poor families in inner-city areas creates an isolated environment so that one family’s behaviors, attitudes, and values are easily influenced by other families who live in the same neighborhood. By residing in a neighborhood where most adults are unemployed, do not have plans for the future, and even conduct criminal activities, parents become tolerant towards unemployment, drug use and crime. In addition, parents may possess a laissez-faire attitude in parenting and have low expectations toward their children because they do not have role models to learn about family management, have low self-efficacy in parenting, or believe parenting is unimportant (Elder et al., 1995). Adolescents who grow up in this kind of neighborhood then have a high possibility of developing problem behaviors.
            This study examines whether adolescent substance use, parenting, and the relationship between parenting and adolescent substance use vary across different types of neighborhoods. The directions or patterns of the relationships are not hypothesized because it is difficult to hypothesize the findings within the context of limited prior theories and empirical studies.

Methods

Data
            The data for this study are obtained from the Family Matters Project, a randomized experimental study designed to determine whether a family-directed intervention prevented adolescent tobacco and alcohol use (Bauman et al., 2001a; Bauman, et al.). Family Matters comprises 1316 parent-adolescent pairs, who were generated by random digit dialing throughout the United States. After finishing baseline interviews (55.4% completed baseline interviews), parents and adolescents were randomly assigned to either experimental or control groups. Family Matters was implemented from July, 1996 to September, 1997. The experimental group received four booklets, which were mailed in sequence. These booklets served as triggers to encourage interaction between parents and adolescents to prevent tobacco and alcohol use. Following each booklet, health educators called the parents to encourage participation and clarify questions. Parents and adolescents were interviewed by phone calls at baseline and at three and twelve months after completing the program (follow-up one and follow-up two). Seventy-nine percent of the baseline adolescents finished the follow-up two interviews. This study uses the responses of adolescents who completed all three interviews (1014 cases) and whose addresses could be matched to census block groups (1237 cases), which generates 924 cases. In the final sample, Whites comprised 78.5%; Blacks comprised 9.6%; Hispanics comprised 7.5%; and others comprised 4.5%. Age ranged from 12 to 14. Half of the sample was male (49.2%). The majority of adolescents’ mothers graduated from high school or had some college education (64.4%); 30% graduated from college; 5.4% did not graduate from high school.
            To understand the influence of attrition on the sample, we compared adolescents who only finished the baseline interview and those who finished all three interviews (Bauman et al., 2001b). Adolescents who only finished baseline interview were more likely to be non-white, have a mother with lower education, live in a single-parent home, and to be baseline cigarette and alcohol users.

Measurement
Neighborhood characteristics
            Neighborhood characteristics are developed from 1990 Dicennial Census for census block groups. Census block groups are used because they provide spatial information about a neighborhood such as geographical and population characteristics. In addition, compared with census tracts, the size of a census block group is closer to residents’ perception about their neighborhood size (Elliott, forthcoming).
            This study includes seven neighborhood dimensions, which are low SES, high SES, residential mobility, immigrant concentration, White and Black racial composition, Hispanic concentration, and rural/suburban/urban. The selection of the first four characteristics is based on the social disorganization theory (Shaw & McKay, 1969), Wilson’s theory (1987) and Brooks-Gunn and her colleagues’ series of studies (1993; 1997). The dimension of high SES is separated from low SES because each dimension represents different potential neighborhood mechanisms. A neighborhood that has a low proportion of high SES residents suggests that adolescents do not have role models to learn about conventional behaviors. On the other hand, a neighborhood that has a high proportion of low SES neighbors suggests that adolescents are more likely to learn problem behaviors from deviant friends. The inclusion of White and Black racial composition follows the suggestion that studies should conceptually separate the effects of SES and race because they represent different origins of a disadvantaged neighborhood (Massey, 1998). The concentration of Blacks in a few neighborhoods reflects the prevalence of residential segregation in U.S. today. These are the results of the persistence of white racial prejudice and discrimination in the housing markets and banking industries (Massey, 1996). Therefore, the proportion of Blacks refers to the extent of residential segregation, which cannot be shown by combing indicators of race and SES in a single neighborhood domain. The inclusion of Hispanic concentration also shows the uneven distribution of resources across neighborhoods due to the racial effects. Because the geographical concentration of Hispanics is poorly correlated with the geographical concentration of Whites and Blacks, I separate Hispanic concentration as another neighborhood domain. The inclusion of rural/suburban/urban is because it provides another unique dimension to classify neighborhoods, which was neglected in previous studies.
            Low SES is assessed by the proportion of residents who have family income less than $12,500 and the proportion of residents who are below the poverty line. High SES is assessed by the proportion of residents who have family income more than $75,000 and the proportion of residents who have professional or managerial occupations. The cut points of family income to represent high SES and low SES follows Coulton and her colleagues’ suggestion (1996). They adapted the federally defined poverty threshold as the cut point of low SES. This threshold was set at $12,674 for a family of four in 1989. They also suggested that the cut point of high SES should represent the top 12% of the family income distribution, which is about $75,000 in 1990 census data. Residential mobility is assessed by the proportion of residents who lived in the same house in 1985 and the proportion of households that were occupied by the owner for more than 10 years. Immigrant concentration is assessed by the proportion of residents who are foreign born. White and Black racial composition is measured by the two items, the proportion of residents who are White and the proportion of residents who are Black. Hispanic concentration is measured by the proportion of residents who are Hispanic.
            Rural/suburban/urban is measured by the proportion of people who live in rural, suburban, or urban areas. The Census Bureau used population size and the nature of surrounding areas to define urban, suburban, or rural areas. An area was defined as an urbanized area if it had population size over 50,000, comprised one or more populous centers, and comprised adjacent densely settled surrounding areas. Outside an urbanized area, a suburban area was defined as any incorporated place or census designated place (CDP) with at least 25,000 people. Incorporated places or census designated places (CDP) were geographical units defined by the Census Bureau, which referred to densely settled population centers that had names and community identities. Territory, population, and housing units that the Census Bureau did not classify in the above two categories were defined as rural. For example, a rural place was any incorporated place or CDP with fewer than 2,500 people (U.S. Department of Commerce and Bureau of the Census, 1994).

Parental characteristics
            Parental characteristics are developed from baseline data. Parental characteristics are created by taking the average of reports about fathers and mothers. Parental closeness measures attachment, affection, and child-centerness of a parent-adolescent relationship. This concept is measured by four indicators: (1) “How often does your mother (father) kiss or hug you?” (2) “How close do you feel toward her (him)?” (3) “ Does your mother (father) spend time just talking with you?” and (4) “Does your mother (father) do fun things with you together?” with responses ranging from “very much/very often” to “not at all” along a 4-point scale (Std. Cronbach ?=0.82). A total score is created by summing up the four items. Higher values indicate higher closeness. Parental monitoring is defined as parental knowledge and awareness about a child’s location and activities. This concept is measured by four indicators: (1) “Does your mother (father) try to know what you do with your free time?” (2) “Does your mother (father) try to know where you are most afternoons after school?” (3) “Does your mother (father) really know what you do with your free time?”, (4) “Does your mother (father) really know where you are most afternoons after school?” with responses ranging from “always” to “not at all” along a 4-point scale (Std. Cronbach ?=0.83). A total score is created by summing up the four items. Higher values represent higher monitoring. Parent smoking is measured by asking adolescents: “About how many cigarettes do you think your mother (father) now smokes in a day” with responses ranging from “more than a pack a day” to “no cigarettes” along a 5-point scale. Parent drinking is measured by asking adolescents: “On the average, about how much alcohol do you think your mother (father) now drinks in a day?” with responses ranging from “4 or more drinks a day” to “none at all” along a 5-point scale. Because the responses to the questions about parent smoking and parent drinking are highly skewed, they were recoded as dichotomized variables, with 0 = does not smoke/drink and 1 = does smoke/drink in a day.

Adolescent cigarette and alcohol use
            Adolescent substance use is developed from the baseline and follow-up two interviews. Adolescent cigarette use is measured by the question: “How much have you ever smoked cigarettes in your life?” Adolescents’ responses range from “more than 20 whole cigarettes” to “none at all, not even a puff” along a 5-point scale. Adolescent alcohol use is measured by the question: “How much alcohol have you ever had in your life?” Adolescents’ responses range from “more than 20 whole drinks” to “none at all, not even a sip” along a 6-point scale. Because adolescent cigarette and alcohol use are highly skewed, they were recoded as 0 = no use of cigarette/alcohol and 1 = use of cigarette/alcohol.

Control variables
            Five control variables are developed from the baseline data. These variables are adolescents’ age, sex, race, and mother’s education as well as treatment condition. Sex was coded as 0 = female and 1 = male. Race is measured by four categories: White, Black, Hispanic and other race/ethnicity. I created three dummy coded variables and used White as the reference group. Mother’s education is measured by a 3-point scale, including from graduated from high school or less, some college education, and college graduates. I created two dummy coded variables and used “graduated from high school or less” as the reference group. Treatment condition is measured by identifying whether the adolescent belongs to the experimental group or the control group. The experimental group was coded as 1 and the control group was coded as 0.

Analysis Plan
            The analysis of this study includes two parts. The first part is to use cluster analysis to create a neighborhood typology. Cluster analysis is a statistical technique designed to divide a heterogeneous sample into more homogeneous subgroups based on some entities (Speece, 1990). There is a series of steps in any cluster analysis. Recommended by Lorr (1983) and Rapkin and Luke (1993), these steps include: (a) examining outliers and missing data; (b) selecting variables; (c) choosing cluster algorithms and similarity measures; (d) determining number of clusters; (e) determining cluster validity and reliability; and (f) interpretation of the results.
            In the first step of cluster analysis, outliers and missing data should be identified and examined. Identifying outliers is especially important because the results of cluster analysis can be strongly affected by outliers. This study first identified outliers and then examined these outliers in order to determine whether these outliers need to be deleted or kept for theoretical reasons. The second step of cluster analysis is to select variables into the clustering. The selection of variables should be based on theories that support the classification (Aldenderfer & Blashfield, 1984). The selection of variables in this study is based on social disorganization theory and Wilson’s theory and Brook-Gunn and her colleagues’ series of studies (1993; 1997). Including too many variables may make the interpretation difficult and also increase the possibility that the presence of non-relevant variables may obscure the cluster structure (Everitt, 1993). Variables with substantially different units of measurements need to be standardized.
            The third step is to choose cluster algorithms. Two kinds of cluster algorithms have been most frequently used in social science research: hierarchical methods and iterative partitioning methods. In hierarchical methods, each subject is treated initially as a single entity. At each successive level during the clustering, two of the clusters are assigned to the same group. The analysis proceeds until a single cluster is formed which contains all the entities. For obtaining the optimal number of clusters, the investigator has to select one of the solutions in the nested sequence of clustering that comprises the hierarchy. The different hierarchical methods reflect different merging rules in the clustering. The most common hierarchical methods include single-linkage, complete-linkage, average-linkage, and Ward’s method. Ward’s method has been recommended to have the best performance in recovering the group structure among hierarchical methods (Milligan & Copper, 1987).
            With iterative partitioning methods, researchers need to begin the clustering with specifying the expected number of clusters and proposing where the centroid (mean) of each cluster might be. Each subject is allocated to the cluster that has the nearest centroid. The new centroids of the clusters are computed again and subjects are reassigned based on the new centroids. The clustering processes iteratively until no subjects change assignments. The strength of iterative partitioning methods is that it allows multiple passes of the data so it can avoid poor initial partitioning, which is the major drawback of hierarchical methods. The disadvantage of using iterative partitioning methods is that the expected number of clusters is sometimes hard to predict.
            K-means is one of the iterative partitioning methods, which is selected to be the clustering method in this study. Previous empirical studies showed that K-means has the best ability to recover the true groupings of the data if a nonrandomized starting point is assigned (Milligan & Cooper, 1987). Since there is not enough information about the expected number of neighborhood types, I used Ward’s method (a hierarchical method), which does not require an initial assigned point, to identify the number of groupings. K-means was then conducted with a starting point obtained from the results of Ward’s method (Chow, 1998; Punj & Stewart, 1983).
            Different clustering methods need different similarity measures to quantify the differences between observations based on the selected variables. The most common similarity measures used in previous studies are Pearson product moment correlation coefficients and Euclidean distance. The choice of similarity measures should depend on the nature of the data and the combined performances with different cluster algorithms. Previous studies demonstrated that Euclidean distance performs well with both Ward’s method and K-means, so that Euclidean distance was chosen to be the similarity measure in this study (Speece, 1990; Milligan & Cooper, 1987).
            The determination of the number of clusters is the fundamental unsolved problem of cluster analysis. The performances of many statistical procedures that determine the number of clusters are still unknown. I used the “inverse scree plot” in conjunction with a cross-validation method to determine the number of clusters. The inverse scree plot graphs number of clusters against fusion coefficient, which is the numerical value at which various cases merge to form a cluster. This plot is analogous to the “scree plot” of factor analysis. A significant jump of fusion coefficient may inform the number of clusters extracted from the data (Aldenderfer & Blashfield, 1984).
            The cross-validation method requires randomizing the total sample into two groups, a test sample and a validation sample (Crow, 1996). I first conducted cluster analysis in the test sample and obtained the centroids (means) of clusters from the result. Then, I conducted the cluster analysis in the validation sample both with and with out specifying the centroids obtained from the test sample. The two results obtained in the validation sample were compared by calculating Kappa coefficients for determining which solution has the higher stability.
            After determining different types of neighborhoods, the second part of the analysis is to use multiple linear regression models or logistic regression models to assess the influences of parenting and other parental behaviors on adolescent substance use. All analyses were processed separately for cigarette and alcohol use. The first set of regression models is to assess whether neighborhood types can determine adolescent follow-up two substance use after controlling demographic variables, treatment condition, and baseline substance use. The second set of regression models is to assess whether neighborhood types can determine parenting and other parental behaviors after controlling demographic variables and treatment conditions. After conducting the first two sets of regression models, adolescent follow-up two substance use was regressed on parental characteristics, control variables, and adolescent baseline substance use under each neighborhood type. For example, under neighborhood type one, the first model contains adolescent smoking as the dependent variable. The independent variables in the same model include parental closeness, parental monitoring, parent cigarette use, control variables, and baseline adolescent smoking. The significance of the relationships between parental characteristics and adolescent substance use under each type of neighborhood can inform me whether the influences of parental characteristics on adolescent substance use vary by neighborhood types. All findings are evaluated at a significance level of .05.

Results

            The neighborhood typology was developed by using cluster analysis with a cross-validation method. I first randomly assigned cases to two samples, a test and a validation sample. Then I conducted Ward’s method in each sample. The Ward’s method shows four, five, six, and eleven potential clusters. To determine which solution has the highest stability, I used K-means to obtain the centroids (means) of clusters in the test sample and used Kappa coefficients to compare the results in the validation sample with and without specifying the centroids obtained from the test sample. The Kappa coefficients for four- and six-cluster solution are 0.99 and 0.96 respectively, which show much higher stability than the other solutions. The six-cluster solution appears to be a subtype of the four-cluster solution. For example, the four-cluster solution identifies an urban type of neighborhood in which the majority of residents are Whites. The six-cluster solution further classifies this type of neighborhood into two types, urban White high SES and urban White middle SES neighborhoods. I believe that the six-cluster solution is more likely to represent the theoretical diversity of neighborhoods, so I decided to adapt the six-cluster solution.

Neighborhood Description

            The six types of neighborhood are: (1) rural low SES neighborhoods; (2) urban middle SES neighborhoods; (3) urban high SES neighborhoods; (4) suburban middle SES neighborhoods; (5) rural middle SES neighborhoods, and (6) urban low SES neighborhoods. Table 3.1 displays the means of neighborhood characteristics by neighborhood type. The first type of neighborhood is located in rural areas. It consists of a high proportion of low SES residents as indicated by a high rate of families with low income, a high rate of residents under the poverty line, and low rates of residents having professional jobs and college education. This type of neighborhood tends to have racially mixed residents including Whites, Blacks, and Hispanics. The residential mobility of this type of neighborhoods is close to the average of all neighborhoods.
            The second type of neighborhood comprises a large proportion of middle SES Whites in urban areas. This type of neighborhood has the highest proportion of foreign-born residents and highest residential mobility among all types of neighborhoods.
            The third type is urban neighborhoods with a majority of residents who are high SES Whites. This type of neighborhood has the highest social economic status as indicated by high rates of residents having professional jobs and college education. This type of neighborhood has a low proportion of Black residents and it has the lowest residential mobility.
            The fourth type of neighborhood is located in suburban areas. It comprises a large proportion of White middle SES residents. This type of neighborhood has the second highest residential mobility among all types of neighborhoods.
            The fifth type of neighborhood is rural neighborhoods with a high rate of middle-class Whites and a low rate of residential mobility. This type of neighborhood differs from other types of neighborhood in that nearly all residents in this type of neighborhood are Whites. It has a low proportion of Blacks and a low proportion of foreign-born residents. This type of neighborhood is a typical type of middle-class neighborhood in that it has low rates of extreme low SES or high SES residents.
            The sixth type of neighborhoods is low SES neighborhoods located in urban areas. Its distinguishing features are that its residents are mainly Blacks and are at the lowest social economic status relative to other neighborhood types. This type of neighborhood differs from rural low SES neighborhoods (type one) in that it has a higher rate of low SES and a higher rate of high SES residents.

Description of Adolescent Cigarette Use, Adolescent Alcohol Use, and Parental Behaviors in each Neighborhood Type

            Table 3.2 summarizes the distribution of sample characteristics in each type of neighborhood. The race and SES of adolescents respectively matches to the race and SES of the majority of residents in the neighborhoods. For example, 17% of adolescents in rural low SES neighborhoods (type one) compared with 39% of adolescents in urban high SES neighborhoods (type three) have mothers who graduated from college. The average of adolescent age is similar across different types of neighborhoods. Approximately the same proportion of males as females exist in each type of neighborhood except for rural low SES neighborhoods (type one) and urban low SES neighborhoods (type six), which have slightly higher rates of male adolescents.
            The primary interest of this paper is to assess whether differences in adolescent cigarette use, adolescent alcohol use, and parental behaviors exist across neighborhood types. Table 3.3 shows the percentage of baseline adolescent cigarette use, adolescent alcohol use, and parental behaviors by neighborhood type. Although there are no significant differences in adolescent cigarette use across neighborhood types, it appears that adolescent cigarette use is more prevalent among those who live in rural low SES neighborhoods (type one). Thirty-seven percent of adolescents in rural low SES neighborhoods (type one) use cigarette. Only 16% of adolescents in urban Black low SES neighborhoods reported using cigarette.
            Although the relationship between neighborhood types and cigarette use is not significant, neighborhood differences do exist for alcohol use (Chi-=20.39, P=0.001). According to Table 3.3, baseline adolescent alcohol use is more common in affluent neighborhoods such that above 60% of adolescents reported alcohol use in high SES (type three) or middle SES (type two, four, and five) neighborhoods. In contrast to adolescents in high and middle SES neighborhoods, adolescent alcohol use is less common in low SES neighborhoods. Fifty percent of adolescents in rural low SES neighborhoods (type one) and 37% of adolescents in urban Black low SES neighborhoods (type six) reported alcohol use.
            Similar to the results of adolescent cigarette and alcohol use, baseline parent alcohol use is significantly associated with neighborhood types but baseline parent cigarette use is not. Table 3.3 shows that parent cigarette use does not significantly vary by neighborhood types but it appears that parent cigarette use is more prevalent in rural low SES neighborhoods (type one) than other types of neighborhood.
            Regarding parent alcohol use, urban Black low SES neighborhoods (type six) have the lowest rate of parent alcohol use. Twenty-Eight percent of parents in this type of neighborhood reported using alcohol. Alcohol use is also less common in suburban middle SES White neighborhoods (type four) such that 63% the parents reported alcohol use. Conversely, more than 75% of parents living in rural low SES (type one), urban White middle SES (type two), and urban White high SES neighborhoods (type three) reported using alcohol.
            Regarding parental monitoring and parental closeness, the mean scores are not significantly different in different types of neighborhoods. Urban White high SES neighborhoods have the lowest mean scores in parental monitoring and closeness.
            From the above findings, significant differences in adolescent and parent alcohol use are found across six types of neighborhood. The next step is to understand which two types of neighborhood are different from each other and whether the differences remain after controlling potential confounding variables. I regressed adolescent alcohol use on neighborhood types by using logistic regression models followed by Tukey-Kramer test for testing pair-wise contrasts. Table 3.4 shows that urban Black low SES neighborhoods (type six) are strongly different from urban White high SES neighborhoods (type three) and borderline different from urban White middle SES neighborhoods (type two) (Chi-=7.90, P=0.055). The same procedure was applied to parent alcohol use. Table 3.4 shows that urban Black low SES neighborhoods (type six) are significantly different from urban White middle SES (type two) and urban White high SES neighborhoods (type three).
            Following these significant contrast tests, I added control variables to the model of adolescent alcohol use and the model of parent alcohol use. The control variables I included in the model of adolescent alcohol use are adolescent race, adolescent age, adolescent sex, mother’s education, treatment condition, and baseline adolescent drinking. The control variables I include in the model of parent alcohol use are adolescent race, mother’s education, and treatment condition. Table 3.4 compares the results before and after adding control variables, which shows that the significant contrasts of neighborhood types disappear after including control variables. The significant control variables are race and age of the adolescents in the model of adolescent alcohol use. Because adolescent age is not different across neighborhood types, adolescent race is largely responsible for the differences in adolescent alcohol use between urban White high SES neighborhoods and urban Black low SES neighborhoods. Specifically, Black adolescents have lower rates of alcohol use than White adolescents, which lead to the differential rates of alcohol use in Black and White neighborhoods. Regarding parent alcohol use, race and mother’s education are responsible for the disappearance of the significant contrasts of neighborhood types on parent alcohol use. Black parents are less likely to use alcohol than White parents as well as parents who graduated from high school are less likely to use alcohol than parents who graduated from college.

Impact of Parenting Behaviors on Adolescent Cigarette and Alcohol Use in each Neighborhood Type

            Although adolescent cigarette use, adolescent alcohol use, and parental behaviors do not vary across neighborhood types, the impacts of parental behaviors on adolescent cigarette and alcohol use do vary by neighborhood type. Table 3.5 and table 3.6 each presents the effects of parental behaviors on adolescent cigarette or alcohol use in each type of neighborhood by using logistic regression models. In each logistic regression model, either adolescent cigarette or alcohol use is regressed on parental closeness, parental monitoring, parent cigarette or alcohol use, and control variables. Because of limited number of cases in some types of neighborhood, I only included the control variables that are significantly associated with either adolescent alcohol use or cigarette use in preliminary analyses. In addition, control variables are re-categorized into fewer categories for reducing the number of dummy coded variables. For example, adolescent race is regrouped into White and non-White and mother’s education is regrouped into less than or equal to high school graduation and more than high school graduation.
            Regarding adolescent cigarette use, adolescents living in suburban White middle SES neighborhoods (type four) are more likely to use cigarette if their parents use cigarette. Specifically, adolescents who have smoking parents are more than four times as likely to smoke as adolescents who do not have smoking parents. Regarding adolescent alcohol use, in urban White middle SES neighborhoods (type two), parent drinking is found to be significantly associated with adolescent alcohol use, which suggests that adolescents who have drinking parents are more than two and two-half times as likely to drink as adolescents who do not have drinking parents. Compared with parents in urban White middle SES neighborhoods, parent use only has borderline significant impacts on adolescent alcohol use in urban White high SES neighborhoods (type three, OR=2.4, p=0.052). However, greater parental monitoring is associated with reduced adolescent drinking in this type of neighborhood. While monitoring protects adolescents from drinking in urban high SES neighborhoods, closeness protects adolescents from drinking in rural White middle SES neighborhoods as those adolescents who have greater closeness with parents are less likely to drink.

Discussion

            This study identified six types of neighborhood. They are: (1) rural neighborhoods with high rates of low SES and racially mixed residents; (2) urban White middle SES neighborhoods with a high rate of residential mobility; (3) urban White high SES neighborhoods with a low rate of residential mobility; (4) suburban White middle SES neighborhoods with a high rate of residential mobility; (5) rural White middle SES neighborhoods, and (6) urban Black low SES neighborhoods with a low rate of residential mobility.
            Two types of neighborhood found in this study are poor neighborhoods, but they show different compositions of neighborhood characteristics. The first type of poor neighborhoods is urban poor neighborhoods, which have low residential mobility and a high proportion of Blacks. Previous studies nearly all focused on this type of neighborhood, which shows a typical profile of inner-city disadvantaged neighborhoods (Gephart, 1997; Leventhal & Brooks-Gunn, 2000, Wilson, 1987). This neighborhood type comprises a group of poor Blacks inhabiting a few neighborhoods in inner-city areas, where local basic institutes are disorganized, conventional norms cannot be maintained, and illicit activities take place. Residential mobility of this type is low, which has negative effects on residents’ psychological well-being. Unlike where residential stability is good for affluent neighborhoods in establishing social ties, residential stability creates frustration and isolation for the residents in this type of neighborhood (Ross, Reynolds, & Gis, 2000).
            The context of a poor neighborhood in rural areas is quite different from that in urban areas. Rural poor neighborhoods have a less extreme profile such that they have lower rates of both poor and affluent residents than urban neighborhoods do. In addition, rural poor neighborhoods are more likely to have residents of different ethnic groups, such as Whites, Blacks, and Hispanics, living in a same neighborhood. In contrast with urban poor neighborhoods where a majority of residents in are Blacks, a large share of poor in rural areas are Whites (Rural Sociological Society, Task Force on Persistent Rural Poverty, 1993).
            The differential rates of adolescent alcohol use between urban White high SES neighborhoods and urban Black low SES neighborhoods are found as confounding effects of individual race. This study found that Black adolescents are less likely to use alcohol than White adolescents, which is consistent with the findings of previous studies. According to the Monitoring the Future Project, Black adolescents have had consistently lower rates of alcohol and cigarette use than White adolescents in the past ten years (Johnston, O’Malley, & Bachman, 1999). Explanations of the racial differences in smoking and drinking are that Black adolescents are more likely to be influenced by religion, have lower vulnerability from the modeling effects of parent use, and have higher rates of invalid reports compared with White adolescents (Bauman & Ennett, 1994; Newcomb & Bentler, 1986; Wallace & Bachman, 1991).
            Previous studies found strong associations between neighborhoods and other adolescent problem behaviors, such as violence. However, why does this study show no differences in adolescent cigarette and alcohol use in different neighborhood context? One explanation may come from the nature of adolescent cigarette and alcohol use. This study measured initiation of smoking and drinking by asking adolescents whether they had used cigarette and alcohol. Experimentation with cigarette and alcohol may be normative behaviors in the process of adolescent development. A large proportion of people try out cigarette and alcohol in the stage of adolescence but only a small proportion of them move to problem smoking and problem drinking. In addition, the social context in which an adolescent experiments with substances is different from the social context in which an adolescent heavily uses substances (Sellers, Winfree, & Griffiths, 1993). While experimentation with cigarette and alcohol may be indicators of problems of adolescent development, problem smoking and problem drinking may represent social problems in local social contexts. Therefore, adolescent initiation of cigarette and alcohol use may not represent social disorganization of neighborhood context, which is usually indicated by severe deviant behaviors such as violence, illicit substance use, and problem smoking and drinking.
            Although adolescent cigarette and alcohol use do not vary by neighborhood types, the impacts of parenting on adolescent cigarette and alcohol use change in different types of neighborhoods. This study found that closeness is protective for adolescents in rural White middle-class neighborhoods and monitoring is protective for adolescents in urban White high SES neighborhoods. Parent drinking has significant effects on adolescents in urban White middle-class neighborhoods and has borderline significant effects on adolescents in urban White high SES neighborhoods. In addition, parent smoking has strong significant effects on adolescent smoking in suburban areas.
            Because individual characteristics are highly correlated with neighborhood characteristics under each neighborhood type, I am unable to attribute the differences in parental effects on adolescent cigarette and alcohol use to neighborhood influences. For example, in rural middle-class neighborhoods that are characterized by a high rate of White residents, more than 90% of adolescents in this study are also White. What can be interpreted from this study is that the differences in parental effects on White adolescent cigarette and alcohol use exist across rural, suburban, and urban neighborhoods. This study found that closeness reduces adolescent alcohol use in rural White neighborhoods but not in urban White neighborhoods. This finding may be explained by observation that rural families usually have stronger kinship ties in close proximity than urban families (Beggs, Haines, & Hurlbert, 1996; Hofferth & Iceland, 1998). The value of closeness in social relationships may be emphasized and reinforced in rural families through the frequent and intimate contacts among kin. Within this social context, rural adolescents may regard closeness with parents as an important part in their lives, which may decrease the possibility that they experiment with cigarette and alcohol use at an early age.
            While closeness reduces adolescent drinking in rural White middle-class neighborhoods, monitoring can protect adolescents from drinking in urban White high SES neighborhoods. Compared with their rural counterparts, urban families are usually exposed to more diverse influences such that a greater variety of beliefs and values exists in urban areas (Coleman et al., 1989). Within this context, parental monitoring may be especially important because parental monitoring directly influences the chances that an adolescent will be exposed to the environments where cigarette and alcohol are available. For example, parental monitoring may reduce adolescent use of cigarette and alcohol through establishing a foundation that adolescents do not affiliate with smoking or drinking friends (Kandel, 1996). Urban areas are also less isolated than rural areas in that public transportation and public facilities are more common in urban areas (Hofferth & Iceland, 1998). Within this context, adolescents may have higher accessibility to cigarette and alcohol. This can explain why parent cigarette and alcohol use can affect adolescents in urban areas because urban adolescents are more likely to respond to parent use by actually having access to these substances.
            The results of this study are limited in several ways. The sample includes only has a few minority neighborhoods, which may reduce the chances of uncovering the true structure of neighborhood clusters. For example, the type of rural low SES neighborhood may have been further categorized into more types if this study had enough number of cases. In addition, the understanding of parental influences on adolescent cigarette and alcohol use in minority neighborhood may also be impeded by the small number of cases. For example, the variances of the odds ratios of parental behaviors in such type of neighborhoods are usually huge, which contribute to the insignificance of the effects of parental behaviors on adolescent cigarette and alcohol use. Another limitation is that this study does not have neighborhood variables that can describe neighborhood social environments, such as informal social control, participation of local organizations, and local social network. Future studies may include these variables in the development of a neighborhood typology for understanding the influences of local social environments.
            In summary, the influence of neighborhoods on adolescent substance use is a complicated phenomenon. This study provides an exploratory understanding of neighborhood influences on adolescent cigarette and alcohol use by developing a neighborhood typology and examining adolescent and parental behaviors under different neighborhood types. This study suggests that adolescent cigarette use does not differ across neighborhood types. However, the extent to which the impacts of parental behaviors on adolescent cigarette use and alcohol use do vary significantly across different types of neighborhoods.
 


BIBLIOGRAPHY

Abdelrahman, A. I., Gloria Rodriguez, John Ryan, John F. French, and Donald Weinbaum. 1998. "The Epidemiology of Substance Use Among Middle School Students: The Impact of School, Familial, Community and Individual Risk Factors." Journal of Child and Adolescent Substance Abuse 8(1):55-75.

Aldenderfer, Mark S. and Roger K. Blashfield. 1984. Cluster Analysis. Newbury Park, CA: Sage Publication. Notes: A Sage University Paper 44

Allison, Kevin W., Isiaah Crawford, Peter E. Leone, Edison Trickett, Alina Perez-Febles, Ree L. Blanc, and Linda M. Burton. 1999. "Adolescent Substance Use: Preliminary Examinations of School and Neighborhood Context." American Journal of Community Psychology 27(2):111-41.

Bandura, Albert. 1986. Social Foundations of Thought and Action: A Social Cognitive Theory. Englewood Cliffs, N. J.: Prentice Hall.

Bauman, Karl E. and Susan E. Ennett. 1994. "Tobacco Use by Black and White Adolescents: the Validity of Self-Reports." American Journal of Public Health 84(3):394-98.

Bauman, Karl E. and Susan T. Ennett. 1996. "On the Importance of Peer Influence for Adolescent Drug Use: Commonly Neglected Considerations." Addiction 91(2):185-98.

Bauman, Karl E., Vangie A. Foshee, Susan T. Ennett, Katherine A. Hicks, Michael Pemberton, and Katherine A. Hicks. 2001a. "Family Matters: A Family-Directed Program Designed to Prevent Adolescent Tobacco and Alcohol Use." Health Promotion Practice 2(1):81-96.

Bauman, Karl E., Vangie A. Foshee, Susan T. Ennett, Michael Pemberton, Katherine A. Hicks, Tonya S. King, and Gary G. Koch. 2001b. "Influence of a Family Program on Adolescent Tobacco and Alcohol Use." American Journal of Public Health 91(4):604-10.

Bearman, P. S., Jones, J., and Udry, J. R. 1997. "The national longitudinal study of adolescent health: research design" [Web Page]. Available at http://www.cpc.unc.edu/projects/addhealth/design.html.

Beggs, John J., Valerie A. Haines, and Jeanne S. Hurlbert. 1996. "Revisiting the Rural-Urban Contrast: Personal Networks in Nonmetropolitan and Metropolitan Settings." Rural Sociology 61(2):306-25.

Bogenschneider, Karen, Ming-yeh Wu, Marcela Raffaelli, and Jenner C. Tsay. 1998. "Parent Influences on Adolescent Peer Orientation and Substance Use: the Interface of Parenting Practices and Values." Child Development 69(6):1672-88.

Brook, Judith S., Carolyn Nomura, and Patricia Cohen. 1989. "A Network of Influences on Adolescent Drug Involvement: Neighborhood, School, Peer, and Family." Genetic, Social, and General Psychology Monographs 115:125-45.

Brooks-Gunn, Jeanne, Greg J. Duncan, Pamela K. Klebanov, and Naomi Sealand. 1993. "Do Neighborhoods Influence Child and Adolescent Development." American Journal of Sociology, 99(2):353-95.

Brooks-Gunn, Jeanne, Greg J. Duncan, Tama Leventhal, and J. L. Aber. 1997. "Lessons Learned and Future Directions for Research on the Neighborhoods in Which Children Lives." Pp. 279-97 in Neighborhood Poverty: Context and Consequences for Children, vol. 1, editors Jeanne Brooks-Gunn, Greg J. Duncan, and
J. L. Aber. New York: Russell Sage Foundation.

Burton, Linda M. and Robin L. Jarrett. 2000. "In the Mix, Yet on the Margins: the Place of Families in Urban Neighborhood and Child Development Research." Journal of Marriage and the Family 62:1114-35.

Burton, Linda M., Townsand Price-spratlen, and Margaret B. Spencer. 1997. "On Ways of Thinking About Measuring Neighborhoods: Implications for Studying Context and Developmental Outcomes for Children." Pp. 132-44 in Neighborhood Poverty: Policy Implications in Studying Neighborhoods, vol. 2, editors Jeanne Brooks-Gunn, Greg. J. Duncan, and J. L. Aber. New York: Russell Sage Foundation.

Case, Anne C. 1991. The Company You Keep: The Effects of Family and Neighborhood on Disadvantaged Youths. Cambridge, MA: National Bureau of Economic Research. Notes: working paper 3705

Chaskin, Robert J. 1998. "Neighborhood As Unit of Planning and Action: A Heuristic Approach." Journal of Planning Literature 13(1):11-30.
———. 1997. "Perspectives on Neighborhood and Community: A Review of the Literature." Social Service Review 71(4):521-47.

Chow, Julian. 1998. "Differentiating Urban Neighborhoods: A Multivariate Structural Model Analysis." Social Work Research 22(3):131-42.

Coleman, James S. 1988. "Social Capital in the Creation of Human Capital." American Journal of Sociology 94(sup):S95-S120.

Coulton, Claudia J., Julian Chow, Edward C. Wang, and Marilyn Su. 1996. "Geographic Concentration of Affluence and Poverty in 100 Metropolitan Areas, 1990." Urban Affairs Review 32(2):186-216.

Coulton, Claudia J., Jill E. Korbin, and Marilyn Su. 1996. "Measuring Neighborhood Context for Young Children in an Urban Area." American Journal of Community Psychology 24(1):5-32.

Crane, Jonathan. 1991. "The Epidemic Theory of Ghettos and Neighborhood Effects on Dropping Out and Teenage Childbearing." American Journal of Sociology 5:1226-59.

Crum, Rosa M., Marsha Lillie-Blanton, and James C. Anthony. 1996. "Neighborhood Environment and Opportunity to Use Cocaine and Other Drugs in Late Childhood and Early Adolescence." Drug and Alcohol Dependence 43:155-61.

Darling, Nancy and Laurence Steinberg. 1997. "Community Influences on Adolescent Achievement and Deviance." Pp. 120-131 in Neighborhood Poverty: Policy Implications in Studying Neighborhoods, vol. 2, editors J. Brooks-Gunn, Greg. J. Duncan, and J. L. Aber. New York: Russell Sage Foundation.

Dembo, Richard, James Schmeidler, William Burgos, and Robert Taylor. 1985. "Environmental Setting and Early Drug Involvement Among Inner-City Junior High School Youths." The International Journal of the Addictions 20(8):1239-55.

Duncan, Greg J., James P. Connell, and Pamela K. Klebanov. 1997. "Conceptual and Methodological Issues in Estimating Causal Effects of Neighborhoods and Family Conditions on Individual Development." Pp. 219-50 in Neighborhood Poverty: Context and Consequences for Children, vol. 1, editor Jeanne Brooks-Gunn, Greg J. Duncan, and J. L. Aber. New York: Russell Sage Foundation.

Duncan, Greg J. and Stephen W. Raudenbush. 1999. "Assessing the Effects of Context in Studies of Child and Youth Development." Educational Psychologist 34(1):29-41.

Earls, F., J. McGuire, and S. Shay. 1994. "Evaluating a Community Intervention to Reduce the Risk of Child Abuse: Methodological Strategies in Conducting Neighborhood Surveys." Child Abuse and Neglect 18:473-85.

Elder, G. H., J. S. Eccles, M. Ardelt, and S. Lord. 1995. "Inner-City Parents Under Economic Pressure: Perspectives on the Strategies of Parenting." Journal of Marriage and the Family 57:771-84.

Ellen, Ingrid C. and Margery A. Turner. 1997. "Does Neighborhood Matter? Assessing Recent Evidence." Housing Policy Debate 8(4):833-66.

Elliott, Delbert S., Scott Menard, Amanda Elliott, Bruce Rankin, David Huizinga, and William J. Wilson. forthcoming. Overcoming Disadvantaged: Successful Youth Development in High-Risk Neighborhoods. Chicago: The University of Chicago Press.

Elliott, Delbert S., William J. Wilson, David Huizinga, Robert J. Sampson, Amanda Elliott, and Bruce Rankin. 1996. "The Effects of Neighborhood Disadvantage on Adolescent Development." Journal of Research in Crime and Delinquency 33(4):389-426.

Ennett, Susan T., Robert L. Flewelling, Richard C. Lindrooth, and Edward C. Norton. 1997. "School and Neighborhood Characteristics Associated With School Rates of Alcohol, Cigarette, and Marijuana Use." Journal of Health and Social Behavior 38:55-71.

Esbensen, Finn-aage and David Huizinga. 1990. "Community Structure and Drug Use: From a Social Disorganization Perspective." Justice Quarterly 7(4):691-709.

Everitt, Brian S. 1993. Cluster Analysis. 3 ed. New York: Halsted Press.

Fellin, Phillip. 1995. The Community and the Social Worker. 2 ed. Itasca, IL: F.E. Peacock Publishers, Inc.

Figueira-McDonough, J. 1991. "Community Structure and Delinquency: A Typology." Social Service Review 1(65):68-91.

Flay, Brian R., Frank B. Hu, Ohidul Siddiqui, L. E. Day, Donald Hedeker, John Petraitis, Jean Richardson, and Steve Sussman. 1994. "Differential Influence of Parental Smoking and Friends' Smoking on Adolescent Initiation and Escalation of Smoking." Journal of Health and Social Behavior 35:248-65.

Frisbie, W. P. 1988. "Spatial Processes." Pp. 629-66 in Handbook of Sociology, editor N. J. Smelser. Newbury Park, CA: Sage Publication.

Fuller, Robert C. 1995. "Religion and Ritual in American Wine Culture." Journal of American Culture 16(1):39-45.

Furstenberg, Frank F. Jr. 1993. "How Families Manage Risk and Opportunity in Dangerous Neighborhoods." Pp. 231-58 in Sociology and Public Agenda, J. N. W. J. Wilson. CA: Sage Publication.

Furstenberg, Frank F. Jr., Thmomas D. Cook, Jacquelynne Eccles, G. H. Elder Jr., and Arnold Sameroff. 1999. Managing to Make It: Urban Families and Adolescent Success. Chicago: The University of Chicago press.

Furstenberg, Frank F. Jr. and Mary E. Hughes. 1997. "The Influence of Neighborhoods on Children's Development: A Theoretical Perspective and a Research Agenda." Pp. 23-47 in Neighborhood Poverty: Policy Implications in Studying Neighborhoods, vol. 2, editors Jeanne Brooks-Gunn, Greg J. Duncan, and J. L. Aber. New York: Russell Sage Foundation.

Galster, G. C. and S. P. Killen. 1995. "The Geography of Metropolitan Opportunity: A Reconnaissance and Conceptual Framework." Housing Policy Debate 6(1):7-43.

Galster, George C. 1986. "What Is Neighborhood? An Externality-Space Approach." International Journal of Urban and Regional Research 10:243-63.

Garbarion, James and Deborah Sherman. 1980. "High-Risk Neighborhoods and High-Risk Families: the Human Ecology of Child Maltreatment." Child Development 51:188-98.

Gephart, Martha A. 1997. "Neighborhoods and Communities As Contexts for Development." Pp. 219-50 in Neighborhood Poverty: Context and Consequences for Children, vol. 1, editor Jeanne Brooks-Gunn, Greg J. Duncan, and J. L. Aber. New York: Russell Sage Foundation.

Groves, Robert M. 1989. Survey Errors and Survey Costs. New York: John Wiley & Sons.

Guest, Avery M. and Barrett A. Lee. 1984. "How Urbanities Define Their Neighborhoods." Population and Environment 7(1):32-56.

Hofferth, Sandra L. and John Iceland. 1998. "Social Capital in Rural and Urban Communities." Rural Sociology 63(4):574-98.

Hoffmann, John P. 1993. "Exploring the Direct and Indirect Family Effects on Adolescent Drug Use." The Journal of Drug Issues 23(3):535-57.

Hogan, Dennis P. and Evelyn M. Kitagawa. 1985. "The Impact of Social Status, Family Structure, and Neighborhood on the Fertility of Black Adolescents." American Journal of Sociology 90(4):825-54.

Hunter, A. 1974. Symbolic Communities: The Persistence and Change of Chicago's Local Communities. Chicago: The University of Chicago Press.
Jarrett, Robin L. 1997. "African American Family and Parenting Strategies in Impoverished Neighborhoods." Qualitative Sociology 20(2):275-88.
Jarrett, Robin L. 1995. "Growing Up Poor: the Family Experiences of Socially Mobile Youth in Low-Income African American Neighborhoods." Journal of Adolescent Research 10(1):111-35.

Jencks, Christopher and Susan E. Mayer. 1990. "The Social Consequences of Growing Up in a Poor Neighborhood." Inner-City Poverty in the United States, editors Lynn L. E. and M. F. H. McGeary. Washington D.C.: National Academy Press.

Johnston, Lloyd D., Patrick M. O'Malley, and Jerald G. Bachman. 2001. Monitoring the Future National Survey Results on Drug Use, 1975-2000. Volume I: Secondary School Students . Bethesda, MD: National Institute on Drug Abuse, c492 pp.

———. 1998. National Survey Results on Drug Use From the Monitoring the Future Study, 1975-1997. Rockville, M.D.: U.S. Government Printing Office.

Jöreskog, Karl and Dag Söborm. 1996. PRELIS 2: User's Reference Guide. Chicago, IL : Scientific Software International, Inc.

Kandel, Denise B. 1996. "The Parental and Peer Contexts of Adolescent Deviance: an Algebra of Interpersonal Influences." Journal of Drug Issues 26(2):289-315.

Karvonen, Sakari and Arja H. Rimpela. 1997. "Urban Small Area Variation in Adolescents' Health Behavior." Social Science Medicine 45(7):1089-98.

Keller, S. 1968. The Urban Neighborhood: A Sociological Perspective. New York: Randon House, Inc.

Klebanov, Pamela K., Jeanne Brooks-Gunn, and Greg J. Duncan. 1994. "Does Neighborhood and Family Poverty Affect Mothers' Parenting, Mental Health, and Social Support?" Journal of Marriage and the Family 56:441-55.

Kline, Rex B. 1998. Principles and Practice of Structural Equation Modeling. New York: The Guilford Press.

Lavrakas, Paul J. 1998. "Methods for Sampling and Interviewing in Telephone Surveys." Handbook of Applied Social Research Methods, editor L. Bickman and D. J. Rog. Thousand Oaks, CA: Sage publication.

Lee, Barrett. A. and Karen. E. Campbell. 1997. "Common Ground? Urban Neighborhoods As Survey Respondents See Them." Social Science Quarterly 78(4):922-36.

Leventhal, Tama and Jeanne Brooks-Gunn. 2000. "The Neighborhoods They Live in : The Effects of Neighborhood Residence on Child and Adolescent Outcomes." Psychological Bulletin 126(2):309-37.

Lorr, M. 1983. Cluster Analysis for Social Scientists. Washington, D. C.: Jossey-Bass.

Massey, Douglas S. 1996. "The Age of Extremes: Concentrated Affluence and Poverty in the Twenty-First Century." Demography 33(4):395-412.
———. 1998. "Back to the Future: The Rediscovery of Neighborhood Context." Contemporary Sociology 27(6):570-572.

Mayer, Susan E. and Christopher Jencks. 1989. "Growing Up in Poor Neighborhoods: How Much Does It Matter?" Science 243:1441-45.

Milligan, Glenn W. and Martha C. Cooper. 1987. "Methodology Review: Clustering Methods." Applied Psychological Measurement 11(4):329-54.

Muthén, Linda K. and Bengt O. Muthén. 1998. Mplus User's Guide. Los Angeles, CA: Muthén & Muthén.

Newcomb, Michael D. and P. M. Bentler. 1986. "Substance Use and Ethnicity: Differential Impact of Peer and Adult Models." Journal of Psychology 120:83-95.

Patterson, Gerald R. and Thomas J. Dishion. 1985. "Contributions of Families and Peers to Delinquency." Criminology 23(1):63-79.

Perry, C. 1929. The Neighborhood Unit: A Scheme of Arrangement for the Family-Life Community. New York: Russell Sage Foundation.

Punj, Girish and David W. Stewart. 1983. "Cluster Analysis and Marketing Research: Review and Suggestions for Application." Journal of Marketing Research 20:134-48.

Rapkin, Bruce D. and Douglas A. Luke. 1993. "Cluster Analysis in Community Research: Epistemology and Practice." American Journal of Community Psychology 21(2):247-77.

Rapoport, Amos. 1997. "The Nature and Role of Neighborhoods." Urban Design Studies 3:93-118.

Ross, Catherine E., John R. Reynolds, and Karlyn J. Geis. 2000. "The Contingent Meaning of Neighborhood Stability for Residents' Psychological Well-Being." American Sociological Review 65:581-97.

Rural Sociological Society Task Force on Persistent Rural Poverty. 1993. Persistent Poverty in Rural America. Boulder, C. O.: Westview Press.

Sampson, Robert J., Stephen W. Raudenbush, and Felton Earls. 1997. "Neighborhoods and Violent Crime: A Multilevel Study of Collective Efficacy." Science 277(15):918-24.

Sellers, Christine S., L. T. Jr. Winfree, and Curt T. Griffiths. 1993. "Legal Attitudes, Permissive Norm Qualities, and Substance Use: a Comparison of American Indian and Non-Indian Youth." The Journal of Drug Issues 23(3):493-513.

Shaw, Clifford R. and Henry D. McKay. 1969. Juvenile Delinquency and Urban Areas: a Study of Rates of Delinquency in Relation to Differential Characteristics of Local Communities in American Cities. Chicago: The University of Chicago Press.

Shiffman, Saul and Thomas A. Wills. 1985. Coping and Substance Use. Orlando: Academic press.

Simcha-Fagan, Ora and Joseph E. Schwartz. 1986. "Neighborhood and Delinquency: an Assessment of Contextual Effects." Criminology 24(4):667-99.

Simon, Ronald L., Christine Johnson, Jay Beaman, Rand D. Conger, and Frederich O. Lorenz. 1997. "Linking Community Context to Quality of Parenting: A Study of Rural Families." Rural Sociology 62(2):207-30.

Simon, Ronald L., Christine Johnson, Jay Beaman, Rand D. Conger, and Les B. Whitbeck. 1996. "Parents and Peer Group As Mediators of the Effect of Community Structure on Adolescent Problem Behavior." American Journal of Community Psychology 24(1):145-71.

Simons, Ronald L., Kuei-Hsiu Lin, Leslie C. Gordon, Gene H. Brody, Velma Murry, and Conger Rand D. forthcoming. "Community Contextual Differences in the Effect of Parental Behavior on Child Conduct Problems: a Multilevel Analysis With an African American Sample." .

Snipp, C. M. 1996. "Understanding Race and Ethnicity in Rural America." Rural Sociology 61(1):125-42.

Specce, Deborah L. 1990. "Methodological Issues in Cluster Analysis: How Clusters Become Real." Learning Disabilities: Theoretical and Research Issues, editors H. L. Swanson and B. Keogh. Hillsdale, N. J.: Lawrence Erlbaum Associates, Inc.

Stern, Susan B. and Carolyn A. Smith. 1995. "Family Processes and Delinquency in an Ecological Context." Social Service Review 29(4):703-31.

Substance Abuse and Mental Health Service Administration. 1999. "Full report: 1999 national household survey on drug abuse" [Web Page]. Available at http://www.health.org/govstudy/bkd376/.

Tienda, Marta. 1991. "Poor People and Poor Places: Deciphering Neighborhood Effects on Poverty Outcomes." Macro-Micro Linkages in Sociology, editor Joan Huber. Newbury Park, CA: Sage Publication.

U.S. Department of Commerce and Bureau of the Census. 1994. Geographic Areas Reference Manual.

Wallace, John M. Jr. and Jerald G. Bachman. 1991. "Explaining Racial/Ethnic Differences in Adolescent Drug Use: the Impact of Background and Lifestyle." Social Problems 38:333-57.

Warren, Rachelle B. and Donald I. Warren. 1977. The Neighborhood Organizer's Handbook. South Bend: University of Notre Dame Press.

Warren, Roland L. 1987. The Community in America. 3 ed. New York: University Press of America.

Wellman, Barry. 1979. "The Community Question: The Intimate Networks of East Yorkers." American Journal of Sociology 84(5):1201-31.

Wilson, William J. 1991b. "Public Policy Research and the Truly Disadvantaged." The Urban Underclass, editor Cristopher Jencks and Paul E. Peterson. Washington D. C.: The Brookings Institution.

———. 1991a. "Studying Inner-City Social Dislocations: The Challenge of Public Agenda Research." American Sociological Review 56:1-14.

———. 1987. The Truly Disadvantaged: The Inner-City , the Underclass and Public Policy. Chicago: The University of Chicago Press.